Datasets:

Hendrik-a

/

arxiv

like 0

we show that the following three systems related to various hydrodynamical approximations : the korteweg de vries equation , the camassa holm equation , and the hunter saxton equation , have the same symmetry group and similar bihamiltonian structures . it turns out that their configuration space is the virasoro group and all three dynamical systems can be regarded as equations of the geodesic flow associated to different right - invariant metrics on this group or on appropriate homogeneous spaces . in particular , we describe how arnold s approach to the euler equations as geodesic flows of one - sided invariant metrics extends from lie groups to homogeneous spaces . we also show that the above three cases describe all generic bihamiltonian systems which are related to the virasoro group and can be integrated by the translation argument principle : they correspond precisely to the three different types of generic virasoro orbits . finally , we discuss interrelation between the above metrics and kahler structures on virasoro orbits as well as open questions regarding integrable systems corresponding to a finer classification of the orbits .

math0210397

one of the main mechanisms of integrability of evolution equations is the presence of two compatible hamiltonian structures . in this paper we compare hamiltonian properties of three extensively studied nonlinear equations of mathematical physics , related to various hydrodynamical approximations : the korteweg - de vries equation @xmath0 the camassa - holm equation @xmath1 derived as a shallow water equation in @xcite ( see also the paper @xcite ) , and the hunter saxon equation @xcite @xmath2 describing weakly nonlinear unidirectional waves . all three equations are known to be bihamiltonian and to possess infinitely many conserved quantities , as well as remarkable soliton or soliton like solutions . a motivation for our paper was the paper @xcite , which described scattering theory for all three equations in a unified way . as we show in this paper , the main reason why such a common treatment is possible is that all these equations have the same symmetry group . it turns out to be the virasoro group , a one - dimensional extension of the group of smooth transformations of the circle . more precisely , the virasoro group serves as the configuration space , and all three equations can be regarded as equations of the geodesic flow related to different right - invariant metrics on this group ( in the case of kdv and ch ) or on an associated homogeneous space ( in the case of hs ) . ( here we will be mostly concerned with the periodic case , though many statements can be extended to the case of rapidly decaying potentials on the real line . ) the main goal of this paper is to give a description of the three equations as _ bihamiltonian _ systems on the dual to the virasoro algebra , and to relate them to the geometry of the virasoro coadjoint orbits . one of the corresponding hamiltonian ( or poisson ) structures is provided by the linear lie - poisson bracket , and it is the same for all three equations . the other poisson structure is constant and can be viewed as a `` linear structure frozen at a point . '' the corresponding `` freezing points '' are different for each equation . we will also see that , in a sense , equations ( [ eq : kdv ] ) , ( [ eq : ch ] ) , and ( [ eq : hs ] ) exhaust all generic possibilities , and among them the camassa - holm equation ( [ eq : ch ] ) is the `` most general '' equation that can be obtained by the `` freezing argument '' method on the dual virasoro space . we tried to make the paper self - contained , including in it necessary background on the euler equations and the classification of virasoro orbits . for additional information we refer the interested reader to the expositions in @xcite or @xcite , as well as to the original papers listed in the bibliography . in @xcite v. arnold suggested a general framework for the euler equations on an arbitrary ( possibly infinite - dimensional ) group , which we recall below . in this framework the euler equation describes a geodesic flow with respect to a suitable one - sided invariant riemannian metric on the given group . in section 2 we show how arnold s approach to the euler equation works for the virasoro group and provides a natural geometric setting for the korteweg de vries and camassa holm equations . in section 3 we give a hamiltonian reformulation of the euler equation . in section 4 we extend this approach to include geodesic flows on homogeneous spaces and then use it to describe the hunter saxton equation and its relatives . this extension might be thought of as a version of the hamiltonian formalism for homogeneous spaces developed in @xcite , @xcite , which is applied to the case of a degenerate metric and in our infinite - dimensional situation . in sections 56 we develop the bihamiltonian formalism for the euler systems . we show why the three equations above represent three main classes ( and exhaust the rotation - invariant ) bihamiltonian systems on the virasoro algebra that can be integrated by means of the `` freezing '' ( called also , the translation of ) argument method . it turns out that the above three equations correspond to three different types of virasoro coadjoint orbits of low codimensions . in our classification of poisson pairs and the corresponding equations we rely heavily on the classical classification of the virasoro orbits , and we recall it in the appendix . an interesting open question is to extend the classification of the equations to orbits of higher codimension , as well as to show how the discrete invariant of virasoro orbits manifests itself in the related bihamiltonian systems . the main objects in our consideration will be the group @xmath3 of all diffeomorphisms of a circle , its lie algebra @xmath4 of vector fields , and their central extensions . the following nontrivial one - dimensional extension of the algebra of vector fields has a special name . the virasoro algebra is an extension of the lie algebra @xmath5 of vector fields on the circle : @xmath6 with the commutator between pairs ( consisting of a vector field and a real number ) given by @xmath7 = \left(~ ( -vw_x + v_x w)(x)\partial_x , \ , c(v\partial_x , w\partial_x)~ \right)\ ] ] where @xmath8 is the gelfand fuchs cocycle . ( here @xmath9 is a coordinate on the circle , the subscript @xmath9 stands for the derivative in @xmath9 , and @xmath10 denotes the vector field @xmath11 on @xmath12 . ) define the following two - parameter family of quadratic forms , `` @xmath13-energies , '' on the lie algebra @xmath14 : @xmath15 the case @xmath16 corresponds to the @xmath17 inner product , while @xmath18 corresponds to @xmath19 . given @xmath20 and any @xmath21 , extend the @xmath13-energy to a right - invariant metric on the virasoro group @xmath22 . this group corresponds to the virasoro algebra and is defined as follows . the virasoro group @xmath22 is the product @xmath23 where the group multiplication between the pairs is given by @xmath24 and @xmath25 is the bott cocycle . ( here the diffeomorphisms of @xmath12 are described by functions , e.g. , @xmath26 . ) having equipped the virasoro group with those right - invariant metrics , one can consider the geodesic flows they generate . [ euler ] 1 ) @xcite the kdv equation is the euler equation , describing the geodesic flow on the virasoro group with respect to the right - invariant @xmath17-metric . \2 ) @xcite the ch equation is the euler equation for the geodesic flow on the same group with respect to the right - invariant sobolev @xmath19-metric . it turns out that one can give a similar description of the hunter - saxton equation as a geodesic flow on a _ homogeneous space _ related to the virasoro algebra . consider the @xmath27 -quadratic form ( which is the @xmath13 -form with @xmath28 ) on the virasoro algebra : @xmath29 although this form is degenerate , as is the corresponding right - invariant metric on the virasoro group , one can define a nondegenerate metric by descending on an appropriate quotient space . [ homeuler ] the hs equation is the equation describing the geodesic flow on the homogeneous space @xmath30 of the virasoro group modulo rotations with respect to the right - invariant homogeneous @xmath27 metric . note that one can also obtain the hs equation by considering the smaller homogeneous space @xmath31 of all diffeomorphisms of the circle modulo rotations , as we explain below . these three equations essentially exhaust the list of integrable systems associated with the virasoro algebra and integrated by the freezing argument method , as we discuss below . note that their degenerations include , e.g. , the inviscid burgers equation ( it corresponds to the @xmath17-metric on the `` centerless virasoro '' group , @xmath3 ) . before proving the theorems , we recall the general set - up for the euler equation on an arbitrary lie group , suggested by v. arnold in @xcite . consider a ( possibly infinite - dimensional ) lie group @xmath32 , which can be thought of as the configuration space of some physical system . ( examples from @xcite : @xmath33 for a rigid body or the group @xmath34 of volume - preserving diffeomorphisms for an ideal fluid filling a domain @xmath35 . ) the tangent space at the identity of the lie group @xmath32 is the corresponding lie algebra @xmath36 . fix some ( positive definite ) quadratic form , the `` energy , '' on @xmath36 . we consider right translations of this quadratic form to the tangent space at any point of the group ( the `` translational symmetry '' of the energy ) . this way the energy defines a right - invariant riemannian metric on the group @xmath32 . the geodesic flow on @xmath32 with respect to this energy metric represents the extremals of the least action principle , i.e. the actual motions of our physical system . to describe a geodesic on the _ lie group _ with an initial velocity @xmath37 , we transport its velocity vector at any moment @xmath38 to the identity of the group ( by using the right translation ) . this way we obtain the evolution law for @xmath39 , given by a ( non - linear ) dynamical system @xmath40 on the _ lie algebra _ @xmath36 . the system on the lie algebra @xmath36 , describing the evolution of the velocity vector along a geodesic in a right - invariant metric on the lie group @xmath32 , is called the euler equation corresponding to this metric on @xmath32 . ( 0 , 64 ) ( 143,210)@xmath41 ( 230,217)@xmath42 ( 150,175)@xmath43 ( 147,135)@xmath44 ( 236,144)@xmath45 ( 208,126)@xmath46 ( 280,124)@xmath47 ( 55,115)@xmath32 -24 mm in particular , the above scheme works for the virasoro group ( see theorem [ euler ] ) and allows one to describe the korteweg de vries and camassa holm equations as geodesic equations on that group . it also can be extended to include geodesic flows on homogeneous spaces and to describe the hunter saxton equation , as we discuss below . we start with preliminaries on lie algebras and poisson structures . the dual space @xmath48 to any lie algebra @xmath36 carries a natural lie poisson structure : @xmath49 , m\rangle\ ] ] for any @xmath50 and any two smooth functions @xmath51 on @xmath48 . ( here the differentials are taken at the point @xmath42 , and @xmath52 is a natural pairing between @xmath36 and @xmath48 . ) in other words , the lie poisson bracket of two linear functions on @xmath48 is equal to their commutator as elements of the lie algebra @xmath36 itself . [ hamfield ] the _ hamiltonian vector field _ on @xmath48 corresponding to a hamiltonian function @xmath53 and computed with respect to the lie - poisson structure has the following form : @xmath54 * proof . * let @xmath55 be the corresponding hamiltonian field . then for any function @xmath56 one has the identities @xmath57 , m\rangle= \langle dg , { \rm ad}^*_{df}m\rangle.\ ] ] this implies that @xmath58 . @xmath59 the differential - geometric description of the euler equation as a geodesic flow on a lie group has a hamiltonian reformulation . fix the notation @xmath60 for the energy quadratic form on @xmath36 which we used to define the riemannian metric . identify the lie algebra and its dual with the help of this quadratic form . this identification @xmath61 ( called the _ inertia operator _ ) allows one to rewrite the euler equation on the dual space @xmath48 , see fig.1 . it turns out that the euler equation on @xmath48 is hamiltonian with respect to the lie poisson structure @xcite . moreover , the corresponding hamiltonian function is minus the energy quadratic form lifted from the lie algebra to its dual space by the same identification : @xmath62 , where @xmath63 . here we are going to take it as the _ definition _ of the euler equation ( we use the proposition above and the observation @xmath64 ) . [ lineuler ] the euler equation on @xmath48 corresponding to the hamiltonian @xmath62 is given by the following explicit formula : @xmath65 as an evolution of a point @xmath50 . the underlying reason for the hamiltonian reformulation is the fact that any geodesic problem in _ riemannian _ geometry can be described in terms of _ symplectic _ geometry . geodesics on @xmath35 are extremals of a quadratic lagrangian ( metric ) on @xmath66 . they can also be described by the hamiltonian flow on @xmath67 for the quadratic hamiltonian function obtained from the lagrangian via the legendre transform . if the manifold is a group @xmath32 with a right - invariant metric then there exists the group action on the tangent bundle @xmath68 , as well as on the cotangent bundle @xmath69 . by taking the quotient with respect to the group action , we obtain from the ( symplectic ) cotangent bundle @xmath69 the lie - poisson structure on the cotangent space @xmath70 , i.e. , on the dual to the lie algebra . the hamiltonian function on @xmath69 is dual to the riemannian metric ( viewed as a form on @xmath68 ) , and its restriction to @xmath48 is the quadratic form @xmath71 , @xmath50 . the geodesics of a left - invariant metric on @xmath32 correspond to the hamiltonian function @xmath72 , while those of a right - invariant metric correspond to @xmath73 . now we are ready to prove theorem [ euler ] on the eulerian nature of the kdv and ch equations in the following slightly more general setting . * ( = [ euler]@xmath74*)[euler ] the euler equation describing the geodesic flow on the virasoro group with respect to the right - invariant @xmath13 -metric with @xmath75 has the form : @xmath76 @xmath77 by choosing @xmath78 one obtains the kdv equation , related to the @xmath17-metric on the virasoro algebra @xcite . similarly , for @xmath18 one recovers a general form of the ch equation @xcite . note that by shifting @xmath79const we get another form of the ch equation , in which the term @xmath80 is replaced by @xmath81 . finally , if @xmath82 then equation ( [ abeuler ] ) becomes the hs ( hunter - saxton ) equation , which we discuss in the next section . the case @xmath83 corresponds to considering the non - extended lie algebra @xmath4 of vector fields rather than the virasoro algebra @xmath84 . depending on the values of @xmath85 and @xmath21 one obtains either the inviscid burgers ( also called , hopf ) equation @xmath86 or the non - extended ch equation . * proof of theorem [ euler ] . * recall that the virasoro coadjoint action can be computed as follows . let @xmath87 be the dual space to the virasoro algebra with the natural pairing given by @xmath88 ( here we denote by @xmath89 ( or , shorter , by @xmath90 ) a quadratic differential on the circle . ) the coadjoint operator is defined by the identity @xmath91 \rangle.\ ] ] using the definition of the virasoro commutator and integrating by parts we obtain that the right - hand - side is equal to @xmath92 thus the coadjoint operator is @xmath93 the @xmath13 -energy ( [ eq : metric ] ) on the virasoro algebra @xmath94 corresponds to the general inertia operator @xmath95 , given by @xmath96 where @xmath97 is a second order differential operator . this operator is non - degenerate on @xmath84 for @xmath75 , while for @xmath98 it has a non - trivial kernel consisting of constant vector fields on @xmath12 . now the euler equation @xmath99 on @xmath100 ( see definition ( [ lineuler ] ) ) assumes the form @xmath101 ( here we substituted @xmath102 into the expression for @xmath103 . ) in terms of @xmath104 ( the first component of ) this equation becomes @xmath105 which is equivalent to the equation ( [ abeuler ] ) , since @xmath106 . for the second component we find that @xmath107 does not change in time : @xmath108 . @xmath59 [ image ] in the proof we assumed that the inertia operator is invertible . in the next section we discuss the precise relation of the geodesic and hamiltonian approaches in the case of a degenerate metric and show what reductions are necessary for the corresponding euler equation to make sense . let @xmath32 be a lie group and @xmath109 its subgroup . consider the space @xmath110 of _ right _ cosets @xmath111 . then the group @xmath32 acts on them on the right . here we are going to develop the formalism for the euler equation , describing the geodesic flow on @xmath110 with respect to a right - invariant metric . one immediately encounters the following difficulty : not every _ right - invariant _ metric on the group @xmath32 , degenerate along @xmath109 at the identity , descends to a metric on the space of right cosets @xmath110 ( see example ( [ exs]a ) below ) . to formulate the condition , which the degenerate metric should satisfy , let us consider the corresponding problem at the level of lie algebras . let @xmath36 be a lie algebra , and @xmath112 a degenerate inertia operator . suppose that the kernel of @xmath43 is a lie subalgebra @xmath113 . ( in other words , the corresponding energy form @xmath114 vanishes for all @xmath115 . ) consider the right - invariant degenerate metric @xmath116 on the group @xmath32 obtained by translating the quadratic form @xmath117 from identity to any point of the group . [ coset ] the right - invariant form @xmath116 on a group @xmath32 descends to a form on the space @xmath110 of right cosets if and only if the quadratic form @xmath117 on the lie algebra @xmath36 vanishes on the lie subalgebra @xmath113 and is @xmath118-invariant with respect to the action of this subalgebra . the condition of ad - invariance for @xmath117 reads as follows : @xmath119 for all @xmath120 and any @xmath121 . the above is an infinitesimal version of the invariance of @xmath117 with respect to the subgroup action : @xmath122 for all @xmath123 . [ exs ] a ) ( rotations of a rod . ) the configuration space of a rod in @xmath124 fixed at its center of mass is @xmath125 . it can be obtained from the configuration space of a rigid body by moding out rotations about one of its axes : @xmath126 . suppose that @xmath127 is a degenerate inertia operator with one vanishing eigenvalue . the corresponding eigenvector generates the 1-dimensional rotation subgroup @xmath12 in @xmath33 . it is not difficult to see that the bi - invariance condition imposes the following restriction on @xmath43 : its non - vanishing eigenvalues must be equal . ( indeed , the @xmath12-action sends one of these two eigenvectors to the other . ) then the corresponding degenerate metric on @xmath33 descends to ( a multiple of ) the standard metric on the sphere , which is the space of cosets : @xmath126 . geodesics with respect to the standard metric on @xmath125 are the great circles . these geodesics describe all free motions of the rod .. for the group of diffeomorphisms we study the right - invariant metrics and the space of right cosets @xmath31 . ] ( we can see that the only parameter of the rod is its length . in terms of the inertia operator , this corresponds to the choice of the nonzero eigenvalue . ) note that the inertia operator @xmath128 with @xmath129 does not correspond to any physical object . the corresponding degenerate metric on @xmath33 is not @xmath12-invariant , and hence it does not descend to @xmath125 . \b ) ( the hunter saxton equation . ) consider the group of diffeomorphisms @xmath3 and its quotient @xmath31 by the subgroup of rotations @xmath130 . consider the degenerate quadratic form on the corresponding lie algebra @xmath4 given by the homogeneous @xmath27-energy : @xmath131 ( similarly , for the virasoro algebra , consider the energy @xmath132 ) this energy vanishes on constant vector fields . those fields generate the subgroup @xmath133 of rotations of the circle @xmath12 . one can see that the form @xmath134 is bi - invariant with respect to the circle action , since the energy is invariant with respect to translations @xmath135 . ( the same holds for the gelfand fuchs cocycle , and the energy @xmath136 on the extended algebra . ) hence the corresponding right - invariant metric on @xmath3 descends to the quotient @xmath31 . we will see that the geodesics with respect to this metric are described by the hs equation . one hopes that a certain modification of this approach can be applied to generalized flows in @xcite , where the fluid particles in 3d can move freely and independently along one coordinate . the corresponding subgroup @xmath109 here might be that of fiberwise diffeomorphisms along a coordinate . * proof of theorem [ coset ] . * first of all we note that the quadratic form @xmath117 on @xmath36 induced from a nondegenerate form on the quotient @xmath137 is degenerate exactly along @xmath113 ( i.e. , @xmath113 is its null subspace ) . let @xmath113 be a lie subalgebra and consider the quotient of the corresponding groups @xmath110 . note that the restriction of the energy form to the subgroup @xmath109 is zero . indeed , the latter is nothing but the right translation from the identity of the energy form on the subalgebra @xmath113 . we would like to compare the right - invariant metric @xmath116 on @xmath32 at two different points @xmath138 and @xmath139 of the same coset @xmath140 . then the element @xmath141 sends @xmath142 to @xmath143 by means of the left translation . this translation also identifies the tangent spaces to @xmath32 along the same coset @xmath140 , see fig.2 . the energy form @xmath116 is invariant under this identification , since it is bi - invariant with respect to the action of elements of @xmath109 . finally , note that the energy @xmath116 is degenerate along cosets . ( indeed , it vanishes on @xmath109 , the `` identity coset , '' and it is invariant with respect to right translations , which shuffle the cosets . ) therefore , the corresponding energy form descends to the coset space . @xmath59 ( 0 , 34 ) ( 146,118)@xmath32 ( 115,33)@xmath110 ( 185,123)@xmath109 ( 220,125)@xmath140 ( 170,75)@xmath46 ( 206,87)@xmath144 ( 240,105)@xmath145 ( 235,70)@xmath146 -13 mm from the hamiltonian point of view , the geodesic picture on a homogeneous space corresponds to a hamiltonian reduction of the non - degenerate case with respect to the subgroup @xmath109 action . more precisely , consider the cotangent bundle @xmath147 of the metric space @xmath110 . similarly to the non - degenerate case , look at the fiber @xmath148 over the `` identity coset '' @xmath109 . this space @xmath148 can be naturally identified with the image @xmath149 of the degenerate inertia operator @xmath61 , for which @xmath150 , ( or , equivalently , with the subspace @xmath151 , the annihilator of @xmath152 in @xmath153 ) . the subgroup @xmath109 , being a stabilizer of the `` identity coset''and hence of @xmath154 , acts on @xmath155 respecting the poisson structure . therefore , there is a natural poisson structure on the quotient @xmath156 . furthermore , one can define the corresponding hamiltonian function on @xmath155 by the same formula as above : @xmath157 this function is @xmath158-invariant and hence is well - defined on the quotient @xmath159 . suppose @xmath43 generates a right - invariant and @xmath160-invariant metric on the group @xmath32 ( i.e. , satisfies the conditions of theorem [ coset ] ) , so that it makes sense to consider geodesics of the corresponding metric on @xmath110 . then the above consideration provides the following limiting ( degenerate ) case of arnold s theorem ( cf . definition [ lineuler ] ) . [ homarnold ] the euler equation , which corresponds to the inertia operator @xmath43 and describes the geodesic flow on the space @xmath110 of right cosets , has the following hamiltonian form on @xmath161 : it is the quotient with respect to the @xmath109-action of the restriction to @xmath162 of the following hamiltonian equation on @xmath48 @xmath163 for @xmath164 . now we are ready to complete the argument showing the eulerian nature of the hunter - saxton equation . * ( = [ homeuler]@xmath74)*[homeuler ] the hs equation @xmath165is a well - defined equation on the equivalence classes of periodic functions @xmath166 it describes the geodesic flow on the homogeneous space @xmath30 of the virasoro group modulo rotations with respect to the right - invariant homogeneous @xmath27 metric . * proof of theorem [ homeuler ] . * the homogeneous @xmath27-metric on the virasoro algebra is related to the degenerate inertia operator @xmath167 sending @xmath168 its image @xmath169 consists of pairs @xmath170 , where functions @xmath171 have zero mean : @xmath172 the action of the subgroup @xmath173 identifies those functions @xmath171 that differ by a rotation : @xmath174 . thus we come to a hamiltonian equation on @xmath159 . for explicit calculations , recall that the present case corresponds to setting @xmath175 , @xmath176 in the proof of theorem [ euler ] . choosing these values in equation ( [ abeuler ] ) , one arrives at the hs equation @xmath177 on @xmath84 . in order to obtain equation describing the evolution of @xmath45 rather than that of @xmath178 we observe that @xmath179 and hence , integrating both sides in @xmath9 , we obtain @xmath180 where @xmath181 is an arbitrary constant . this constant is uniquely determined by the condition that the right - hand - side is a complete derivative , i.e. , by @xmath182 . then @xmath183 where @xmath184 is an arbitrary constant . whence @xmath185 ( and hence the evolution of @xmath45 ) is defined only up to addition of a parameter @xmath184 and a multiple of @xmath186 . ( the latter is the velocity of the rotation subgroup : @xmath187 . ) this manifests the fact that the evolution of @xmath45 is defined on the equivalence classes @xmath188 . the inertia operator @xmath43 sends this equation on classes to the hamiltonian equation on the quotient @xmath159 . note that the equivalence classes above absorb the term @xmath189 ( respectively , @xmath190 for @xmath191 : consider a shift @xmath192 ) and one obtains the hs equation in its standard form ( [ eq : hs ] ) . @xmath59 one should notice that setting @xmath83 corresponds to the euler equation on the quotient @xmath31 . thus in the homogeneous case the consideration of the central extension does not give anything new , since the euler equations for @xmath31 and @xmath193 are equivalent . to formulate our next result we need to recall some generalities on bihamiltonian systems . assume that a manifold @xmath35 is equipped with two poisson structures @xmath194 and @xmath195 . they are said to be compatible ( or , form a poisson pair ) if all of their linear combinations @xmath196 are also poisson structures . a dynamical system @xmath197 on @xmath35 is called bi - hamiltonian if the vector field @xmath198 is hamiltonian with respect to both structures @xmath194 and @xmath195 . consider the dual space @xmath48 to a lie algebra @xmath36 . as we discussed above , it is equipped with the _ lie - poisson structure _ : @xmath49 , m\rangle\ ] ] where @xmath50 and @xmath51 are two arbitrary functions on @xmath48 . now fix a point @xmath199 in @xmath48 . one can associate to this point another poisson bracket on @xmath48 as follows , see @xcite . the constant poisson bracket associated to a point @xmath200 is the bracket @xmath194 on the dual space @xmath48 defined by @xmath201 , m_0\rangle\ ] ] for any two smooth functions @xmath51 on the dual space , and any @xmath50 . the differentials @xmath202 of the functions @xmath51 are taken , as above , at the point @xmath42 and are regarded as elements of the lie algebra itself . the constant bracket depends on the choice of the `` freezing '' point @xmath199 , while the lie poisson bracket is defined by the lie algebra structure only . note that the brackets @xmath203 and @xmath194 coincide at the point @xmath199 itself , and , moreover , the bivector defining the constant bracket @xmath194 is the same at all points @xmath42 . the brackets @xmath203 and @xmath194 are compatible for every `` freezing '' point @xmath199 . * proof . * indeed , any linear combination @xmath204 is again a poisson bracket , since it is just the linear lie poisson structure @xmath203 translated from the origin to the point @xmath205 . @xmath59 [ rem : hamconst ] explicitly , the hamiltonian equation on @xmath153 with the hamiltonian function @xmath53 and computed with respect to the constant poisson structure frozen at a point @xmath206 has the following form : @xmath207 as a modification of proposition [ hamfield ] shows . now we can formulate another main result . [ thm:2-h ] the euler equation ( [ abeuler ] ) for the @xmath13 -metric ( with @xmath75 ) on the virasoro group is bihamiltonian on the dual @xmath100 of the virasoro algebra . the corresponding `` freezing '' point in @xmath208 is @xmath209 . in appendix we show that the dual space @xmath210 can be thought of as the space of hill s operators @xmath211 . in these terms the above theorem can be stated as follows : the @xmath13 -metric on @xmath84 given by the inertia operator @xmath212 is bihamiltonian on @xmath100 with `` freezing '' at the point @xmath213 . [ rem:2-hh ] the kdv and ch equations are bihamiltonian on the virasoro dual . the corresponding `` freezing '' points @xmath214 in @xmath208 are @xmath215 for the ch equation and @xmath216 for the kdv equation , see fig.3 , as they are related to the @xmath19- and @xmath17-energies , respectively . to describe bihamiltonian nature of the hs equation on the reduced space @xmath159 , discussed in theorems [ homarnold][homeuler ] , one should consider the following analog of the constant poisson structure . take the lie poisson structure on @xmath208 `` frozen '' at the point @xmath217 and then push it forward to the corresponding quotient space for the hs equation . an alternative way to show integrability ( rather than the bihamiltonian property ) of this equation is to use integrability of ch and an infinite - dimensional version of the formalism developed in @xcite , @xcite . which metrics on the virasoro group ( or which quadratic forms on the virasoro algebra ) correspond to the bihamiltonian system on @xmath208 with `` freezing '' at a point @xmath218 for non - constant @xmath219 ? , @xcite . ] for which @xmath219 are these metrics positive definite ( i.e. , riemannian rather than pseudo - riemannian ) ? ( 0,0 ) ( 240,102)ch ( 330,82)@xmath220 ( 200,142)@xmath221 ( 146,142)@xmath222 ( 150,42)@xmath223 ( 150,96)@xmath224 ( 170,102)hs ( 240,48)kdv -5 mm * proof of theorem [ thm:2-h ] . * let @xmath225 be a function on @xmath208 and let @xmath226 be a ( variational ) derivative of @xmath198 at @xmath170 . then the hamiltonian equation with hamiltonian function @xmath198 computed with respect to the constant poisson structure `` frozen '' at @xmath227 has the form @xmath228 ( here we used remark [ rem : hamconst ] , the explicit form ( [ virad * ] ) of @xmath229 for the virasoro algebra , and the fact that @xmath230 . ) for the first component of this equation one has @xmath231 setting @xmath232 and @xmath233 we obtain @xmath234 and this simplifies the equation to @xmath235 to prove the theorem one needs to show that for any @xmath236 and any @xmath237 the euler equation ( [ abeuler ] ) can also be expressed in the form ( [ eq : lam ] ) for an appropriate hamiltonian function @xmath198 . next , consider the hamiltonian function @xmath198 of the form @xmath238 ( the operator @xmath239 is invertible for @xmath236 . ) by definition , the variational derivative @xmath240 of the functional @xmath198 is determined by the following identity satisfied for any @xmath241 : @xmath242 since we need only the partial variational derivative @xmath243 , we compute : @xmath244 @xmath245 @xmath246 thus , we have found that @xmath247 now , we substitute the variational derivative @xmath248 into equation ( [ eq : lam ] ) and then rewrite the obtained equation on the algebra @xmath84 , rather than on its dual @xmath100 . the latter corresponds to rewriting the equation in terms of @xmath249 , i.e. , in terms of an unknown function @xmath250 and setting @xmath251 . finally , applying @xmath252 to the equation we obtain @xmath253 recalling that @xmath254 and collecting the terms we recover equation ( [ abeuler ] ) . @xmath59 to explain in what sense the above equations are generic bihamiltonian systems on @xmath100 ( obtained by the freezing argument method ) , we need to consider symplectic leaves of the above poisson structures . recall that the symplectic leaves , i.e. , maximal non - degenerate submanifolds , of the lie poisson structure are the coadjoint orbits of the group action on @xmath48 ( see , e.g. , @xcite or this also follows from proposition [ hamfield ] ) . therefore the functions constant on symplectic leaves ( called _ casimir functions _ ) of the lie poisson bracket are those functions on the dual space @xmath48 that are invariant under the coadjoint action . the tangent plane to the group coadjoint orbit at the point @xmath199 , as well as all the planes in @xmath48 parallel to this tangent plane are the symplectic leaves of the constant bracket frozen at the point @xmath199 . the codimension of the coadjoint orbit passing through @xmath199 will be called the codimension of the poisson pair @xmath194 and @xmath203 . it turns out that there are no poisson pairs of codimension 0 or 1 in the ( smooth ) virasoro dual @xmath100 , and that the poisson pairs of codimension 2 can all be classified . [ pairclass ] all poisson pairs @xmath194 and @xmath203 on @xmath100 of codimension 2 belong to one of three classes according to the orbit type of the `` freezing '' point @xmath214 . these classes can be represented by the points a ) @xmath255 , b ) @xmath256 , and c ) @xmath257 . * proof . * first we observe that for any lie algebra @xmath36 the list of coadjoint orbits in @xmath48 provides the list of normal forms for the constant and lie poisson pairs as well . indeed , let a point @xmath258 be a normal form for all points @xmath199 belonging to the same coadjoint orbit as @xmath259 . this means that @xmath199 can be mapped by the group coadjoint action to its normal form : @xmath260 . the group action @xmath261 is a linear operator on @xmath48 , which preserves the lie poisson bracket on @xmath41 . it also maps the constant bracket frozen at @xmath199 to that frozen at @xmath259 . thus the group action @xmath262 sends one poisson pair to the other . the proof of theorem [ pairclass ] is based on the virasoro orbit classification , which we recall in appendix . notice that a `` cocentral value '' @xmath263 is invariant on the orbits in @xmath100 . this allows us to fix @xmath263 and consider the orbits in the hyperplane @xmath264 . as shown in corollary [ normforms ] of appendix , for @xmath265 there are exactly two types of virasoro orbits of codimension 1 in this hyperplane that correspond to either a generic or a jordan @xmath266 block holonomy for hill s operator @xmath267 . if we discard the discrete invariant of these orbits ( see appendix ) , representatives for their normal forms can be chosen as stated in the a ) and b ) parts of the theorem . if @xmath268 there is just one orbit type of codimension 1 in the corresponding hyperplane , and represented by c ) , see remark [ kdvfreez ] . note that in the whole dual space those orbits have codimension 2 , taking extra dimension @xmath222 into account . @xmath269 given a poisson pair one can generate a bihamiltonian dynamical system , by producing a sequence of hamiltonians in involution , according to the following _ lenard scheme_. let @xmath270 be the poisson bracket on a manifold @xmath35 for any @xmath271 . denote by @xmath272 its casimir function on @xmath35 parameterized by @xmath271 . this means that @xmath273 for any function @xmath53 . expand @xmath272 in a power series : @xmath274 , where each coefficient @xmath275 is a function on @xmath35 . the following theorem is well - known . [ biham ] the functions @xmath276 are hamiltonians of a hierarchy of bihamiltonian systems . in other words , each function @xmath275 generates the hamiltonian field @xmath277 with respect to the poisson bracket @xmath195 ( i.e. , @xmath277 satisfies @xmath278 for any @xmath279 ) , which is also hamiltonian for the other bracket @xmath194 with hamiltonian function @xmath280 ( i.e. , @xmath281 for any @xmath279 ) . other functions @xmath282 are first integrals of the corresponding dynamical systems @xmath277 . in other words , the functions @xmath283 are in involution with respect to each of the two poisson brackets @xmath194 and @xmath195 . * proof . * substituting the power series for @xmath272 into the casimir condition we obtain : @xmath284 collecting the terms at @xmath285 we obtain a sequence of identities : @xmath286 for any function @xmath53 . the first identity expresses the fact that @xmath287 is a casimir function for the bracket @xmath194 . the next one says that the hamiltonian field for @xmath288 with respect to @xmath194 coincides with the hamiltonian field for @xmath289 and the bracket @xmath195 , and so on . to see that every function @xmath290 is a first integral for the equation generated by @xmath275 with respect to each bracket , we check that @xmath291 indeed , e.g. , if @xmath292 @xmath293 since we finally obtain the bracket ( either @xmath194 or @xmath195 ) of one of the functions @xmath294 with itself . @xmath269 thus the choice of casimir functions @xmath272 determines the corresponding ( hierarchy of ) dynamical systems . by combining theorems [ pairclass ] and [ biham ] we get the following [ classpairs ] the three types of poisson pairs of codimension 2 on the virasoro algebra correspond to the three integrable systems : ch , kdv , and hs . these three systems represent all generic hamiltonian systems on @xmath100 ( modulo the ambiguity in the choice of casimir ) , which can be integrated by the freezing argument method . [ ell - hyp ] the corresponding `` freezing '' points in @xmath100 represent all three types of the virasoro coadjoint orbits of codimension 2 . if the `` freezing '' point @xmath214 is generic , one obtains an equation `` equivalent '' to the ch equation . in this sense , the ch equation is the most general equation , which is encountered by applying the freezing argument method of integration ; this is the case a ) in theorem [ pairclass ] . two other equations can be recovered by confining the `` freezing '' point to special hypersurfaces in @xmath100 . ( in turn , these hypersurfaces are foliated into coadjoint orbits . those orbits are of codimension 1 in the hypersurfaces , and hence of total codimension 2 in @xmath100 . the classification of the orbits will be discussed in detail in appendix . ) a generic point on the hyperplane @xmath268 produces the kdv equation ; see the case b ) in theorem [ pairclass ] . the case c ) in the same theorem corresponds to the hs equation if we consider the cone - like virasoro orbits in the @xmath265-case ( see appendix and fig.4 ) . the latter ( e.g. , `` freezing '' at the point @xmath295 ) corresponds to the euler equation with a degenerate metric on the group , which we discussed in section [ homsect ] . one could also consider a more subtle virasoro orbit classification , where one distinguishes between the two types of generic orbits in @xmath100 : hyperbolic and elliptic ones , according to the eigenvalues of the monodromy , as well as between the orbits which differ by the discrete invariant ( see corollary [ normforms ] ) . while all elliptic orbits with arbitrary values of the discrete invariant can be represented by hill s operators with _ constant coefficients _ , just one of the hyperbolic and one of the jordan block classes has such representatives , while others do not . note that the bihamiltonian equations corresponding to elliptic orbits with different discrete invariants are almost the same : these are the ch equations with different coefficients . it would be very interesting to see whether an analogous similarity holds for the hyperbolic and jordan block orbits with different discrete invariants . when the symplectic leaves of the bracket @xmath296 are of codimension 1 , then the choice of a casimir function is essentially unique for every @xmath271 . ( any two casimir functions for every fixed @xmath271 are functionally dependent . ) therefore , the choice of the poisson pair itself defines the bihamiltonian system ( modulo the mentioned functional dependence of the initial hamiltonian ) , provided that the symplectic leaves are hypersurfaces . this is indeed the case for the virasoro coadjoint orbits discussed above , which have codimension 1 for a fixed cocentral value @xmath222 . it turns out that a natural casimir function @xmath297 corresponding to the kdv poisson pair on the dual space to the virasoro algebra is the trace of the monodromy operator associated to hill s operator @xmath298 . it generates the first integrals of the kdv equation , see remark [ kdvint ] in appendix . similarly , one can expand casimir functions for the two other integrable cases , the ch and hs equations . note that for orbits of higher codimension one can start with several casimirs and consider several lenard schemes to generate the sequences of hamiltonians . a generic virasoro coadjoint orbit @xmath299 can be equipped with a complex structure and a two - parameter family of compatible ( pseudo- ) kahler metrics @xcite . this family of kahler metrics has a simple origin : a generic virasoro orbit has codimension 2 , i.e. , it is locally included in a two - parameter family of orbits , each equipped with its own symplectic structure compatible with the complex structure . alternatively , one could consider a two - parameter family of symplectic structures on the same orbit , given by the hamiltonian operators @xmath300 . it turns out that the restriction of the two - parameter family of @xmath13 -metrics on @xmath100 to a generic virasoro orbit @xmath299 coincides with the family of kahler metrics on the orbits . proof is achieved by comparison with formula ( 7 ) in @xcite for those homogeneous metrics at one point of the orbit . this is yet another fact manifesting a special role of the @xmath13 -metrics in virasoro geometry . in this section we recall the classification result for virasoro coadjoint orbits ( see , e.g. , @xcite , @xcite or the book @xcite ) . the dual spaces to the infinite - dimensional lie algebras considered below are always understood as smooth duals , i.e. identified with appropriate spaces of smooth functions . _ a. classification of quadratic differentials . _ we start with the non - extended group of diffeomorphisms of the circle . let @xmath3 be the group of all orientation - preserving diffeomorphisms of @xmath12 and let @xmath4 be its lie algebra . [ quaddif ] @xcite the dual space @xmath301 is naturally identified with the space of quadratic differentials @xmath302 on the circle . the pairing is given by the formula : @xmath303 for any vector field @xmath304 . the coadjoint action coincides with the action of a diffeomorphism on the quadratic differential : for a diffeomorphism @xmath305 the action is @xmath306 hence , for instance , if @xmath307 for all @xmath308 then the square root @xmath309 transforms under a diffeomorphism as a differential 1-form . in particular , @xmath310 is a casimir function ( i.e. , an invariant of the coadjoint action ) . one can see that there is only one casimir function in this case , since the corresponding orbit has codimension 1 in the dual space @xmath301 . indeed , a diffeomorphism action sends the quadratic differential @xmath311 to the constant quadratic differential @xmath312 , where the constant @xmath313 is the average value of the 1-form @xmath314 on the circle : @xmath315 on the other hand , if a differential @xmath311 changes sign on the circle , then the integral @xmath316 , evaluated between any two consecutive zeros @xmath222 and @xmath107 of the function @xmath317 , is invariant . in particular , the coadjoint orbit of such a differential @xmath311 has necessarily codimension higher than 1 . [ kdvfreez ] in our study of the kdv equation we pick the `` freezing '' point in the dual space @xmath301 to be @xmath318 . ( actually , we consider the dual space to the virasoro algebra , but we choose the cocentral term equal to zero , so that the `` freezing '' point @xmath319 belongs to the dual space to the lie algebra of vector fields . ) other values of @xmath313 give equivalent equations , different from the kdv by scaling only . _ b. virasoro dual and hill s operators . _ let @xmath84 be the virasoro algebra . we can think of its dual space as the space of pairs @xmath320 consisting of a quadratic differential and a real number ( cocentral term ) . it is more convenient , however , to regard such pairs as hill s operators , i.e. differential operators @xmath321 , as we will see below . this group action on hill s operators : @xmath325 has the following geometric interpretation ( see , e.g. , @xcite , @xcite , @xcite ) . fix @xmath326 and consider hill s operators of the form @xmath327 . let @xmath53 and @xmath328 be two independent solutions of the corresponding differential equation @xmath329 for an unknown function @xmath330 . although the equation has periodic coefficients , the solutions need not necessarily be periodic , but instead are defined over @xmath331 . consider the ratio @xmath332 . * proof . * note that the wronskian @xmath335 is constant , since it should satisfy the differential equation @xmath336 . here we normalize @xmath337 by setting @xmath338 . this additional condition allows one to find the potential @xmath171 from the ratio @xmath333 . indeed , first one reconstructs the solutions @xmath339 from the ratio @xmath333 by differentiating : @xmath340 therefore , @xmath341 given two solutions @xmath53 and @xmath328 , one immediately finds the corresponding differential equation they satisfy by writing out the following @xmath342-determinant : @xmath343 = 0\ ] ] since @xmath53 and @xmath328 satisfy the equation @xmath344 , one obtains from the determinant above that @xmath345\,.\ ] ] the explicit formula for @xmath171 expressed in terms of @xmath333 turns out to be one half of the schwarzian derivative of @xmath333 . @xmath59 * proof . * indeed , for a given potential @xmath171 the solutions @xmath53 and @xmath328 of the corresponding differential equation are not defined uniquely , but up to a transformation of the pair @xmath349 by a matrix from @xmath350 . then the ratio @xmath333 changes by a mbius transformation . @xmath59 * proof . * we look at the corresponding infinitesimal action on the solutions of the differential equation @xmath352 . for @xmath353 close to the identity , consider the infinitesimal virasoro action of @xmath324 on the potential @xmath317 : @xmath354 ( cf . formula ( [ virad * ] ) for @xmath355 ) . it is consistent with the following action on a solution @xmath330 of the above differential equation : @xmath356 the consistency means that @xmath357 . note that the action @xmath358 is an infinitesimal version of the following action of the diffeomorphism @xmath353 on @xmath330 : @xmath359 thus solutions to hill s equation transform as forms of degree @xmath360 . therefore the ratio @xmath333 of two solutions transforms as a function under a diffeomorphism action . @xmath59 in short , to calculate the coadjoint action on the potential @xmath171 one can first pass from this potential to the ratio of two solutions , then change the variable in the ratio , and finally take the schwarzian derivative of the new ratio to reconstruct the new potential @xmath361 . @xcite , @xcite the coadjoint virasoro orbits ( for a given cocentral term @xmath362 ) are enumerated by the conjugacy classes in @xmath363 , the universal covering of @xmath350 without the identity and modulo the @xmath364-action . * proof . * for a periodic potential @xmath171 the solutions of @xmath365 are quasiperiodic . in other words , the boundary values of the fundamental set of solutions @xmath366 on @xmath367 $ ] are related by a holonomy matrix @xmath368 : @xmath369 . ( similarly , one can consider the `` projective solution '' @xmath370 , which consists of the solution ratio @xmath333 and its derivative @xmath371 with the correponding holonomy @xmath372 now in @xmath373 . ) this holonomy matrix @xmath35 changes to a conjugate matrix if @xmath374 is replaced by any point @xmath375 or if @xmath198 is replaced by another system of solutions . now regard the ratio @xmath376 as a map @xmath377\to{\mathbb rp}^1 $ ] describing a motion ( `` rotation '' ) along the circle @xmath378 . one can see that the condition @xmath379 is equivalent to the condition @xmath380 on the wronskian . choosing the negative sign of the wronskian , @xmath381 , we can assume that the rotation always goes in the positive direction : @xmath382 . by a diffeomorphism change of the parameter @xmath383 , one can always turn the map @xmath377\to{\mathbb rp}^1 $ ] into a _ uniform _ rotation along @xmath384 , while keeping the boundary values of @xmath385 on the segment @xmath367 $ ] satisfying the holonomy relation @xmath386 . furthermore , the number of rotations ( the winding number " ) for the map @xmath377\to{\mathbb rp}^1 $ ] does not change under a reparametrization by a diffeomorphism @xmath324 . thus the orbits of the maps @xmath333 ( or , equivalently , of the potentials @xmath220 ) are described by the conjugacy classes of matrices in the universal covering of @xmath350 . the choice in the sign of the wronskian reflects the @xmath364-action on this universal covering . finally , note that the identity matrix in the universal covering @xmath387 ( or in its projectivization ) can not be obtained as a holonomy matrix for the maps @xmath377\to{\mathbb rp}^1 $ ] . indeed , any map @xmath333 starts at the identity and goes in the positive direction . thus , no matter how slow the rotation , one always moves out from the identity . @xmath59 [ normforms ] the virasoro orbits in the hyperplane @xmath388 with fixed @xmath362 are classified by the jordan normal form of matrices in @xmath350 and a positive integer parameter , winding number . matrices in the group @xmath350 split into three classes , whose normal forms are the exponents of the following three classes : @xmath389\,,~~ ii)~~\left[\begin{array}{cc } 0&\pm 1\\ 0&0 \end{array } \right]\,,~~ { { \rm } { and } ~~~~ } iii)~~\left[\begin{array}{cc } 0&0\\ 0&0 \end{array } \right]\,\ ] ] in the corresponding lie algebra @xmath390 , see fig.4 . the virasoro orbit containing hill s operator @xmath391 has the codimension in the hyperplane @xmath392 that is equal to the codimension in @xmath350 of ( the conjugacy class of ) the holonomy matrix @xmath35 corresponding to this hill s operator . this codimension is 1 for the classes @xmath393 and @xmath394 , and it is 3 for @xmath395 independ of the integer parameter . note that the class of real matrices whose exponents have the normal form @xmath393 decomposes into rotation matrices ( with @xmath396 ) and hyperbolic rotations ( with @xmath397 ) , the elliptic and hyperbolic cases , cf . remark [ ell - hyp ] . in the paper we regard these cases as belonging to the same general class ( and similarly we do not distinguish between the cases @xmath398 in type @xmath394 ) , since we are interested in the algebraic ( rather than geometric ) question of constructing the corresponding integrable equations . one should also mention that the equality of codimension of the virasoro coadjoint orbits in @xmath100 and ( the conjugacy class of ) the corresponding holonomy matrices in in @xmath350 can be seen by checking the smooth dependence on a parameter in the above classification . ( the versal deformations of the orbits can be given in terms of the jordan arnold normal forms of the holonomy matrices depending on a parameter , cf . alternatively , one can find the dimension of the corresponding stabilizers , see @xcite , @xcite . regarded as homogeneous spaces , the orbits of type @xmath393 are often denoted by @xmath299 , the notation @xmath399 stands for @xmath394 ( and sometimes for the case @xmath400 in @xmath393 ) , and @xmath401 corresponds to @xmath395 . for applications to bihamiltonian systems we would like to describe all points in @xmath100 belonging to orbits of codimension at most 2 . as we have shown above , in the smooth dual there are no orbits of codimension 0 or 1 , as @xmath222 is a casimir function , and in each hyperplane the orbits are of codimension at least 1 . besides smooth elements we also admit singular ones , cf . @xcite . in this paper we consider the classification of the smooth dual elements only . ] for @xmath265 the orbits are represented by the hill s operators , whose holonomy matrices were classified above , while for @xmath268 they are quadratic differentials . our choices of representatives for the orbits of codimension 1 for a fixed @xmath263 ( i.e. of total codimension 2 in @xmath100 ) will be as follows . \a ) for a generic point representing the case @xmath393 above we take hill s operator @xmath404 . it corresponds to the differential equation @xmath405 and has the holonomy matrix @xmath406 , the exponent of the type @xmath393 . this freezing point corresponds to the ch equation . \b ) the matrix of type @xmath394 can be encountered in a generic 1-parameter family of matrices in @xmath390 . its exponent , a jordan @xmath266-block with the eigenvalue 1 , can be represented as the holonomy matrix by the hill s operator @xmath407 . the latter point in @xmath100 corresponds ( after an appropriate reduction ) to the hs equation . \c ) the hyperplane @xmath268 in the virasoro dual is the dual space @xmath408 to the non - extended lie algebra of vector fields . the orbits of codimension 1 in the space @xmath408 are represented , e.g. , by the quadratic differential @xmath409 ( i.e. , by the point @xmath410 ) as we discussed in section @xmath43 . freezing the poisson structure at the latter point leads to the kdv equation . [ kdvint ] recall that the holonomy matrix @xmath35 of hill s operator @xmath411 changes to a conjugate one under the virasoro action . this implies that @xmath412 is a casimir function on @xmath100 . one can use it to generate the kdv hierarchy via the lenard scheme described in theorem [ biham ] . recall that for the kdv equation the freezing point for the constant poisson structure @xmath194 is @xmath410 . therefore , the casimir function of the bracket @xmath413 has the form @xmath414 , where @xmath415 is the holonomy of the operator @xmath416 . the expansion of the function @xmath272 in @xmath271 produces the first integrals of the kdv equation : @xmath417 where @xmath418 and @xmath419 for @xmath420 . one can see that the hamiltonian @xmath421 is quadratic in @xmath171 and coincides with the `` energy '' hamiltonian of the kdv equation , regarded as an euler equation . ( note that the kdv hamiltonians @xmath275 are differential polynomials whose degree increases with @xmath422 . the latter follows from the recurrence relation @xmath423 for hamiltonians @xmath275 ( cf . theorem [ biham ] ) for the constant and linear poisson brackets on @xmath100 . ) much more details on the kdv structures can be found in @xcite . we are grateful for the hospitality to the isaac newton institute ( cambridge , uk ) and the fields institute ( toronto , canada ) , where most of this work was done . we are indebted to p. pushkar and d. novikov for drawing the figures . special thanks go to i. zakharevich for many useful remarks , as well as for illuminating discussions on theorem [ homeuler ] . v. arnold , _ sur la gomtrie diffrentielle des groupes de lie de dimension infinie et ses applications lhydrodynamique des fluides parfaits _ , ann . fourier ( grenoble ) * 16 * ( 1966 ) , 319361 ; _ mathematical methods of classical mechanics _ , springer - verlag , new york 1989 , xvi+508 pp . r. beals , d.h . sattinger and j. szmigielski , _ acoustic scattering and the extended korteweg - de vries hierarchy _ , adv . in math . , * 140 * ( 1998 ) , 190206 ; _ inverse scattering solutions of the hunter saxton equations _ , preprint ( 2001 ) . a. kirillov , _ the orbits of the group of diffeomorphisms of the circle , and local lie superalgebras _ , funct . * 15 * ( 1981 ) , 135136 ; _ the orbit method , ii : infinite - dimensional lie groups and lie algebras _ , contemp . math , * 145 * ( 1993 ) , 3363 . g. misioek , _ a shallow water equation as a geodesic flow on the bott - virasoro group _ , j. geom * 24 * ( 1998 ) , 203208 ; _ classical solutions of the periodic camassa - holm equation _ , to appear in geom . and funct . anal . g. segal , _ unitary representations of some infinite - dimensional groups _ , comm . * 80 * ( 1981 ) , 301342 ; _ the geometry of the kdv equation _ , in `` topological methods in quantum field theory '' ( trieste , 1990 ) , internat . j. modern phys . a 6 ( 1991 ) , * 16 * , 28592869

we carry out an analysis of the cosmological perturbations in general relativity for three different models which are good candidates to describe the current acceleration of the universe . these three set - ups are described classically by perfect fluids with a phantom nature and represent deviations from the most widely acceptedcdm model . in addition , each of the models under study induce different future singularities or abrupt events known as ( i ) big rip , ( ii ) little rip and ( iii ) little sibling of the big rip . only the first one is regarded as a true singularity since it occurs at a finite cosmic time . for this reason , we refer to the others as abrupt events . with the aim to find possible footprints of this scenario in the universe matter distribution , we not only obtain the evolution of the cosmological scalar perturbations but also calculate the matter power spectrum for each model . finally , we constrain observationally these models using several measurements of the growth rate function , more precisely , and compare our results with the observational ones reaching the conclusion that even if these three models are very similar at present there are small differences that could allow us to distinguish them .

1611.00392

cosmology has made a long way on the last years with the impressive amount of observations and theoretical advancements . yet , cosmology faces many challenging questions like the fundamental cause of the recent acceleration of the universe as predicted by sneia observations almost twenty years ago @xcite , and afterwards confirmed by several type of cosmological and astrophysical observations ( see for example @xcite for a recent account on this issue ) . the simplest approach is to assume a cosmological constant that started recently dominating the late - time energy density budget of the universe which is in agreement with the current observations @xcite . but then the issues _ of why is it so tiny ? _ and _ why this cosmological constant has begun to be important only right now ? _ have to be addressed as well ( see for example : @xcite ) . another , equally important , issue is what happens if the cosmological constant is not quite constant ? this has lead to a great interest in exploring other possible scenarios to explain the late - time acceleration of the cosmos by invoking either an additional matter component in the universe , which we name dark energy ( de ) @xcite , or by modifying appropriately the laws of gravity ( for a recent account on this issue see , for example , @xcite and the extensive list of references provided therein ) . we will focus on the third question : _ what happens if the cosmological constant is not quite constant ? _ more precisely , we will address this question on the framework of the cosmological perturbations and for some de models whose equations of state ; i.e. , the ratio between its pressure and its energy density , deviate slightly from the one corresponding to a cosmological constant . before proceeding let us remind the following well known fact : if the equation of state ( eos ) of de deviates from -1 , the universe fate might be quite different from the one corresponding to an empty de sitter universe . in particular , if the equation of state of dark energy is smaller than -1 , i.e. , de is apparently ( and at least from an effective point of view ) not fulfilling the null energy condition , several future singularities or abrupt events might correspond to the cosmic doomsday of the universe . amazingly , some of these models are in accordance with current data @xcite . on the other hand , the theory of cosmological perturbations is a cornerstone of nowadays cosmology . it provides us with a theoretical framework which allows us to determine , for example , the cmb predicted from an early inflationary era or compute the matter power spectrum and the growth rate of matter in order to make a comparison with the observational results . in addition , it allows us to compute the evolution and possible clustering of de perturbations and investigate their effect on the growth of dark matter ( dm ) . even though no perturbations of de have so far been detected , and are in fact absent in the @xmath0cdm model , the existence of a great number of experiments aiming to probe the physics of the late - universe , like the dark energy survey @xcite and the euclid mission @xcite , suggests that a thorough study and characterization of such effects can be proven to be important to understand the nature of this mysterious fluid that drives the acceleration of the universe . with this mindset , on this work we analyse the perturbative effects of phantom de models and look for observational fingerprints that could be used as a mean to favour or disregard such models . there are three genuinely phantom de fates ; i.e. , which happens if and only if a phantom de component is present : * big rip ( br ) singularity : it happens at a finite cosmic time with an infinite scale factor where the hubble parameter and its cosmic time derivative diverge @xcite . * little rip ( lr ) : this case corresponds to an abrupt event , i.e. , it is not strictly speaking a future space - time singularity , as it takes place at an infinite cosmic time @xcite . the radius of the universe , the hubble parameter and its cosmic time derivative all diverge at an infinite cosmic time @xcite . * little sibling of the big rip ( lsbr ) : this behaviour again corresponds to an abrupt event rather than a future space - time singularity . at this event , the hubble rate and the scale factor blow up but the cosmic derivative of the hubble rate does not @xcite . consequetly , this abrupt event takes place at an infinite cosmic time where the scalar curvature diverges . these three cases share in common the fact that in the ( far ) future all the structures in the universe would be ripped apart in a finite cosmic time @xcite . the classical singular asymptotic behaviour of these dark energy models has lead to a quantum cosmological analysis of these setups @xcite . in these works , it was concluded that once the universe enters in a genuinely quantum phase ; i.e. , where coherence and entanglement effects are important , the universe would evade a doomsday _ la rip _ and this applies even to the smoother version of these singular behaviours corresponding to a lsbr @xcite ( see also @xcite ) . in this paper , we will analyse the cosmological perturbations of de models that induces a br , lr or lsbr . while the background analysis of the phantom de scenario has been widely analysed this has not been the case of its cosmological perturbations @xcite . in @xcite a kinematical approach was assumed , i.e. , a dependence of the scale factor as a function of the cosmic time was considered for flrw universes with future singularities at a finite cosmic time . within this setup and using approximated equations for the growth of the perturbations at late - time , the authors obtained the dm and de perturbations @xcite . furthermore , in @xcite , de perturbations are disregarded and only the growth rate of matter perturbations is calculated . in this work , we will rather assume a dynamical model , i.e. , we assume a given eos for de . this is the approach employed in ref @xcite , where the future behaviour of the linear scalar perturbations is presented for a type of model that , depending on the value of the parameters , can lead to a br or a big freeze singularity @xcite . in our analysis , we use the full theory of linear perturbation to study how the perturbations of dm and de , as well as the gravitational potential , evolve for a range of different scales . our numerical integrations start from well inside the radiation era and continue till the far future . in fact , in order to see the behaviour of the phantom de models , we extend our numerical calculations till the universe is roughly @xmath2 times larger than at present , i.e. , roughly @xmath3 . the paper is organised as follows : in sect . [ review ] , we briefly review de models that induce a br , lr or lsbr ; i.e. , we review perfect fluid that can describe de from a phenomenological point of view and leading to the above mentioned singularity or abrupt events . in sect . [ perturbed equations ] , we present the framework for studying the scalar linear perturbations . the evolution equations obtained employ no approximations and are therefore valid for all relevant modes and redshifts , as long as the linear theory is itself valid . a phenomenological mapping of the de fluid to a phantom scalar field is employed to avoid classical instabilities in the perturbations of de . we present our numerical results in sect . [ results ] , where we show the evolution of different perturbed quantities related to dm and de . we present as well the matter power spectrum for the different models . we equally constrain these models using several measurements of the growth rate function , more precisely @xmath1 . finally , in sect . [ conclusions ] , we present our conclusions . in this section we briefly review the different models that , at the background level , lead to distinct future cosmological abrupt events : ( i ) big rip ( br ) , ( ii ) little rip ( lr ) and ( iii ) little sibling of the big rip ( lsbr ) . we will consider a homogeneous and isotropic universe described by the friedmann - lematre - robertson - walker ( flrw ) metric : @xmath4 \,,\end{aligned}\ ] ] where @xmath5 is the scale factor and @xmath6 for open , flat and closed spatial geometry , respectively . we will focus on the spatially flat case @xmath7 , for which the friedmann and raychaudhuri equations read @xmath8 @xmath9 here , @xmath10 is the hubble parameter , a dot represents a derivative with respect to the cosmic time , @xmath11 , @xmath12 is the cosmological constant and @xmath13 and @xmath14 are the total energy density and pressure of all the matter content of the universe . in this work , we will consider the universe to be filled by radiation , dust ( cold dark matter and baryons ) , and de . as such , we can decompose @xmath13 and @xmath14 as @xmath15 where @xmath16 , @xmath17 , and @xmath18 correspond to the energy density of radiation , matter ( cold dark matter and baryonic ) and de . similarly , @xmath19 , @xmath20 , and @xmath21 are the pressure of radiation @xmath22 , matter @xmath23 , and de @xmath24 . we will not take into account interactions between the individual matter components . consequently , each fluid @xmath25,@xmath26,@xmath27 verifies the usual conservation equation : @xmath28 for latter convenience , we define the fractional energy density of the individual mater components as @xmath29 and the individual parameters of eos @xmath30 the de parameter of eos , @xmath31 , will be fixed later for each individual model . from ( [ back1 ] ) and ( [ back2 ] ) we can obtain the total parameter of eos , @xmath32 , from the individual @xmath33 as : @xmath34 in the following subsections , we present the three models which will be studied in this work . from this point onward a @xmath35-subscript denotes the present value of a given quantity and we set @xmath36 . a br singularity @xcite can be induced by a perfect fluid whose eos is constant and smaller than @xmath37 . @xmath38 therefore , using the conservation equation ( [ conservindiv ] ) , the energy density evolves with the scale factor as @xmath39 finally , the asymptotic evolution of the scale factor is @xmath40^{\frac{2}{3\left(1+w_\textrm{d}\right ) } } \,.\end{aligned}\ ] ] in this kind of future singularity , the scale factor , the hubble parameter and its cosmic time derivatives blow up at a finite cosmic time . for the rest of the work , for the br model we fix the free parameters of the models to the best fit in accordance with the planck data for wcdm model @xcite : @xmath41 , @xmath42 and @xmath43 ( please , cf . page 687 of @xcite ) . the case of a lr @xcite can be caused by a perfect fluid whose eos fulfils @xcite @xmath44 where the constant @xmath45 is positive and has dimensions of inverse length square . applying the conservation equation ( [ conservindiv ] ) , the energy density in terms of the scale factor reads @xmath46 ^ 2 \,.\end{aligned}\ ] ] finally , the asymptotic future evolution of the scale factor can be written as @xcite @xmath47\right\}\right ) \,,\end{aligned}\ ] ] where @xmath48 . in this kind of abrupt event , the scale factor , the hubble parameter and its cosmic time derivatives blow up at an infinite cosmic time . to fix the values of the parameters of the model we choose the best fit found in @xcite , where the authors used the data of @xcite to constrain the lr model observationally from a background point of view with respect to its evolution . the results obtained in @xcite were @xmath49 , @xmath50 and @xmath51 , and these are the values that we will use in this work . the lsbr can be induced by a perfect fluid whose eos deviates from that of a cosmological constant as @xcite @xmath52 where @xmath53 is a positive constant and whose dimensions are length to the fourth power . making use of the conservation equation ( [ conservindiv ] ) , the energy density on this case reads @xmath54 where @xmath55 plays the role of a cosmological constant . finally , the future asymptotic growth of the scale factor with respect to the cosmic time can be written as @xcite @xmath56 ^ 2 - 1\right\}\right ) \,,\end{aligned}\ ] ] where we have introduced the dimensionless constant @xmath57 . in this type of abrupt event , the scale factor and the hubble parameter blow up at infinite cosmic time . however , the cosmic time derivative of the hubble parameter remains constant . for practical purposes , we will choose the numerical values taken in @xcite : @xmath58 . for @xmath59 and @xmath60 , we choose the same values used in the model ( i ) according to the planck results @xcite , i.e. @xmath42 being @xmath43 . in this section , we review the theory of linear perturbations for multi - fluid components . we choose the newtonian gauge and work with the corresponding gauge invariant perturbation quantities . for a flrw universe , the perturbed line element is @xcite @xmath61 \,,\end{aligned}\ ] ] where @xmath62 is the comoving time , @xmath63 , the latin index denote purely spatial coordinates , and @xmath64 and @xmath65 are the gauge invariant bardeen potentials @xcite . the transformation rule @xmath66 , where a prime represents a derivative with respect to the conformal time , allows us to write @xmath10 and @xmath67 in terms of the conformal hubble parameter , @xmath68 , and its derivative , @xmath69 , as @xmath70 drawing from the line element ( [ lineelem ] ) , the inverse of the metric tensor can be obtained applying a taylor expansion up to first order . once we define the christoffel symbols , we can compute the perturbation of the ricci tensor , @xmath71 , and of the curvature scalar , @xmath72 , in order to obtain the perturbed einstein tensor @xmath73 . in addition , the perturbed einstein equations read @xmath74 where @xmath75 is the linear perturbation of the total energy momentum tensor . the individual components of eq . ( [ einstten ] ) can be written as @xcite @xmath76 the @xmath75 on the right hand side of ( [ einstten ] ) is the sum of the perturbations of the energy momentum tensor of radiation , @xmath77 , non - relativistic matter ( cold dark matter and baryons ) , @xmath78 , and de , @xmath79 . for each fluid , we can write the individual components of @xmath80 ( @xmath81 ) as @xmath82 where @xmath83 , @xmath84 , @xmath85 , and @xmath86 are , respectively , the perturbation of the energy density , the perturbation of the pressure , the peculiar velocity potential and the anisotropic stress tensor of the fluid @xmath53 . in this work we consider that none of the fluids considered introduce anisotropies at the linear level of scalar perturbations , therefore , from this point onward we will set @xmath87 . from eq . , we find that this implies the equality of the metric potentials @xmath88 . replacing this equivalence and eqs . in the first three equations of , we obtain @xcite @xmath89 where we have introduced the total energy density perturbation , @xmath90 , total pressure perturbation , @xmath91 , and total velocity potential , @xmath92 . these can be written in terms of the individual fluid variables in eqs . through the relations @xmath93 following @xcite , we decompose the pressure perturbation as . ] @xmath94 where @xmath95 are , respectively , the effective squared speed of sound defined in the rest frame ( r.f . ) of the fluid and the adiabatic speed of sound . in the following analysis , we will replace the energy density perturbation @xmath83 by the fractional energy density perturbation @xmath96 . the total perturbation @xmath97 can be obtained from eq . and reads @xmath98 the perturbed conservation equations of the energy - momentum tensor for each fluid read : @xmath99 where @xmath100 is the perturbation of the christoffel symbol and @xmath101 is the background value of the energy momentum tensor . using eqs . , , , and , we can compute the temporal and spatial components of eq . and obtain the evolution equations for the fractional energy density perturbation @xmath102 and the velocity potential @xmath85 @xmath103v_a + 3\left(1+w_a\right)\psi ' \,,\\ & v_{a}'=\left(3c^2_{sa}-1\right)\mathcal{h}v_a-\frac{c^{2}_{sa}}{1+w_a}\delta_a - \psi \ , . \end{split}\end{aligned}\ ] ] in summary , from the perturbed einstein equation , we obtain eqs . which relate the metric perturbations to the total perturbed matter quantities . on the other hand , from the perturbed conservation equations we obtain eqs . which dictate the dynamics for the individual energy density and velocity perturbations . in order to study the evolution of the linear perturbations , we have conveniently applied a fourier transformation , where we decompose a given function @xmath104 into its fourier components @xmath105 as @xmath106 therefore , for practical purposes , we make the substitution @xmath107 in all the evolution equations . on the other hand , we will apply the following change of variable : @xmath108 where the subscript @xmath109 denotes a derivative with respect to @xmath109 . by applying the fourier decomposition and the transformation to the sets of equations and , we obtain the evolution equations for each mode of the energy density and velocity perturbations of radiation , dust and dark energy @xmath110v_\textrm{d}+3\psi_x\right\}+3\left(w_\textrm{d}-1\right)\delta_\textrm{d } \,,\\ \left(v_{\textrm{d}}\right)_x & = -\frac{1}{\mathcal{h } } \left(\frac{1}{1+w_\textrm{d}}\delta_\textrm{d}+\psi\right ) + 2v_\textrm{d } \,.\\ \end{split}\end{aligned}\ ] ] and for the metric potential @xmath111\psi_x-3w\psi & = \frac{3}{2}\frac{\delta p}{\rho } \,.\\ \end{split}\end{aligned}\ ] ] so far , in this work we have described all the individual matter components as perfect fluids with a barotropic equations of state @xmath112 . since a barotropic fluid is adiabatic , for such fluid the effective and adiabatic squared speeds of sound are the same ( cf . while for radiation and matter such a representation works well , for fluids with negative eos , in particular for fluids playing the role of de , there might be some problems if the squared speed of sound becomes negative , which would lead to instabilities at the perturbative level . as a matter of extra - clarification we discuss in the appendix [ appb ] how instabilities at the linear level in perturbations arise in fluids with a negative adiabatic squared speed of sound . it is therefore necessary to take into account additional considerations for the de fluid . to avoid this problem we employ the same strategy used in @xcite and map the de fluid to a phantom scalar field . since a canonical scalar field with positive potentials can not achieve the values @xmath113 , here we consider that our de fluid corresponds to a phantom scalar field @xcite . ] , @xmath114 , whose energy density and pressure perturbations can be written as @xmath115 in the rest frame of the phantom scalar field , which is defined by the surfaces with constant value of @xmath114 , i.e. @xmath116 @xcite , we find that @xmath117 and therefore the effective squared speed of sound of de , as defined in , is equal to unity . for each de model that we consider , we will therefore set @xmath118 in eqs . , while @xmath119 is given by eq . . as we will show below , this mapping of the de fluid to a phantom scalar field removes any instabilities in the energy density of the scalar perturbations of de overcoming through this phenomenological approach the classical perturbative pathologies associated to some phantom de models . once the system of equations of the perturbed quantities is defined , we need to impose proper initial conditions in order to compute the cosmological evolution of the perturbations ( cf . for example @xcite for a detailed discussion on the initial conditions on de models ) . for this goal , we will take into account the following considerations . first , we assume that at an initial moment , @xmath120 ( which roughly corresponds to @xmath121 ) , the universe is completely dominated by radiation , so that all relevant quantities of the total matter fluid are those of a perfect fluid with @xmath122 . secondly , we note that at such moment the wave - length of all the relevant modes is small when compared with the comoving hubble parameter ( @xmath123 ) , i.e. they are outside the horizon . with these two approximations , we can combine the first and third equations of and obtain a closed evolution equation for the metric potential in the asymptotic past @xcite @xmath124 the dominant solution of this equation is a constant solution @xmath125 . applying this result to the set of equations ( [ gtequations3n ] ) , we find that initially @xmath126 assuming initial adiabatic conditions , we can relate the initial values of the individual fluid perturbed variables to the total perturbation in eq . through @xcite @xmath127 this allows us to write the initial values of @xmath102 in terms of @xmath128 as @xmath129 by imposing the adiabatic condition on the comoving gauge , we obtain @xmath130v_a\left(x_*\right)}{1+w_a\left(x_*\right ) } = \frac{\delta_b\left(x_*\right)-3\mathcal{h}\left(x_*\right)\left[1+w_b\left(x_*\right)\right]v_b\left(x_*\right)}{1+w_b\left(x_*\right ) } \,,\end{aligned}\ ] ] which then leads to the initial values of the peculiar velocities @xmath85 : @xmath131 we note that the conditions and coincide with the ones presented in @xcite in the absence of neutrinos . making use of the linearity of eqs . and , we can first compute the evolution of the perturbation quantities using the initial conditions , and for @xmath132 ( which implies @xmath133 ) and then multiply all the solutions obtained by the physical value of @xmath134 , which we will take from the planck observational fit to single field inflation and given as @xcite @xmath135 here , @xmath136 and @xmath137 are defined as the amplitude and spectral index of the primordial inflationary power spectrum corresponding to a previously selected pivot scale @xmath138 . in this work , we will use the values @xmath139 , @xmath140 , and @xmath141 in accordance with planck observational data , @xmath137 and the pivot scale to those of wmap7 . ] @xcite . the matter power spectrum describes how galaxies are distributed along the universe and provides us with a method to compare the theoretical predictions with the observational data . in this work , we will compute the linear matter power spectrum for each model studied in sec . [ review ] and we will try to detect deviations from the predictions of @xmath0cdm . the whole framework presented in the previous sections provide us the necessary tools to obtain the aimed results . notice , however , that the correct definition of the matter power spectrum uses the fractional energy density perturbation in the comoving gauge @xcite , while the analysis carried out in this work has been done in the newtonian gauge . using the variables employed in the previous sections , we can resolve this gauge difference by expressing the matter power spectrum as @xmath142 another method we will use to constrain the models presented in section [ review ] is based in computing the growth rate of the matter perturbations for the different models . by definition , the growth rate of the matter perturbations is given by the formula @xcite @xmath143 for dm - de models that closely mimic @xmath0cdm , it was found that the growth rate at late - time can be approximated reasonably well by the formula @xcite @xmath144 where @xmath145 for @xmath0cdm . the next leading order of ( [ f_approx ] ) can be found in @xcite . in this work , instead of using any approximated parametrisation , we opt to calculate the evolution of the growth rate @xmath146 for each de model using the full eqs . - and compare the results with observations . we note , however , that in most cases , the observational data refers not to the growth rate @xmath146 directly , but to the combination @xmath1 , where @xmath147 is the root mean square mass fluctuation amplitude in spheres of size @xmath148 which is used to normalise the matter power spectrum . this combination has the advantage that it avoids the degeneracy in the parameter space regarding @xmath147 and the linear bias , @xmath149 , between the perturbations of dark matter and density of galaxies @xcite . we calculate the temporal evolution of @xmath147 by the formula @xcite @xmath150 where @xmath151 hmpc@xmath152 is the wave - length of the mode corresponding to distances of @xmath148 . for each of the de models considered in this work , we will calculate the evolution of @xmath1 using the numerical solutions of eqs . - and the relations and . for the models ( i ) and ( iii ) we use @xmath153 @xcite as the present day value of @xmath147 , while for the model ( ii ) we use the value @xmath154 @xcite . we compare the results obtained with the available observational data @xcite to check whether the predictions of the models are within the observational constraints . since at the background level these models are very similar to @xmath0cdm till the present time , we expect that the deviations from @xmath0cdm in the evolution of the growth rate to be small . in this figure the first column ( i ) shows the results of the br model , the second column ( ii ) shows the results of the lr model , and the third column ( iii ) shows the results of the lsbr model . the first row corresponds to the dm perturbations while the second row corresponds to the de perturbations . in both cases the initial value of the perturbations is given by eqs . and . the third row corresponds to the perturbation of the gravitational potential normalised to the initial value @xmath155 . each colour represent a mode where the values of the corresponding wave - number are : @xmath156 ( purple ) , @xmath157 ( dark blue ) , @xmath158 ( light blue ) , @xmath159 ( green ) , @xmath160 ( orange ) , @xmath161 ( red ) . all perturbations are represented versus @xmath162 which varies from values well inside the radiation era ( @xmath163 ) till the far future ( @xmath164 ) . the value @xmath165 corresponds to the present time . ] this figure shows the matter power spectrum at redshift @xmath166 . on the left hand side panel , we plot the matter power spectrum of the models ( i ) and ( iii ) together with the matter power spectrum given by @xmath0cdm using the parameters provided by planck @xcite . there is no observable difference between these three models , consequently , they look as a single curve . on the right hand side panel , we plot the matter power spectrum corresponding to the model ( ii ) ( dashed red ) , together with the @xmath0cdm model ( dashed black ) in order to distinguish potential deviations between the two models . here , the numerical values are chosen accordingly with the fit obtained in @xcite using the wmap7 data @xcite . ] this figure shows the relative deviation of each model in comparison with the corresponding @xmath0cdm model . the red curve corresponds to the model ( i ) , the blue curve corresponds to the model ( ii ) and the black curve corresponds to the model ( iii ) . the model that induces a br singularity shows the highest deviation while the model that induces a lsbr abrupt event presents the smallest deviation respect to the @xmath0cdm model . the information encoded on this figure has to be taken with care as we assumed an @xmath167 corresponding to the planck data @xcite for the models ( i ) and ( iii ) and to wmap7 data @xcite for the model ( ii ) . this choice for the model ( ii ) was motivated by the fact that the only available cosmological constraint for such model @xcite was done using the wmap7 data.,width=377 ] [ growthrate1 ] this figure presents , in terms of the redshift , the evolution of @xmath1 for each model ( black curve ) together with the corresponding @xmath0cdm model ( red curve ) . the theoretical results obtained in this work are plotted against the observational values ( blue dots with error bar ) according with the data given in table [ tablegrowth ] . our comments in fig . [ diflambda ] about the susceptibility of the results with the value of @xmath167 taken from different datasets applies equally here and can be extended to the plots in this figure.,title="fig : " ] in this section , we present and discuss the results obtained for the evolution of the cosmological perturbations in the three models discussed in sec [ review ] that contain distinct future cosmological abrupt events : a. a big rip singularity ( sec . [ br ] ) , b. a little rip event ( sec . [ lr ] ) c. a little sibling of the big rip event ( sec . [ lsbr ] ) . for each of these models , the evolution of the matter perturbations @xmath168 , @xmath169 , @xmath170 , @xmath171 , @xmath172 , and @xmath173 was obtained by numerically integrating the set of eqs . after substituting @xmath174 and @xmath175 given in . after carrying this numeric integration , the gravitational potential @xmath174 and its derivative @xmath175 can be obtained from the first two equations in . the integration was performed since an initial moment deep inside the radiation epoch ( @xmath120 ) , when all the relevant modes are outside the horizon , and till a point in the distant future ( @xmath176 ) . at the initial moment , the values of the variables were fixed according to eqs . , and . in addition , for each model this integration was repeated for 40 different modes with wave - numbers ranging from @xmath177 , which corresponds to the mode that is exiting the hubble horizon at the present time , to a @xmath178 . notice that for @xmath179 the validity of the linear perturbation theory breaks down as non - linear effects start to become dominant in the evolution of the perturbations . in fact , @xmath180 should be at most @xmath181 . on the plots we included the the higher value @xmath178 to amplify visually the effect and evolution on the largest modes . in fig . [ perturbationsgraphs ] , we illustrate the evolution of the cosmological perturbations , since the initial moment and till a point in the future evolution of the universe . the first and second rows present , respectively , the evolution for the fractional energy densities of dm , @xmath168 , and de , @xmath172 , while the third row presents the evolution of the gravitational potential , @xmath174 , which has been normalised with respect to its initial value , @xmath155 . for each of these quantities , we plot the results for 6 different wave - numbers : @xmath156 ( purple ) , @xmath157 ( dark blue ) , @xmath158 ( light blue ) , @xmath159 ( green ) , @xmath160 ( orange ) , and @xmath161 ( red ) . in terms of evolution , we can distinguish three different behaviours , according to the range of the wave - numbers : * large k : 0.30 hmpc@xmath152 ( red ) and @xmath182hmpc@xmath152 ( orange ) . * medium k : @xmath183 hmpc@xmath152 ( green ) and @xmath184hmpc@xmath152 ( light blue ) . * small k : @xmath185 hmpc@xmath152 ( dark blue ) and @xmath186 hmpc@xmath152 ( purple ) . in addition , the results for the three models studied are separated in individual columns , with the first column corresponding to the br model , i.e. ( i ) , the second column to the lr model , i.e. ( ii ) , and the third to the lsbr model ; i.e. ( iii ) ( c.f . sec [ review ] ) . as can be seen on the first row of fig . [ perturbationsgraphs ] , the matter perturbations for the different models present an almost identical behaviour . during the initial radiation dominated epoch , each individual mode remains constant until it enters the hubble horizon . after this point , the gravitational collapse leads to the growth of @xmath168 , which becomes exponential in @xmath109 during the matter era . once the de starts to become dominant , the growth of the matter perturbations slows down as @xmath168 seems to converge to a constant value in the asymptotic future . notice that since the modes with larger @xmath187 enter the horizon at earlier time , they correspond to the curves with higher values of @xmath168 in fig . [ perturbationsgraphs ] . on the second row of fig . [ perturbationsgraphs ] , we present the evolution of the fractional energy density of de for different wave - numbers . in contrast with the perturbations of dm , here we observe some small differences between the three models . first , we note that while in models ( i ) and ( iii ) all modes start with an almost constant behaviour , in model ( ii ) we can detect a slight decay of @xmath172 during the radiation epoch , even when the modes are outside the horizon . furthermore , from one model to the other , the amplitude of @xmath172 for the same wave - number can vary more than two orders of magnitude . these differences can be explained by the great sensibility of the evolution equations ( [ seteq1each ] ) and the initial adiabatic conditions ( [ adiab ] ) to values of the eos parameter @xmath188 . this sensibility amplifies the small differences in the behaviour of @xmath31 for the three models , which is then propagated to both the initial value of @xmath172 and its evolution during the early radiation epoch . at late time , we find that the differences in magnitude of the initial conditions of @xmath172 in each model are maintained and can therefore serve as an imprint of the de model . similarly to what happens with the dm perturbations , we can characterise the qualitative evolution of @xmath172 in terms of the range of the wave - number of the mode : * large @xmath187 : these modes are the first to enter the horizon during the radiation epoch . after the horizon crossing their amplitude decays and we observe a damped oscillatory behaviour till the radiation - matter equality . at this point the modes stabilise and present a plateau that lasts until the end of the matter era . once de starts to dominate they start to decay rapidly once more . * medium @xmath187 : in this range , the modes present a behaviour of transition . while they enter the horizon only in the matter era , and therefore present no early oscillations , we still observe the existence of a plateau during the matter dominated era . the length of the plateau depends on the horizon crossing time - for the modes with smaller @xmath187 the amplitude does not have time to stabilise before the de dominated epoch . at this point , the amplitude starts to decay but , much like what happens for modes in the small @xmath187 range , it stabilises once the accelerated expansion shrinks the hubble radius and leads the modes to exit the horizon . * small @xmath187 : these modes are outside of the horizon for most of their evolution and therefore present little to no variation in amplitude till close to the present time , when they enter the horizon . at this point we observe a difference in behaviour , with the modes with smallest wave - numbers being amplified during the initial stages of the de dominated epoch , while the largest modes in this range present a slight decay . once the de fluid completely dominates , however , the modes exit the horizon and their amplitude stabilises once more . despite all the differences in behaviour and amplitude between distinct models and modes , we always find that the amplitude of de perturbations is extremely small when compared to dm . this validates the usual assumption of a de smooth fluid . on the third row of fig . [ perturbationsgraphs ] , we present the evolution of the gravitational potential @xmath174 . as can be seen , for all the ranges of @xmath187 the perturbations start with a constant value at the radiation dominated epoch , during matter dominated era they reach a second plateau and finally vanish in the far future when de gains importance . nevertheless , we can identify some qualitative differences in the evolution of @xmath174 , depending on the range of @xmath187 : * large @xmath187 : initially , during the radiation epoch we observe a constant behaviour , these modes decay and oscillate once they enter the hubble horizon . then , during the matter dominated era the oscillations are suppressed and the perturbations reach a constant value till de gains importance . at this point , @xmath174 starts to decay and vanishes asymptotically . * medium @xmath187 : the gravitational potential remains constant during the radiation dominated epoch , decays around the radiation - matter equality and reaches a second plateau in the matter dominated era . the decay observed in the transition between the two epochs is scale dependent and affects mostly the modes with higher wave - number . finally , when dark energy takes over , the modes decay rapidly , vanishing in the far future . * small @xmath187 : as in the previous case , the modes corresponding to the smallest wave - numbers show a constant behaviour during the radiation dominated and the matter dominated epochs and a decay around the radiation - matter equality . here , however , the decay is scale independent and for all the modes the amplitude of @xmath189 during the matter era is @xmath190 of its initial value , as follows from the theoretical prediction in the limit @xmath191 @xcite . once de begins to dominate , we observe a rapid decay of @xmath174 , with the gravitational potential vanishing in the future . here , we mention the exceptional cases of the model ( i ) and ( ii ) for which the value of @xmath174 for the smaller modes does not appear to vanish . instead , it evolves to negative values in the future , an effect which is particularly visible in the model ( ii ) . this results indicates that gravity becomes repulsive which is in agreement with the rip of the cosmological structures . in fig . [ matterpowerspectrum ] we compare the theoretical matter power spectrum @xmath192 obtained for each of the considered models with the one predicted by @xmath0cdm . as mention in sec . [ review ] , the cosmological parameters for the models ( i ) and ( iii ) were adjusted with the recent planck data @xcite while in the model ( ii ) the wmap7 data @xcite was used @xcite ( see the paragraph after eq . ) . as such , we present on the left panel of fig . [ matterpowerspectrum ] , the matter power spectra computed for the models ( i ) and ( iii ) and for @xmath0cdm with @xmath42 and @xmath193 @xcite , while on the right panel we present the matter power spectrum computed for the model ( ii ) and for @xmath0cdm with @xmath49 and @xmath194 . for all the cases , we find an almost perfect superposition of the results of the models ( i ) , ( ii ) and ( iii ) with the ones of @xmath0cdm . this reflects the close resemblance in terms of evolution of all the models up to the present time and shows that observables like the matter power spectrum may not be able to distinguish dark energy models that do not differ significantly from @xmath0cdm till today . in order to draw a comparison between the models , we compute the relative error with the corresponding @xmath0cdm model in each case . as can be seen in fig . [ diflambda ] , for large values of the wave - number @xmath187 , all models present a constant difference respect to the @xmath0cdm model while for small values of @xmath187 the deviation is smaller . in all cases , the relative error is less than @xmath195 . the model ( i ) shows the largest deviation while the model ( iii ) is better adjusted to the @xmath0cdm model . notice , that the model ( iii ) has not been observationally constrained and we took suitable values for the model parameters such that the model do not deviate too much from @xmath0cdm . we have fixed the value of the parameter @xmath53 to be small enough to ensure a good approximation , as was done in ref . @xcite . for @xmath196 , we recover the @xmath0cdm model ( see sec [ review ] ) , i.e. the smaller is the value of @xmath53 , the closer we are from the @xmath0cdm model . finally , in fig . [ growth ] we present the evolution of @xmath1 of the three models ( black curve ) against several observational points in the range @xmath197 ( cf . table [ tablegrowth ] ) . for comparison , we represent as well the evolution of @xmath1 for the corresponding @xmath0cdm model ( red curve ) . notice that , just like in the case of the matter power spectrum , the curve of the models ( i ) , ( ii ) and ( iii ) are almost indistinguishable from the respective curve for @xmath0cdm . we find that all models are within the error bars for almost all of the points . nevertheless , there seems to be some tension between the theoretical predictions and the observational data , in particular for the models ( i ) and ( iii ) , as the @xmath1 curves are systematically above most of the data points for redshifts up to @xmath198 . this higher tension between the theoretical predictions and the data in the models ( i ) and ( iii ) seems to come from the fact that for those models we used the planck data which has a higher value of @xmath59 than the wmap7 data used for the model ( ii ) we note that the formation of structure is more intense , i.e. , @xmath146 is larger , for higher values of @xmath59 ( cf . . a more detailed discussion on this topic can be found in @xcite . in order to better understand which of these three models fits better the observational data , we compute for each of them the value of @xmath199 , which can be written as @xmath200^{2}}{\sigma_{i}^{2 } } \,.\end{aligned}\ ] ] here , @xmath201 and @xmath202 are , respectively , the observational and the theoretical growth rates at redshift @xmath203 , while @xmath204 is the corresponding error for each measurement . the results obtained are : @xmath205 for the model ( i ) , @xmath206 for the model ( ii ) , and @xmath207 for the model ( iii ) . we thus find that the model ( ii ) presents the smallest @xmath199 , while the models ( i ) and ( iii ) have higher but similar values , with the model ( iii ) presenting the smallest @xmath199 between the two . this confirms the visual analysis of fig . [ growthrate1 ] which suggests that the model ( ii ) , is the one that better fits the observational data . .85 c0.3 c.7 c2.6 c0.4 & & * survey * & * ref . * + @xmath208 & @xmath209 & & @xcite + @xmath210 & @xmath211 & 6df galaxy survey & @xcite + @xmath212 & @xmath213 & sdss dr7 mgs & @xcite + @xmath214 & @xmath215 & 2df galaxy redshift survey & @xcite + @xmath216 & @xmath217 & wigglez dark energy survey & @xcite + @xmath218 & @xmath219 & sdss ii lrg & @xcite + @xmath220 & @xmath221 & sdss i / ii lrg + sdss iii boss cmass & @xcite + @xmath222 & @xmath223 & sdss iii boss dr12 lowz & @xcite + @xmath224 & @xmath225 & sdss dr5 lrg & @xcite + @xmath226 & @xmath227 & sdss ii lrg & @xcite + @xmath228 & @xmath229 & sdss iii boss dr12 & @xcite + @xmath230 & @xmath231 & sdss i / ii lrg + sdss iii boss cmass & @xcite + @xmath232 & @xmath233 & wigglez dark energy survey & @xcite + @xmath234 & @xmath235 & wigglez dark energy survey + alcock - paczynski distortion & @xcite + @xmath236 & @xmath237 & sdss i / ii lrg + sdss iii boss cmass & @xcite + @xmath238 & @xmath239 & sdss iii boss dr12 & @xcite + @xmath240 & @xmath241 & sdss iii boss dr12 cmass & @xcite + @xmath242 & @xmath243 & sdss iii boss dr12 cmass & @xcite + @xmath244 & @xmath245 & wigglez dark energy survey & @xcite + @xmath246 & @xmath247 & sdss i / ii lrg + sdss iii boss cmass & @xcite + @xmath244 & @xmath248 & wigglez dark energy survey + alcock - paczynski distortion & @xcite + @xmath249 & @xmath250 & sdss iii boss dr12 & @xcite + @xmath251 & @xmath252 & wigglez dark energy survey + alcock - paczynski distortion & @xcite + @xmath253 & @xmath254 & vimos - vlt deep survey & @xcite + @xmath255 & @xmath256 & wigglez dark energy survey & @xcite + @xmath257 & @xmath258 & vimos public extragalactic redshift survey & @xcite + @xmath259 & @xmath260 & fastsound & @xcite + in this work , we analyse the cosmological perturbations within the framework of gr , taking into full account the presence of de at the perturbative level . the de component is described by three different models , where each one of them behaves almost as @xmath0cdm model at present but induces a unique doomsday scenario in the future : model ( i ) leads to a br ; model ( ii ) leads to a lr ; model ( iii ) leads to a lsbr . at late time , the parameter of eos of de for each of these models is very close to but slightly smaller than @xmath37 , thus corresponding to a phantom - like behaviour of de . despite the small variations of the parameter of the eos for the three models , the asymptotic behaviour of the universe is quite different from the one in @xmath0cdm , with the unavoidable rip of all the structure in the universe no matter the interaction that bound them . the cosmological parameters of the models are fixed as follows : the parameters for the model ( i ) are chosen according to the best fit in the planck2015 data @xcite ; for the model ( ii ) we use the results of ref . @xcite , where a fit to the cosmological parameters of the model was performed using the results of wmap7 @xcite ; finally , for the model ( iii ) we use the same parameters given by planck2015 data for the model ( i ) , while setting @xmath53 small enough ( @xmath261 ) to ensure a good approximation to @xmath0cdm until the present time @xcite . an improved analysis would incorporate in the calculations an observationally constrained value for the parameter @xmath53 , obtained by constraining the homogeneous and isotropic evolution of the model using standard candles like sneia . for each of the models studied , we analyse the evolution of the linear cosmological perturbations and , in particular , we compute numerically the evolution of the matter density contrast for the dm and de components , together with the evolution of the gravitational potential . the integrations are performed from well inside the radiation era till the far future . the outcome , which we present in fig . [ perturbationsgraphs ] , shows that in all the models there is a very similar behaviour in the evolution of the perturbations . the largest difference seems to lie in the magnitude of de perturbations , which can vary by up to two orders of magnitude . this fact can be explained by the sensibility of the adiabatic initial conditions , which are imposed during the radiation epoch , to small deviations in the value of the parameter of eos of de , @xmath262 . this difference in the initial values is then propagated along the whole evolution as a consequence of the linearity of the dynamical equations ( [ seteq1each ] ) and ( [ gtequations3n ] ) . using the results of the numerical integrations , we obtain for each model the theoretical prediction for the matter power spectrum , as observed today , and the late - time evolution of the observable combination @xmath1 . in all three models the deviations of the matter power spectrum to the results of @xmath0cdm are within the @xmath263 margin , cf . [ diflambda ] , with the model ( iii ) presenting the smallest deviation . this result can be seen as a consequence of the value selected for @xmath53 , which makes the lsbr model almost indistinguishable from @xmath0cdm at the background level at the present time . we compare the evolution of @xmath1 for low reshift ( @xmath264 ) with the latest observations and find that , for all the models , the curve of @xmath1 is within the error bars for most data points . we quantify the deviation from the observations by calculating the corresponding @xmath265 , which allows us to make a preliminary comparison of the results for the three models . we find that the model ( ii ) is the most favoured by observations . we thus conclude that although the @xmath0cdm model gives the best fit to the observations , we can not exclude other models like the ones analysed in this work . the obtained results are in agreement with the observations and a further examination is necessary to find the footprints that can further differentiate each model . however , we stress that as we did not make an observational fit of these models ( from a homogeneous and isotropic point of view and for the same set of observational data ) , all our results should be taken as a guideline for a more accurate study that we hope to carry in the future . while the classical cosmological pertubations of these models at first order are well defined , as we have shown , there are still some fundamental and intrinsic problems related to phantom dark energy models . in fact , as discussed in @xcite , when a particle - physics description of the phantom dark energy is attempted , some instabilities may rise in the theory due to higher order effects . in the particular case of the phantom scalar field it was found that an infinite decay rate of dark energy particles into gravitons would occur unless an appropriate cut - off is introduced at some energy scale below the planck scale , rendering the phantom scalar field as an effective description up to that enegy scale . in all three models presented in this work , this kind of effects can potentially become more problematic as the universe evolves into one of the cosmological events considered , as the energy density becomes increasingly high . though a thorough examination of these effects and the compatibility of the phantom dark energy models with a particle physics description is outside the scope of this paper , we will take this question into account in a future work . the authors are grateful to j. beltrn jimnez and v. salzano for enlightening comments on a previous version of the paper . the work of ia is supported by a santander - totta fellowship `` bolsas de investigao faculdade de cincias ( ubi ) - santander totta '' . the work of mbl is supported by the portuguese agency `` fundao para a cincia e tecnologia '' through an investigador fct research contract , with reference if/01442/2013/cp1196/ct0001 . she also wishes to acknowledge the partial support from the basque government grant no . it592 - 13 ( spain ) and fondos feder under grant fis2014 - 57956-p ( spanish government ) . this research work is partially supported by the grant uid / mat/00212/2013 . jm is thankful to upv / ehu for a phd fellowship and acknowledges the support from the basque government grant no . it592 - 13 ( spain ) and fondos feder , under grant fis2014 - 57956-p ( spanish government ) . the authors acknowledge the cost action ca15117 ( cantata ) . here , we review the derivation of eq . which relates the pressure perturbation of a fluid @xmath53 with its relative energy density perturbation and peculiar velocity . we begin by separating the pressure perturbation into its adiabatic , @xmath266 , and non - adiabatic , @xmath267 , components @xcite @xmath268 under a general gauge transformation @xmath269 the energy density and pressure perturbations and the peculiar velocity in the final @xmath270 and initial @xmath271 gauges are related through @xcite @xmath272 where @xmath45 is related to the metric perturbation @xmath273 and is set to zero in the newtonian gauge @xcite . from eq . we arrive to the conclusion that the non - adiabatic term in eq . is gauge invariant . since by definition , we have @xmath274 in the rest frame of the fluid @xmath53 , we can write @xmath267 as @xcite @xmath275 next , we note that in order to change from the rest frame gauge ( @xmath276 , @xmath277 ) to the newtonian one ( @xmath278 ) we have to set @xmath279 , cf . . therefore , we find that in the newtonian gauge the previous equation can be written as @xcite @xmath280 on the following appendix , we remind why a universe filled with a dark energy fluid with a negative speed of sound , like it is the case of a phantom fluid with a constant equation of state , the linear perturbations of dark energy blow up . we begin by recalling the evolution equations for the pair @xmath102 and @xmath85 in fourier space : @xmath281v_a + 3\left(1+w_a\right)\psi ' \,,\\ \label{a2 } & v_{a}'=\left(3c^2_{sa}-1\right)\mathcal{h}v_a-\frac{c^{2}_{sa}}{1+w_a}\delta_a - \psi \,.\end{aligned}\ ] ] we can combine these equations into a single second order inhomogeneous differential equation for @xmath102 by differentiating and then using and to eliminate @xmath85 and @xmath282 @xmath283 \mathcal{h}\delta_a ' \nonumber\\&~ -3\left [ \mathcal{h } ' \left(w_a - c_{sa}^2\right ) + \mathcal{h } \left(w_a - c_{sa}^2\right ) ' + \mathcal{h}^2 \left ( 1 - 3w_a\right)\left ( w_a - c_{sa}^2 \right ) + 3\mathcal{h}^2 w_a \left ( c_{aa}^2 - c_{sa}^2 \right ) \right . \nonumber\\ & \left.\phantom{-3\left[\right . } + \mathcal{h}^2\left ( c_{sa}^2 - w_a\right)\frac{\mathcal{h } \left(c_{sa}^2 - c_{aa}^2\right ) ' + 2\mathcal{h } ' \left(c_{sa}^2 - c_{aa}^2\right)}{\mathcal{h}^2 \left(c_{sa}^2 - c_{aa}^2\right ) + \frac{k^2}{9 } } -\frac{k^2}{3}c_{sa}^2 \right]\delta_a \nonumber\\ & ~= 3\left(1+w_a\right ) \left\ { \psi '' + \left [ \left(1 - 3c^2_{sa}\right ) - \frac{\mathcal{h } \left(c_{sa}^2 - c_{aa}^2\right ) ' + 2\mathcal{h } ' \left(c_{sa}^2 - c_{aa}^2\right)}{\mathcal{h}^2 \left(c_{sa}^2 - c_{aa}^2\right ) + \frac{k^2}{9 } } \right ] \mathcal{h}\psi ' - \left[3\mathcal{h}^2\left(c_{sa}^2-c_{aa}^2\right ) + \frac{k^2}{3}\right ] \psi \right\ } \,.\end{aligned}\ ] ] this is a wave equation with a variable damping term , @xmath284 , and mass , @xmath285 , and with an external source , @xmath286 : @xmath287 where @xmath288 \mathcal{h } \nonumber\\ m_a^2 & ~= k^2 c^{2}_{sa } -3\bigg [ \mathcal{h } ' \left(w_a - c_{sa}^2\right ) + \mathcal{h } \left(w_a - c_{sa}^2\right ) ' + \mathcal{h}^2 \left ( 1 - 3w_a\right)\left ( w_a - c_{sa}^2 \right ) + 3\mathcal{h}^2 w_a \left ( c_{aa}^2 - c_{sa}^2 \right ) \bigg . \nonumber\\ & ~\bigg.\phantom{= k^2 c^{2}_{sa}-3\bigg[\bigg . } + \mathcal{h}^2\left ( c_{sa}^2 - w_a\right)\frac{\mathcal{h } \left(c_{sa}^2 - c_{aa}^2\right ) ' + 2\mathcal{h } ' \left(c_{sa}^2 - c_{aa}^2\right)}{\mathcal{h}^2 \left(c_{sa}^2 - c_{aa}^2\right ) + \frac{k^2}{9 } } \bigg ] \nonumber\\ \mathcal{s}_a & ~= 3\left(1+w_a\right ) \left\ { \psi '' + \left [ \left(1 - 3c^2_{sa}\right ) - \frac{\mathcal{h } \left(c_{sa}^2 - c_{aa}^2\right ) ' + 2\mathcal{h } ' \left(c_{sa}^2 - c_{aa}^2\right)}{\mathcal{h}^2 \left(c_{sa}^2 - c_{aa}^2\right ) + \frac{k^2}{9 } } \right ] \mathcal{h}\psi ' - \left[3\mathcal{h}^2\left(c_{sa}^2-c_{aa}^2\right ) + \frac{k^2}{3}\right ] \psi \right\ } \end{split}\end{aligned}\ ] ] to leading order in @xmath289 , we have @xmath290 . therefore , for sufficiently large @xmath187 , the mass becomes imaginary if @xmath291 and the solutions of the homogeneous equation comprise a decaying and a growing mode , with the latter leading to the instabilities . as such , barotropic fluids with negative adiabatic squared speed of sound , for which @xmath292 , are stricken by instabilities at the linear level in perturbations . the growth of @xmath102 can be particularly intense during the matter era , when most relevant modes are inside the hubble horizon , i.e. , @xmath293 , and @xmath294 . m. bouhmadi - lpez and c. y. chen , _ accepted for publication in jcap _ [ https://arxiv.org/abs/1609.00700[arxiv:1609.00700 [ gr - qc ] ] ] . a. kamenshchik , c. kiefer and b. sandhfer , phys . d * 76 * ( 2007 ) 064032 [ https://arxiv.org/abs/0705.1688[arxiv:0705.1688 [ gr - qc ] ] ] . https://lambda.gsfc.nasa.gov/product/map/dr4/params/lcdm_sz_lens_wmap7_bao_h0.cfm[http://lambda.gsfc.nasa.gov/product/map/dr4/params/lcdm_sz_lens_wmap7_bao_h0.cfm ] h. kurki - suonio _ cosmological perturbation theory _ ( lecture notes , university of helsinki , 2003 ) y. wang _ et al . _ , mon . not . * 409 * ( 2010 ) 737 [ https://arxiv.org/abs/1006.3517[arxiv:1006.3517 [ astro - ph ] ] ] . m. j. hudson and s. j. turnbull , astrophys . j. * 751 * ( 2013 ) l30 [ https://arxiv.org/abs/1203.4814[arxiv:1203.4814 [ astro-ph.co ] ] ] .

motivated by the local lifting problem for galois covers of curves , this paper investigates galois branched covers of the open-adic disc . using the field of norms functor of fontaine and wintenberger , we show that the special fiber of a galois cover is determined by arithmetic and geometric properties of the generic fiber and its characteristic zero specializations . as applications , we derive a characteristic zero reformulation of the abelian local lifting problem , and give a new proof of the liftability of-cyclic covers .

0807.0619

the main result of this paper ( theorem [ main ] ) says that the special fiber of a galois branched cover of the open @xmath0-adic disc is determined by characteristic zero fibers . the motivation for such a theorem comes from the _ global lifting problem for galois covers of curves _ : if @xmath1 is a finite group , @xmath2 is an algebraically closed field of characteristic @xmath3 , and @xmath4 is a finite @xmath1-galois branched cover of smooth projective @xmath2-curves , does there exist a lifting of @xmath5 to a @xmath1-galois branched cover of smooth projective @xmath6-curves , where @xmath6 is a discrete valuation ring of mixed characteristic with residue field @xmath2 ? this problem has been much studied ; see e.g. @xcite , @xcite , @xcite , @xcite , @xcite , @xcite . as a brief survey of the subject , we mention the following results : * ( grothendieck , @xcite ) if @xmath5 is at most tamely ramified , then a lifting exists over @xmath6 , where @xmath6 is any complete dvr with residue field @xmath2 . moreover , the lifting is unique once we fix a lifting of @xmath7 and the branch locus of @xmath5 . * not all wildly ramified covers are liftable : there exist curves of genus @xmath8 in positive characteristic whose automorphism groups are too large to be automorphism groups of genus @xmath9 curves in characteristic zero , by the hurwitz genus bounds . if @xmath10 is such a curve , it is then clear that @xmath11 is not liftable to mixed characteristic . * there are examples of non - liftable @xmath0-elementary abelian covers . see @xcite for an example with @xmath12 for @xmath13 . * oort , sekiguchi , and suwa showed in @xcite that cyclic covers lift if @xmath0 divides @xmath14 at most once . their method was global in nature , involving a group scheme degeneration of kummer theory to artin - schreier theory . * using local methods ( see below ) , green and matignon proved in @xcite that cyclic covers lift if @xmath0 divides @xmath14 at most twice . * pagot has shown in @xcite that klein - four covers always lift . * @xmath15-covers are liftable by @xcite , where the method of differential data is used . * green has shown in @xcite that the @xmath16-cyclic covers that occur as automorphisms of lubin - tate formal groups are liftable . the guiding conjecture in the subject was provided by f. oort ( @xcite , i.7 ) who suggested the cyclic covers always lift . in section 5 , we derive an arithmetic reformulation of a stronger form of this conjecture , which specifies the ring @xmath6 over which the lifting should occur : if @xmath4 is cyclic of order @xmath17 with @xmath18 , then @xmath5 lifts over @xmath19 $ ] , where @xmath20 denotes the witt vectors of @xmath2 . a major breakthrough in the subject came when green and matignon discovered that the obstructions to lifting are not global in nature , but rather local . indeed , using either rigid patching @xcite or deformation theory @xcite , the global lifting problem reduces to a local one , due to the _ ( @xcite section iii , @xcite corollaire 3.3.5 ) _ let @xmath21 be a ramification point for the @xmath1-galois cover @xmath4 , and consider the @xmath22-galois cover @xmath23 obtained by completion , where @xmath22 is the inertia subgroup at @xmath24 . suppose that for each such ramified @xmath21 , the map @xmath25 can be lifted to a @xmath22-galois cover of open @xmath0-adic discs , @xmath26 , where @xmath27)$ ] . then the local liftings , @xmath28 , can be patched together to yield a lifting of @xmath5 to a @xmath1-galois cover of smooth projective @xmath6-curves . since @xmath29 $ ] , we are led to consider the following _ local lifting problem for galois covers of curves _ : given a finite @xmath1-galois extension of power series rings @xmath30|k[[z]]$ ] , does there exist a lifting to a @xmath1-galois extension @xmath31|r[[z]]$ ] , where @xmath6 is a mixed characteristic dvr with residue field @xmath2 ? one could also consider the ( weaker ) _ birational local lifting problem for galois covers of curves _ : given a finite @xmath1-galois extension of laurent series fields @xmath32 , does there exist a lifting to a @xmath1-galois extension of normal rings @xmath33 $ ] ? by a lifting in this case , we mean that @xmath34 is an integral domain ( where @xmath35 is a uniformizer for @xmath6 ) , and the fields @xmath36 and @xmath37 are isomorphic as @xmath1-galois extensions of @xmath38 . in particular , in the birational version of the local problem , we do not require @xmath39 to be smooth : in terms of geometry , this corresponds to allowing the curve @xmath10 to acquire singularities . as mentioned above , the local lifting problem does not always have a positive solution . on the other hand , garuti has shown in @xcite that the birational local lifting problem _ does _ always have a positive solution , so the birational problem is indeed weaker than the local lifting problem . it is clearly advantageous to work with laurent series fields rather than power series rings , however , and the following criterion for good reduction ensures that we may do so without sacrificing the smoothness of our liftings . _ ( @xcite section 5 , @xcite 3.4 ) _ let @xmath40 be a normal integral local ring , which is also a finite @xmath41$]-module . assume moreover that @xmath34 is reduced and @xmath42 is separable . let @xmath43 be the integral closure of @xmath44 , and define @xmath45 . also , setting @xmath46 , denote by @xmath47 the degree of the different of @xmath48\otimes k)$ ] , and by @xmath49 the degree of the different of @xmath42 . then @xmath50 , and if @xmath51 , then @xmath52 $ ] . using this criterion , we obtain the suppose that @xmath30|k[[z]]$ ] is a @xmath1-galois extension of power series rings . then a @xmath1-galois extension of normal integral local rings , @xmath33 $ ] , is a lifting of @xmath30|k[[z]]$ ] if and only if it is a birational lifting of @xmath32 , and @xmath53 . hence , the local lifting problem can be reformulated as follows : given a @xmath1-galois extension @xmath32 , does there exist a @xmath1-galois birational lifting @xmath33 $ ] which preserves the different , i.e. such that @xmath53 ? it is this last formulation of the local lifting problem that provides our motivation for studying galois covers of the open @xmath0-adic disc . in particular , given such a @xmath1-galois branched cover @xmath54 , we are interested in determining geometric and arithmetic properties of the special fiber @xmath55 ( such as irreducibility , separability , and the degree of the different @xmath49 ) from the corresponding properties of the generic fiber @xmath56 and its specializations at various points @xmath57 . our main result ( theorem [ main ] ) provides precisely such a characterization of the special fiber in terms of characteristic zero data . roughly speaking , our result says that the special fiber of a galois cover of the open @xmath0-adic disc `` wants '' to be the field of norms of the characteristic zero fibers , and the degree to which this fails is the phenomenon of inseparability . our work can be regarded as a concrete investigation of the class field theory of the open @xmath0-adic disc , and our main result suggests that the local lifting problem would be answered by a grunwald - wang type theorem for the open disc , with control over the generic different . in section 2 of this paper we review the basic structure of the open @xmath0-adic disc , and then in section 3 we describe the theory of the field of norms due to fontaine and wintenberger , which plays a major role in our main result . section 4 contains the proof of our main theorem characterizing the special fiber of a galois branched cover of the open @xmath0-adic disc in terms of the characteristic zero fibers of the cover . an arithmetic reformulation of the oort conjecture is deduced in section 5 , together with a new proof of this conjecture in the @xmath0-cyclic case . let @xmath58 be a complete discretely valued field with residue characteristic @xmath3 . we make the following notational conventions : * @xmath59 denotes the valuation ring of @xmath58 ; * @xmath60 denotes the maximal ideal of @xmath59 ; * @xmath61 denotes the residue field of @xmath59 ; * @xmath62 denotes the normalized discrete valuation on @xmath58 , so that @xmath63 ; * @xmath64 is the absolute value on @xmath58 induced by @xmath62 , normalized so that @xmath65 . * if @xmath66 is the completion of an algebraic extension of @xmath58 , then we also denote by @xmath62 ( resp . @xmath64 ) the unique prolongation of @xmath62 ( resp . @xmath64 ) to @xmath66 . let @xmath58 be a complete discretely valued @xmath0-adic field , with valuation ring @xmath67 . then the _ open @xmath0-adic disc _ ( over @xmath58 ) is defined to be @xmath68\otimes_r k)$ ] , and its smooth integral model is denoted by @xmath27)$ ] . the key result for understanding the structure of the open @xmath0-adic disc is the _ ( @xcite vii.3.8 , prop . 6)_[wpt ] suppose that @xmath69 $ ] has a nonzero _ reduced series _ @xmath70 $ ] , of valuation @xmath71 . then @xmath72 can be written uniquely as @xmath73 where all @xmath74 and @xmath75 is a unit in @xmath41 $ ] . the degree @xmath76 is called the _ weierstrass degree _ of @xmath72 . polynomials @xmath77 with coefficients from @xmath78 as in the proposition are called _ distinguished polynomials _ , and we see that the ring @xmath41 $ ] has the following properties : it is a 2-dimensional regular local ring with maximal ideal @xmath79 , where @xmath35 is a uniformizer for @xmath6 . moreover , if @xmath80 is a height 1 prime of @xmath41 $ ] , then either @xmath81 , or @xmath82 for some irreducible distinguished polynomial in @xmath83 $ ] . it follows that @xmath41\otimes k$ ] is a dedekind domain whose maximal ideals are in one - to - one correspondence with the irreducible distinguished polynomials over @xmath6 . finally , the geometric points of @xmath84 can be described as : @xmath85 \ \textrm{irred . distinguished}\right\}\\ & = & \left\{\alpha\in \overline{k } \ | \ |\alpha|_k<1 \right\},\end{aligned}\ ] ] which explains the name of @xmath84 . as a consequence of the weierstrass preparation theorem , we see that an arbitrary nonzero power series @xmath69 $ ] can be written in the form @xmath86 , where @xmath87 , the polynomial @xmath88 is distinguished of degree @xmath89 , and @xmath75 is a unit . in the course of our investigation , we will often have to work with the ring @xmath41_{\mathfrak{m}}$ ] , the local ring of @xmath90 at the generic point of the special fiber @xmath91)\subset d$ ] . the previous remarks imply that an arbitrary nonzero element @xmath92_{\mathfrak{m}}$ ] has the form @xmath93 where the @xmath94 are distinguished polynomials . in particular , the denominator @xmath95 will be relatively prime to almost all height one primes of @xmath41 $ ] , so if @xmath96 is one of these primes , we will have @xmath92_\mathcal{p}$ ] , and it will make sense to look at the image of @xmath97 in @xmath41_\mathcal{p}/\mathcal{p}\cong k(\alpha)$ ] , where @xmath98 is a root of @xmath99 in @xmath100 . when we have chosen a particular root @xmath98 , we will refer to the image of @xmath97 in @xmath101 as the _ specialization _ of @xmath102 at the point @xmath103 , and denote it by @xmath104 . more generally , if @xmath105 is a polynomial with coefficients in @xmath41_\mathfrak{m}$ ] , then we can apply the previous reasoning to each of the finitely many coefficients @xmath106 . we conclude that for almost all points @xmath103 , we can specialize to obtain the polynomial @xmath107.\ ] ] in what follows , we will refer to this argument ( which allows us to specialize polynomials almost everywhere ) as the _ weierstrass argument_. suppose that @xmath108 is a sequence of points corresponding to a sequence @xmath109 with each @xmath110 being a uniformizer for the discrete valuation field @xmath111 . moreover , suppose that @xmath112 as @xmath113 , so that the points @xmath114 are approaching the boundary of @xmath84 . equivalently , we are assuming that the ramification index @xmath115 goes to @xmath116 with @xmath117 . given @xmath118_\mathfrak{m}$ ] , we can consider the specialization of @xmath102 at @xmath119 for @xmath120 ( by the weierstrass argument ) . we find that @xmath121 note that for any @xmath122 we have @xmath123 . hence , if @xmath124 is the weierstrass degree of @xmath94 , then for @xmath120 we have @xmath125 for all @xmath122 , since @xmath126 . it follows that @xmath127 , so @xmath128 thus , we see that the normalized valuations of the specializations @xmath129 are bounded below by @xmath130 . moreover , if @xmath131 ( i.e. if @xmath132 ) , then @xmath133 as @xmath134 . finally , if @xmath135 , then @xmath136 for @xmath120 , even if @xmath137 . the preceding remarks can also be applied to the finitely many coefficients of a polynomial @xmath138_{\mathfrak{m}}$ ] as in ( [ spoly ] ) . we thereby obtain a uniform lower bound on the normalized valuations of the coefficients of @xmath139 , independently of @xmath117 . moreover , it is easy to check whether these specialized coefficients are integral , and whether their valuations remain bounded as @xmath134 . we will refer to this argument ( which yields information on the valuations of specializations ) as the _ ramification argument _ in the sequel . in this section we briefly recall the definitions and important properties of the upper and lower ramification filtrations . see @xcite , chapter iv for a complete treatment . let @xmath58 be a complete discretely valued field , and @xmath140 a finite galois extension . then we define a function @xmath141 by @xmath142 the function @xmath143 just defined differs slightly from the function @xmath144 defined by serre in @xcite : @xmath145 for @xmath146 . we prefer to use the function @xmath143 in order to match the notations of @xcite . let @xmath140 be a finite galois extension with group @xmath1 . then for an integer @xmath147 , @xmath148 is called the _ @xmath149th ramification subgroup in the lower numbering_. we extend the indexing to the set of real numbers @xmath150 by setting @xmath151 for @xmath152 the normal subgroups @xmath153 form a decreasing and separated filtration of @xmath1 , with @xmath154 and @xmath155 the inertia subgroup of @xmath1 . the _ herbrand function of @xmath140 _ , @xmath156 , is given by @xmath157 the function @xmath158 is an increasing , continuous , piecewise - linear bijection , and hence has an increasing , continuous , and piecewise - linear inverse @xmath159 . using the inverse function @xmath159 , we define a new ramification filtration on the galois group @xmath1 . for a real number @xmath160 , define @xmath161 the decreasing and separated filtration @xmath162 is called _ the ramification filtration in the upper numbering_. note that @xmath163 and @xmath164 for @xmath165 . the importance of the upper ramification filtration stems from the fact that it behaves well under the formation of quotient groups : [ upperram ] let @xmath166 be a normal subgroup of @xmath167 , with fixed field @xmath168 . then the upper ramification filtration on @xmath169 is induced by the upper ramification filtration of @xmath1 : @xmath170 now suppose that @xmath140 is an infinite galois extension . then proposition [ upperram ] allows us to define an upper ramification filtration on the profinite group @xmath146 : for a real number @xmath160 , define @xmath171 where the limit is over all finite galois subextensions @xmath172 of @xmath140 . the groups @xmath162 form a decreasing , exhaustive , and separated filtration of @xmath1 by closed normal subgroups , called the _ upper ramification filtration_. we say that a real number @xmath173 is a _ jump _ for the upper ramification filtration if @xmath174 for all @xmath175 in particular , if @xmath176 is a separable closure of @xmath58 , then the absolute galois group @xmath177 is equipped with its upper ramification filtration @xmath178 . the field of norms construction applies to a certain type of field extension , which we now describe . the basic reference for this material is @xcite . let @xmath58 be a complete discrete valuation field with perfect residue field @xmath61 of characteristic @xmath3 , and @xmath179 a fixed separable closure . then an extension @xmath140 contained in @xmath176 is called _ arithmetically profinite ( apf ) _ if for all @xmath180 , the group @xmath181 is open in @xmath182 . if we set @xmath183 , then this definition means simply that @xmath184 is a finite extension of @xmath58 for all @xmath180 . since the upper ramification filtration is separated , it follows that @xmath185 , which implies that @xmath186 . a concrete example of an infinite apf extension is @xmath187 , and in this case @xmath188 . the apf property allows us to define an ( inverse ) herbrand function : indeed , if @xmath140 is a ( possibly infinite ) apf extension , then we set @xmath189 and define @xmath190 then @xmath159 is increasing , continuous , and piecewise - linear , with inverse @xmath158 which is also increasing , continuous , and piecewise - linear . of course , when @xmath140 is finite , @xmath158 coincides with our previous definition of the herbrand function . an important quantity attached to an apf extension @xmath140 is @xmath191 in terms of the ramification subextensions @xmath192 , the quantity @xmath193 is the supremum of the indices @xmath194 such that @xmath195 . in the case where @xmath140 is galois with group @xmath196 , we have @xmath197 , and @xmath193 is the first jump in the upper ramification filtration on @xmath1 . note that @xmath198 if and only if @xmath140 is totally ramified , and @xmath199 if and only if @xmath140 is totally wildly ramified . because the inverse herbrand function @xmath159 and the quantity @xmath193 will be essential for our later work , we include here the [ apfprops ] let @xmath200 and @xmath201 be two extensions of @xmath58 contained in @xmath179 with @xmath202 . then 1 . if @xmath203 is finite , then @xmath204 is apf if and only if @xmath205 is apf ; 2 . if @xmath205 is finite , then @xmath204 is apf if and only if @xmath203 is apf ; 3 . if @xmath204 is apf then @xmath203 is apf ; 4 . if @xmath204 is apf , then @xmath206 , and if in addition @xmath203 is finite , then @xmath207 . parts i ) and ii ) of this proposition say that the apf property is insensitive to finite extensions of the top or bottom , while part iii ) says that the apf property is inherited by subextensions . part iv ) says that the quantity @xmath208 can only increase in subextensions or under a finite extension of the base @xmath203 . in the latter case , we get a lower bound on the increase of @xmath208 in terms of the inverse herbrand function @xmath209 . given an infinite apf extension @xmath140 , let @xmath210 denote the set of finite subextensions of @xmath140 , partially ordered by inclusion . the key technical fact about the extension @xmath140 is the following property of the quantity @xmath208 , which generalizes part iv ) of the previous proposition : [ apfinfty ] the numbers @xmath211 for @xmath212 tend to @xmath116 with respect to the directed set @xmath210 . having discussed some general properties of infinite apf extensions , we are now ready to describe the field of norms construction , following @xmath213 : given an infinite apf extension @xmath140 , set @xmath214{}}e^*,\ ] ] the transition maps being given by the norm @xmath215 for @xmath216 . then define @xmath217 thus , a nonzero element @xmath98 of @xmath218 is given by a norm - compatible sequence @xmath219 . we wish to endow this set with an additive structure in such a way that @xmath218 becomes a field , called the _ field of norms _ of @xmath140 . this is accomplished by the following if @xmath220 , then for all @xmath212 , the elements @xmath221 converge ( with respect to the directed set @xmath222 ) to an element @xmath223 . moreover , @xmath224 is an element of @xmath218 . with this definition of addition , the set @xmath218 becomes a field , with multiplicative group @xmath225 . moreover , there is a natural discrete valuation on @xmath218 . indeed , if @xmath226 denotes the maximal unramified subextension of @xmath140 ( which is finite over k by apf ) , then @xmath227 does not depend on @xmath228 . in fact ( @xcite , thorme 2.1.3 ( ii ) ) , @xmath218 is a complete discrete valuation field with residue field isomorphic to @xmath229 ( which is a finite extension of @xmath61 ) . the isomorphism of residue fields @xmath230 comes about as follows . for @xmath231 , let @xmath232\in l^0 $ ] denote the teichmller lifting . that is , @xmath233:k_l\rightarrow l^0 $ ] is the unique multiplicative section of the canonical map @xmath234 . note that @xmath235 is of @xmath0-power degree for all @xmath236 , so @xmath237}\in k_l$ ] for all such @xmath238 , since @xmath229 is perfect . the element @xmath239}])_{e\in\mathcal{e}_{l|l^1}}$ ] is clearly a coherent system of norms , hence ( by cofinality ) defines an element @xmath240 . the map @xmath241 is a field embedding which induces the isomorphism @xmath242 mentioned above . the following result will be used several times in the proof of our main result , theorem [ main ] . before stating it , we make a for any subfield @xmath243 , define @xmath244 [ approxapf ] let @xmath140 be an infinite apf extension and @xmath245 be any finite extension of @xmath246 contained in @xmath66 . then 1 . for any @xmath247 , there exists @xmath248 such that @xmath249 2 . for any @xmath250 , we have @xmath251 the construction just described , which produces a complete discrete valuation field of characteristic @xmath252 from an infinite apf extension @xmath140 is actually functorial in @xmath66 . precisely , @xmath253 can be viewed as a functor from the category of infinite apf extensions of @xmath58 contained in @xmath179 ( where the morphisms are @xmath58-embeddings of finite degree ) to the category of complete discretely valued fields of characteristic @xmath0 ( where the morphisms are separable embeddings of finite degree ) . moreover , this functor preserves galois extensions and galois groups . fixing an infinite apf extension @xmath140 , the functorial nature of @xmath253 allows us to define a field of norms for _ any _ separable algebraic extension @xmath254 . namely , given such an @xmath200 , define the directed set @xmath255<\infty\}$ ] , and note that @xmath256{}}l'.\ ] ] then we define the field of norms @xmath257{}}x_k(l').\ ] ] with this definition , we can consider @xmath258 as a functor from the category of separable algebraic extensions of @xmath66 to the category of separable algebraic extensions of @xmath218 . [ cats ] the field of norms functor @xmath258 is an equivalence of categories . in particular , @xmath259 is a separable closure of @xmath218 , and we have an isomorphism @xmath260 . since @xmath218 is a complete discrete valuation field with residue field @xmath229 , it follows that any choice of uniformizer @xmath261 for @xmath218 yields an isomorphism @xmath262 , defined by sending @xmath263 to @xmath264 . via this isomorphism , an element @xmath265 corresponds to a power series @xmath266 $ ] . the following lemma describes the relationship between @xmath267 and the coherent system of norms @xmath268 in terms of the chosen uniformizer @xmath261 . first we need to introduce some notation . given a power series @xmath269,\ ] ] define for each @xmath236 a new power series @xmath270}}]z^i= \sum_{i=0}^\infty ( f_{l|k}(a_i))_e z^i\in r_e[[z]].\ ] ] [ specialcong ] for all @xmath271 , we have @xmath272 for all @xmath236 . by definition of the isomorphism @xmath273 , if @xmath274 , then @xmath275 now by proposition [ approxapf ] , for any @xmath236 we have @xmath276 thus we see that @xmath277 the congruence of the previous lemma can be replaced by an equality if one is willing to restrict attention to lubin - tate extensions of local fields . this is a theorem of coleman ( @xcite , theorem a ) , to which we now turn . a special class of infinite apf extensions are the lubin - tate extensions , which we now briefly recall ( see @xcite and @xcite chapter v ) . let @xmath166 be a finite extension of @xmath278 , and @xmath279 a lubin - tate formal group associated to a uniformizer @xmath35 of @xmath166 . then @xmath279 is a formal @xmath280-module , and interpreting the group @xmath279 in @xmath281 makes this ideal into an @xmath280-module ( here @xmath281 is the maximal ideal of the valuation ring of @xmath282 ) . let @xmath283 be the @xmath284-torsion of this @xmath280-module . then @xmath285 for all @xmath117 , and in particular @xmath286 is finite . now define @xmath287 , which is an infinite totally ramified abelian extension of @xmath166 , the _ lubin - tate extension of @xmath166 associated to @xmath35_. let @xmath58 be a complete unramified extension of @xmath166 with frobenius element @xmath288 , and define @xmath289 . then @xmath140 is an infinite abelian apf extension , with ramification subfields @xmath290 ( @xcite , corollary v.5.6 ) . moreover , we have @xmath291=q^{m-1}(q-1)$ ] , where @xmath292 . we will refer to extensions of this type as lubin - tate extensions , despite the fact that they are really the compositum of a lubin - tate extension with an unramified extension . now fix a primitive element @xmath293 for the tate module @xmath294 that is , @xmath295 is a generator for @xmath286 as an @xmath280-module , and if @xmath296_{\gamma}$ ] denotes the endomorphism of @xmath279 corresponding to @xmath35 , then @xmath296_{\gamma}(\omega_{m+1})=\omega_m$ ] for all @xmath297 . note that the frobenius @xmath298 acts coefficient - wise on the ring @xmath299[\frac{1}{z}]$ ] . [ co ] for all @xmath300 , there exists a unique @xmath301[\frac{1}{z}]^*$ ] such that for all @xmath297 , @xmath302 coleman s theorem has the following consequence in our situation ( we omit the proof , since we will not make use of this result in what follows ) : [ colemanlifting ] let @xmath140 be a lubin - tate extension as described above , and suppose that @xmath261 is a uniformizer for @xmath218 which is also a primitive element for the tate module @xmath303 ( this is always possible if @xmath304 ) . as described previously , @xmath264 determines an isomorphism @xmath305\rightarrow r_{x_k(l)}$ ] by sending @xmath263 to @xmath264 . consider a power series @xmath306 $ ] corresponding to @xmath307 under the isomorphism above , where @xmath308 . then @xmath309 where @xmath310 $ ] is the coleman power series for @xmath98 coming from theorem [ co ] . hence , a choice of uniformizer ( that is also a primitive element ) defines a lifting of the multiplicative monoid @xmath311}\mathbb{f}_{q^l}[[z]])-\{0\}$ ] to @xmath299^*$ ] . of course , if @xmath312 is finite then the first monoid above is simply @xmath305-\{0\}$ ] . note that the multiplicativity of this lifting is guaranteed by the uniqueness in coleman s theorem . we extend this to a lifting @xmath313}\mathbb{f}_{q^l}[[z]])\rightarrow r_k[[z]]$ ] by setting @xmath314 , and we note that this _ coleman lifting _ provides an alternative to the more obvious teichmller lifting ( coefficient - wise ) of power series . the use of @xmath10 allows for some simplification in the proof of theorem [ main ] in the case @xmath304 and @xmath264 a primitive element . however , theorem [ main ] holds for @xmath315 and also for an arbitrary choice of uniformizer @xmath264 , and the proof in the general case given below has recourse to the usual teichmller lifting , @xmath316 . given a totally ramified infinite apf extension @xmath140 , we have seen how any choice of a uniformizer @xmath317 determines an isomorphism @xmath318 defined by sending @xmath263 to @xmath264 ( here we set @xmath319 ) . we would now like to explicitly describe a connection between the field of norms @xmath218 and the open @xmath0-adic disc @xmath68\otimes k)$ ] that will underly the rest of our investigation ( here @xmath320 ) . namely , the special fiber of the smooth integral model @xmath321)$ ] is @xmath91)$ ] , with generic point @xmath322 . via the isomorphism above coming from the choice of uniformizer @xmath264 , we can thus identify @xmath323 with @xmath324 . on the other hand , each component @xmath325 of @xmath264 is a uniformizer in @xmath238 , and in particular has absolute value @xmath326 . hence , each @xmath325 corresponds to a point @xmath327 with residue field @xmath238 . in terms of the dedekind domain @xmath41\otimes k$ ] , the point @xmath328 corresponds to the maximal ideal @xmath329 generated by the minimal polynomial of @xmath325 over @xmath6 . thus , the uniformizer @xmath264 defines a net of points @xmath330 which approaches the boundary . in summary , we have the following picture : @xmath331\otimes k\\ \qquad\qquad \ z \qquad\qquad @ > > > \ \qquad\qquad\pi=(\pi_e)_e \qquad \leftrightsquigarrow \qquad \quad \{x^e\}_e \end{cd}$ ] let @xmath140 be a lubin - tate extension as described in section [ coleman ] , with residue field @xmath319 . hence , there exists a @xmath0-adic local field @xmath166 such that @xmath312 is unramified and @xmath332 , where @xmath333 is an honest lubin - tate extension , associated to a formal group @xmath279 . as usual , we let @xmath334 , and we recall that @xmath335=\#(k_h)^{m-1}=q^{m-1}.$ ] choose a uniformizer @xmath317 , which yields the identification @xmath336 as well as the net of points @xmath330 as described in the last section . consider a @xmath1-galois regular branched cover @xmath337 , with @xmath338 normal . we consider this cover to be a family over @xmath339 , and we introduce the following notations : * @xmath55 denotes the special fiber of the cover , obtained by taking the fibered product with @xmath340 ; * @xmath56 denotes the generic fiber , obtained by taking the fibered product with @xmath341 ; * for each @xmath212 , we denote by @xmath342 the fiber of @xmath343 at @xmath327 ; * if @xmath344 is an affine scheme , then @xmath345 denotes the total ring of fractions of @xmath344 , obtained from the ring of global sections , @xmath346 , by inverting all non - zero - divisors ; * if the special fiber @xmath347 is reduced , then @xmath348 is a product of @xmath349 copies of a field @xmath350 , which is a finite normal extension of @xmath351 ; * since only finitely many of the points @xmath328 are ramified in the cover @xmath56 , for @xmath238 large the fiber @xmath342 is reduced and we have an isomorphism @xmath352 , where @xmath353 is a finite galois extension of fields ; * @xmath354 denotes the degree of the different of @xmath353 ; * @xmath355 denotes the compositum of the fields @xmath66 and @xmath356 in @xmath179 . [ main ] let @xmath337 be a @xmath1-galois regular branched cover of the open @xmath0-adic disc , with @xmath338 normal and @xmath347 reduced . then 1 . if @xmath55 is generically separable , then there exists a cofinal set @xmath357 such that for @xmath358 large , we have @xmath359 and @xmath360 as subfields of @xmath361 . moreover , for these @xmath238 , the functor @xmath258 induces an isomorphism @xmath362 which respects the ramification filtrations . in particular , if @xmath49 is the degree of the different of @xmath363 , then @xmath364 . if @xmath347 is irreducible , then @xmath55 is generically inseparable if and only if @xmath365 . 3 . if @xmath1 is abelian , then @xmath366 for @xmath238 large , independently of any separability assumption . in particular , @xmath347 is irreducible if @xmath342 is irreducible for @xmath238 large . if @xmath2 is a finite field , say @xmath367 , then we can take the cofinal set @xmath368 in i ) of the theorem to be @xmath369 . that is , in the case of a finite residue field , @xmath368 is independent of the particular cover @xmath337 . the knowledgeable reader will note that much of our proof of i ) is inspired by the proof in @xcite of the essential surjectivity statement in proposition [ cats ] . the main difficulty is to spread the construction of @xmath213 over the open @xmath0-adic disc . let @xmath370 , so that @xmath33 $ ] is a @xmath1-galois extension of normal rings ( here @xmath67 ) . we are assuming that @xmath34 is reduced , where @xmath35 is a uniformizer of @xmath6 . moreover , we have @xmath371 for @xmath238 large ( here @xmath329 is the maximal ideal of @xmath41\otimes k$ ] corresponding to @xmath328 ) . i ) : suppose that @xmath55 is generically separable , which means that the field extension @xmath372 is separable , hence galois . by the primitive element theorem , there exists @xmath373 such that @xmath374 $ ] . moreover , we can choose @xmath375 to be integral over @xmath376 $ ] , say with minimal polynomial @xmath377[t]$ ] . further , since @xmath38 is infinite , we can choose @xmath349 different primitive elements @xmath378 such that the corresponding minimal polynomials @xmath379[t]$ ] are distinct . even more , by krasner s lemma , we may assume that each @xmath380[t]$ ] , so that in fact @xmath381 $ ] for some @xmath382 . having fixed this @xmath383 , we take @xmath384 . setting @xmath385 , the chinese remainder theorem implies that we have an isomorphism @xmath386/(f(t))\cong\prod_{j=1}^{n_s}k((z))[\chi_j]= \prod_{j=1}^{n_s}\mathcal{k}\cong f(y_k)\ ] ] let @xmath387 be the element of @xmath388 corresponding to @xmath389 under this isomorphism , and choose a lifting , @xmath390 , of @xmath387 to @xmath391 . denote the minimal polynomial of @xmath390 over @xmath41_{(\varpi)}$ ] by @xmath392 , so that @xmath393 . then @xmath392 has the form @xmath394_{(\varpi)}[t].\ ] ] now by the weierstrass preparation theorem , the coefficients of @xmath392 have the form @xmath395 where @xmath69 $ ] and the denominator is a distinguished polynomial . moreover , because @xmath396[t]$ ] , it follows that each @xmath397 $ ] , which implies that either @xmath398 in @xmath41 $ ] ( in which case @xmath399 ) , or the weierstrass degree of @xmath72 is greater than @xmath400 ( the degree of the denominator ) . now by the weierstrass argument described in section [ warg ] , for @xmath238 large we can specialize the polynomial @xmath392 at the point @xmath401 to obtain the polynomial @xmath402 $ ] . for @xmath238 large , the specialized polynomial @xmath403 lies in @xmath404 $ ] , where @xmath405 is the valuation ring of @xmath238 . this follows immediately from the previous remarks and the ramification argument applied to @xmath406 in the notation of section [ rarg ] . since @xmath377[t]=r_{x_k(l)}[t]$ ] is separable , we have @xmath407 in @xmath218 . since @xmath408 , we know by proposition [ apfinfty ] that @xmath409 , so there exists @xmath410 such that for @xmath411 we have @xmath412 now @xmath413[t].\ ] ] under our fixed identification of @xmath38 with @xmath218 , each coefficient @xmath414 corresponds to a coherent system of norms @xmath415 . hence , we can write @xmath416.\ ] ] now let @xmath417 $ ] be the polynomial obtained from @xmath418 by selecting the @xmath238th component from each coefficient : @xmath419.\ ] ] [ disc ] for @xmath358 large , we have @xmath420 . let @xmath421[t]$ ] be the teichmller lifting of @xmath418 : @xmath422 then @xmath1 and @xmath423 both reduce mod @xmath35 to @xmath5 , hence @xmath424 for some @xmath425_{(\varpi)}[t]$ ] . specializing at @xmath401 for @xmath358 now yields the equation @xmath426 for some @xmath427 $ ] . indeed , by lemma [ specialcong ] , we have @xmath428 but @xmath429=q^{lt_e}=(q^l)^{t_e}$ ] for @xmath358 . the operation of raising to the @xmath430th power on @xmath431 is the identity , and since the coefficients of @xmath414 lie in @xmath431 , it follows that @xmath432 . hence @xmath433 , from which equation ( [ specialeq ] ) follows immediately . consider @xmath434 now by proposition [ approxapf ] , @xmath435 . moreover , for @xmath238 large we have @xmath436 indeed , the ramification argument ( section [ rarg ] ) applied to @xmath437 shows that the valuations of the coefficients of @xmath438 go to infinity as @xmath238 gets large , and we know that @xmath439 . it follows that @xmath440 for @xmath358 large , which implies by @xmath441 that @xmath442 for @xmath358 large . now let @xmath443 be a root of @xmath403 in @xmath444 , and define @xmath445 . then @xmath446 is apf by proposition [ apfprops ] , and we set @xmath447 . moreover , note that for @xmath238 large we have @xmath448 and thus @xmath449 ( recall that @xmath450 ) . this follows from the fact that there exists @xmath451 $ ] such that @xmath452 ( for example , take @xmath9 to be the product of the denominators of the coefficients @xmath106 of @xmath453_{(\varpi)}[t]$ ] ) . then the conductor of the subring @xmath454\otimes k)_g[\xi]\subset(\mathcal{a}\otimes k)_g$ ] defines a closed subset of @xmath343 , and if @xmath328 lies outside the image of this set in @xmath84 , then the splitting of @xmath392 mod @xmath329 determines the fiber @xmath342 ( see @xcite , prop . i.8.3 ) . in particular , @xmath455 is galois over @xmath66 . at this point we introduce the following lemma from @xmath213 , which it is easy to check holds in our situation , since it is a direct consequence of lemma [ disc ] : [ lindis ] for @xmath456 large , the extensions @xmath457 and @xmath458 are linearly disjoint . moreover , we have @xmath459 since for @xmath460 large , @xmath457 is totally wildly ramified , it follows from this lemma that @xmath461 so @xmath462 is totally wildly ramified . hence , proposition [ approxapf ] says that there exists @xmath463 such that @xmath464 our immediate goal is to prove the following lemma about the polynomial @xmath465 $ ] . [ limit ] @xmath466 . note that @xmath467 moreover , as mentioned above , @xmath462 is totally ramified , hence @xmath468 denote by @xmath469 $ ] the polynomial obtained by replacing each coefficient of @xmath470\subset x_k(\tilde{l}_e)[t]$ ] by its component in @xmath471 . then by the linear disjointness of @xmath457 and @xmath458 , it follows that @xmath472 and we have @xmath473 by proposition [ approxapf ] , we have @xmath474 on the other hand , by the ramification argument ( again applied to @xmath437 ) and equation [ specialeq ] @xmath475 where @xmath476 is the maximal degree of the denominators in the coefficients of @xmath451_\mathfrak{m}[t]$ ] . thus , we see that @xmath477 since @xmath478 . together with the fact that @xmath479 , this implies that @xmath480 thus we have shown that @xmath481 but @xmath482 for @xmath238 large , and since @xmath476 is a constant , we also have @xmath483 . it follows that @xmath484 so that @xmath466 as claimed . replacing the net @xmath485 by a subnet , we may assume that it converges to a root @xmath486 of @xmath5 . but then @xmath486 is conjugate to one of the roots @xmath487 from the beginning of this proof , and since @xmath372 is galois , we have that @xmath488 . moreover , by krasner s lemma , @xmath489 for @xmath358 large . this implies that @xmath490 for @xmath358 large , and i claim that this inclusion is actually an equality . for this we need a simple preliminary lemma . note that if @xmath491 , then @xmath492 and we have by definition @xmath493 [ galnorm ] given @xmath491 , suppose that @xmath494 is such that @xmath495 using proposition [ approxapf ] , choose an element @xmath496 such that @xmath497 then @xmath498 this follows from a straightforward computation using proposition [ approxapf ] . we wish to apply this lemma with @xmath499 and @xmath500 , so we compute @xmath501 for @xmath238 large by lemma [ disc ] . since @xmath502 , it follows that @xmath443 satisfies the hypothesis of lemma [ galnorm ] for @xmath238 large , and we conclude that @xmath503 . this immediately implies that @xmath504 , because if the inclusion were proper , then there would exist @xmath505 in @xmath506 such that @xmath507 , which is a contradiction since @xmath508 . thus , in order to show that @xmath509 , we just need to show that @xmath510 . but the net @xmath511 converges to @xmath486 , and ( ) shows that the krasner radii @xmath512 for some constant @xmath10 independent of @xmath238 . hence for @xmath238 sufficiently large so that @xmath513 , krasner s lemma tells us that @xmath510 as required . thus , we have shown that @xmath514 for @xmath515 large . it now follows from the fundamental equality that @xmath516 : @xmath517}=\frac{\deg{f}}{[l_e : l ] } = \frac{\deg{f}}{[e':e]}=n_e.\ ] ] it remains to prove the statement about the galois groups . by the general theory of the field of norms , we have @xmath518 moreover , since @xmath519 , and @xmath457 and @xmath353 are linearly disjoint , it follows that @xmath520 thus , we just need to show that the ramification filtrations are preserved under these isomorphisms . first note that for all @xmath521 sufficiently large , we have @xmath522 , since by the preceding proof we have that @xmath523 and @xmath258 is an equivalence of categories . denote this common field by @xmath524 . [ rambehavior ] for @xmath525 and @xmath238 large , we have @xmath526 . since the lower ramification filtration is determined by the function @xmath149 , it follows that the isomorphism @xmath527 induced by @xmath258 preserves the ramification filtrations . since the degree of the different depends only on the ramification filtration , it follows that @xmath364 for @xmath238 large . this completes the proof of i ) . + ii ) : note that by part i ) , if @xmath365 , then the special fiber must be generically inseparable , without any irreducibility hypothesis . now suppose that @xmath347 is irreducible and @xmath55 is generically inseparable . let @xmath528 be the first ramification group at the unique prime of @xmath40 lying over @xmath529 . taking @xmath528-invariants , we obtain the tower @xmath530 . since @xmath528 is a nontrivial @xmath0-group , it has a @xmath0-cyclic quotient . hence @xmath531 has a @xmath0-cyclic subcover @xmath532 , and we have the tower @xmath533 . now consider the associated tower of special fibers @xmath534 which corresponds ( by considering the generic points ) to a chain of field extensions @xmath535 . here the extension @xmath536 is purely inseparable of degree @xmath0 , defined by @xmath537 . note that there is no extension of constants in this tower because the cover @xmath337 was assumed to be regular . the extension @xmath538 is separable and totally ramified , so the minimal polynomial of @xmath539 over @xmath38 is eisenstein : @xmath540[t],\ ] ] where @xmath541 is a unit . it follows that @xmath542 is the minimal polynomial of @xmath543 over @xmath38 . now let @xmath390 be a lifting of @xmath543 to the localized ring of global sections @xmath544 . then @xmath390 is integral over @xmath391 and we let @xmath545 $ ] be its minimal polynomial . since @xmath546 and @xmath390 is a lifting of @xmath543 , it follows that the degree of @xmath547 is also @xmath548 , and @xmath549 modulo @xmath35 . using the teichmller lifting @xmath550\rightarrow r[[z]]$ ] we find that : @xmath551 for some polynomial @xmath552_{(\varpi)}[t]$ ] of degree at most @xmath553 in @xmath554 . setting @xmath401 , we get the specialized polynomial @xmath555,\ ] ] which for @xmath238 sufficiently large is eisenstein by the ramification argument applied to @xmath556 . letting @xmath557 denote the image of @xmath390 in @xmath558 , it follows by degree considerations that @xmath559 is irreducible for @xmath238 large and @xmath557 is a uniformizer for the field @xmath558 . we obtain the chain of field extensions @xmath560 , and can compute the different as follows : @xmath561 but @xmath562 and the latter quantity goes to @xmath116 with @xmath238 . by multiplicativity of the different in towers we conclude that @xmath563 so @xmath564 goes to @xmath116 with @xmath238 as claimed . + iii ) : we now assume that @xmath1 is abelian , but make no separability assumption on the special fiber @xmath55 . since @xmath1 is abelian , the decomposition groups at the @xmath349 primes of @xmath391 lying over @xmath565)$ ] all coincide . call this decomposition group @xmath566 . taking @xmath566-invariants , we observe that @xmath567 is a @xmath568-galois regular branched cover with totally split special fiber : @xmath569 in particular , there is no further splitting in the special fiber @xmath570 . hence , we can apply part i ) to the cover @xmath567 , and a review of the beginning of the proof of i ) shows that we may take @xmath571 . we conclude that @xmath572 for @xmath238 large ( here @xmath573 is the number of components of @xmath574 ) . since @xmath575 is surjective , it follows that @xmath576 as claimed . this completes the proof of part iii ) , and hence of theorem [ main ] . using theorem [ main ] , we deduce a local lifting criterion for finite abelian extensions of @xmath577 . for this , consider a lubin - tate extension @xmath140 as described in section [ coleman ] , with @xmath578 for some finite extension @xmath579 . then choose a uniformizer @xmath317 , which defines an isomorphism @xmath580 as well as a net of points @xmath330 ( see section [ disc ] ) . we will be interested in the cofinal sequence of ramification subfields @xmath581 , and will use the simplified notation @xmath582 , etc . [ arithcrit ] suppose that @xmath1 is a finite abelian group , and let @xmath254 be a @xmath1-galois extension , corresponding to the @xmath1-galois extension @xmath583 via the field of norms functor . suppose that @xmath337 is a @xmath1-galois regular branched cover with @xmath338 normal and @xmath347 reduced . then @xmath337 is a smooth lifting of @xmath583 if and only if there exists @xmath382 such that for @xmath120 and @xmath584 , we have @xmath585 as @xmath1-galois extensions of @xmath66 , and @xmath586 . first suppose that @xmath337 is a smooth lifting of @xmath583 . then @xmath55 is generically separable , so by the proof of part i ) of theorem [ main ] there exists @xmath382 such that for @xmath120 and @xmath584 , we have @xmath587 and @xmath588 . since @xmath253 is an equivalence of categories , we conclude that @xmath589 for these values of @xmath117 . moreover , by the local criterion for good reduction ( see introduction ) , we have @xmath53 , which implies that @xmath586 for @xmath120 and @xmath584 , as claimed . now suppose that there exists @xmath382 so that @xmath585 and @xmath586 for @xmath120 and @xmath584 . then by part iii ) of theorem [ main ] , @xmath347 is irreducible , and then by part ii ) , @xmath55 is generically separable . hence we may apply part i ) to conclude that there exists @xmath590 such that @xmath591 and @xmath592 for @xmath120 and @xmath593 . but the two arithmetic progressions @xmath594 and @xmath595 have a common subsequence . it follows that @xmath596 and @xmath53 , so @xmath337 is a birational lifting of @xmath583 which preserves the different . by the local criterion for good reduction , it follows that @xmath337 is actually a smooth lifting . in particular , we obtain an arithmetic reformulation of the ring specific oort conjecture concerning the liftability of cyclic covers over @xmath597 . set @xmath598 , and let @xmath599 . then @xmath140 is lubin - tate for @xmath600 and @xmath601 . moreover , if @xmath10 is a finite cyclic group , define @xmath602\subset r_l$ ] . then we have the following arithmetic form of the ring specific oort conjecture from the introduction : it can be shown by standard techniques of model theory that the ring specific oort conjecture over @xmath597 implies the ring specific oort conjecture over @xmath2 , where @xmath2 is an arbitrary algebraically closed field of characteristic @xmath0 . we conclude this paper by giving a direct proof of the arithmetic form of the ring specific oort conjecture for @xmath0-cyclic covers over @xmath597 . we begin by recalling some facts about @xmath0-cyclic extensions for which we omit the straightforward proofs . recall that @xmath598 , and @xmath599 . [ minpoly ] suppose that @xmath606 is a uniformizer for @xmath218 , so that @xmath607 . let @xmath608 be the minimal polynomial of @xmath609 over @xmath610 . then for any @xmath611 , there exists @xmath612 so that if @xmath613 , the coefficients of @xmath614 $ ] satisfy @xmath615 for @xmath616 . we will also need a result about the stability of the ramification filtration under base change by subfields of an infinite apf extension @xmath140 . this result is actually an immediate consequence of lemma [ rambehavior ] , which we used in the proof of theorem [ main ] . [ stable ] suppose that @xmath617 is a finite @xmath1-galois extension , obtained by base change to @xmath66 from a @xmath1-galois extension @xmath618 . let @xmath619 be an element of the galois group , and for @xmath620 define @xmath621 . then there exists @xmath622 such that for @xmath623 we have @xmath624 . consequently , the canonical isomorphism @xmath625 preserves the ramification filtrations on these groups . in particular , the conductor of @xmath626 is equal to the conductor of @xmath627 for @xmath623 . [ kummer ] let @xmath254 be a @xmath0-cyclic field extension . then there exists @xmath628 and a principal unit @xmath629 such that @xmath630 , where @xmath631 . moreover , @xmath632 may be chosen so that the conductor of @xmath618 is stable under base change by @xmath633 ( lemma [ stable ] ) . modifying @xmath194 by a @xmath0-power in @xmath634 , we may assume that @xmath635 where @xmath636 is a unit in @xmath634 , @xmath637 , and @xmath638 is the artin conductor of @xmath618 let @xmath254 be a @xmath0-cyclic extension , and choose @xmath632 as described in lemma [ kummer ] , so that @xmath639 , where @xmath640 now set @xmath641 and increase @xmath632 if necessary so that @xmath642 . by proposition [ approxapf ] , there exists a unit @xmath643 such that @xmath644 . hence we can write @xmath645 where @xmath646 is a unit of index at least @xmath201 . now @xmath647 , so by hensel s lemma @xmath648 has a @xmath649th root in @xmath218 . then replace the uniformizer @xmath606 by @xmath650 . again calling this uniformizer @xmath264 , we find that @xmath651 where @xmath652 is a unit of index at least @xmath201 , say @xmath653 where the @xmath654 are teichmller representatives . now set @xmath655^\times$ ] and consider the extension of normal rings @xmath656 $ ] defined generically by the kummer equation @xmath657 by @xcite , proposition 1.4 , the special fiber of this extension , @xmath658 , is reduced . hence , we can apply theorem [ main ] to the @xmath0-cyclic cover @xmath659 . setting @xmath660 yields the integral equation @xmath661 the right hand side has @xmath638 zeros , yielding @xmath638 ramified points , each with ramification index @xmath0 . it follows that the degree of the generic different is @xmath662 . specializing the equation at @xmath663 yields @xmath664 and so all of the specializations @xmath665 are field extensions with conductor @xmath638 , hence degree of different @xmath666 . theorem [ main ] says the special fiber is separable and irreducible , and @xmath51 . hence , the lifting is smooth , and it remains to verify that @xmath667 for @xmath120 . but at level @xmath632 we have @xmath668 , hence @xmath669 . for @xmath670 , the extension @xmath665 is defined by adjoining a @xmath0th root of @xmath671 , so by kummer theory it suffices to show that this element differs from @xmath672 by a @xmath0th power in @xmath66 . the following lemma thus completes the proof of the @xmath0-cyclic case of the arithmetic form of the ring specific oort conjecture over @xmath597 . in fact , i claim that @xmath675 is a @xmath0th power in @xmath676 . for simplicity , we will assume that @xmath304 in the computation that follows ( the case @xmath315 is similiar ) . moreover , since any principal unit in @xmath676 of index greater than @xmath677 is a @xmath0th power in @xmath676 , we will work in the quotient ring @xmath678 to simplify the computation . by lemma [ minpoly ] we have @xmath679 and we may assume @xmath680 for each @xmath149 . then since @xmath681 has index at least @xmath682 in @xmath610 , we have the following congruence in @xmath683 : @xmath684 multiplying by a @xmath0th power and remembering that @xmath685 yields @xmath686 finally , we compute @xmath687 indeed , @xmath688 for some @xmath689 , and so @xmath690 hence , the quantity @xmath691 is a @xmath0th power in @xmath676 as required .

a coexistent phase of spin polarization and color superconductivity in high - density qcd is investigated using a self - consistent mean - field method at zero temperature . the axial - vector self - energy stemming from the fock exchange term of the one - gluon - exchange interaction has a central role to cause spin polarization . the magnitude of spin polarization is determined by the coupled schwinger - dyson equations with a superconducting gap function . as a significant feature , the fermi surface is deformed by the axial - vector self - energy and then rotation symmetry is spontaneously broken down . the gap function results in being anisotropic in the momentum space in accordance with the deformation . as a result of numerical calculations , it is found that spin polarization barely conflicts with color superconductivity , but almost coexists with it . = -2.0 cm = 1.5em

hep-ph0304223

recently much interest is given for high - density qcd , especially for quark cooper - pair condensation phenomena at high - density quark matter ( called as color superconductivity ( csc ) ) , in connection with , e.g. , physics of heavy ion collisions and neutron stars @xcite . its mechanism is similar to the bcs theory for the electron - phonon system @xcite , in which the attractive interaction of electrons is provided by phonon exchange and causes the cooper instability near the fermi surface . as for quark matter , the quark - quark interaction is mediated by colored gluons , and is often approximated by some effective interactions , e.g. , the one - gluon - exchange ( oge ) or the instanton - induced interaction , both of which give rise to the attractive quark - quark interaction in the color anti - symmetric @xmath0 channel . csc leads to spontaneous symmetry breaking of color @xmath1 into @xmath2 as a result of condensation of quark cooper pairs @xcite . in this paper we would like to address another phenomenon expected in quark matter : spin polarization or ferromagnetism of quark matter . we examine the possibility of the spin - polarized phase with csc in quark matter . as far as we know , interplay between the color superconducting phase and other phases characterized by the non - vanishing mean fields of the spinor bilinears @xmath3 has not been explored except for the case of chiral symmetry breaking @xcite . our main concern here is to investigate the possibility of the quark cooper instability under the axial - vector mean - field , @xmath4 which is responsible for spin polarization of quark matter . it would be worth mentioning in this context that ferromagnetism ( or spin polarization ) and superconductivity are fundamental concepts in condensed matter physics , and their coexistent phase has been discussed for a long time @xcite . as a recent progress , a superconducting phase have been discovered in ferromagnetic materials and many efforts have been made to understand the coexisting mechanism @xcite . besides being interesting in its own right , the coexistence problem may be related to some physical phenomena . recently , a new type of neutron stars , called as `` magnetars '' , with a super strong magnetic field of @xmath5 g ) has been discovered @xcite . they may raise an interesting question for the origin of the magnetic field in compact stars , since its strength is too large to regard it as a successor from progenitor stars , unlike canonical neutron stars @xcite . since hadronic matter spreads over inside neutron stars beyond the nuclear density ( @xmath6@xmath7@xmath8@xmath9 ) , it should be interesting to consider the microscopic origin of the magnetic field in magnetars . in this context , a possibility of ferromagnetism in quark matter due to the oge interaction has been suggested by one of the authors ( t.t . ) within a variational framework @xcite ; a competition between the kinetic and the fock exchange energies gives rise to spin polarization , similarly to bloch s idea for itinerant electrons . salient features of spin polarization in the relativistic system are also discussed in ref . thus , it might be also interesting to examine the possibility of the spin - polarized phase with csc in quark matter , in connection with magnetars . we investigate spin polarization in the color superconducting phase by a self - consistent framework , in which quark cooper pairs are formed under the axial - vector mean - field . we shall see that this phenomenon is a manifestation of spontaneous breaking of both color @xmath1 and rotation symmetries . we adopt here the oge interaction as an effective quark - quark interaction . since the fermi momentum is very large at high density , asymptotic freedom of qcd implies that the interaction between quarks is very weak @xcite . so it may be reasonable to think that the oge interaction has a dominant contribution for the quark - quark interaction . in the framework of relativistic mean - field theories , the axial - vector and tensor mean - fields , which stem from the fock exchange terms , @xmath10 and @xmath11 , may have a central role to split the degenerate single - particle energies of the two spin states , and then leads to spin polarization , e.g. , see @xcite for discussion in nuclear matter . as for quark matter , several types of the color singlet mean - fields appear after the fierz transformation in the fock exchange terms , but we retain only the axial - vector mean - field as the origin of spin polarization , because the oge interaction by no means holds the tensor mean - field due to chiral symmetry in qcd , unlike nuclear matter @xcite . presence of the axial - vector mean - field deforms the quark fermi seas according to their spin degrees of freedom , and thereby the gap function should be no more isotropic in the momentum space . we assume here an anisotropic gap function @xmath12 on the fermi surface by a physical consideration and solve the coupled schwinger - dyson equations self - consistently by way of the nambu formalism to find the axial - vector mean - field @xmath13 and the superconducting gap function @xmath12 . thus we discuss the interplay between spin polarization and superconductivity in quark matter . in section 2 we give a framework to deal with the present subject . the explicit structure of the anisotropic gap function @xmath12 in the color , flavor , and dirac spaces is carefully discussed there and in the appendix b and appendix c. numerical results about @xmath13 and @xmath12 are given in section 3 , where phase diagram of spin polarization and color superconductivity is given in the mass - baryon number density plane . section 4 is devoted to summary and concluding remarks . in this section we present our formalism to treat csc and spin polarization . we consider quark matter with flavor @xmath2 and color @xmath1 symmetries , and assume that the interaction action is described by the oge interaction as @xmath14 d_{\mu \nu}(x , y ) \left[\bar{\psi}(y)\gamma^\nu \frac{\lambda_a}{2 } \psi(y)\right ] , \end{aligned}\ ] ] where @xmath15 is the quark field , @xmath16 is the gauge boson ( gluon ) propagator , and @xmath17 are the @xmath1 gell - mann matrices . using the nambu formalism @xcite the effective action is given within the mean - field approximation as @xmath18 with the inverse quark green function @xmath19 where @xmath20 with the chemical potential @xmath21 . @xmath22 is a self - energy and @xmath12 is the gap function for the quark cooper pair ; both terms @xmath22 and @xmath12 should be provided by the fock exchange terms of the oge interaction . we define here @xmath23 and @xmath24 as @xmath25 with the charge conjugation matrix @xmath26 which is explicitly given by @xmath27 in dirac representation . the green function @xmath28 can be written straightforwardly from eq.([fullg ] ) as @xmath29 with @xmath30^{-1 } \label{g11i } \\ g_{21}(p ) & = & -\left ( \sla{p}-m-\sla{\mu}+\overline{v}(p ) \right)^{-1 } \delta(p ) g_{11}(p ) . \label{g21i}\end{aligned}\ ] ] following nambu s argument @xcite , we impose the self - consistency condition to obtain the hartree - fock ground state such that the self - energy by the residual interaction , @xmath31 , vanishes : @xmath32 where @xmath33 is defined by @xmath34 , \end{aligned}\ ] ] and @xmath35 is given by the use of the oge interaction . within the first - order approximation in @xmath36 , @xmath35 renders @xmath37 which is nothing else but the fock exchange energy by the oge interaction . using eqs . ( [ res ] ) - ( [ self11 ] ) , we obtain the self - consistent equation for @xmath38 by the use of the diagonal component of the full green function ( [ g11i ] ) : @xmath39 \gamma_\mu \frac{\lambda_\alpha}{2 } [ -ig_{11}(p ) ] \gamma_\nu \frac{\lambda_\alpha}{2 } \label{self1}. \end{aligned}\ ] ] the gap equation is also obtained from the off - diagonal component as @xmath40 \gamma_\mu \frac{-(\lambda_\alpha)^t}{2 } [ -ig_{21}(p ) ] \gamma_\nu \frac{\lambda_\alpha}{2}. \label{gap1}\ ] ] in the following sections , we present explicit forms of @xmath41 and @xmath42 and then solve their coupled equations ( [ self1 ] ) and ( [ gap1 ] ) . we , hereafter , take the static approximation for the gauge - boson propagator as @xmath43 where @xmath44 is an effective gauge boson mass originated from the debye screening @xmath45 @xcite . since typical momentum transfer @xmath46 at high density is of the order of the chemical potential , we may further introduce the zero - range approximation @xcite for the propagator as @xmath47 with a typical momentum scale @xmath48 of @xmath49 . this approximation corresponds to the stoner model @xcite , which is popular in solid - state physics , and stands on the same concept of the njl model @xcite as well . to proceed , we assume , without loss of generality , that total spin expectation value is oriented to the negative @xmath50-direction in the spin - polarized phase which is caused by the finite axial - vector mean - field along the @xmath50-axis - axis without loss of generality . ] . as shown in ref . @xcite , rotation symmetry is spontaneously broken down in this phase while axial symmetry around the @xmath50-axis is preserved . then two fermi seas of the different spin states are deformed accordingly . applying the fierz transformation for the fock exchange energy term ( 14 ) we can see that there appear the color - singlet scalar , pseudoscalar , vector and axial - vector self - energies ( appendix d ) . in general we must take into account these self - energies in @xmath22 , @xmath51 with the mean - fields @xmath52 . here we introduce an ansatz : the femri distribution holds the reflection symmetry with respect to the @xmath53 plane , and only the mean - field parts @xmath54 , @xmath55 and @xmath56 are retained in @xmath22 . later we will see that the self - consistent solution is obtained with the zero - range approximation ( [ prop - zero ] ) under this ansatz . in this paper , furthermore , we disregard the scalar mean - field @xmath54 and the time component of the vector mean - field @xmath57 for simplicity since they are irrelevant for the spin degree of freedom ; @xmath57 has only a role to shift the total energy ot the chemical potential , and may not affect any other physical properties . on the other hand , @xmath54 may significantly influence the spin - polarization properties through changing the quark effective mass . instead of introducing the scalar mean - field explicitly , however , we treat the quark mass as a variable parameter , and discuss its effect in the next section . according to the above assumptions and considerations the self - energy @xmath22 in eq.([fullg ] ) renders @xmath58 with the axial - vector mean - field @xmath13 . then the diagonal component of the green function @xmath59 is written as @xmath60^{-1}\ ] ] with @xmath61 where @xmath62 and @xmath63 is the green function with the axial - vector mean - field @xmath13 which is determined self - consistently by way of eq . ( [ self1 ] ) . before constructing the gap function @xmath12 , we first find the single - particle spectra and their eigenspinors in the absence of @xmath12 , which is achieved by diagonalization of the operator @xmath64 . in the usual case of no spin polarization this procedure gives nothing but the free energy spectra and plane waves . then we choose a gap structure on the basis of a physical consideration as in the usual bcs theory . from the condition that @xmath65 one can obtain four single - particle energies @xmath66 ( positive energies ) and @xmath67 ( negative energies ) , which are given as @xmath68 where the sign factor @xmath69 being in front of @xmath13 indicates the energy splitting between different spin states due to the presence of the axial - vector self - energy , which corresponds to the _ exchange splitting _ in the non - relativistic electron system @xcite . in the following , we call the `` spin''-up ( -down ) states for the states @xmath70 ( @xmath71 ) . ( [ eig ] ) also shows that each fermi sea for the `` spin''-up ( -down ) state should undergo a deformation and lose rotation symmetry , once @xmath13 is finite . this is a genuine relativistic effect @xcite ; actually the exchange splitting never produces deformation of the fermi sea in the non - relativistic ferromagnetism , e.g. in the stoner model @xcite . here , it would be interesting to see the peculiarities of the quark fermi seas in the presence of the axial - vector self - energy . in fig . [ fs ] we sketch the profile of the fermi seas projected onto the @xmath72 plane ( @xmath73 ) for the cases of ( a ) @xmath74 , ( b ) @xmath75 and ( c ) @xmath76 . as is already mentioned these seas still hold the axial symmetry around the @xmath50-axis and the reflection symmetry with respect to the @xmath77 plane . the region surrounded by the outer line show the fermi sea of `` spin''-down quarks , and the shaded region is that of `` spin''-up quarks . = 13.5 cm = 6.8 cm we can see in fig . [ fs]a that the fermi seas for the `` spin''-down and `` spin''-up states are deformed in the prolate and oblate shapes , respectively , where the minimum of the single - particle energy still resides at the origin @xmath78 . when @xmath75 as shown in fig . [ fs]b , there appear two minima at the points @xmath79 for the `` spin''-down quark . hence in the massless limit , @xmath80 , the fermi sea is described by two identical spheres with radii @xmath21 in the momentum space , which are centered at the points @xmath81 ( see fig . [ fs]c ) . in what follows we use subscript ` @xmath82'@xmath83 for notational convenience as @xmath84 which means @xmath85 . we define the spinor @xmath86 that satisfies the equation @xmath87 , which corresponds to the eigenspinor with the single - particle energy @xmath84 in the absence of the quark cooper pairing . the spinor @xmath86 is explicitly given as @xmath88(p_x-{\rm i } p_y)\\ p_t^2 p_z \end{array } \right ) , \label{spinor } \ ] ] where @xmath89@xmath90\left [ \epsilon_n+u_a+(-1)^n \beta_p \right]/ ( \epsilon_n \beta_p)}/(2p_t^2 p_z)$ ] and @xmath91 . it is to be noted that the spinors @xmath92 do not return to the eigenspinors of spin operator @xmath93 even when @xmath94 , but become mixtures of them , see appendix a. introducing the projection operator @xmath95 with properties @xmath96 and @xmath97 , we can recast @xmath63 in the spectral representation into @xmath98 in this subsection , we give the explicit form of the gap function @xmath12 in the dirac , color and flavor spaces , and then calculate the diagonal component of the full green function @xmath59 in eq . ( [ g11i ] ) , provided that only the axial - vector self - energy is taken for @xmath41 in eq . ( [ self1 ] ) . in general various types of the gap structures are possible in the dirac , color and flavor spaces ; they depend on the form of interaction and the quark mass @xcite , especially on the strange quark mass @xcite . here we suppose a simple gap structure from a physical consideration , disregarding the finite mass effect . using the the spinor @xmath86 we assume that the gap function @xmath12 in eq . ( [ gap1 ] ) has a following form in the color and flavor spaces : @xmath99 with the operator @xmath100 , @xmath101 where the subscript ` @xmath102'(@xmath103 ) indicates that the single - particle energy in the spinor is replaced by that of opposite sign , @xmath104 , without change of `` spin '' . one can easily see what kind of quark pairs the gap function @xmath12 ( [ aa ] ) represents . utilizing the property , @xmath105@xmath106@xmath107 , one can find for the general spinor @xmath108 with arbitrary coefficients @xmath109 , @xmath110 this equation clearly shows that two quarks included in the cooper pairing have opposite momenta to each other and belong to the same energy eigenstate as illustrated in fig . [ fs]b . now we should note that the antisymmetric nature of the fermion self - energy imposes a constraint on the gap function @xcite , @xmath111 since @xmath112 satisfies the relation @xmath113 , @xmath114 must be a symmetric matrix in the spaces of internal degrees of freedom . taking into account the property that the most attractive channel of the oge interaction is the color antisymmetric @xmath0 one , it must be the flavor singlet state . thus we can choose the form of the gap function as @xmath115_{\alpha \beta , ~i j}= \epsilon_{\alpha \beta 3 } \epsilon_{i j } \delta_n({{\mbox{\boldmath $ p$ } } } ) , \label{2sc}\ ] ] where @xmath116 and @xmath117 are indices in three - color and two - flavor spaces , respectively . the form of gap function ( [ 2sc ] ) in the color and flavor spaces is familiar for two - flavor csc @xcite . using the properties of @xmath118 and @xmath119 , we then obtain an explicit form of @xmath59 as @xmath120_{\alpha \beta , i j } & = & \left\ { \sum_n \left [ ( p_0+\mu-\epsilon_n ) - \frac{\delta_n^\dagger \delta_n}{p_0+\epsilon_n-\mu } \right ] \gamma_0 \lambda_n \right\}^{-1}_{\alpha \beta , i j } \nonumber \\ & = & \sum_n \frac{p_0-\mu+\epsilon_n } { p_0 ^ 2-(\epsilon_n-\mu)^2-\frac{1}{2}{\rm tr}[\delta_n^\dagger \delta_n ] ( 1-\delta_{3 \alpha})+i\eta } \lambda_n \gamma_0 \ , \delta_{\alpha \beta}\ , \delta_{i j } \label{g11 } \end{aligned}\ ] ] with @xmath121 where @xmath122 is a positive infinitesimal . the quasiparticle energies are obtained by looking for the poles of @xmath59 : @xmath123 the quark number density @xmath124 is also given as @xmath125 \label{bn } \\ & = & n_f \sum_{n=1,2 } \int \frac{{\rm d}^3 p}{(2\pi)^3 } \left [ \theta(\mu-\epsilon_{n } ) + 2 v_{n}^2({{\mbox{\boldmath $ p$ } } } ) - 2 \left(1-v_{-n}^2({{\mbox{\boldmath $ p$}}})\right ) \right ] \label{bn2}\end{aligned}\ ] ] with @xmath126 where the first two terms in eq . ( [ bn2 ] ) show the quark contributions , while the last term the anti - quark contribution ; @xmath127 is the occupation probability of the quark pairs with momentum @xmath128 and represents diffuseness of the momentum distribution . similarly we can know the self - consistent solutions satisfy our ansatz about the mean - fields in @xmath22 . from the above solutions we can easily obtaine that @xmath129=0 $ ] , @xmath130\propto p_i$ ] , @xmath131\propto p_z$ ] and @xmath132\propto p_x , p_y$ ] . hence the pseudoscalar mean - field @xmath133 , the space - component of vector mean - field @xmath134 , the axial - vector mean - fields @xmath135 and @xmath136 are vanished after the integration over angles . using eq . ( [ g11 ] ) , the off - diagonal component of the full green function @xmath28 , given in eq . ( [ g21i ] ) , can be represented in the similar way as @xmath137 where @xmath138 is the pauli matrix in the two - flavor space . substituting eq . ( [ g21 ] ) into the gap equation ( [ gap1 ] ) and using the identity @xmath139 , we obtain @xmath140 \delta_n({{\mbox{\boldmath $ p$ } } } ) , \label{gap1i}\ ] ] where the factor @xmath141 is simply the fierz coefficient for the color and flavor degrees of freedom ( appendix d ) . furthermore multiplying both sides of eq . ( [ gap1i ] ) by @xmath142 and taking trace with respect to the dirac indices , the coupled equations for the gap functions @xmath143 are obtained after @xmath144 integration , @xmath145 where the function @xmath146 is defined as @xmath147 = ( \bar{\phi}_{-n'}({{\mbox{\boldmath $ k$ } } } ) \gamma^\mu \phi_{-n}({{\mbox{\boldmath $ p$ } } } ) ) ( \bar{\phi}_{n}({{\mbox{\boldmath $ p$ } } } ) \gamma^\nu \phi_{n'}({{\mbox{\boldmath $ k$ } } } ) ) \label{mn},\end{aligned}\ ] ] a decomposition of @xmath100 in terms of gamma matrices and its properties are given in appendix b. here we take the zero - range approximation in eq . ( [ prop - zero ] ) . in terms of the polar coordinates @xmath148 , we can consider that the gap function @xmath149 does not depend on the horizontal angle @xmath150 due to axial - symmetry around the @xmath151-axis . thus we can explicitly perform the integration with respect to the angle @xmath150 in the gap equation ( [ gap1j ] ) : @xmath152 with the effective coupling constant @xmath153 . as seen from the above equation , each of the gap functions couples with others by the function @xmath154 defined as @xmath155 , \label{tss1}\ ] ] where @xmath156@xmath157@xmath158 and @xmath151@xmath157@xmath159 and the same for @xmath160 and @xmath161 . the term proportional to @xmath151 in eq . ( [ tss1 ] ) will disappear after the integration over @xmath162 . using eqs . ( [ qusie ] ) and ( [ cof ] ) , @xmath59 is recasted in the form , @xmath163_{\alpha \beta , ~i j } = \left[\sum_n \left(\frac{1-v_n^2({{\mbox{\boldmath $ p$}}})}{p_0-e_n+i\eta}+ \frac{v_n^2({{\mbox{\boldmath $ p$}}})}{p_0+e_n - i\eta } \right ) { \rm e}^{ip_0 \eta } \lambda_n({{\mbox{\boldmath $ p$ } } } ) \gamma_0 \right ] \delta_{\alpha \beta } \delta_{i j}. \end{aligned}\ ] ] substituting the above equation into eq . ( [ self1 ] ) , and integrating with respect to @xmath144 , we obtain the self - consistent equation for @xmath13 in the zero - range approximation : @xmath164 s_n({{\mbox{\boldmath $ p$ } } } ) , \label{ua1 } \end{aligned}\ ] ] where the factor @xmath165 stems from the fierz coefficient of the color - singlet axial - vector channel of the oge interaction ( appendix d ) , and @xmath166 is the expectation value of the spin operator , @xmath93@xmath157@xmath167 , with respect to the spinor @xmath86 : @xmath168 thus @xmath13 is related to the expectation value of @xmath93 summing over the state with momentum @xmath128 . an effect of the cooper pairing enters into eq . ( [ ua1 ] ) through the function @xmath169 . in this subsection we consider a high - density limit , which means the weak coupling limit due to asymptotic freedom of qcd , and then disregards the anti - quark pairing and contributions from the negative - energy sea ( the dirac sea ) in eq . ( [ gap1 ] ) . actually it costs more energy to form the anti - quark pairing than the quark pairing for a large chemical potential . taking the approximation , we also disregard the contribution from anti - quarks to calculate the quark number density in eq . ( [ bn2 ] ) and the axial - vector mean - field eq . ( [ ua1 ] ) consistently . to @xmath170 , which correspond to the positive - energy states with different `` spins '' specified by the subscript @xmath171 . ] in the following calculations we define gap functions of the quark pairing by subscript @xmath172 whcih corresponds to the `` spin''-up ( -down ) of positive - energy states as @xmath173 and @xmath174 . the other symbols with the subscript @xmath172 have the same meaning , e.g. , @xmath175 . in addition , we assume that only quarks near the fermi surface form the cooper pairs , and thereby replace the gap function by an approximated form , @xmath176 where @xmath177 is a cut - off parameter around the fermi surface . the function @xmath178 is also regarded as a form factor to regularize the integration in the gap equation @xcite . the step - function form factor mimics the asymptotic freedom ; inner particles in fermi sea costs large kinetic energy to create the pairing and takes large momentum transfer which indicates that coupling of this inner - process is small . there , however , might be more realistic form factors for finite density qcd , which are smoother functions of momentum and @xmath21 than ours , we think that they makes little change on qualitative results of the csc and spin polarization . there are models with other form factors or cut - off functions @xcite . here we examine the polar - angle dependence of the anisotropic gap function at the fermi surface @xmath187 . the fermi momentum @xmath188 of each `` spin '' eigenstate is given as @xmath189^{1/2 } , \end{aligned}\ ] ] where the subscript @xmath172 corresponds to the `` spin''-up ( -down ) state again . substituting the above formula into the gap function ( [ para ] ) , we get @xmath190.\ ] ] note that this form exhibits a @xmath191- wave pairing nature : it is a genuine relativistic effect by the dirac spinors ( appendix b ) . we show a schematic view of the above gap functions in fig . [ delta ] . as characteristic features , both the gap functions vanish at poles ( @xmath192 ) and take maximal values near equator ( @xmath193 ) , keeping the relation , @xmath194.- pairing in liquid @xmath195 @xcite or nuclear matter @xcite . ] suppression of @xmath196 and enhancement of @xmath197 at @xmath193 for the case of @xmath198 ( fig . [ delta]b ) are originated from a finite value of @xmath180 , while they vanish if quark is taken to be massless ( fig . [ delta]a ) . the anisotropic gap functions give rise to the different diffuseness in the momentum distribution of the two `` spin '' eigenstates , and thereby make some effects on spin polarization , unlike in the normal phase . the anisotropic diffuseness has two effects that it obscures the deformation in the momentum distribution due to their angle dependence and enlarges the difference of the state density between the two `` spin '' eigenstates through the relation @xmath194 . in this section we solve the coupled equations ( [ ua1 ] ) , ( [ eqf ] ) and ( [ eqr ] ) and investigate the effects of the superconducting gap on spin polarization . before going to numerical calculations of @xmath13,@xmath180 and @xmath181 , each of which is coupled with others by the self - consistent equations , it is instructive to see their parameter dependence by treating one of them as an input parameter . first we show @xmath180 and @xmath181 as functions of @xmath13 in fig . [ parau ] where @xmath199 mev and @xmath200 . = 6.8 cm @xmath180 starts from zero and almost linearly increases with @xmath13 ( fig . [ parau]a ) , which is understood by seeing that @xmath180 is proportional to the difference , @xmath201 , due to finite @xmath13 , see eq . ( [ para ] ) or appendix c. thus @xmath180 is induced by @xmath13 and closely coupled with it . as for the behavior of @xmath181 , it is barely affected by @xmath13 ( slight decreasing with @xmath13 in the numerical value ) ( fig . [ parau]b ) . as seen from the dependence on @xmath21 the magnitude of @xmath181 is almost determined by the volume of the phase space in the gap equation , that is , by @xmath21 and @xmath177 . this reflects the fact that @xmath181 is related to the sum , @xmath202 ( appendix c ) . thus we expect that @xmath181 increases with density when other parameters are fixed . from the above results we have found that @xmath181 is not so much influenced by @xmath13 . next we examine the behavior of @xmath13 and @xmath180 when @xmath181 is treated to be an input parameter . in fig . [ paraf ] we show the parameter dependence of @xmath13 and @xmath180 on @xmath181 , where @xmath203 mev and we use three values of the cut - off parameter @xmath204 , @xmath205 and @xmath206 , and add the result of @xmath13 in the normal phase ( @xmath207 ) . = 6.8 cm comparing the dependence of @xmath13 on @xmath181 ( fig . [ paraf]a ) with that in the normal phase , we see a characteristic behavior for different values of @xmath177 : there are regions where @xmath13 is larger than that in the normal phase for relatively small @xmath181 , and this region seems to extend with @xmath177 . on the other hand results from the self - consistent calculations show that @xmath181 becomes larger with @xmath177 so that its value corresponds to a region where @xmath13 is comparable with or slightly less than that in the normal phase , for any value of the chemical potential . this situation seems to be qualitatively unchanged , once the ratio of the effective coupling constants in the axial - vector channel @xmath208 and the diquark channel @xmath209 is kept , @xmath210 , which comes from the fierz transformation for color and flavor ( appendix d ) . however , if the coupling constant in each channel is taken independently , our results might be changed qualitatively . seeing the results for @xmath180 in fig . [ paraf]b , we find that @xmath180 increases with @xmath177 due to the growth of the phase space and is hardly affected by @xmath181 except the region of small @xmath181 where @xmath13 varies rapidly as shown in fig . [ paraf]a : it also shows that @xmath180 is closely related to @xmath13 . these parameter dependences also suggest that the regularization scheme for the gap equation , i.e.,the sharp momentum cut - off function , the form factor , etc . , will give rise to a qualitative change to @xmath13 . in the present cut - off function , @xmath211 , @xmath13 ( spin polarization ) coexists with csc , except a slight competition , as will be shown later . we demonstrate some self - consistent solutions here . since we have little information to determine the values of the parameters @xmath212 and @xmath177 ( there may be other more reasonable form factors than the present cut - off function ) , and our purpose is to figure out qualitative properties of spin polarization in the color superconducting phase , we mainly set in the following calculations them as @xmath213 mev@xmath214 and @xmath215 , for example , which is not so far from the couplings in njl - like models @xcite . we first examine spin polarization in the absence of csc . in fig . [ ua ] we show the the axial - vector mean - field @xmath13 , with @xmath216 being set to be zero , as a function of baryon number density @xmath217 relative to the normal nuclear density @xmath218 @xmath219 for @xmath220 mev ( dashed lines ) . = 6.9 cm it is seen that the axial - vector mean - field ( spin polarization ) appears above a critical density and becomes larger as baryon number density gets higher . moreover , the results for different values of the quark mass show that spin polarization grows more for the larger quark mass . this is because a large quark mass gives rise to much difference in the fermi seas of two opposite `` spin '' states , which leads to growth of the exchange energy in the axial - vector channel . = 6.9 cm next we solve the coupled equations ( [ ua1 ] ) , ( [ eqf ] ) and ( [ eqr ] ) . results for @xmath13 , @xmath180 and @xmath181 are shown in fig . [ ua ] ( solid lines ) and fig . [ fr ] , for values of the quark mass @xmath220 mev . it is found again , by comparing these cases of the quark mass , that @xmath13 is very sensitive to the quark mass and increases with it as in the absence of csc ( fig . [ ua ] ) . for the behavior of the gap functions , @xmath180 is induced by @xmath13 and both of @xmath181 and @xmath180 increase with @xmath221 due to the growth of the fermi surface ( fig . it is also seen that @xmath181 is not sensitive to the quark mass . to see the bulk behavior of pairing gap as a function of baryon number density , we also show , in fig . [ dpm ] , their mean values with respect to the polar angle on the fermi surface ; @xmath222 the mean values @xmath223 begin to split with each other at a density where @xmath13 becomes finite . this reflects that @xmath180 is induced by @xmath13 and then has a negative ( positive ) contribution to @xmath196 ( @xmath197 ) . = 6.8 cm here we would like to comment on the magnitude of @xmath223 . these should be compared with the usual uniform gap function , and may look very large values of @xmath224(gev ) in our case . however these values would be largely reduced by taking a smooth form factor which models asymptotic freedom of qcd@xcite ; it further reduces the integral value in the gap equation , compared with our sharp cut - off function . in fig . [ spin ] we show the expectation value of the spin operator per quark , @xmath225 , as a function of @xmath226 with and without the superconducting gap . the critical density becomes lower as the quark mass increases , and the peak positions of @xmath225 are located at relatively lower densities in each quark mass . the magnitude of @xmath225 is to be compared with @xmath227 for a free quark , because @xmath228 at the rest frame for the free spinor @xmath229 . we arrange the results of three quark masses @xmath230 mev by @xmath227 mev in fig . [ spin]a to show a high sensitivity of spin polarization to the quark mass , which implies that the exchange energy from the attractive axial - vector interaction is strongly enhanced by the quark mass to produce the large axial - vector mean - field . the exchange energy is also enhanced by larger chemical potential and the resulting axial - vector mean - field increases with it ( see fig.[ua ] ) . but the spin expectation value per quark , which is relative to the axial - vector mean - field per quark ( @xmath231 ) , has an upper limit since the increase of @xmath232 is far superior to that of @xmath13 for larger chemical potential , which gives rise to the peak positions in fig . [ spin ] . = 15.7 cm = 7.8 cm the quark mass is very important in relation to the breaking of chiral symmetry in qcd . models incorporating chiral dynamics have indicated that the dynamical mass becomes smaller as chiral symmetry is restored at a high density , while the current quark mass is small and explicitly breaks it @xcite . in our model , on the other hand , we treat the quark mass @xmath233 as a variable parameter so that we may simulate a change of the dynamical mass . in order to further examine the effect of the quark mass on spin polarization , we show the mass dependence at densities @xmath234 5@xmath6 , @xmath234 10@xmath6 and @xmath234 15@xmath6 for the cases with and without the superconducting gap in fig . [ spinmass ] . = 6.9 cm spin polarization increases with the quark mass in all the three densities . in the figure we exhibit only the results for a narrow region of the mass parameter ( @xmath235 mev ) , while as for larger masses of @xmath236 ( order of the strange quark mass ) spin polarization monotonically increases without singular oscillations . critical values of the quark mass at which spin polarization disappears become smaller as density increases in both cases . in relation of @xmath13 to @xmath233 we can derive an exact result in the massless limit , @xmath80 . in the normal phase where @xmath237 , eq . ( [ ua1 ] ) becomes @xmath238 with @xmath239 the right - hand side of the above equation can be analytically integrated to give @xmath240 here we have assumed that @xmath241 . in the massless limit , the fermi sea is described by two complete spheres in the momentum space with radii @xmath21 , whose centers are located at @xmath81 ( see fig . [ fs]c ) . the momentum distribution for quarks in the `` spin''-down state occupies these two spheres , while the `` spin''-up state does their overlap region ( shaded ) . in the above integration for the `` spin''-down quarks , the expectation value of the spin operator given by quarks with @xmath242 is canceled with that of quarks with @xmath243 . the remaining contribution from the region , @xmath244 , cancels with that by the `` spin''-up quarks . thus we can see that spin polarization disappears as @xmath80 in the absence of csc . this analytical result that @xmath94 as @xmath80 can be also understood as follows . the eigenstates of non - interacting massless fermions are classified by the definite helicity states : the positive energy state is right - handed ( left - handed ) with positive ( negative ) helicity , while the negative energy state those with negative ( positive ) helicity . this property is not spoiled by introducing the axial - vector mean - field , when we extend the meaning of helicity ; the dirac equations for the `` left- '' and `` right - handed '' positive - energy fermion fields @xmath245 are now given as @xmath246 which give the eigenvalues , @xmath247 for @xmath248 and @xmath249 for @xmath250 . @xmath248 ( @xmath250 ) is the eigenstate of generalized helicity @xmath251 projected onto the shifted momentum @xmath252 . if @xmath253 they form the spherical fermi seas , see fig . [ fs]c . here it would be interesting to compare these eigenvalues with the limit form of @xmath254 in eq . ( [ eig ] ) , @xmath255 then we can see the relations : @xmath256 , which clearly show that the two fermi seas of the eigenspinors ( [ spinor ] ) give the same fermi seas of @xmath245 for a given chemical potential @xmath21 . thus we can take an alternative view of the fermi seas in terms of the definite helicity states by rearranging the eigenspinor ( [ spinor ] ) properly in the massless limit . in each fermi sea for @xmath245 the particle number with the definite @xmath257 becomes the same , and thereby the total spin - expectation value becomes vanished . in the color superconducting phase , on the other hand , the situation is different because the momentum distribution becomes diffused due to the creation of the cooper pairs near the fermi surface . for @xmath80 and @xmath258 , eq . ( [ ua1 ] ) becomes @xmath259 where @xmath260 indicates the diffused part of the momentum distribution , defined in eq . ( [ cof ] ) . here we should note that the gap functions @xmath261 are still non zero even at @xmath76 . the diffused part , however , give no contribution to the spin polarization from the viewpoint of the helicity eigenstates . the gap function in the massless limit becomes @xmath262 see fig . [ delta]a . the diffuseness from the above gap function has an equal contribution to the two complete fermi spheres of chirality . thus the total spin , which is obtained by summing up these momentum distributions , should be zero in the massless limit even if the csc is taken into account . to summarize we show a phase diagram for the quark mass and baryon number density in fig . [ pd ] where we add the result of @xmath263mev@xmath214 to see the dependence on the coupling constant . = 6.9 cm the lines indicate the critical mass at a fixed density ( at regions above the lines spin polarization arises ) . we can confirm that the critical mass becomes smaller with the increase of the density , and spin polarization occurs at moderate densities ( @xmath264 ) if the coupling is strong enough even though quark mass is taken to be smaller as a simulation for change of dynamical mass ( restoration of chiral symmetry ) . here we would like to understand how the gap function affects spin polarization and brings about a slight reduction of it . in the spin - polarized phase , the momentum distribution is deformed from the simple spherical shape . as mentioned in ref . @xcite , the deformation is induced by finite @xmath13 and feeds back to @xmath13 in a self - consistent manner . in the color superconducting phase , diffuseness caused by the cooper pairing in the momentum distribution depends on the polar angle and then has an influence on the deformation . as can be expected from the polar - angle dependence of the gap function , diffuseness tends to obscure the deformation . from the consideration of the spin expectation values by spinors ( [ spinor ] ) near the fermi surfaces ; @xmath265 where @xmath266 for two `` spins '' . the difference of the spin expectation value between two `` spin '' states @xmath267 is largely affected by high-@xmath151 regions or regions near both poles ( @xmath192 ) . thus the large deformation along the @xmath50-axis seems to enhance spin polarization . in order to specify to what extent the fermi sea is deformed , we calculate the quadrupole deformation of the momentum distribution defined by @xmath268 in fig . [ q2 ] , we show @xmath269 as a function of @xmath13 at @xmath203 mev in the normal phase and in the color superconducting phase in which the gap functions are given by their equations for fixed @xmath13 . = 6.7 cm from this result of @xmath269 deformation , we can see that the diffused part near the fermi surface obscures the deformation then gives an opposite effect against @xmath269 , and thus reduces spin polarization . nevertheless the gap function has another effect on spin polarization . it is to be noted that the qualitative relation , @xmath194 , is always retained as seen from fig . [ delta ] and then has a effect to enlarge the difference of the state density between the two `` spin '' states . this effect is expected to enhance spin polarization since the difference of the spin expectation value by each spinor ( [ nearfermi ] ) near the equator ( @xmath193 ) , so that @xmath270 , seems to depend only on the difference of the state density . to see it in both the normal and color superconducting phases , we define that @xmath271 ( @xmath272 ) is the state density of the `` spin''-up ( -down ) state and show their difference by @xmath273 , only in the first two colors , as a function of @xmath13 in fig . [ frac ] at @xmath203 mev . the result indicates that the gap functions slightly enhance @xmath274 than normal phase . = 6.7 cm from the above discussions spin polarization is significantly influenced by both the deformation and the state density in each `` spin '' state . as a result of the self - consistent calculation in the color superconducting phase , the reduction effect on the deformation is slightly superior to the enhancement effect from the difference of the state densities , and the pairing effect finally reduces spin polarization than in the normal phase . it , however , should be noted that this qualitative conclusion about whether csc enhances spin polarization than normal phase or not is very delicate and may be changed depending on the regularization scheme , as already mentioned . moreover other types of pairing which are not considered here , e.g. pairing of the `` spin '' -up and -down states , may gives rise to qualitatively different results , while it is very difficult to see which type of pairing is energetically favored . finally we would like to comment on the coupling of the spin polarized quark matter with the external magnetic field ; quark fields couple with the magnetic field through its anomalous magnetic moment . the magnetic interaction is described by the gordon identity for the gauge coupling term : @xmath275 where @xmath276 is a form factor and @xmath277 an effective charge . in quark matter a magnetic moment is given as the expectation value @xmath278 with respect to the ground state . in our model only @xmath279 is nonzero , and the magnetic moment per quark is given as @xmath280 \bar{\phi}_n({{\mbox{\boldmath $ p$ } } } ) \sigma_{12 } \phi_n({{\mbox{\boldmath $ p$ } } } ) . \label{mz}\ ] ] note that the expectation value of @xmath281 by the spinor does not depend on @xmath13 ; @xmath282 , so that @xmath283 reflects only the asymmetry in the momentum distribution due to the axial - vector mean - field . = 15.5 cm in fig . [ mag ] , @xmath283 is given as a function of baryon number density . this indicates that resulting ground state also holds ferromagnetism ( spontaneous magnetization ) . in this paper we have examined spin polarization in quark matter in the color superconducting phase . we have introduced the axial - vector self - energy and the quark pair field ( the gap function ) , whose forms are derived from the one - gluon - exchange interaction by way of the fierz transformation under the zero - range approximation . within the relativistic hartree - fock framework we have evaluated their magnitudes in a self - consistent manner by way of the coupled schwinger - dyson equations . as a result of numerical calculations spontaneous spin polarization occurs at a high density for a finite quark mass in the absence of csc , while it never appears for massless quarks as an analytical result . in the spin - polarized phase the single - particle energies corresponding to spin degrees of freedom , which are degenerate in the non - interacting system , are split by the exchange energy in the axial - vector channel . each fermi sea of the single - particle energy deforms in a different way , which causes an asymmetry in the two fermi seas and then induces the axial - vector mean - field in a self - consistent manner . in the superconducting phase , however , spin polarization is slightly reduced by the pairing effect ; it is caused by competition between reduction of the deformation and enhancement of the difference in the phase spaces of opposite `` spin '' states due to the anisotropic diffuseness in the momentum distribution . in connection of the deformation with superconductivity it has recently been reported @xcite that in the superconducting asymmetric nuclear matter the fermi sea may undergo a deformation even in the spin - saturated system due to the difference of the fermi surface between neutrons and protons ; the momentum distributions of neutrons and protons may deform respectively to enlarge the overlapped region in the phase space , which effectively contributes to the @xmath284- pairing . they have shown the possibility of the deformation in a variational way ; the fermi sea of the majority of nucleons deforms in a prolate shape , while the minority in an oblate shape . thus the deformation property of the fermi seas looks very similar to our case . nevertheless , note that our deformation is produced by the relativistic effect . anyway it would be interesting to look further into the common feature . it is to be noted that if the effective coupling constant is strong enough to lower the critical quark mass , spin polarization ( magnetization ) has potential to appear at rather moderate densities such as in the core of neutron stars , even though csc weakly works against it . from the above observations it is suggested that spin polarization does not compete with csc but can coexist with it , unlike in ordinary superconductors of the electron system with the @xmath285-wave and spin - singlet pairing . this reflects the fact that internal degrees of freedom of the quark field , e.g.the color , flavor and dirac indices , have rich structures to satisfy the antisymmetric constraint on the quark - pair field . the possibility of the coexistent phase might also give a clue as for the origin of the superstrong magnetic field observed in magnetars . we roughly estimate the expected magnetic field when magnetars are assumed to be quark stars . the maximum dipole magnetic field at the star surface reads @xmath286 with @xmath287 and @xmath288 being the quark magnetic moment and the quark number density , respectively;e.g . , for @xmath289 and @xmath290 , we find @xmath291 , which is comparable to that observed in magnetars @xcite . in the present paper we have not taken into account chiral symmetry , which is one of the basic concepts in qcd . if chiral symmetry is restored at finite baryon number density , the quark mass becomes drastically smaller as density increases . in order to simulate it we have examined the quark mass dependence on spin polarization . in the future work we would like to consider an effect of the dynamical mass on the axial - vector self - energy . * acknowledgements . * the present research of t.t . and t.m . is partially supported by the reimei research resources of japan atomic energy research institute , and by the japanese grant - in - aid for scientific research fund of the ministry of education , culture , sports , science and technology ( 11640272 , 13640282 ) . in this appendix we rewrite the spinor ( [ spinor ] ) in terms of the free quark one and the remainder characterized by @xmath13 . we employ the free spinor @xmath292 in which the two - component pauli spinors are given as eigenvectors of the spin matrix @xmath93 : @xmath293 where @xmath294 and @xmath295 are the pauli spin matrices . the spinor @xmath296 for the `` spin''-up state is decomposed as follows : @xmath297 where @xmath298 , @xmath299 and @xmath300 \\ -(p_x+ip_y)(\beta_p+m ) \left [ \beta_p({\rm \delta}\epsilon -2u_a -2\epsilon_0 ) + m\delta\epsilon \right ] \\ \delta\epsilon [ ( \epsilon_+ -\beta_p - u_a)p_z^2 + 2\beta_p^2(\beta_p+m ) ] -2\beta_p(\beta_p+m ) \left [ ( \epsilon_- -u_a)(\epsilon_+-u_a ) -\epsilon_0 ^ 2 \right ] \\ p_z(p_x+ip_y)(\beta_p+m ) \delta\epsilon \end{array } \right ) . \nonumber\ ] ] note that the term @xmath301 vanishes in the limit , @xmath94 . thus one can fined that @xmath302 is a mixture of the free spinors even when @xmath303 . a decomposition of the spinor @xmath304 for the `` spin''-down state can also be done in the similar way : @xmath305 where @xmath306 \\ ( p_x+ip_y)(\beta_p - m ) \left [ \beta_p({\rm \delta}\epsilon -2u_a -2\epsilon_0 ) + m\delta\epsilon \right ] \\ \delta\epsilon [ ( \epsilon_- + \beta_p - u_a)p_z^2 -2\beta_p^2(\beta_p - m ) ] + 2\beta_p(\beta_p - m ) \left [ ( \epsilon_- -u_a)(\epsilon_+-u_a ) -\epsilon_0 ^ 2 \right ] \\ -p_z(p_x+ip_y)(\beta_p - m ) \delta\epsilon \end{array } \right ) . \nonumber\ ] ] the operator @xmath119 in eq . ( [ bn ] ) consists of some gamma matrices ; it is a linear combination of @xmath308 , @xmath309 , @xmath310 , @xmath311 and @xmath312 , in the diquark field @xmath313 . the last two matrices give the tensor diquark fields , while these terms have no influence on the gap equation ( [ gap1i ] ) due to axial symmetry of the fermi seas around the @xmath151 axis : the integration of @xmath100 with respect to the azimuthal angle @xmath150 in eq . ( [ gap1i ] ) gives @xmath314 . \label{deco1}\ ] ] thus tensor terms disappear because they are proportional to @xmath315 in @xmath119 . the first term in the right hand side has also no contribution after symmetric integration with respect to @xmath151 . the remainders , @xmath316 , imply the pseudo - scalar ( @xmath317 ) and vector ( @xmath318 ) diquark pairings in terms of the notation in ref . please note that the csc gap ( [ deco1 ] ) results in a linear combination of different angular momentum pairs @xmath319 and @xmath320 because of the lack of rotation symmetry . since the diquark fields @xmath321 contain the off - diagonal matrices which connect the lower component with the upper one of the dirac spinors , these pairings vanish in the non - relativistic limit or in the limit @xmath322 . hence @xmath323 resembles @xmath191- wave pairing as is seen in eq . ( 50 ) , although it has no correspondence in the non - relativistic limit : the gap function for ( [ deco1 ] ) has the nodes ( vanishing at @xmath192 ) due to the factor @xmath156 , which is similar to @xmath324-pairing in the liquid @xmath325he - a phase , but these nodes are entirely attributed to the genuine relativistic effect . this property survives even in the limit , @xmath94 . from eq . ( [ deco1 ] ) we can also obtain the relation appearing in eq . ( [ mn ] ) ; @xmath326 in this appendix we derive the parameterization ( [ gap1 ] ) . the gap equation ( [ para ] ) is expanded as @xmath327 . \end{aligned}\ ] ] introducing @xmath328 through the equation , @xmath329 we obtain the `` gap '' equation for @xmath328 , @xmath330 \label{c1}.\ ] ] then we find the following properties , @xmath331 where @xmath181 ( @xmath180 ) is a constant which characterize the symmetric ( asymmetric ) combinations of the gap functions @xmath332 . thus we can further parameterize @xmath333 as @xmath334 substituting the above formula into eq . ( [ c1 ] ) , one can obtain the coupled equations for @xmath181 and @xmath180 , eqs . ( [ eqf ] ) and ( [ eqr ] ) . we present the fock exchange energy term by the oge interaction by the use of the fierz transformation . the green function with vertices in the right - hand side of eq . ( [ self1 ] ) can be expanded as @xmath335 where @xmath336 are coefficients of a fierz transformation ( [ fierz1 ] ) for the dirac matrices , the identity matrix in the flavor space and the gell - mann matrices in the color space ; @xmath337\\ ( \lambda_c)_{ii ' } ( \lambda_c)_{j'j } & = & \frac{2}{n_c^2 } ( n_c^2 - 1)\delta_{ij } \delta_{j'i ' } -\frac{1}{n_c } ( \lambda_c)_{ij } ( \lambda_c)_{j'i'}.\end{aligned}\ ] ] it is to be noted that there appears no tensor term in eq . ( [ notensor ] ) due to chiral symmetry in qcd . thus , e.g. , the coefficient for the color - singlet axial - vector self - energy reads @xmath338 for @xmath339 and @xmath340 . we also present a fierz transformation for diquark fields . the right hand side of eq . ( [ gap1 ] ) can be expanded , in the similar way for @xmath341 ; @xmath342 where @xmath343 are coefficients of a fierz transformation ( [ fierz2 ] ) and are explicitly given as @xmath344\\ ( \lambda_c)_{i'k } ( \lambda_c)_{j'j}&= & \left ( 1-\frac{1}{n_c } \right ) \left [ \frac{2}{n_c } ( \delta)_{i'j ' } ( \delta)_{jk } + ( \lambda_c^s)_{i'j ' } ( \lambda_c^s)_{jk } \right ] -\left ( 1+\frac{1}{n_c } \right ) ( \lambda_c^a)_{i'j ' } ( \lambda_c^a)_{jk},\end{aligned}\ ] ] where @xmath345 are symmetric ( antisymmetric ) matrices of @xmath346 . the present gap function , @xmath347 which is a linear combination of the gamma matrices , can be obtained by taking projection on @xmath323 . m. iwasaki and t. iwado , phys . lett . * b350 * ( 1995 ) 163 ; r. rapp , t. schfer , e. shuryak and m. velkovsky , phys . * 81 * ( 1998 ) 53 ; m. alford , k.rajagopal and f. wilczek , phys * b422 * ( 1998 ) 247 ; d.t . son , phys . rev . * d59 * ( 1999 ) 094019 ; n. evans , s. hsu and m. schwetz , phys . lett . * b449 * ( 1999 ) 281 ; r. d. pisarski and d. h. rischke , phys . rev . * d61 * ( 2000 ) 074017 ; w.e . brown , j.t . liu and h. ren , phys . rev . * d61 * ( 2000 ) 114012 ; v. a. miransky , i. a. shovkovy and l.c.r . wijewardhana , phys . rev . * d62 * ( 2000 ) 085025 ; for a recent review of color superconductivity , m. alford , ann . nucl . part . 51 ( 2001 ) 131 , and referenses therein . m. alford , k. rajagopal and f. wilczek , nucl . b537 * ( 1999 ) 443 ; j. berges and k. rajagopal , nucl . phys . * b538 * ( 1999 ) 215 ; t. m. schwarz , s. p. klevansky and g. papp , phys . rev . * c60 * ( 1999 ) 055205 ; d.ebert , v.v.khudyakov , v.ch.zhukovsky and k.g.klimenko , phys . rev . * d65 * ( 2002 ) 054024 . b. paczyski , acta . astron . * 41 * ( 1992 ) 145 ; r.c.duncan and c. thompson , astrophys . j. * 392 * ( 1992 ) l19 ; c. thompson and r.c.duncan , mon . not . * 275 * ( 1995 ) 255 . c. kouveliotou et al . , nature * 393 * ( 1998 ) 235 ; k. hurley et al . , astrophys . j. 510 ( 1999 ) l111 .

in this note , we study energy estimates for einstein vacuum equations in order to prove the formation of black holes along evolutions . the novelty of the paper is that , we completely avoid using rotation vector fields to establish the global existence theorem of the solution . more precisely , we use only canonical null directions as commutators to derive energy estimates at the level of one derivatives of null curvature components . we show that , thanks to the special cancelations coming from the null structure of non - linear interactions , desirable estimates on curvatures can be derived under the short pulse ansatz due to klainerman and rodnianski ( which is originally discovered by christodoulou ) .

1105.2630

penrose singularity theorem states that if in addition to the dominant energy condition , the space - time has a trapped surface , then the space - time contains _ singularities_. the _ weak cosmic censorship _ conjecture asserts that under reasonable physical assumptions , singularities should be hidden from an observer at infinity by the event horizon of a black hole . thus , by combining these two claims , to predict the existence of black holes , it suffices to exhibit one trapped surface in a space - time , . in other words , although many supplementary conditions are required , we regard the existence of a trapped surface as the presence of a black hole . a major challenge in general relativity is to understand how trapped surfaces actually form due to the focusing of gravitational waves . in a recent breakthrough @xcite , demetrios christodoulou gave an answer to this long standing problem . he discovered a mechanism which is responsible for the dynamical formation of trapped surfaces in vacuum space - times . in the monograph @xcite , in addition to the minkowskian flat data on a incoming null hypersurface , christodoulou identified an open set of initial data ( this is the _ short pulse _ ansatz ) on a outgoing null hypersurfaces . based on the techniques developed by himself and klainerman in the proof of the global stability of the minkowski space - times @xcite , he managed to understand the whole picture of how various geometric quantities interact along the evolution . once the estimates on curvatures are established in a large region of the space - time , the actual formation of trapped surfaces is easy to demonstrate . christodoulou also proved a version of the same result for the short pulse data prescribed on past null infinity . this miraculous work provides the first global _ large data _ result in general relativity ( without symmetry assumptions ) and opens the gate for many new developments on dynamical problems related to black holes . in @xcite , klainerman and rodnianski extended aforementioned result of christodoulou . they significantly simplified the proof of christodoulou ( from about six hundred pages to one hundred and twenty ) . they also enlarged the admissible set of initial conditions and show that the corresponding propagation estimates of connection coefficients and curvatures are much easier to derive . the relaxation of the propagation estimates are just enough to guarantee that a trapped surface still forms . based on the trace estimates developed in a sequence of work of the authors towards the critical local well - posedness for einstein vacuum equations , they reduced the number of derivatives needed of christodoulou in the argument from two derivatives of the curvature ( in christodoulou s proof ) to just one . more importantly , klainerman and rodnianski introduced a parabolic scaling in @xcite which is incorporated into lebesgue norms and sobolev norms . these new techniques allow them to capture the hidden _ smallness _ of the nonlinear interactions among different small or large components of various geometric objects . the result of klainerman and rodnianski can be easily localized with respect to angular sectors , has the potential for further developments , see @xcite . we remark that klainerman and rodnianski only concentrated on the problem on a finite region . the question from past null infinity can be solved in a similar manner as in @xcite once one understand the picture on a finite region . the problem from past null infinity has been studied in a recent work by reiterer and trubowitz , @xcite . one common feature of the proofs in @xcite and @xcite is that , in order to derive energy estimates on one or higher derivatives of curvature components , they all constructed three _ angular momentum vector fields _ @xmath0 , @xmath1 and @xmath2 which essentially captured almost rotational symmetry of the space - time . as far as the author aware , the main reason of using @xmath3 s is that the energy estimates on @xmath4 behave well because one can take advantage of cancelations from the pseudo - symmetry of @xmath3 s . here , the @xmath5 is the _ modified lie derivative _ defined in @xcite and @xmath6 is the curvature of the space - time . we can observe this advantage in the proof of stability of minknowski space - time @xcite where @xmath4 yields better decay estimates . there is one obvious defect of the modified lie derivative . as usual lie derivatives , the @xmath5 is _ not _ tensorial in @xmath3 . in fact , it evolves one derivative of @xmath3 . in other words , if we use modified lie derivatives , we may lose immediately one derivative . we have two remedies to this loss of derivatives : in @xcite , one relies on higher order derivative estimates ; in @xcite , one makes use of more subtle trace estimates . the above discussion can be summarized as follows : roughly speaking , a good estimate on lie derivatives relies on the almost symmetries @xmath3 s , but lie derivatives causes a loss of derivative as we just explained . hence , to avoid this loss , we shall give up the use of the pseudo - symmetries . in this paper , we propose an approach to derive energy estimates on curvatures without constructing rotational vector fields at all . in particular , in stead of using modified lie derivatives , we work with covariant derivatives to save one derivative . of course , we have to pay a a price of controlling much more error terms . nevertheless , this significantly simplifies the proofs compared to either @xcite or @xcite . in particular , compared to @xcite , we use only one derivative in curvature ; compared to @xcite , instead of using trace inequality , all the estimates are derived from the classical sobolev inequalities . we also want to mention that in the thesis of l. bieri , see @xcite , based on a more general asymptotic assumptions , she gave a simplified proof of the stability of minkowski space - time . she managed to derive decay from the time vector field and the conformal scaling of the space - time which allowed her to circumvent rotational vector fields in that situation . but the current situation is different from @xcite : we do not use lie derivatives at all and the lower regularity forces us to explore more structures from the einstein equations . the main observation arises from the second bianchi identities . roughly speaking , they explicitly show how one expresses angular or rotational derivatives of curvature components @xmath7 in terms of some null derivatives of curvature components plus lower order non - linear terms . namely , they can be written schematically as @xmath8 or @xmath9 where we use lower order terms @xmath10 to collect all nonlinear interactions and @xmath11 , @xmath12 are two standard null directions under the framework of double null foliations . thus , up to lower order corrections , to obtain the estimates on @xmath7 rotational derivatives on curvatures , it suffices to control @xmath13 or @xmath14 null derivatives of curvatures . thus , we identify our main targets to be @xmath15 and @xmath16 and we shall derive energy estimates for them . according the above idea , we use the modified short pulse ansatz proposed klainerman and rodnianski in @xcite . we remark that this ansatz allows more large components than the original short pulse data discovered by christodoulou in @xcite . we shall derive the energy estimates based on bel - robinson tensors associated to @xmath17 and @xmath18 . to deal with error terms , namely terms @xmath19 , @xmath20 and @xmath21 in the proof in following sections , we have to take account of the special structure of those terms . in reality , some generic terms in error term may cause a loss of @xmath22 which prevent us from closing the bootstrap argument . to avoid this loss , there are typically three techniques to use : 1 . we bound a product of two term in @xmath23 by two @xmath24 estimates on each term instead of one @xmath25 estimate on one of them and one @xmath23 estimate on the other . in most of the situation , this trick saves a @xmath26 . 2 . when we integrate a product of terms on some null hypersurface or on a domain in the space - time , we use integration by parts to move a bad derivative , typically @xmath27 or @xmath28 , from one term ( for whom this bad derivative may cause a loss of @xmath22 ) to another ( for whom this bad derivative is not bad at all , namely , there is no loss in @xmath29 ) . combined with bianchi identities , this procedure may save a @xmath30 . 3 . for some generic term in error estimates mentioned above , though it appears that it causes a loss of @xmath29 ( which can not be retrieved by using the trick 1 ) and 2 ) ) , we can use in fact either signature considerations or a precise computations to show that this term does not show up at all in the error estimates . this manifests the special cancelations in the error terms . the whole proof is to combine these three tricks . the paper is organized as follows : in next section , we recall basic definitions and estimates from @xcite and we state the main theorem ; in following sections , we derive energy estimates for derivatives of curvature components in the following order : @xmath31 , @xmath32 and then @xmath33 for @xmath34 and @xmath35 for @xmath36 . we briefly recall the double null foliation formalism , see @xcite for more precise definitions . we use @xmath37 to denote the underlying space - time and use @xmath38 to denote the background metric . we assume that @xmath39 is spanned by a double null foliation generated by two optical functions @xmath40 and @xmath41 and they increase towards the future , @xmath42 and @xmath43 . we use @xmath44 / @xmath45 to denote the outgoing / incoming null hypersurfaces generated by the level surfaces of @xmath40 / @xmath41 . we use @xmath46 to denote the space - like two surface @xmath47 . we denote by @xmath48 the region of @xmath44 defined by @xmath49 ; similarly , we can define @xmath50 . the shaded region on the right represents the domain @xmath51 with @xmath42 . the function @xmath40 is in fact defined from @xmath52 to @xmath29 . when @xmath53 , this part of @xmath54 is assumed to be a flat light cone in minkowski space with vertex located at @xmath55 . we use @xmath56 to measure the maximal radius of the flat part of @xmath54 . in @xcite , the trapped surface forms at @xmath57 and @xmath58 . let @xmath59 be the null geodesic generators of the double null foliation and we define the lapse function @xmath60 by @xmath61 . the normalized null pair @xmath62 is defined by @xmath63 . on sphere @xmath46 we choose an arbitrary orthonormal frame @xmath64 . we call @xmath65 a _ null frame_. to denote an index from @xmath66 to @xmath67 and latin letters @xmath68 to denote an index from @xmath66 to @xmath69 . we also @xmath70 to denote the null direction either @xmath11 or @xmath12 . repeated indices are always understood as taking sums ] we use @xmath71 to denote levi - civita connection defined by the metric @xmath38 and we define the _ connection coefficients _ as follows , @xmath72 where @xmath73 . on @xmath74 , @xmath75 is the induced connection ; @xmath27 and @xmath28 are the projections to @xmath74 of the covariant derivatives @xmath76 and @xmath77 . given a weyl field @xmath78 , we introduce its null decomposition with respect to the given null frame , @xmath79 where @xmath80 is the space - time hodge dual of @xmath78 . when @xmath78 is the weyl curvature tensor , we use @xmath81 to denote its null components . we recall the null structure equations for the einstein vacuum space - times ( see @xcite ) . the originally einstein field equations are @xmath82 where @xmath83 is the ricci curvature of the underlying space - times . we express this tensorial equations by using the null frame , by definition , this yields the null structure equations . we only list the transport type null structure equations which are relevant to the current work . @xmath84 @xmath85 @xmath86 @xmath87 @xmath88 @xmath89 @xmath90 @xmath91 @xmath92 we also express the bianchi equations relative to the null frame to derive null bianchi equations . by @xmath93 . ] @xmath94 @xmath95 @xmath96 @xmath97 @xmath98 @xmath99 @xmath100 @xmath101 @xmath102 @xmath103 we review our scheme for energy estimates on weyl fields , see @xcite for the original resource . assume a weyl field @xmath104 and its hodge dual @xmath105 solve following divergence equations with source terms @xmath106 @xmath107 and @xmath108 are called _ weyl currents_. for vacuum , the curvature tensor @xmath6 is a weyl field with zero currents @xmath109 the bel - robinson tensor @xmath110_{\alpha\beta\gamma\delta}$ ] associated to @xmath104 is defined as follows for @xmath110 $ ] , @xmath111 for @xmath112 , @xmath113 for @xmath114 , @xmath115 for @xmath116 , ... , if there is no confusion in the context . ] @xmath117_{\alpha\beta\gamma\delta}=w_{\alpha\mu\gamma\nu}w_{\beta}{}^{\mu}{}_{\delta}{}^{\nu}+\wstar_{\alpha\mu\gamma\nu}{}\wstar_{\beta}{}^{\mu}{}_{\delta}{}^{\nu}.\ ] ] it is fully symmetric and traceless in all pair of indices . moreover , it satisfies the dominant energy condition which allows one to recover estimates for weyl field @xmath78 . in pratical terms , this condition can be expressed by formulas , @xmath118 we also list other null components of @xmath119 for future use , @xmath120 in view of , @xmath119 enjoys the following divergence equations @xmath121 given vector fields @xmath122 , @xmath123 and @xmath124 , we define the current @xmath125(x , y , z)_{\alpha}$ ] associated to @xmath122 , @xmath123 , @xmath124 and @xmath78 by @xmath126(x , y , z)_{\alpha } = q_{\alpha\beta\gamma\delta}x^\beta y^\gamma z^\delta.\ ] ] thus , @xmath127 where @xmath128 is the deformation tensor of @xmath122 defined by @xmath129 and @xmath130 we integrate on @xmath131 to derive the fundamental energy identity and @xmath12 are corresponding normals of the null hypersurfaces @xmath44 and @xmath45 . ] @xmath132 we list non - zero components of deformation tensors of @xmath11 and @xmath12 as well as the non - zero component of @xmath133 and @xmath133 for future use and @xmath134 in order to avoid @xmath135 and @xmath136 which do not have certain @xmath137 estimates , see @xcite for details . ] : @xmath138 we consider one derivative of curvature @xmath6 in a null direction @xmath70 . one easy but important observations is that @xmath139 is still a weyl field . we can commute @xmath140 with to derive commutes with hodge @xmath141 operator . ] @xmath142 where @xmath143 and @xmath144 we briefly recall the notions of _ signature _ and _ scale _ introduced by klainerman and rodnianski in @xcite . let @xmath145 be either a null component of curvature or a connection coefficient , we use @xmath146 , @xmath147 and @xmath148 to denote the number of times @xmath149 , respectively @xmath150 and @xmath134 appearing in the definition of @xmath145 . the _ signature _ of @xmath145 , @xmath151 , and the _ scale _ of @xmath145 , @xmath152 , are defined as @xmath153 we list the signatures and scales for all connection coefficients and curvature components , [ cols="^,^,^,^,^,^,^,^,^",options="header " , ] we impose following rules on signatures , @xmath154 we define the _ scale invariant norms _ for @xmath145 . along null hypersurfaces @xmath155 or @xmath156 , @xmath157 on a two dimensional surface @xmath46 , @xmath158 those norms are obviously related by formulas , @xmath159 those scale invariant norms come up naturally with a small parameter @xmath29 . roughly speaking , it captures the _ smallness _ of the non - linear interaction . we have hlder s inequality in scale invariant form , @xmath160 similar estimates hold along null hypersurfaces . [ gain or not gain ] the rule of thumb for treating the nonlinear terms is , whenever one has a product of two terms , gains a @xmath30 . we do have cases that does not gain any power in @xmath29 . in fact , if @xmath161 is a bounded ( in usual sense ) scalar function ( say , bounded by a universal constant ) , the best we can hope is @xmath162 . in particular , in this paper , for @xmath161 can be @xmath163 where @xmath164 , we have to pay special attentions to the appearance of @xmath165 , see @xcite for more detailed descriptions . we introduce a family of scale invariant norms for connection coefficients where @xmath166 or @xmath167 , ] @xmath168 as well as for curvature components , @xmath169 finally , we introduce total norms . we define @xmath170 and @xmath171 ; similarly , we can define supremum norms for other scale invariant norms . the total norms are defined as follows , @xmath172 we use @xmath173 and @xmath174to denote total norms on the initial hypersurface @xmath54 . in the above definitions , all norms are scale invariant except for @xmath175 , @xmath176 , @xmath177 , @xmath178 and @xmath179 . those terms are understood to cause a loss of @xmath22 . notice also @xmath113 on incoming hypersurfaces @xmath45 is scale invariant . by abuse of language , we call those terms _ anomalies _ or _ anomalous _ if they cause a loss of @xmath180 in @xmath181 norm . notice also all the connection coefficients are not anomalous in @xmath182 norms . inspired by this , we use @xmath183 ( g ` for good ) to denote some ( good ) connection coefficient in @xmath184 and @xmath185 to denote an arbitrary connection coefficient . we use @xmath186 to denote some curvature component in @xmath187 and @xmath188 to denote an arbitrary connection coefficient . we shall also put a g ' as a lower index to other quantities in order to indicate that this quantity is not anomalous . for example , we can write @xmath189 because @xmath190 . for einstein equations with characteristic data prescribed on @xmath191 ( where the data is trivial ) and @xmath54 , we can freely specify the conformal geometry on @xmath54 . in other words , we can specify @xmath192 freely along @xmath54 to fix an initial data for the evolutionary problem . we remark that , in contrast to the case where the initial data is give on a space - like hypersurface , there is no constraints and the data can be freely specified . we require the initial data @xmath193 subject to the following norm is finite , @xmath194,\end{aligned}\ ] ] i.e. our main assumption on the initial data is the following short pulse ansatz , @xmath195 notice that the size of @xmath192 can be as large as @xmath22 so the problem is far away from small data regime . the ansatz was introduced by klainerman and rodnianski in @xcite . this initial data set is larger than those of christodoulou s original short pulses . in fact , allows more components than those of christodoulou s to be as large as @xmath22 . in this ansatz , we allow @xmath75 behaves as @xmath22 ; in christodoulou s ansatz , @xmath75 behaves as @xmath66 . under the ansatz , with the help of null structure equations , we can easily derive the following estimates on initial outgoing surfaces @xmath54 , under the ansatz , along the initial outgoing hypersurface @xmath54 , if @xmath29 is sufficiently small , there is a constant @xmath196 depending only on @xmath197 , such that @xmath198 thanks to this proposition , we shall replace by . and we omit the proof and refer the reader to @xcite or chapter 2 of @xcite . the next proposition claims that we can control connection coefficients provided that we have bound on curvatures . this is _ theorem a _ in @xcite and we omit the proof . [ theorema ] assume that @xmath199 , @xmath200 and @xmath201 are finite in @xmath202 . then there is a constant @xmath203 depending only on @xmath204 and @xmath201 such that to denote a constant depending only on @xmath204 and @xmath201 . ] @xmath205 we now state our main theorem . this is the following propagation estimates , which asserts that if holds on initial hypersurface @xmath54 , thus on the whole @xmath202 we can bound the curvature norms @xmath200 and @xmath201 by a function depending only on initial data . [ theoremc ] assume the short pulse ansatz hence , thus if @xmath29 is sufficiently small , we have @xmath206 where @xmath196 is a constant depending only on the size of the initial data . the main consequence of our estimates is the following global existence theorem , [ theoremd ] given initial data @xmath192 satisfying , if @xmath29 is sufficiently small , we can construct an unique solution of the einstein vacuum equations @xmath207 on the whole region @xmath202 . once we have the main estimates , the proof of the theorem is a typical continuity argument . we refer the reader to chapter 16 of @xcite for the detailed proof . we shall not repeat this argument . we want to remark that the formation of trapped surfaces can not be derived from the initial data ansatz . we have to add two more modifications to make sure that @xmath54 is free of trapped surface and @xmath208 is a trapped surface . more precisely , in addition to , we also assume that @xmath209 for sufficiently small @xmath210 such that @xmath211 and we also assume @xmath192 satisfies , @xmath212 where @xmath213 is a universal constant and @xmath56 measure the maximal radius of the flat part of @xmath54 . these condition guarantees the dynamical formation of trapped surfaces , we refer the readers to @xcite for details . we use a bootstrap argument to prove the * main estimates*. to be more precisely , we assume that @xmath200 and @xmath201 are finite . this assumption holds initially near @xmath54 . our main task is to carry out the following estimates under this assumption , @xmath214 thus , if @xmath29 is sufficiently small , we have @xmath215 this yields the * main estimates * . the derivation for consists of two steps . the first step is to derive energy estimates for curvature components ; the second step is to derive energy estimates for one derivatives of curvature components . the first step is done in @xcite : [ energy estimates for curvature ] if @xmath29 is sufficiently small , we have @xmath216 as we explained in the introduction , our main target is the second step , i.e. to derive energy estimates for one derivatives of curvature components in the form of without constructing rotational vector fields @xmath217 s . this is done in the rest of the paper . in this section , we collect some estimates either already derived in @xcite or relatively easy to prove directly . we have @xmath218 estimates for curvature on each leaf @xmath219 . [ l2 estimates for curvature ] if @xmath29 is sufficiently small , we have @xmath220 we also have @xmath24 estimates on curvature components . we recall again that the constant @xmath203 depends on @xmath200 and @xmath201 which has information on the one horizontal derivatives of the curvature . [ l4 estimates for curvature ] if @xmath29 is sufficiently small , we have @xmath221 this proposition is proved by virtue of null bianchi equations - and following scale invariant sobolev trace type inequalities stated as proposition 4.15 in @xcite . [ sobolev trace ] given an arbitrary tensor field @xmath145 , we have @xmath222 @xmath223 integrating along null hypersurfaces , we derive [ l4 estimates for curvature on null hypersurfaces ] if @xmath29 is sufficiently small , we have @xmath224 we also collect some more precise estimates on @xmath192 , @xmath225 and @xmath226 . they are essentially developed in section 5 of @xcite provided we bound @xmath111 by proposition [ energy estimates for curvature ] . [ pricise on chi chib ] if @xmath29 is sufficiently small , we have @xmath227 the aim is to compare null components of derivatives of curvature and derivatives of null components of curvature . the first lemma compares @xmath228 with @xmath229 where @xmath230 or @xmath12 . [ comparison lemma dn commute with psi ] for @xmath231 or @xmath12 , we have @xmath232 by direct computations , in schematic form , we have the following commutator formula , denotes a sum of products of this form . we recall once again that @xmath185 denote a generic connection coefficient and @xmath188 denote a generic curvature component . the subindex @xmath38 means this component is not anomalous . ] @xmath233 to estimate @xmath234 on some null hypersurface ( say @xmath235 without loss of generality ) , we use @xmath24 estimates on both factors and notice that @xmath188 may cause a loss of @xmath26 , thus we have @xmath236 since @xmath237 and @xmath238 . this completes the proof . the next lemma , by investigating null bianchi equations , compares @xmath228 with @xmath239 . [ lemma estimates on nabla_n psi ] we have following estimates , + ( 1 ) . for @xmath240 , we have @xmath241 ( 2 ) . for @xmath242 , except @xmath243 on @xmath244 and @xmath245 on @xmath44 , we have @xmath246 we can not estimate on @xmath247 or @xmath248 at the moment . the reason is that we do not have any knowledge on either @xmath249 or @xmath250 . but they are controlled in next section when we derive estimates on @xmath31 and @xmath32 . we only sketch one estimate ( which may be regarded as the most difficult one ) . the rest is similar so we skip . for @xmath240 , according to , we have @xmath251 we can bound @xmath252 by @xmath253 and bound last two terms by @xmath254 thanks to hlder s inequality . for @xmath255 , in view of proposition [ energy estimates for curvature ] , it is bounded by @xmath256 . combining all the above , we have the desired estimates for @xmath240 . this completes the proof . combining previous two lemmas , we derive the comparison estimates for null components of @xmath257 and @xmath258 . [ comparison proposition ] ( 1 ) . for @xmath259 , we have @xmath260 ( 2 ) . for @xmath261 , except @xmath262 on @xmath244 and @xmath263 on @xmath44 , we have @xmath264 we now consider horizontal derivatives and compare @xmath265 with @xmath266 . [ comparison proposition horizontal version ] if @xmath267 , we have @xmath268 the proof is quite similar to lemma [ comparison lemma dn commute with psi ] . notice that , if @xmath267 , then @xmath269 the product term is bounded by @xmath270 as before and last term is bounded by @xmath271 by definition . thus we have the first inequality . the second follows immediately . we turn to the anomalous term @xmath272 . first of all , notice that @xmath273 thus , the first term @xmath274 causes anomalous behavior . hence we have @xmath275 we see that @xmath276 , @xmath277 , @xmath259 and @xmath278 are _ anomalies _ since their estimates cause a loss of @xmath22 . we shall distinguish @xmath259 and @xmath278 from @xmath276 and @xmath277 . heuristically , @xmath259 and @xmath278 are anomalous in zero order ; @xmath276 and @xmath277 are anomalous in first order . we observe that @xmath259 and @xmath278 are anomalous because they can be written schematically as @xmath279 this amounts to say that their anomalies come from the zero order term @xmath274 . we shall call @xmath259 and @xmath278 _ mild anomalies_. by the above comparison lemmas , @xmath240 is also a mild anomaly . in what follows , we use this observation to replace @xmath259 and @xmath278 by the worst term @xmath111 . we have two advantages of performing this substitution : the first is that we have already controlled @xmath111 in proposition [ energy estimates for curvature ] ; the second is that we can use @xmath24 estimates on @xmath111 ( but not on either @xmath259 or @xmath278 ) . the other terms are much easier to control in all the contexts since they are not anomalous at all . this becomes much clear and important when we derive estimates in following sections . we take @xmath280 in and @xmath281 in to derive @xmath282_{\mu444 } + { } ^{(l)}\!\pi_{\mu\nu}q^{\mu\nu}{}_{44}.\ ] ] in view of lemma [ comparison lemma dn commute with psi ] , by ignoring the term @xmath262 on the left hand side and passing to scale invariant norms , we derive @xmath283_{\mu444 } + { } ^{(l)}\!\pi_{\mu\nu}q^{\mu\nu}{}_{44}\\ & \lesssim \delta^{-1}\izero^2 + ( i + j ) + k.\end{aligned}\ ] ] we split the bulk integral into three terms @xmath19 , @xmath20 and @xmath21 in such a way that the integrands of @xmath284 are exactly the terms appearing in @xmath285_{\mu444}$ ] and the integrands of @xmath21 are exactly the terms appearing in @xmath286 . [ remark forget delta ] in the above terms @xmath19 , @xmath20 and @xmath21 , there is a @xmath29 to certain power in front of the integral which reflects the fact that the inequality is scale invariant . in what follows , this phenomenon appears constantly . to simplify expressions , we shall always ignore this @xmath29 to certain power since they can be easily restored by investigating the signatures . according to and , the integrand in @xmath287 has the following schematic form , @xmath288 with @xmath289 . here we use the upper index @xmath290 s to indicate the signature . we require the integrand of @xmath19 is of the form @xmath291 and the integrand of @xmath20 is of the form @xmath292 . we remark that @xmath293 . , @xmath20 and @xmath21 in the same way . and we shall not explain when one encounters this situation later . ] we now examine the possible anomalies in the integrand of @xmath19 . for @xmath294 on @xmath44 , according to lemma [ comparison proposition ] , the only possible anomaly is @xmath276 which has @xmath295 ; for @xmath296 and @xmath297 , the only possible anomaly is @xmath111 with signature @xmath69 . thus , in view of the signature consideration , at least one of the terms is not anomalous . we say that , in this situation , we do not have _ triple anomalies_. in view of corollary [ l4 estimates for curvature on null hypersurfaces ] , we estimate @xmath19 as follows @xmath298 for @xmath21 , according to signature considerations , we can also avoid triple anomalies . thus , @xmath299 we dominate @xmath183 and @xmath185 in @xmath300 norm to derive @xmath301 for @xmath20 , it possesses most of the features of @xmath21 . we still need to investigate the structure of the integrand . in fact , the integrand is of the following form @xmath302 thus , since @xmath293 , we can bound @xmath20 by @xmath303 combining the estimates for @xmath19 , @xmath20 , @xmath21 and young s inequality , we derive @xmath304 in conclusion , we have the desired anomalous estimates on @xmath31 @xmath305 the above analysis also yields the following estimates which is absent in _ proposition [ comparison proposition ] _ , @xmath306 we take @xmath308 in and @xmath309 in to derive @xmath310_{\mu333 } + { } ^{(\lb)}\!\pi_{\mu\nu}q^{\mu\nu}{}_{33}.\ ] ] in view of lemma [ comparison lemma dn commute with psi ] , by ignoring the term @xmath262 on the left hand side and passing to scale invariant norms , we derive @xmath311_{\mu333 } + { } ^{(\lb)}\!\pi_{\mu\nu}q^{\mu\nu}{}_{33}\\ & \lesssim \delta^{-1}\izero^2 + i + j+ k.\end{aligned}\ ] ] we split the bulk integral into @xmath19 , @xmath20 and @xmath21 in the obvious way as before . the integrands in @xmath19 have the following schematic form , @xmath312 with @xmath313 . according to signature considerations , none of @xmath314 and @xmath315 can be anomalous since the signature for anomalous curvature component has signature @xmath69 . thus , we estimate @xmath19 as follows @xmath316 we now deal with @xmath21 . again , by counting signatures , the anomalous terms @xmath276 , @xmath259 and @xmath278 do not appear in the integrand since their signatures are at least @xmath69 . thus the only possible anomaly is @xmath277 . hence , if their is a double anomaly in curvature components , they must be two @xmath277 . this forces the connection coefficient to have signature @xmath66 . in particular , it can not be @xmath317 . thus , the integrands have the following schematic form is a constant with worse estimates in @xmath182 than usual connection coefficients . ] @xmath318 hence , bounding @xmath185 in @xmath182 yields , @xmath319 the analysis for @xmath20 is exactly the same as @xmath21 , hence , we have @xmath320 combining those estimates for @xmath19 , @xmath20 and @xmath21 , as we have done for @xmath31 in last subsection , we derive @xmath321 as a byproduct , we also have the following estimates which is absent in _ proposition [ comparison proposition ] _ , @xmath322 in this section , we derive energy estimates for @xmath323 . we first take @xmath231 and all possible @xmath324 , except for the case @xmath325 , in to derive , @xmath326 and @xmath327 to denote either @xmath11 or @xmath12 coming from @xmath122 , @xmath123 and @xmath124 . ] @xmath328_{\mu xyz } + { } ^{(n_0)}\!\pi_{\mu\nu}q[d_4 r]^{\mu\nu}{}_{n_1 n_2}.\end{aligned}\ ] ] we then multiply both sides by @xmath329 and sum all those inequalities , in view of remark [ remark forget delta ] , we have @xmath330 = \izero^2 + i + j + k\ ] ] where @xmath331_{\mu xyz } , \quad k=\doubleint_{\d(u,\ub)}{}^{(n_0)}\!\pi_{\mu\nu}q[d_4 r]^{\mu\nu}{}_{n_1 n_2}.\ ] ] the sum in is over on all possible @xmath332 s except @xmath276 and @xmath333 . the definition for @xmath19 , @xmath20 is in an obvious way as in the previous section . the integrand of @xmath19 can be written schematically as @xmath334 with total signature @xmath335 . we shall further split @xmath19 into @xmath336 where @xmath337 denotes the collection of the terms whose integrand has exactly @xmath338 anomalies . we observe that @xmath339 according to signature considerations . otherwise , for a triple anomaly , thus @xmath340 , @xmath341 and @xmath342 which contradicts the fact that @xmath343 . we also notice that the only possibility for @xmath294 to be anomalous is @xmath344 . the control of @xmath345 is much easier than other terms . we further split @xmath347 into two terms @xmath348 where the corresponding integrands for @xmath349 and @xmath350 are @xmath351 respectively . @xmath349 can be bounded as follows , @xmath352 @xmath350 requires an extra integration by parts . before going into details , we first show how to replace @xmath276 by @xmath31 . according to the proof of lemma [ comparison lemma dn commute with psi ] , @xmath353 . thus , @xmath354 for @xmath355 , since the nonlinearity gains one more term , we gain an extra @xmath30 thanks to scale invariant hlder s inequality . thus we can bound it as follows , @xmath356 [ remark switch ] the above analysis allows us to freely switch between @xmath229 and @xmath228 . the difference contributes at most @xmath254 to energy estimates . in the rest of the paper , we shall ignore this @xmath254 and we shall _ not _ distinguish @xmath229 from @xmath228 . we move on to @xmath350 . @xmath357 for the bulk integral , in view of lemma [ lemma estimates on nabla_n psi ] , we bound it as follows @xmath358 for boundary integrals , we bound them as follows @xmath359 thus , we conclude @xmath360 . combining the estimates for @xmath349 and @xmath350 , we derive @xmath361 we split @xmath362 into two terms @xmath363 where the corresponding integrands for @xmath364 and @xmath365 are @xmath366 respectively . we first show that @xmath367 . in view of , , and , the integrand of @xmath364 has the following expression @xmath368 we write only one term in . but this term is generic and the rest can be treated in the same way . if we look for @xmath369 , none of @xmath370 or @xmath371 can be 3 or 4 ; this forces @xmath372 . thus reduces to @xmath373 if @xmath374 or @xmath375 , it vanishes ; if @xmath376 , the the total signature is @xmath377 , it is impossible . hence @xmath364 must vanish . for @xmath365 , since total signature is at most @xmath378 , thus @xmath379 . thus the integrand of @xmath365 must be and @xmath380 are traceless , @xmath381 . ] @xmath382 in view of remark [ remark switch ] , we have @xmath383 for the bulk integral , we use bianchi equation to replace @xmath384 . one can argue as in remark [ remark switch ] , only the top order term @xmath385 in is relevant . the quadratic terms in contribute at most a @xmath254 . by performing an extra integration by parts , we can move @xmath75 to @xmath111 to eliminate the anomaly of @xmath111 . more precisely , we proceed as follows @xmath386 to control the boundary integral , we need to improve proposition [ l4 estimates for curvature ] , [ l4 alpha precise ] if @xmath29 is sufficiently small , then @xmath387 in view of proposition [ sobolev trace ] , we have @xmath388 combined with proposition [ energy estimates for curvature ] and , thus @xmath389 we then bound @xmath390 by @xmath200 to complete the proof . thus , combined with proposition [ energy estimates for curvature ] , the boundary integral is controlled as follows @xmath391 thus , we have the following estimates for @xmath365 and @xmath392 , @xmath393 combining the estimates for @xmath394 and @xmath392 , we derive @xmath395 the integrand in @xmath21 can be written schematically as @xmath396 with total signature @xmath335 . the appearance of @xmath317 is due to @xmath397 . we split @xmath21 into @xmath398 where @xmath399 denotes the collection of terms with exactly @xmath338 @xmath276 s . we observe that @xmath400 according to the signature considerations . we first split @xmath401 into two terms @xmath402 where the corresponding integrands for @xmath403 and @xmath404 are @xmath405 respectively . @xmath403 can be bounded as follows , @xmath406 @xmath404 can be bounded in a similar way , although we lose a @xmath22 due to @xmath317 , @xmath407 once we plug @xmath404 into , in view of gronwall s inequality , it can be absorbed by the left hand side of . hence we can ignore @xmath404 at this stage and simply write @xmath408 first of all , we remark that @xmath317 does not appear as connection coefficients in the integrand of @xmath409 . otherwise , in view of , the appearance of @xmath317 is through @xmath410 since @xmath276 appears , this forces @xmath411 . thus , the integrand can be written as @xmath412 since @xmath111 and @xmath413 are traceless , the @xmath317 will be canceled . this argument also shows @xmath414 is absent in the integrand , otherwise they must be either @xmath415 or @xmath416 with total signature @xmath417 . this is impossible . we further split @xmath409 into two terms @xmath418 where the corresponding integrands for @xmath419 and @xmath420 are @xmath421 respectively . for @xmath419 , as we just observed , @xmath422 or @xmath423 . we only treat the first possibility . the second one can be estimated exactly in the same way . in view of the remark [ remark switch ] and , we have @xmath424 for @xmath425 , we reduce the number of anomalies by an extra integration by parts . @xmath426 to estimate @xmath427 , we use null structure equation , @xmath428 to estimate @xmath429 , we compute the commutator @xmath430 \beta = -\chi \cdot \nabla \beta + { } ^*\!\beta \cdot \beta + \frac{1}{2}(\eta + \etab)\nabla_4 \beta + \etab \cdot \beta \cdot \chi,\ ] ] we claim , @xmath431 \beta| \lesssim c \delta^{\frac{1}{4}}.\ ] ] in fact , the worst term should be @xmath432 . we can use @xmath182 norm on @xmath225 , @xmath24 norm on two @xmath113 s and @xmath23 on @xmath111 to dominate it . thus , up to an error of size @xmath270 @xmath433 for the first integral , @xmath434 for the second integral , @xmath435 hence , @xmath436 to estimate @xmath437 , we proceed as follows , @xmath438 combining the estimates for @xmath427 , @xmath429 and @xmath437 , we derive @xmath439 we now turn to @xmath440 . @xmath441 to estimate @xmath442 , we use to derive @xmath443 to estimate @xmath444 , we proceed as follows @xmath445 thus , we have @xmath446 . hence , @xmath447 it remains to control @xmath420 whose integrand is @xmath448 . the key estimates for @xmath449 except for the case @xmath450 . ] is to observe @xmath451 . otherwise , since total signature is at most @xmath378 , it forces the integrand to be @xmath452 . once again , we investigate the origin of this term @xmath453 by direction computation , we see that @xmath454 never appears in @xmath455 , hence we arrive at a contradiction . based on the above observation , since we have @xmath456 estimates on @xmath457 , we can proceed exactly as we did for @xmath419 to derive @xmath458 in this case , since @xmath183 is not anomalous , the estimates are much easier to derive . we skip details . combining the estimates for @xmath419 and @xmath420 , we derive @xmath459 combining the estimates for @xmath401 and @xmath409 , we derive @xmath460 the integrand of @xmath20 can be written schematically as @xmath461 with total signature @xmath335 . we split @xmath20 into @xmath462 where @xmath463 denotes the collection of terms with exactly @xmath338 anomalous curvature components . we make the following observation : the only anomaly for @xmath464 is @xmath276 ; things are different for @xmath465 , it has multiple choices : @xmath276 , @xmath277 , @xmath259 and @xmath278 . we also recall section [ mild anomaly ] which asserts that last two anomalies are mild in the sense that they all come from zero order term @xmath111 . this will be extremely crucial in our proof . the integrand of @xmath466 , by definition , has the following schematic expressions @xmath467 thus , we bound @xmath466 as follows , @xmath468 we split @xmath469 into @xmath470 . for @xmath471 , @xmath472 ; for @xmath473 , @xmath474 is not anomalous . the integrand of @xmath471 has the following form ( recall that we ignore the hodge dual part ) @xmath475 since @xmath472 , it forces @xmath476 , @xmath477 , @xmath478 and @xmath479 . thus , the integrand is reduced to @xmath480 since @xmath481 , we see @xmath482 . if @xmath483 , we use bianchi identities , up to lower order terms , to convert @xmath484 into the form @xmath485 . hence , the only possible forms of the integrand are either @xmath486 or @xmath487 ( @xmath451 ) . we then split @xmath471 as @xmath488 notice that @xmath489 has exactly the same form as @xmath420 term in the previous section , thus @xmath490 the estimates for @xmath491 is quite similar to @xmath419 . first of all , according to signature consideration , @xmath492 , thus @xmath493 to estimate @xmath491 , we use null structure equation , @xmath494 hence , @xmath495 to estimate @xmath496 , we compute commutator @xmath430 \betab = -\chi \cdot \nabla \betab + { } ^*\!\beta \cdot \betab + \frac{1}{2}(\eta + \etab)\nabla_4 \betab + \etab \cdot \betab \cdot \chi,\ ] ] using this formula , we can easily derive @xmath497 \betab| \lesssim c \delta^{\frac{1}{4}}.\ ] ] hence , up to an error of size @xmath254 , we have @xmath498 to estimate @xmath499 , we proceed as follows , @xmath500 putting things together , we have @xmath501 we move on to the estimates for @xmath473 where @xmath465 must be anomalous . thus , because @xmath502 , it must be either @xmath276 or @xmath503 . we can assume @xmath504 . otherwise , by signature considerations , those terms ( of the form @xmath505 ) have already been treated in @xmath471 , so we ignore them at this stage . the anomalous term must be @xmath503 . as we emphasized at the beginning of the section , its anomaly comes from the lower derivative term @xmath111 . hence , up to an error of size @xmath270 , we can replace the integrand by @xmath506 where @xmath507 according to signature considerations . the second term can be estimated easily by placing @xmath456 estimates on @xmath183 and @xmath111 ( which saves an extra @xmath26 ) , thus we bound @xmath473 as follows @xmath508 finally , we can bound @xmath469 by @xmath509 we claim @xmath511 . since both curvature terms in the integrand @xmath512 are anomalous , it must be @xmath513 since total signature is @xmath378 at this situation , the second anomaly must be either @xmath259 or @xmath278 . their signature must be @xmath69 . once again , since @xmath514 , thus the second anomaly must be @xmath278 . thus , @xmath515 . finally , since the signature of @xmath516 is at most @xmath517 , it can not be anomalous . this shows @xmath511 . combining the estimates for @xmath466 and @xmath469 , we finally derive estimates for @xmath20 as follows , @xmath518 we combine , , and , we conclude that @xmath519 \lesssim c(\izero ) ( 1 + ( \r + \rb)^{\frac{3}{2 } } ) + c\delta^{\frac{1}{8}}.\ ] ] considering @xmath520 in , we have @xmath521 in view of , modulo quadratic terms which gives an error of size @xmath270 , we have @xmath522 according to the standard elliptic estimates for hodge systems , we derive @xmath523 considering @xmath524 in , similarly , we derive @xmath525 considering @xmath526 in and , we derive @xmath527 considering @xmath528 , @xmath529 in and as well as , we derive @xmath530 similarly , we have @xmath531 putting all them all together , we have @xmath532 we take @xmath308 , @xmath533 or @xmath534 , in to derive @xmath535_{\mu xyz } + { } ^{(n_0)}\!\pi_{\mu\nu}q[d_3 r]^{\mu\nu}{}_{n_1 n_2}|.\end{aligned}\ ] ] we multiply both sides by @xmath329 and sum those two inequalities , in view of remark [ remark forget delta ] , we have @xmath536 where @xmath537_{\mu xyz}| , \quad k=|\doubleint_{\d(u,\ub)}{}^{(n_0)}\!\pi_{\mu\nu}q[d_3 r]^{\mu\nu}{}_{n_1 n_2}|.\ ] ] in this section , the possible anomalies for @xmath538 are @xmath277 and @xmath259 ; the possible anomalies for @xmath465 are @xmath276 , @xmath277 , @xmath259 and @xmath278 . we still emphasize that the anomalies for @xmath259 and @xmath278 are mild in nature . they come from the lower derivative anomaly @xmath111 . for example , in applications , we always use the following expressions , @xmath539 the integrand of @xmath19 can be written schematically as @xmath312 with total signature @xmath540 or @xmath295 . we split @xmath19 into @xmath336 where @xmath337 denotes the collection of terms with exactly @xmath338 anomalies . according to the signature considerations , @xmath541 , otherwise @xmath542 . to estimate @xmath545 , we observe that none of the @xmath186 s is @xmath380 . otherwise , since the total signature is at least @xmath69 , this force another @xmath186 to be @xmath111 which is impossible(in this case , we only assume @xmath347 contains one anomaly ) . thus , up to an error of size @xmath270 , @xmath549 for the bulk integral , we bound it as follows @xmath550 for boundary integrals , we bound them as follows @xmath551 hence , we bound @xmath552 . combined with the estimates for @xmath349 and @xmath350 , we derive @xmath553 since the total signature is @xmath69 or @xmath295 , thus @xmath555 . the estimates for @xmath556 and @xmath557 are similar , we shall only concentrate on @xmath558 . hence , we further split @xmath392 into @xmath559 to control @xmath364 , we use move the derivative to a better position , @xmath560 in view of , up to an error of size @xmath270 , the bulk integral is @xmath561 then we can place @xmath24 norm on @xmath111 to save an @xmath26 to derive @xmath562 to control @xmath365 , we still integrate by parts , @xmath563 exactly as for @xmath364 , we can bound each term by @xmath254 easily . thus , @xmath564 the integrand of @xmath21 can be written schematically as @xmath566 with total signature @xmath567 or @xmath295 . we split @xmath21 into @xmath398 where @xmath399 denotes the collection of terms with exactly @xmath338 anomalous curvature components . @xmath404 can be bounded as follows , @xmath570 according to gronwall s inequality , @xmath404 can be absorbed by the left hand side of the sum of and . thus , we can ignore @xmath571 at this stage . hence , @xmath572 one can argue exactly as in the beginning of section [ estimates for k_1 ] , @xmath317 does not appear as the connection coefficients in the integrand of @xmath409 . we split @xmath409 into two terms @xmath418 where the corresponding integrands for @xmath419 and @xmath420 are @xmath573 respectively . to estimate @xmath579 , observe that the appearance of @xmath192 is through the term @xmath580 since @xmath277 also appears , this forces @xmath581 . direct computations show @xmath582 or @xmath583 . without loss of generality , we may assume @xmath584 we use to replace @xmath585 . up to an error of size @xmath254 , we have @xmath586 for the boundary integral @xmath587 , @xmath588 for @xmath589 , we compute commutator @xmath590 \betab = -\chib \cdot \nabla \betab + { } ^*\!\betab \cdot \betab + \frac{1}{2}(\eta + \etab)\nabla_3 \betab + \chib \cdot \eta \cdot \betab.\ ] ] all terms are quadratic except the first one which is essentially linear in @xmath591 . thus , it is routine to derive @xmath592 \betab| \lesssim |\doubleint_{\d(u,\ub ) } \chih \cdot \alphab \cdot \nabla \betab|+c \delta^{\frac{1}{4 } } \lesssim c \delta^{\frac{1}{4}}.\ ] ] hence , up to an error of size @xmath270 , we have @xmath593 hence , we eliminate all the anomalies . thus , it is routine to bound both integrals by @xmath254 as well as @xmath594 for @xmath595 , we use to replace @xmath596 , thus , @xmath597 for the first integral , we simply use @xmath25 on @xmath192 to overcome the anomaly ; for the second one , we control it as follows , @xmath598 this yields the control of @xmath599 . combined with the estimates for @xmath589 and @xmath600 , we have @xmath601 to estimate @xmath602 , we observe that in its integrand @xmath603 , we have @xmath604 . otherwise the total signature is less than @xmath69 . this allow @xmath605 to have @xmath24 estimates ( while @xmath136 does not , see @xcite ) . we can proceed exactly ( much easier in reality ) as we did for @xmath606 to show that @xmath607 . combined with the estimates for @xmath419 , we derive @xmath608 we claim @xmath400 . the integrand of @xmath609 may have the following three types of terms @xmath610 since the total signature is either 2 or 3 , this rules out the first two cases immediately . to rule out the third one , notice that its appearance is originally through the following term @xmath611^{\mu\nu}{}_{n_1 n_2}\ ] ] here , @xmath326 and @xmath327 are null . once again , direct computations show that @xmath612 never appears in @xmath613 . the integrand of @xmath20 can be written schematically as @xmath615 with total signature @xmath567 or @xmath295 . we split @xmath20 into @xmath462 where @xmath463 denotes the collection of terms with exactly @xmath338 anomalous curvature components . the integrand of @xmath466 must be one of the following forms @xmath616 thus , those terms can be dealt with exactly as we have done for @xmath401 is last subsection ( of the current section ) . thus , modulo the terms which will be removed eventually by gronwall s inequality , we derive @xmath617 according to the position of the anomaly , we split @xmath469 into @xmath618 where the corresponding integrands for @xmath471 , @xmath473 and @xmath619 are @xmath620 respectively . we use subindex @xmath621 to denote the anomaly ( b for bad ) . to estimate @xmath471 , since the total signature is at most @xmath295 , thus the signature of @xmath622 is at most @xmath66 . using bianchi equations , up to an error of size @xmath270 , we can replace the curvature term @xmath623 by @xmath485 and we aslo replace the mild @xmath259 anomaly by @xmath111 . now we only consider the worst scenario . it should contain @xmath317 . in this case , according to signature considerations , @xmath624 . hence we control the worst scenario for @xmath471 as follows , @xmath625 to estimate @xmath473 , recall that the integrand is originally from ( modulo the hodge dual part ) @xmath626 it has to be @xmath627 if @xmath628 , we know @xmath629 . we also notice @xmath630 , otherwise @xmath469 contains a double anomaly in curvature components . hence , neither @xmath225 or @xmath317 appears as connection coefficients . furthermore , @xmath226 is also absent . thus , @xmath473 is essentially @xmath602 in last subsection ( of the current section ) . hence , @xmath631 to estimate @xmath619 , we need to carefully analyze the structure of @xmath626 as we pointed out at the beginning of the section , the possible anomalies for last term are @xmath276 , @xmath277 , @xmath259 and @xmath278 . first of all , @xmath276 can not occur since @xmath632 . thus , we have three cases left to deal with . if @xmath637 or @xmath278 , those anomalies are mild hence can be replaced by @xmath111 . combined with bianchi equations if necessary , up to an error of size @xmath270 , the corresponding integrand looks like @xmath638 those terms have been treated in @xmath471 . thus , we can derive estimates for @xmath619 . combined with @xmath471 and @xmath473 , we have @xmath639 in this case , both the curvature components in the integrand @xmath626 are anomalous . according to the different anomalies of @xmath640 , we decompose @xmath641 into @xmath642 for @xmath643 , @xmath644 ; for @xmath645 , @xmath646 . to estimate @xmath643 , we observe that @xmath647 because the total signature is at most @xmath295 . thus , up to an error of size @xmath270 , @xmath648 for @xmath649 , according to , up to an error term of size @xmath270 , we can replace @xmath650 by @xmath115 , thus , @xmath651 we can then repeat the proof of lemma [ l4 alpha precise ] to derive @xmath652 thus , @xmath653 the last integral will be eventually removed by gronwall s inequality , thus we could write @xmath654 for @xmath655 , in view of proposition [ pricise on chi chib ] , we have @xmath656 we derive @xmath657 for @xmath658 , we switch the order of @xmath27 and @xmath28 acting on @xmath111 , @xmath659 \alpha\end{aligned}\ ] ] we ignore the commutator term since it is bounded by @xmath270 . the proof is routine . thus , up to an error of size @xmath270 , @xmath660 the last two integrals are bounded by @xmath254 . in fact , we can replace the mild anomaly @xmath240 by @xmath111 . then the proof becomes routine . we now move on to the most dangerous term in @xmath658 , @xmath661 it contains @xmath135 which does not have @xmath456 estimates . we can only rely on its @xmath662 estimates . it requires an extra integration by parts . up to an error of size @xmath270 , we use to derive @xmath663 the second integral is bounded by @xmath270 , because we can put @xmath23 on @xmath135 and @xmath24 on curvature components . we move around the derivatives in the first integral , @xmath664\beta + \doubleint_{\d(u,\ub)}\omega \cdot \alphab \cdot \nabla \nabla_4 \beta\\ & = \doubleint_{\d(u,\ub)}\omega \cdot \alphab \cdot [ \nabla_4 , \nabla ] \beta + \doubleint_{\d(u,\ub ) } \nabla \omega \cdot \alphab \cdot \nabla_4 \beta + \doubleint_{\d(u,\ub)}\omega \cdot \nabla \alphab \cdot \nabla_4 \beta\end{aligned}\ ] ] thus , we removed all the anomalies . this yields the bound for @xmath658 . together with @xmath649 and @xmath655 , we derive @xmath665 we now estimate @xmath645 . its integrand is in the following form @xmath627 since @xmath666 and @xmath667 is an anomaly , the integrand of @xmath645 is reduced to the following form @xmath668 we further decompose @xmath645 into @xmath669 such that the corresponding integrands of @xmath670 and @xmath671 are @xmath672 respectively . for @xmath670 , up to an error term of size @xmath270 , we replace @xmath673 by @xmath674 to derive @xmath675 for @xmath676 , @xmath677 the last two integrals are bounded by @xmath270 . the proof is routine . for the first term , we use to replace @xmath678 , thus @xmath679 for @xmath680 , we have @xmath681 combining with and , each of the above integrals generates the most difficult term @xmath682 in view of proposition [ pricise on chi chib ] , @xmath683 thus , in view of lemma [ l4 alpha precise ] we have @xmath684 thus , yields the estimates on @xmath680 . combined with @xmath676 , we have @xmath685 the last mission we need to complete is to show that @xmath671 vanishes . in fact , the expression of the integrand @xmath686 is not sufficient to show @xmath671 vanishes . and we can not control it . in view of the formula , we have to also investigate the part from the hodge dual to see the cancelation . we now show that @xmath687 we also have @xmath692 , @xmath693 and @xmath694 . thus , @xmath695 \varepsilon_{ca}\varepsilon_{bd}\beta(d_c \!^ * r)_d=\alphab(d_3 r)^*_{ab } \varepsilon_{ca}\varepsilon_{bd}\!^*\beta(d_c r)_d\\ & = \alphab(d_3 r)^*_{ae } \varepsilon_{eb}\varepsilon_{ca}\varepsilon_{bd}\!^*\beta(d_c r)_d = -\alphab(d_3 r)^*_{ae } \varepsilon_{ca}\delta_{ed}\!^*\beta(d_c r)_d\\ & = - \alphab(d_3 r)_{ab } \varepsilon_{ca}\!^*\beta(d_c r)_b.\end{aligned}\ ] ] thus , @xmath696 . this cancelation can be explained in another way . to derive energy estimates at this situation , i.e. @xmath308 and @xmath697 , by using bel - robinson tensors , it is equivalent to first commute , and with @xmath27 , and then multiply the top order terms and integrate by parts directly on @xmath698 . in this way , we see directly that @xmath671 type term do not appear . putting all the estimates together , we derive @xmath699 combined with the estimates for @xmath700 and @xmath469 , we have @xmath701 combined again with , and , we derive @xmath702 considering @xmath703 in , combined and standard elliptic estimates for hodge systems , we have @xmath704 considering @xmath585 and @xmath705 in , combined , and elliptic estimates , we derive @xmath706 hence , @xmath707

we calculate the hard - scattering kernels relevant to the negative - helicity decay amplitude in decays to two vector mesons in the framework of qcd factorisation . we then perform a comprehensive analysis of the 34 decays , including decays and the complete set of polarisation observables . we find considerable uncertainties from weak annihilation and the non - factorisation of spectator - scattering . large longitudinal polarisation is expected with certainty only for a few tree - dominated colour - allowed modes , which receive small penguin and spectator - scattering contributions . this allows for an accurate determination of the ckm angle ( or ) from resulting in we also emphasize that the system is ideal for an investigation of electroweak penguin effects . pitha 06/13 + hep - ph/0612290 + december 21 , 2006 * branching fractions , polarisation and asymmetries + of decays * martin beneke , johannes rohrer and deshan yang + _institut fr theoretische physik e , rwth aachen , + d52056 aachen , germany +theoretical particle physics group ( eken ) , nagoya university , + furo - cho , chikusa - ku , nagoya 464 - 8602 , japan _

hep-ph0612290

the variety of accessible final states in @xmath12 decays to two mesons provides an abundant source of information on cp violation and flavour - changing processes . when the final state consists of two vector mesons , an angular analysis of the vector mesons decay products also provides insight into the spin structure of the flavour - changing interaction . for the @xmath13 coupling of the standard model , a specific pattern of the three helicity amplitudes is expected @xcite , such that the longitudinal polarisation fraction @xmath14 should be close to 1 . since @xmath15 was first observed @xcite for penguin - dominated strangeness - changing decays , many theoretical papers addressed the question whether this result could be explained as a strong - interaction effect , or whether it could be reproduced within specific `` new physics '' scenarios @xcite . in this paper we revisit this question using the qcd factorisation framework @xcite to deal with the strong interaction in the amplitude calculation . our study goes beyond previous ones in several respects . on the theoretical side we provide the first complete results for the hard - scattering kernels relevant to vector - vector ( @xmath16 ) final states , correcting several errors in the literature . ( in fact , the only correct calculation is @xcite . ) we also provide a more detailed discussion of the factorisation structure and power counting for the various amplitudes . it seems to have escaped attention so far that , contrary to the longitudinal polarisation amplitude and those relevant to @xmath17 and @xmath18 final states , the transverse polarisation amplitudes do not factorise even at leading power in the heavy - quark expansion . this , together with the high sensitvity to penguin weak - annihilation @xcite , implies that the calculation of polarisation observables stands on a much less solid footing than the calculation of @xmath19 decays . on the phenomenological side , we provide estimates for all @xmath20 decays ( including @xmath21 decays ) and for all parameters that enter the angular analysis . previous studies concentrated on single or a few decay modes and considered the longitudinal polarisation fraction @xmath14 only , making it often difficult to distinguish general patterns from the consequences of particular parameters choices . the organisation of the paper is as follows : in section [ sec : defs ] we summarise the definitions for the helicity amplitudes , angular variables and polarisation observables . the calculation of the @xmath20 decay amplitudes in the qcd factorisation framework is briefly reviewed in section [ sec : factorisation ] . we then discuss a few aspects of the transverse polarisation amplitudes that allow for an understanding of the main characteristics of @xmath20 phenomenology . one important conclusion from this discussion is that the analysis of @xmath1 decays will be much less rigorous and much more uncertain than the corresponding analysis of @xmath22 and @xmath1 modes @xcite . the technical results of the calculation are summarised in an appendix . section [ sec : inputs ] provides the list of input parameters , an overview of the flavour amplitude parameters with theoretical uncertainties , and a classification of the 34 @xmath1 decay channels , which guides the subsequent numerical analysis . we begin the analysis in section [ sec : tree ] with a discussion of branching fractions , cp asymmetries and polarisation observables of the nine tree - dominated decays . among these the four colour - allowed modes can be well predicted . in particular , we show that the time - dependent cp asymmetry measurement in @xmath23 leads to one of the most accurate determinations of the ckm angle @xmath4 . in section [ sec : penguin ] we turn to the 14 colour - allowed penguin - dominated decay modes . it will be seen that theoretical calculations allow for large transverse polarisation within large uncertainties . this suggests to determine the transverse penguin amplitude from data using the well - measured @xmath24 modes . this approach is used to sharpen the predictions for the remaining decay modes in this class . the analysis concludes in section [ sec : others ] with a brief discussion of the remaining penguin - dominated modes , and decays that occur only through weak annihilation . section [ sec : conclusion ] summarises our main results and conclusions . we consider a @xmath0 meson with four - momentum @xmath25 and mass @xmath26 decaying into two light vector mesons @xmath27 , @xmath28 with masses @xmath29 of order @xmath30 . the decay amplitude can be decomposed into three scalar amplitudes @xmath31 according to @xmath32 with convention @xmath33 alternatively , one can choose a basis of amplitudes describing decays to final state particles with definite helicity @xmath34 or use the transversity amplitudes , where @xmath35 are replaced by @xmath36 and @xmath37 , corresponding to linearly polarised final states . we choose @xmath38 to be directed in positive @xmath39-direction in the @xmath0 meson rest frame , and the polarisation four - vectors of the light vector mesons such that in a frame where both light mesons have large momentum along the @xmath39-axis , they are given by @xmath40 , and @xmath41 , @xmath42 . here and thoughout the paper we neglect corrections @xmath43 quadratic in the light meson masses . thus @xmath44 with @xmath45 . our conventions imply the identity @xmath46 experimentally , the magnitudes and relative phases of the various amplitudes are extracted from the angular distributions of the vector resonance decay products . the full angular dependence of the cascade where both vector mesons decay into pseudoscalar particles is given by @xcite @xmath47 \\&\mathrel{\hphantom{\propto } } + \frac{1}{2 } \sin^2\vartheta_1\sin^2\vartheta_2 \,{\mathrm{re}}\left({\mathrm{e}}^{2 i\varphi } { \ensuremath{\mathcal{a}}}_+ { \ensuremath{\mathcal{a}}}_-^*\right ) , \end{split } \label{angulardist}\ ] ] where @xmath48 measures the angle between the decay planes of the two vector mesons in the @xmath0 meson rest frame , and @xmath49 are the angles between the direction of motion of one of the @xmath50 pseudoscalar final states and the inverse direction of motion of the @xmath0 meson as measured in the @xmath51 rest frame , see figure [ fig : kinematics ] . the omitted proportionality factor is such that one obtains the decay rate @xmath52 when @xmath53 and @xmath54 are integrated from @xmath55 to @xmath56 , and @xmath48 from 0 to @xmath57 . . ] thus , any given @xmath20 decay allows us to define five observables corresponding to the three magnitudes and two relative phases of the helicity amplitudes , or the five angular coefficients in ( [ angulardist ] ) . in experimental analyses , observables are preferably defined in terms of the transversity amplitudes as they have definite cp transformation properties . a typical set of observables consists of the branching fraction , two out of the three polarisation fractions @xmath58 , and two phases @xmath59 , @xmath60 , where @xmath61 it is conventional to combine the five observables of some @xmath20 decay with those of its cp - conjugate @xmath62 decay , and to quote the ten resulting observables as cp - averages and cp - asymmetries . we denote @xmath62 decay helicity amplitudes as @xmath63 and define the corresponding transversity amplitudes as @xmath64 , so that @xmath65 in the absence of cp violation . observables @xmath66 are then defined as in ( [ eq : polobs ] ) , and cp averages and asymmetries are calculated by @xmath67 ( @xmath68 ) for the polarisation fractions and @xmath69 ( @xmath70 ) for the phase observables @xmath71 and @xmath72 . the implicit definition ( [ phasedef ] ) ensures that the cp - averaged phase is the geometrical bisection of the _ acute _ angle enclosed by @xmath73 and @xmath74 ; the magnitude of this angle is @xmath75 . more explicitly , the averaged quantities can be obtained as @xmath76 our phase convention for the amplitudes and definition of observables is compatible with that used in the relevant publications of the babar and belle collaborations ( for example in @xcite ) , except for the sign of @xmath77 relative to the transverse amplitudes , which leads to an offset of @xmath78 for @xmath79 . we favour the above convention , because it implies @xmath80 and @xmath81 at leading order , where all strong phases are zero . the decay amplitudes follow from the matrix elements @xmath82 of the effective hamiltonian ( conventions as in @xcite ) @xmath83 with @xmath84 and @xmath85 . a quark model @xcite or naive factorisation analysis indicates a hierarchy of helicity amplitudes @xmath86 for @xmath62 meson decays . ( for @xmath0 decays exchange @xmath87 . ) this is a consequence of the left - handedness of the weak interaction and the fact that high - energy qcd interactions conserve helicity . in naive factorisation one considers only the four - quark operators in @xmath88 and approximates their matrix elements by the matrix elements of two currents @xcite . the helicity amplitudes @xmath89 are proportional to @xmath90 in this approximation . evaluating this expression ( conventions for the form factors as in @xcite ) we obtain @xmath91 with the definitions @xmath92 the transverse amplitudes @xmath93 are suppressed by a factor @xmath94 relative to @xmath95 . in addition , the axial - vector and vector contributions to @xmath96 cancel in the heavy - quark limit , due to an exact form factor relation @xcite . thus @xmath97 , @xmath98 , and ( [ eq : hierarchy ] ) follows . the dominance of the longitudinal amplitude indicated by ( [ eq : hierarchy ] ) leads to the well - known expectation that @xmath14 should be close to unity . experimental data for penguin - dominated @xmath0 decays is in conflict with this expectation thus motivating theoretical studies beyond the naive - factorisation approximation . we use the qcd factorisation approach @xcite to compute the matrix elements @xmath99 of the effective hamiltonian . in this framework they can be expressed ( at leading power in an expansion of the amplitude in @xmath100 ) in terms of form factors , meson light - cone distribution amplitudes and perturbatively calculable hard scattering kernels . in condensed notation , the factorisation formula reads @xmath101 \right ) + t^{ii}_i { \ast}f_b \phi_{{{\ensuremath{{b } } } } } { \ast}f_{v_1}\phi_{v_1 } { \ast}f_{v_2}\phi_{v_2 } , \end{split}\ ] ] where the star products imply an integration over light - cone momentum fractions . in addition the framework contains estimates of some power corrections , which usually can not be computed rigorously . we follow closely the scheme developed in @xcite for @xmath102 decays to match contributions to the hard - scattering kernels @xmath103 on terms involving products of flavour coefficients @xmath104 and factorised matrix elements @xmath105 . the longitudinal amplitude @xmath106 can be deduced from the results given in @xcite . for the analysis of the present paper we calculated the transverse helicity amplitudes . in the following we describe the basic results and main differences with respect to the longitudinal amplitude ; the expressions for the hard - scattering functions are given in the appendix . non - leptonic decay amplitudes are sums of products of ckm factors , wilson coefficients from ( [ eq : heff ] ) and matrix elements ( [ eq : qcdfformula ] ) of operators with different flavours . it is convenient to organise the amplitudes according to flavour . thus , one writes for example , @xmath107 \nonumber\\ & & + \ , a^h_{\bar k^*\rho } \big [ \delta_{pu}\,\alpha_2^h + \frac{3}{2}\alpha_{3,{\rm ew}}^{p , h } \big]\bigg\}. \label{ampexample}\end{aligned}\ ] ] in naive factorisation , the flavour coefficients @xmath104 are linear combinations of wilson coefficients @xmath108 . in qcd factorisation , they include non - factorisable loop effects and spectator - scattering . the @xmath109 coefficients parameterise weak annihilation amplitudes . the decomposition of the amplitudes for the 34 @xmath16 final states in terms of these quantities follows from the @xmath18 expressions given in @xcite with obvious replacements of pseudoscalar by vector mesons . the @xmath110 relate to the coefficients @xmath111 used in the older factorisation literature as follows ( helicity indices and @xmath112 arguments suppressed ) : @xmath113 where we have used the notation @xmath114 the explicit expressions for the negative - helicity coefficients @xmath115 and the transverse weak annihilation amplitudes are collected in the appendix . beyond leading order the @xmath115 are sums of vertex corrections , penguin contractions , and spectator - scattering contributions , see ( [ ai ] ) . most of the relevant hard - scattering functions have been calculated before . in @xcite results for all three of these contributions have been given . we do not find agreement with these results , however . as far as we understand , the origin of the discrepancy is that the authors of these papers use an incorrect projection on the light - cone distribution amplitudes of transversely polarised vector mesons , which neglects the transverse - momentum derivative terms in ( [ eq : mperp ] ) . an exception is @xcite , which does state the correct projector , but the results still differ from ours , particularly for the spectator - scattering contributions . kagan @xcite has calculated the qcd penguin contractions , spectator - scattering terms as well as the weak annihilation amplitudes , but did not consider the vertex contractions . we confirm his results on the penguin contractions and qcd penguin annihilation , for which explicit expressions were given in the paper . the nlo calculation of the negative - helicity amplitude is quite similar to the calculation of the longitudinal amplitude . the result exhibits , however , some qualitative differences which have important consequences for the phenomenology of @xmath20 decays . in this section we explain the non - factorisation of the negative - helicity amplitude ; that the positive - helicity amplitude can not be calculated in an analogous manner ; that the amplitude hierarchy ( [ eq : hierarchy ] ) is violated by electromagnetic effects ; that penguin annihilation is comparatively more significant for transverse polarisation than longitudinal polarisation penguin amplitudes . the factorisation formula ( [ eq : qcdfformula ] ) contains two structurally different terms , the first of which is dominated by soft interactions within the @xmath116 transitions . these are absorbed into the qcd form factor . the second term stands for interactions where a hard ( more precisely , hard - collinear ) interaction with the spectator quark in the @xmath0 meson takes place . both terms are of the same order in the heavy - quark expansion . this remains true for the transverse polarisation amplitudes , but now one finds that the convolution integrals over the light - cone distribution amplitudes are logarithmically divergent due to the occurence of the integral @xmath117 see ( [ eq : hm1 ] ) , ( [ eq : hm2 ] ) . this endpoint divergence at @xmath118 signals that the presumed factorisation of spectator - scattering does not hold _ even at leading power_. a similar effect occurs in the @xmath19 and longitudinal @xmath20 amplitude only when one attempts to calculate power corrections by applying the light - cone projection including twist-3 terms . it is perhaps not surprising that this divergence is obtained at leading - power for the transverse amplitude , since the entire amplitude is formally a twist-3 term . ( this is the origin of the power suppression of the transverse amplitudes relative to the longitudinal amplitude . ) factorisation - violation at the leading power implies that the calculation of transverse polarisation amplitudes is on a much less solid footing than of the other amplitudes , and often should be considered more as an estimate . in practice , we find that the non - factorisation of spectator - scattering is only significant for the colour - suppressed tree and the flavour - singlet qcd penguin amplitudes , where the ( regulated ) divergent integral is multiplied by a large wilson coefficient . the fact that there are endpoint divergences in the spectator - scattering contribution to the transverse helicity amplitudes has been observed in previous calculations , but its significance for the theoretical status of the factorisation approach and its phenomenological implications have not been sufficiently emphasized . the calculation of the kernels @xmath119 in ( [ eq : qcdfformula ] ) can be interpreted as matching the operators @xmath120 to four - quark operators with field content @xmath121 [ \bar\xi h_v]$ ] in soft - collinear effective theory @xcite . the field @xmath122 describes collinear quarks moving in the direction of @xmath123 , the meson that does not pick up the spectator quark , and satisfies @xmath124 . the leading quark bilinears that have non - vanishing overlap with @xmath125 are @xmath126 the subscript @xmath127 denotes projection of a lorentz vector on the plane transverse to the two light - cone vectors @xmath128 . the first operator overlaps only with the longitudinal polarisation state of @xmath123 , the second only with a transverse vector meson . however , the second operator is not generated by the @xmath13 interactions of the standard model , at least up to the one - loop level . hence the transverse amplitudes arise from power - suppressed operators @xmath129 [ \bar\xi h_v]$ ] . ( in terms of the light - cone projector ( [ eq : mperp ] ) this statement implies that the leading term in the first line does not contribute for @xmath13 interactions , leaving the twist-3 terms in the second and third line . since @xmath130 contains transverse momentum derivatives , one must keep the transverse momenta of partons collinear to @xmath123 ; this explains why the transverse - momentum derivative terms in the projector are required . ) the left - handedness of the weak interactions implies that operators of this form contribute only to the negative - helicity amplitude . the positive - helicity amplitude appears first in yet higher - dimensional operators such as @xmath131 [ \bar\xi d_\perp h_v]$ ] . to match to such operators one must keep the transverse momentum of the quark lines collinear to both mesons non - zero such a calculation has not yet been done , and therefore all calculations of the positive - helicity amplitude in the literature must be regarded as incomplete . it is even possible that for the positive - helicity amplitude no useful factorisation formula holds even for the non - spectator - scattering terms in ( [ eq : qcdfformula ] ) . it follows that the positive - helicity amplitude is power - suppressed relative to the negative one , and should be set to zero in the absence of any consistent calculation of this power correction . within this approximation , @xmath132 , there are only two rather than four independent polarisation observables , since @xmath133 similarly identities for the corresponding cp asymmetries hold . it should be noted that these identities are non - trivial consequences of the @xmath13 nature of the weak interactions _ and _ of factorisation , and it is therefore worthwhile to test them experimentally . in our analysis we proceed as follows : we assume the naive - factorisation expression for the positive - helicity amplitude @xmath134 and allow the form factor to vary within the range @xmath135 . thus , we allow a small variation of @xmath134 around zero to estimate the error from neglecting this power correction . we note that qcd sum rule results for the form factors give @xmath136 values consistent with zero @xcite . in the previous paragraphs we explained the origin of the amplitude hierarchy ( [ eq : hierarchy ] ) . however , when electromagnetic effects are included , a transverse polarisation amplitude can be generated by a short - distance transition to a vector meson and a photon with small virtuality which subsequently converts to a vector meson @xcite . this transition is enhanced by a factor @xmath137 due to the large photon propagator , resulting in the parametric relation @xmath138 thus , formally , the negative - helicity amplitude is leading in the heavy - quark limit . technically , this is related to the existence of the operator @xmath139 [ \bar\xi h_v]$ ] , which contributes only to the transverse amplitude , and which is enhanced relative to @xmath121 [ \bar\xi h_v]$ ] . in the standard model the leading effect of this type involves the electromagnetic - dipole operator in the effective hamiltonian . as a consequence the colour - allowed electroweak penguin amplitude @xmath140 is completely different from its naive - factorisation value . our calculations include this contribution , which has already been discussed specifically in @xcite . weak annihilation is a power correction not included in ( [ eq : qcdfformula ] ) , since it does not factorise due to endpoint divergences in the convolution integrals . the effect is often estimated by a parameterisation suggested in @xcite , where the endpoint - divergences are regulated by a cut - off . in this model one finds that the most important annihilation effect is a penguin annihilation amplitude that is phenomenologically indistinguishable from the qcd penguin amplitude . these general observations also hold for the transverse polarisation amplitudes . in particular , the weak annihilation contribution to the negative - helicity amplitude is a power correction relative to the leading , factorisable contributions to this amplitude . yet , as found in @xcite , the effect is numerically much larger than in @xmath19 decays , and perhaps so large that a theoretical calculation of the negative - helicity qcd penguin amplitude is no longer possible . to explain this point , we consider the qcd penguin amplitude @xmath141,\ ] ] and compare the @xmath106 and @xmath142 amplitudes . here @xmath143 are the qcd penguin contributions , and @xmath144 the penguin annihilation contributions . for the longitudinal amplitude @xmath145 is @xmath146 suppressed relative to @xmath147 , but it turns out that numerically the largest effect arises from a @xmath148 term , which has a large colour factor and wilson coefficient . this particular contribution is not suppressed by the extra factor of @xmath146 in the negative - helicity amplitude . one finds @xmath149 , while @xmath150 due to the power suppression of @xmath151 with respect to @xmath152 . thus , relative to @xmath143 the numerical effect of @xmath144 is a factor of @xmath153 larger for the negative - helicity amplitude than for the longitudinal amplitude ( but still power - suppressed since the suppression was @xmath148 for @xmath106 ) . the following numerical estimates illustrate this point . we consider the @xmath154 , @xmath155 helicity amplitudes for @xmath7 , and also the @xmath156 amplitude for comparison . the imaginary parts of the amplitudes are neglected , since they are not important for this discussion . we then find @xmath157\big\ } , \\[0.2 cm ] & \alpha_4^{c0}(\rho \bar k^*)+\beta_3^{c0}(\rho \bar k^ * ) & = \;\ ; -0.03 - \big\{0.00 \ , [ -0.00,0.00]\big\ } , \\[0.2 cm ] & \alpha_4^{c-}(\rho \bar k^*)+\beta_3^{c-}(\rho \bar k^ * ) & = \;\ ; -0.05 - \big\{0.03 \ , [ -0.04,0.10]\big\}. \end{aligned}\ ] ] the numbers in curly brackets refer to @xmath158 . the first number in brackets is the default value , while the interval provides the range allowed by the parameterisation adopted in @xcite . we observe a ( presumably accidental ) cancellation in the longitudinal annihilation amplitude . what is significant is the difference in the range of the interval relative to @xmath159 for the negative - helicty @xmath16 amplitude vs. the @xmath17 amplitude . in particular , the annihilation contribution @xmath160 may be significantly larger than the qcd penguin amplitude @xmath161 for the negative - helicity amplitude . we can also compare the @xmath142 and @xmath106 amplitudes , @xmath162}{0.12},\ ] ] where we used @xmath163 . this shows that @xmath164 could be as large as @xmath165 , if annihilation is maximal . thus , for penguin - dominated decays , a longitudinal polarisation fraction @xmath14 around 0.5 is not ruled out . let us summarise these and a few further observations on the role of weak annihilation in @xmath20 decays : * the annihilation contribution to the longitudinal penguin amplitude is small , perhaps due to an accidental cancellation . * the annihilation contribution to the negative - helicity penguin amplitude can ( but need not ) be very large , possibly leading to significant transverse polarisation in penguin - dominated decays . * no such enhancement is observed for the annihilation contribution to the tree amplitudes , hence tree - dominated decays should be predominantly longitudinally polarised . * we also calculated the weak annihilation contribution to the positive - helicity amplitude , and find that the large contribution to @xmath166 is absent . hence there is no evidence for large corrections to ( [ pluszero ] ) even for penguin - dominated decays . it should be clear that these statements assume that the parameterisation adopted in @xcite reproduces correctly the qualitative features of the weak annihilation amplitudes . ccccc + @xmath167 & @xmath168 & @xmath169 & @xmath170 & @xmath171 + 0.225 & 4.2 & @xmath172 & @xmath173 & 0.0413 + + + @xmath174 & @xmath175 & @xmath176 & @xmath4 & + @xmath177 & @xmath178 & @xmath179 & @xmath180 & + + & @xmath181 & @xmath182 & @xmath183 + & @xmath184 [ ps ] & 1.64 & 1.53 & 1.46 + & @xmath185 [ mev ] & & @xmath186 + & @xmath187 [ mev ] & & @xmath188 + + + & @xmath189 & @xmath190 & @xmath191 & @xmath192 + @xmath193 [ mev ] & @xmath194 & @xmath195 & @xmath196 & @xmath197 + @xmath198 [ mev ] & @xmath199 & @xmath200 & @xmath199 & @xmath200 + @xmath201 , @xmath202 & 0 & @xmath203 & 0 & 0 + @xmath204 , @xmath205 & @xmath206 & @xmath206 & @xmath207 & @xmath207 + .summary of theoretical input parameters . all scale - dependent quantities refer to @xmath208gev unless indicated otherwise . [ cols="^,^,^,^,^,^ " , ] 1.2= in this paper we performed a comprehensive analysis of the 34 @xmath12 decays to two vector mesons ( and their cp conjugates ) . together with @xcite , where the @xmath17 and @xmath18 final states were discussed , this completes the phenomenology of two - body decays in qcd factorisation in next - to - leading order . in comparison with the @xmath17 and @xmath18 final states , the @xmath16 ones are much more uncertain . this is due to a potentially large negative - helicity penguin weak - annihilation amplitude pointed out in @xcite , but also due to the non - factorisation of spectator - scattering for the transverse amplitudes , which has a particularly large effect on colour - suppressed partial amplitudes . our main results are summarised as follows . we obtain a very good description of the @xmath210 system including the @xmath211 final state . as a general rule , all colour - suppressed tree decays are poorly predicted due to the non - factorisation of transverse spectator - scattering . the observed large longitudinal polarisation @xmath212 suggests , however , that this effect is not as large as it could be . in contrast , the colour - allowed tree - dominated decays are the theoretically best predicted @xmath1 modes , and should all show @xmath14 near 1 . we also find a small longitudinal qcd penguin amplitude , which makes the time - dependent cp asymmetry @xmath5 an ideal observable to determine the ckm angle @xmath4 . we find @xmath213 where the theoretical error alone is only @xmath214 . the penguin - dominated decays are plagued by the weak - annihilation uncertainty . while it is natural to obtain an equal amount of transverse and longitudinal polarisation , many observables can only be predicted if at least some information is taken from data . in our analysis , we explored the possibility of replacing the calculated negative - helicity penguin amplitude by a fit from @xmath24 data . while weak annihilation remains the most plausible dynamical explanation for significant transverse polarisation , other options exist : spectator - scattering may enhance the transverse flavour - singlet qcd penguin amplitude , and the electroweak penguin amplitude receives a large contribution from the electromagnetic dipole operator @xcite . the latter effect is expected to be most prominent in the @xmath7 system , and we propose to measure certain helicity - specific decay rate ratios to isolate it . 0.5cmfinally , we comment on the possibility to uncover the helicity structure of the weak interactions through polarisation studies in @xmath1 decays . for instance , the presence of tensor operators changes the helicity - amplitude hierarchy ( [ eq : hierarchy ] ) . according to the theoretical picture that emerges from our study , this appears to be very difficult as far as the hierarchy between the negative - helicity and longitudinal amplitude is concerned , since it is already weak or violated in the qcd penguin amplitudes due to standard model qcd dynamics . the tree amplitudes can not receive large new contributions , since similar effects should then be seen in semi - leptonic decays . this leaves the colour - allowed electroweak penguin amplitude , which is theoretically well - controlled , providing further motivation for the investigation of the @xmath7 system . regarding the hierarchy between the negative- and positive - helcity amplitudes , there is currently no indication of its violation , neither from the @xmath24 data nor from theory . however , one should be aware that the factorisation properties of the positive - helicity amplitude are virtually unknown . this work is supported by the deutsche forschungsgemeinschaft , sfb / tr 9 `` computergesttzte theoretische teilchenphysik '' , and by the german - israeli foundation for scientific research and development under grant no . i - 781 - 55.14/2003 . d.y . acknowledges support from the japan society for the promotion of science . after publication of the preprint version of this paper , we have been made aware of talks by a. kagan , where a fit to the @xmath215 system similar to the one discussed in section [ phikstsec ] was reported , arriving at similar conlcusions on the magnitude of transverse weak annihilation . in this appendix we collect the formulae for the hard - scattering functions relevant to the negative - helicity amplitudes . up to twist-3 , six two - particle light - cone distribution amplitudes are relevant for vector mesons . we neglect three - particle @xmath216 amplitudes . consistency then requires that one adopts the wandzura - wilczek relations among the two - particle amplitudes , reducing the input to two amplitudes @xmath217 and @xmath60 @xcite . the two - particle light - cone projection operator on the longitudinal polarisation state of the vector meson is given as @xmath218 given in @xcite , section 2.3 . to obtain the projector on the transverse polarisation states in the helicity basis , we use the result for @xmath219 from @xcite , and insert @xmath220 to obtain @xmath221 where @xmath222 is the momentum of the vector meson , and we assigned the parton momenta @xmath223 to the quark and antiquark constituents , respectively . the collinear approximation @xmath224 must be applied only after the projection . using the wandzura - wilczek relations @xcite and defining @xcite @xmath225 the transverse helicity projectors simplify to @xmath226\\ & \hspace*{2 cm } + \epsilon^*_{\mp\nu } \,\phi_b(u ) \left[\gamma^\nu ( 1\pm\gamma_5 ) -\bar u\!{\not\!p } \,(1\pm \gamma_5 ) \frac{\partial}{\partial l_{\perp\nu}}\right ] \bigg\}\bigg|_{l_\perp=0}. \end{split}\ ] ] many previous calculations of transverse polarisation amplitudes @xcite used an incorrect expression for the projector , where the transverse - derivative terms are dropped . the authors of @xcite used ( [ longprojector ] ) without employing the wandzura - wilczek relations . the projector is not given explicitly in @xcite . the leading order results for the coefficients @xmath227 , needed to calculate amplitudes according to ( [ ampexample ] ) , ( [ eq : alphaa ] ) , reproduce the results from naive factorisation . at next - to - leading order they receive contributions from one - loop vertex corrections , penguin and dipole operator insertion topologies and hard spectator interaction terms . we assemble these in the form @xmath228 + p^{h , p}_i(v_2 ) , \end{split } \label{ai}\ ] ] where the upper ( lower ) signs apply when @xmath229 is odd ( even ) , and it is understood that the superscript ` p ' is to be ommited for @xmath230 . the results for @xmath106 correspond to those given in @xcite for @xmath18 final states with obvious replacements of @xmath231 by @xmath232 , so we will only give explicit results for the negative - helicity amplitude . the leading - order coefficient , corresponding to naive factorisation , is simply @xmath233 as is well - known , there is no leading - order contribution from @xmath234 operators for vector mesons . the negative - helicity vertex corrections read @xmath235 , & \qquad i \in \{1,2,3,4,9,10\ } \\[1em ] \displaystyle \int_0 ^ 1 { d}y\ ; \phi_{a2}(y ) \left [ -12\ln\frac{m_b}{\mu } + \hphantom{0}6 - g_t(\bar{y } ) \right ] , & \qquad i \in \{5,7\ } \\[1.7em ] \displaystyle 0 , & \qquad i \in \{6,8\ } \end{array } \right.\ ] ] with ( @xmath236 ; similarly for other convolution variables below ) @xmath237 \right ) \nonumber \\ & = & g(y)+\frac{\ln y}{\bar y}. \end{aligned}\ ] ] the function @xmath238 differs from the corresponding function @xmath239 in the longitudinal amplitude only by a single term . corrections from penguin contractions and dipole - operator insertions are present for @xmath240 at order @xmath241 ( @xmath242 for the electroweak penguin amplitudes @xmath243 ) . we find @xmath244 , and @xmath245\\&\hphantom{=\frac{{\ensuremath{\alpha_{\mathrm{s}}}}{\ensuremath{c_{\mathrm{f}}}}}{4\pi { \ensuremath{n_{\mathrm{c}}}}}\bigg\ { } + c_3 \left[\frac{4}{3}\ln\frac{m_b^2}{\mu^2 } + \frac{4}{3 } - g^{-}_{v_2}(0 ) - g^{-}_{v_2}(1 ) \right]\\&\hphantom{=\frac{{\ensuremath{\alpha_{\mathrm{s}}}}{\ensuremath{c_{\mathrm{f}}}}}{4\pi { \ensuremath{n_{\mathrm{c}}}}}\bigg\ { } + ( c_4+c_6 ) \left[\frac{10}{3}\ln\frac{m_b^2}{\mu^2 } - 3 g^{-}_{v_2}(0 ) - g^{-}_{v_2}(s_c ) - g^{-}_{v_2}(1 ) \right ] \bigg\ } , \end{split}\\[0.3 cm ] p_{7}^{-,p}(v_2 ) & = p_{9}^{-,p}(v_2 ) = -\frac{{\ensuremath{\alpha_{\mathrm{em}}}}}{3\pi } c^{\mathrm{eff}}_{7\gamma}\ , \frac{m_b m_b}{m_2 ^ 2 } \nonumber\\ & \hspace*{2.4cm}+\frac{2 { \ensuremath{\alpha_{\mathrm{em}}}}}{27\pi}\,(c_1+{\ensuremath{n_{\mathrm{c}}}}c_2 ) \left[\delta_{pc } \ln\frac{m_c^2}{\mu^2 } + \delta_{pu } \ln\frac{\nu^2}{\mu^2 } + 1\right],\\[0.3 cm ] p_{10}^{-,p}(v_2 ) & = \frac{{\ensuremath{\alpha_{\mathrm{em}}}}}{9\pi { \ensuremath{n_{\mathrm{c } } } } } \bigg\{(c_1+{\ensuremath{n_{\mathrm{c}}}}c_2)\left [ \frac{2}{3}\ln\frac{m_b^2}{\mu^2 } + \frac{2}{3 } -g^{-}_{v_2}(s_p ) \right ] \bigg\}\end{aligned}\ ] ] with @xmath246 where @xmath247 is the usual penguin function ( see , for instance , @xcite ) . notice that unlike the longitudinal case , there is no contribution from the dipole operators @xmath248 and @xmath249 to @xmath250 , @xmath251 , for which we confirm the result given in @xcite . the two terms @xmath252 , @xmath253 exhibit an enhancement by @xmath254 proportional to @xmath255 due to the factor @xmath256 , which alters the naive power - counting ( [ eq : hierarchy ] ) to ( [ eq : emhierarchy ] ) . see also @xcite . we neglect the small contributions from the electroweak penguin operators @xmath257 to the penguin coefficients . the spectator - scattering contribution is given by @xmath258 for @xmath259 . our expressions for the spectator - scattering kernels are simpler than previously published results @xcite , and differ even when the wandzura - wilzcek relations are used to simplify those results . the @xmath0 meson light - cone distribution amplitude enters the kernels via the parameter @xmath187 defined in @xcite . the factors in front of the integrals are of order 1 in the @xmath146 counting except for @xmath260 , where there is an extra @xmath261-enhancement . thus , the hard - spectator scattering contribution to @xmath262 is formally leading over the naive - factorisation amplitude , although not numerically , because @xmath263 are multiplied by small wilson coefficients . since @xmath262 contribute to a decay amplitude in the product @xmath264 , and since @xmath265 according to ( [ rchidef ] ) , we conclude that the suppression of the ( pseudo)scalar penguin amplitudes relative to the @xmath13 ones , @xmath266 , is absent for the negative helicity amplitude . another qualitative difference to the longitudinal amplitude is that the @xmath267-integrals in ( [ eq : hm1 ] ) , ( [ eq : hm2 ] ) are divergent , since the integrand is too singular near @xmath118 . thus factorisation for the negative - helicity amplitude breaks down even at leading order in the heavy - quark expansion due to the non - factorisation of spectator - scattering . to estimate this contribution , we extract the logarithmic divergence by applying a `` plus - prescription '' to the integrand , and replace the large logarithm by a phenomenological parameter @xmath268 @xcite . to estimate the endpoint behaviour , we note that the asymptotic distribution amplitudes are @xmath269 , @xmath270 , @xmath271 . we can then write @xmath272 } _ { \text{finite}}.\ ] ] as specified in ( [ xhparam ] ) , we use a simple model where we treat @xmath273 as an unknown complex parameter universal to all @xmath274 with magnitude around @xmath275 because the logarithmic infrared divergence has its origin in a soft gluon interaction with the spectator quark and can therefore be expected to be regulated at a physical scale of order @xmath276 . weak - annihilation is parameterised by a set of amplitudes @xmath278 . the leading contributions can be assembled from a few basic building blocks as shown in @xcite . the corresponding formulae also hold for the helicity amplitudes in @xmath1 decays , hence we only summarise these building blocks here . for the longitudinal case @xmath106 , only a few signs change in comparison with the known results for @xmath279 or @xmath18 . we find @xmath280 -r_\chi^{v_1 } r_\chi^{v_2 } \phi_{v1}(x)\phi_{v2}(y ) \frac{2}{x\bar{y } } \right\},\\[0.4 cm ] { a^{\mathrm{i},0}}_2 & = \pi{\ensuremath{\alpha_{\mathrm{s}}}}\int_0 ^ 1{d}{x}{d}{y}\left\ { \phi_{v_1}(x ) \phi_{v_2}(y ) \left [ \frac{1}{\bar{y}(1-\bar{x}y)}+\frac{1}{x^2\bar{y } } \right ] -r_\chi^{v_1 } r_\chi^{v_2 } \phi_{v1}(x)\phi_{v2}(y ) \frac{2}{x\bar{y } } \right\},\\[0.4 cm ] { a^{\mathrm{i},0}}_3 & = \pi{\ensuremath{\alpha_{\mathrm{s}}}}\int_0 ^ 1{d}{x}{d}{y}\left\ { r_\chi^{v_1 } \phi_{v1}(x ) \phi_{v_2}(y ) \frac{2\bar{x}}{x\bar{y}(1-\bar{x}y ) } + r_\chi^{v_2 } \phi_{v_1}(x ) \phi_{v2}(y ) \frac{2y}{x\bar{y}(1-\bar{x}y ) } \right\},\\[0.4 cm ] { a^{\mathrm{f},0}}_3 & = \pi{\ensuremath{\alpha_{\mathrm{s}}}}\int_0 ^ 1{d}{x}{d}{y}\left\ { r_\chi^{v_1 } \phi_{v1}(x ) \phi_{v_2}(y ) \frac{2(1+\bar{y})}{x\bar{y}^2 } - r_\chi^{v_2 } \phi_{v_1}(x ) \phi_{v2}(y ) \frac{2(1+x)}{x^2\bar{y } } \right\ } , \end{split}\ ] ] and @xmath281 . the non - vanishing transverse building blocks are @xmath282 \bigg\},\\[0.4 cm ] { a^{\mathrm{i}-}}_2 & = \pi { \ensuremath{\alpha_{\mathrm{s}}}}\ , \frac{2m_1 m_2}{m_b^2 } \int_0 ^ 1 { d}{x}{d}{y } \ ; \left\ { \phi_{a1}(x)\phi_{a2}(y ) \left [ \frac{x+y}{x^2\bar{y}^2 } + \frac{1}{(1-\bar{x}y)^2 } \right ] \right\},\\[0.4 cm ] { a^{\mathrm{i}-}}_3 & = \pi { \ensuremath{\alpha_{\mathrm{s}}}}\int_0 ^ 1 { d}{x}{d}{y } \ ; \left\ { \frac{2m_1}{m_2 } r_\chi^{v_2 } \frac { \phi_{a1}(x ) \phi_2^\perp(y ) } { x \bar{y } ( 1-\bar{x}y ) } - \frac{2m_2}{m_1 } r_\chi^{v_1 } \frac { \phi_1^\perp(x ) \phi_{b2}(y ) } { x \bar{y } ( 1-\bar{x}y ) } \right\ } , \\[0.4 cm ] { a^{\mathrm{f}-}}_3 & = \pi { \ensuremath{\alpha_{\mathrm{s}}}}\int_0 ^ 1 { d}{x}{d}{y } \ ; \left\ { \frac{2m_1}{m_2 } r_\chi^{v_2 } \frac{\phi_{a1}(x)\phi_2^\perp(y ) } { x\bar{y}^2 } + \frac{2m_2}{m_1 } r_\chi^{v_1 } \frac{\phi_1^\perp(x)\phi_{b2}(y ) } { x^2 \bar{y } } \right\},\\[0.4 cm ] { a^{\mathrm{i}+}}_1 & = \pi { \ensuremath{\alpha_{\mathrm{s}}}}\ , \frac{2 m_1 m_2}{m_b^2 } \int_0 ^ 1 { d}{x}{d}{y } \ ; \bigg\ { \phi_{a1}(x ) \phi_{a2}(y ) \left [ \frac{2}{x\bar{y}^3}-\frac{y}{(1-\bar{x}y)^2 } -\frac{y}{\bar{y}^2(1-\bar{x}y ) } \right ] \bigg\},\\[0.4 cm ] { a^{\mathrm{i}+}}_2 & = \pi { \ensuremath{\alpha_{\mathrm{s}}}}\ , \frac{2 m_1 m_2}{m_b^2 } \int_0 ^ 1 { d}{x}{d}{y } \ ; \left\ { \phi_{b1}(x)\phi_{b2}(y ) \left [ \frac{2}{x^3\bar{y } } - \frac{\bar{x}}{(1-\bar{x}y)^2 } -\frac{\bar{x}}{x^2(1-\bar{x}y ) } \right ] \right\}. \end{split}\ ] ] again , the convolution integrals exhibit logarithmic and even linear infrared divergences , which we extract into unknown complex quantities using the prescriptions @xmath283 as for @xmath273 , we assume both @xmath284 and @xmath285 to be universal to all @xmath16 final states with magnitudes around @xmath286 and @xmath153 , respectively . since all non - vanishing building blocks contain such divergences , making the treatment of annihilation rather model dependent , we further simplify our results by evaluating the convolution integrals with asymptotic distribution amplitudes @xmath287 , @xmath288 , @xmath289 and obtain the expressions @xmath290,\\ { a^{\mathrm{i},0}}_3 & \approx 18\pi{\ensuremath{\alpha_{\mathrm{s}}}}\left(r_\chi^{v_1}+r_\chi^{v_2}\right ) \left(-x_a^2 + 2x_a - 4 + \frac{\pi^2}{3 } \right),\\[0.1 cm ] { a^{\mathrm{f},0}}_3 & \approx 18\pi{\ensuremath{\alpha_{\mathrm{s}}}}\left(r_\chi^{v_1}-r_\chi^{v_2}\right ) \left(2x_a-1\right)\left(2-x_a\right ) \end{split}\ ] ] for the nonvanishing longitudinal contributions , and @xmath291 { a^{\mathrm{i}-}}_1 \approx { a^{\mathrm{i}-}}_2 & \approx 18\pi{\ensuremath{\alpha_{\mathrm{s}}}}\frac{m_1 m_2}{m_b^2}\left ( \frac{1}{2}x_l + \frac{5}{2 } - \frac{\pi^2}{3 } \right ) , \\[0.1 cm ] { a^{\mathrm{i}-}}_3 & \approx 18\pi{\ensuremath{\alpha_{\mathrm{s}}}}\left(\frac{m_1}{m_2}r_\chi^{v_2}-\frac{m_2}{m_1 } r_\chi^{v_1}\right ) \left ( x_a^2 - 2x_a + 2 \right),\\[0.1 cm ] { a^{\mathrm{f}-}}_3 & \approx 18\pi{\ensuremath{\alpha_{\mathrm{s}}}}\left(\frac{m_1}{m_2}r_\chi^{v_2}+\frac{m_2}{m_1 } r_\chi^{v_1}\right ) \left ( 2x_a^2 - 5 x_a + 3 \right ) \end{split}\ ] ] for the transverse ones . as expected , all annihilation contributions are suppressed by at least one power of @xmath100 compared to the form factor contributions , with @xmath292 and @xmath293 carrying an additional explicit suppression factor . where a comparison is possible , these results agree with @xcite . j. g. krner and g. r. goldstein , phys . lett . * b79 * , 105 ( 1979 ) . babar collaboration , b. aubert _ et al . _ , * 91 * , 171802 ( 2003 ) , [ hep - ex/0307026 ] . belle collaboration , k. f. chen _ et al . _ , lett . * 91 * , 201801 ( 2003 ) , [ hep - ex/0307014 ] . cheng and k .- c . yang , phys . lett . * b511 * , 40 ( 2001 ) , [ hep - ph/0104090 ] . x. q. li , g .- lu and y. d. yang , phys . rev . * d68 * , 114015 ( 2003 ) , [ hep - ph/0309136 ] . a. l. kagan , phys . lett . * b601 * , 151 ( 2004 ) , [ hep - ph/0405134 ] . p. colangelo , f. de fazio and t. n. pham , phys . * b597 * , 291 ( 2004 ) , [ hep - ph/0406162 ] . hou and m. nagashima , hep - ph/0408007 . m. ladisa , v. laporta , g. nardulli and p. santorelli , phys . rev . * 114025 ( 2004 ) , [ hep - ph/0409286 ] . j. rohrer , rwth aachen diploma thesis ( 2004 ) h .- li and s. mishima , phys . rev . * d71 * , 054025 ( 2005 ) , [ hep - ph/0411146 ] . yang , r .- m . wang and g .- lu , phys . rev . * d72 * , 015009 ( 2005 ) , [ hep - ph/0411211 ] . k. das and k .- c . yang , phys . * d71 * , 094002 ( 2005 ) , [ hep - ph/0412313 ] . * b622 * , 63 ( 2005 ) , [ hep - ph/0411305 ] . c. s. kim and y .- d . yang , hep - ph/0412364 . zou and z .- j . xiao , phys . rev . * d72 * , 094026 ( 2005 ) , [ hep - ph/0507122 ] . et al . _ , phys . rev . * d73 * , 014011 ( 2006 ) , [ hep - ph/0508080 ] . s. baek , a. datta , p. hamel , o. f. hernandez and d. london , phys . rev . * d72 * , 094008 ( 2005 ) , [ hep - ph/0508149 ] . huang , p. ko , x .- h . wu and y .- d . yang , phys . rev . * d73 * , 034026 ( 2006 ) , [ hep - ph/0511129 ] . m. beneke , j. rohrer and d. yang , phys . lett . * 96 * , 141801 ( 2006 ) , [ hep - ph/0512258 ] . chen , hep - ph/0601019 . q. chang , x .- q . li and y .- d . yang , hep - ph/0610280 . m. beneke , g. buchalla , m. neubert and c. t. sachrajda , phys . lett . * 83 * , 1914 ( 1999 ) , [ hep - ph/9905312 ] . m. beneke , g. buchalla , m. neubert and c. t. sachrajda , nucl . phys . * b591 * , 313 ( 2000 ) , [ hep - ph/0006124 ] . m. beneke and m. neubert , nucl . phys . * b675 * , 333 ( 2003 ) , [ hep - ph/0308039 ] . babar collaboration , b. aubert _ et al . _ , * 93 * , 231804 ( 2004 ) , [ hep - ex/0408017 ] ; babar collaboration , b. aubert _ et al . _ , hep - ex/0610073 a. gritsan , hep - ex/0409059 . belle collaboration , k. f. chen _ et al . _ , lett . * 94 * , 221804 ( 2005 ) , [ hep - ex/0503013 ] . belle collaboration , k. abe _ et al . _ , hep - ex/0408141 . m. beneke , g. buchalla , m. neubert and c. t. sachrajda , nucl . b606 * , 245 ( 2001 ) , [ hep - ph/0104110 ] . d. fakirov and b. stech , nucl . phys . * b133 * , 315 ( 1978 ) . m. beneke and t. feldmann , nucl . phys . * b592 * , 3 ( 2001 ) , [ hep - ph/0008255 ] . g. burdman and g. hiller , phys . rev . * d63 * , 113008 ( 2001 ) , [ hep - ph/0011266 ] . j. chay and c. kim , nucl . * b680 * , 302 ( 2004 ) , [ hep - ph/0301262 ] . p. ball and v. m. braun , phys . * d58 * , 094016 ( 1998 ) , [ hep - ph/9805422 ] . p. ball and r. zwicky , phys . rev . * d71 * , 014029 ( 2005 ) , [ hep - ph/0412079 ] . p. ball , phys . b * 644 * , 38 ( 2007 ) , [ hep - ph/0611108 ] . v. m. braun , d. y. ivanov and g. p. korchemsky , phys . rev . * d69 * , 034014 ( 2004 ) , [ hep - ph/0309330 ] . a. khodjamirian , t. mannel and n. offen , phys . lett . * b620 * , 52 ( 2005 ) , [ hep - ph/0504091 ] . p. ball and r. zwicky , phys . * d71 * , 014015 ( 2005 ) , [ hep - ph/0406232 ] . babar collaboration , b. aubert _ _ , phys . rev . * d74 * , 051102 ( 2006 ) , [ hep - ex/0605017 ] . babar collaboration , b. aubert _ et al . _ , 97 * , 261801 ( 2006 ) , [ hep - ex/0607092 ] . babar collaboration , b. aubert _ _ , hep - ex/0607098 . babar collaboration , b. aubert _ et al . _ , hep - ex/0612021 . belle collaboration , j. zhang _ et al . _ , . lett . * 91 * , 221801 ( 2003 ) , [ hep - ex/0306007 ] . a. somov _ et al . lett . * 96 * , 171801 ( 2006 ) , [ hep - ex/0601024 ] . m. beneke and s. jger , hep - ph/0610322 . belle collaboration , k. abe _ et al . _ , hep - ex/0702009 . heavy flavor averaging group ( hfag ) , hep - ex/0603003 , with online update at http://www.slac.stanford.edu/xorg/hfag . babar collaboration , b. aubert _ et al . _ , hep - ex/0607107 . belle collaboration , k. f. chen _ et al . _ , lett . * 98 * , 031802 ( 2007 ) , [ hep - ex/0608039 ] . m. beneke , talk at beauty 2006 , september 24 - 28 , 2006 , oxford , u.k . , to appear in the proceedings , hep - ph/0612353 . m. beneke , m. gronau , j. rohrer and m. spranger , phys . lett . * b638 * , 68 ( 2006 ) , [ hep - ph/0604005 ] . babar collaboration , b. aubert _ lett . * 97 * , 201801 ( 2006 ) , [ hep - ex/0607057 ] . belle collaboration , k. abe _ et al . lett . * 95 * , 141801 ( 2005 ) , [ hep - ex/0408102 ] . cleo collaboration , r. godang _ et al . _ , phys . lett . * 88 * , 021802 ( 2002 ) , [ hep - ex/0101029 ] . m. neubert and j. l. rosner , phys . b441 * , 403 ( 1998 ) , [ hep - ph/9808493 ] . a. j. buras and r. fleischer , eur . j. * c11 * , 93 ( 1999 ) , [ hep - ph/9810260 ] . t. yoshikawa , phys . rev . * d68 * , 054023 ( 2003 ) , [ hep - ph/0306147 ] . m. gronau and j. l. rosner , phys . b * 572 * , 43 ( 2003 ) , [ hep - ph/0307095 ] . a. j. buras , r. fleischer , s. recksiegel and f. schwab , eur . j. * c32 * , 45 ( 2003 ) , [ hep - ph/0309012 ] . l , y .- l . shen and w. wang , chin . * 23 * , 2684 ( 2006 ) . p. ball , v. m. braun , y. koike and k. tanaka , nucl . b529 * , 323 ( 1998 ) , [ hep - ph/9802299 ] .

we introduce and analyze a phenomenological model for anti - angiogenic therapy in the treatment of metastatic cancers . it is a structured transport equation with a nonlocal boundary condition describing the evolution of the density of metastasis that we analyze first at the continuous level . we present the numerical analysis of a lagrangian scheme based on the characteristics whose convergence establishes existence of solutions . then we prove an error estimate and use the model to perform interesting simulations in view of clinical applications . + nous introduisons et analysons un modle phnomnologique pour les thrapies anti - angiogniques dans le traitement des cancers mtastatiques . cest une quation de transport structure munie dune condition aux limites non - locale qui dcrit lvolution de la densit de mtastases . au niveau continu , des estimations a priori prouvent lunicit . nous prsentons lanalyse numrique dun schma lagrangien bas sur les caractristiques , do nt la convergence nous permet dtablir lexistence de solutions . nous dmontrons ensuite une estimation derreur et utilisons le modle pour produire des simulations intressantes au regard de possibles applications cliniques . _ ams 2010 subject classification : 35f16 , 65m25 , 92c50 _ * keywords * : anticancer therapy modelling , angiogenesis , structured population dynamics , lagrangian scheme .

1009.3173

during the evolution of a cancer disease , a fundamental step for the tumor consists in provoking proliferation of the surrounding blood vessels and migration toward the tumour . this process , called tumoral neo - angiogenesis establishes a proper vascular network which ensures to the tumour supply of nutrients and allow the tumor to grow further than 2 - 3 mm diameter . it is also important in the metastatic process by making possible the spread of cancerous cells to the organism which then can develop in secondary tumors ( metastases ) . thus , an interesting therapeutic strategy first proposed by j. folkman @xcite in the seventies consists in blocking angiogenesis with the goal to starve the primary tumor by depriving it from nutrient supply . this can be achieved by inhibiting the action of the vascular endothelial growth factor molecule either with monoclonal antibodies or tyrosine kinase inhibitors . although the concept of the therapy seems perfectly clear , the practical use of the anti - angiogenic ( aa ) drugs leaves various open questions regarding to the best temporal administration protocols . indeed , aa treatments lead to relatively poor efficacy and can even provoke deleterious effects , especially on metastases @xcite . regarding to these therapeutic failures , it seems that the _ scheduling _ of the drug plays a major role . indeed , as shown in the publication @xcite , different schedules for the same drug can lead to completely different results . moreover , aa drugs are never given in a monotherapy but always combined with cytotoxic agents ( also named chemotherapy ) which act directly on the cancerous cells . again , the scheduling of the drugs seems to be highly relevant @xcite and the optimal combination schedule between these two types of drugs is still a clinical open question . thus , the complex dynamics of tumoral growth and metastatic evolution have to be taken into account in the design of temporal administration protocols for anti - cancerous drugs . + in order to give answers to these questions , various mathematical models are being developed for tumoral growth including the angiogenic process . we can distinguish between two classes of models : mechanistic models ( see for instance @xcite ) try to integrate the whole biology of the processes and comprise a large number of parameters ; on the other hand phenomenological models aim to describe the tumoral growth without taking into account all the complexity levels ( see @xcite for a review and @xcite ) . most of these models deal only with growth of the primary tumor but in 2000 , iwata et al . @xcite proposed a simple model for the evolution of the population of metastases , which was then further studied in @xcite . this model did not include the angiogenic process in the tumoral growth and thus could not integrate a description of the effect of an aa drug . we combined it with the tumoral model introduced by hahnfeldt et al . @xcite which takes into account for angiogenesis . the resulting partial differential equation is part of the so - called structured population dynamics ( see @xcite for an introduction to the theory ) : it is a transport equation with a nonlocal boundary condition . its mathematical analysis is not classical because the structuring variable is two - dimensional ; as far as we know such models have only been studied in the case where one structuring variable is the age and thus has constant velocity ( see @xcite ) . this is not the case in our situation and the theoretical analysis of the model without treatment ( autonomous case ) was performed in @xcite . + in this paper , we present some mathematical and numerical analysis of the model in the non - autonomous case that is , integrating both cytotoxic and aa treatments and with a general growth field @xmath0 satisfying the hypothesis that there exists a positive constant @xmath1 such that @xmath2 where @xmath3 is the normal to the boundary . we first simplify the problem by straightening the characteristics of the equation . we perform some theoretical analysis first at the continuous level ( uniqueness and a priori estimates ) using the theory of renormalized solutions . then we introduce an approximation scheme which follows the characteristics of the equation ( lagrangian scheme ) . the introduction of such schemes in the area of size - structured population equations can be found in @xcite for one - dimensional models . here , we go further in the lagrangian approach by doing the change of variables straightening the characteristics and discretizing the simple resulting equation , in the case of a general class of two - dimensional non - autonomous models . we prove existence of the weak solution to the continuous problem through the convergence of this scheme _ via _ discrete a priori @xmath4 bounds and establish an error estimate in the case of more regular data . + finally , we use this scheme to perform various simulations demonstrating the possible utility of the model . first , as a predictive tool for the number of metastases in order to refine the existing classifications of cancers regarding to metastatic aggressiveness . secondly , the model can be used to test various temporal administration protocols of aa drugs in monotherapy or combined with a cytotoxic agent . the model is based on the approach of @xcite to describe the evolution of a population of metastases represented by its density @xmath5 with @xmath6 being the structuring variable , here two - dimensional @xmath7 with @xmath8 the size ( = number of cells ) and @xmath9 the so - called angiogenic capacity . it is a partial differential equation of transport type . the behavior of each individual of the population ( metastasis ) , that is the growth rate @xmath10 of each tumor is taken from @xcite and is designed to take into account for the angiogenic process , as well as the effect of both anti - angiogenic ( aa ) and cytotoxic drugs ( ct ) . its expression will be established in the following subsection . the model writes where @xmath11 , the birth rate @xmath12 , the repartition along the boundary @xmath13 and the source term @xmath14 will be specified in the sequel , @xmath15 is a positive time and @xmath3 is the unit external normal vector to the boundary @xmath16 . let @xmath17 denote the size ( number of cells ) of a given tumor at time @xmath18 . the growth of the tumor is modeled by a gompertzian growth rate modified by a death term describing the action of a ct . the equation is : @xmath19 where @xmath20 is a parameter representing the velocity of the growth , @xmath9 the carrying capacity of the environment , and the term @xmath21 stands for the effect of a cytotoxic drug , where @xmath22 is the concentration of the ct , @xmath23 is a minimal size for the drug to be effective ( @xmath24 ) and the function @xmath25 is a regularization of the heaviside function ( for example @xmath26 , with @xmath27 being a parameter controlling the slope at @xmath28 ) , in order to avoid regularity issues in the analysis . the idea is now to take @xmath9 as a variable of the time , representing the degree of vascularization of the tumor and called `` angiogenic capacity '' . the variation rate for @xmath9 derived in @xcite is : @xmath29 where the terms @xmath30 and @xmath31 represent respectively the endogenous stimulation and inhibition of the vasculature and @xmath32 is the effect of an anti - angiogenic drug . the factor @xmath33 comes from the analysis of @xcite which concluded that the ratio of the stimulation rate over the inhibition one should be homogeneous to the tumoral radius to the square . in the figure [ plansphase ] , we present some numerical simulations of the phase plan of the system , . ) . in green , the nullclines . in both , a. parameters from @xcite : @xmath35 . b. @xmath36[plansphase],title="fig:",width=226 ] 0.9 cm ) . in green , the nullclines . in both , @xmath34 . a. parameters from @xcite : @xmath35 . b. @xmath36[plansphase],title="fig:",width=226 ] following @xcite , we assume a one compartmental pharmacokinetic for the aa and do the same for the ct ( in @xcite there is no ct ) . we also assume that the drugs are administered as boli . this gives @xmath37 where the @xmath38 are the administration times of the aa , @xmath39 is the administered dose and @xmath40 the clearance . the expression for the ct is the same , with c instead of a. we denote @xmath7 and @xmath41 . we define @xmath42 and @xmath43 where @xmath44 is the maximal reachable size and angiogenic capacity for @xmath45 solving the system , with initial size @xmath46 ( see @xcite for a study of this system without the ct term ) . we consider that each tumor is a particle evolving in @xmath11 with the velocity @xmath0 . writing a balance law for the density @xmath5 we have @xmath470,t[\times\om\ ] ] that we endow with an initial condition @xmath48 . + metastasis do not only grow in size and angiogenic capacity , they are also able to emit new metastasis . we denote by @xmath49 the birth rate of new metastasis with size and angiogenic capacity @xmath50 by metastasis of size @xmath8 and angiogenic capacity @xmath9 , and by @xmath14 the term corresponding to metastasis produced by the primary tumor . expressing the equality between the number of metastasis arriving in @xmath11 per unit time ( l.h.s in the following equality ) and the total rate of new metastasis created by both the primary tumor and metastasis themselves ( r.h.s . ) , we should have for all @xmath51 @xmath52 we assume that the emission rate of the primary and secondary tumors are equal and thus take @xmath53 where @xmath54 represents the primary tumor and solves the ode system - . we also assume that the new metastasis created have size @xmath55 and that there is no metastasis of maximal size @xmath44 nor maximal or minimal angiogenic capacity because they should come from metastasis outside of @xmath11 since @xmath0 points inward all along @xmath16 . an important feature of the model is to assume that the vasculature of the neo - metastasis is independent from the one which emitted it . this means that @xmath56 with @xmath13 having its support in @xmath57 and describing the angiogenic distribution of the metastasis at birth . we assume it to be uniformly centered around a mean value @xmath58 , thus we take @xmath59}$ ] , with @xmath60 a parameter of dispersion of the new metastasis around @xmath58 . following the modeling of @xcite for the colonization rate @xmath61 we take @xmath62 with @xmath63 the colonization coefficient and @xmath64 the so - called fractal dimension of blood vessels infiltrating the tumor . the parameter @xmath64 expresses the geometrical distribution of the vessels in the tumor . for example , if the vasculature is superficial then @xmath65 is assigned to @xmath33 thus making @xmath66 proportional to the area of the surface of the tumor ( assumed to be spheroidal ) . else if the tumor is homogeneously vascularised , then @xmath64 is supposed to be equal to 1 . assuming the equality of the integrands in in order to have the equality of the integrals , we obtain the boundary condition of . in the autonomous case , that is when @xmath0 depends only on @xmath6 and there is no treatment , the analysis of the equation has been performed in @xcite . it was proven the existence , uniqueness , regularity and asymptotic behavior of solutions . we present now some analysis on the equation with a more general growth field @xmath0 than the one defined in the section [ equationrenouv ] . + let @xmath11 be a bounded domain in @xmath67 , with @xmath16 being piecewise @xmath68 except in a finite number of points . let @xmath69 be a @xmath68 vector field . we make the following assumption on @xmath0 : @xmath70 we do the following assumptions on the data : @xmath710,t[\times \d\om).\ ] ] in the case of @xmath0 being the one of the section [ equationrenouv ] if there is no treatment ( that is , if @xmath72 , or @xmath73 ) then we do nt have @xmath74 all along the boundary since @xmath0 vanishes at the point @xmath75 . but since the problem was solved in this case ( see @xcite ) we consider that the time @xmath28 is the starting time of the treatment and that @xmath76 or @xmath77 is positive , which makes the assumption true . + we say that @xmath780,t[\times \om)$ ] is a weak solution of the problem if for all test function @xmath79 in @xmath80\times\bom)$ ] with @xmath81 @xmath82dxdt & + \int_\om \rho^0(x)\phi(0,x)dx \notag\\ & + \int_0^t\int_{\d\om}\left\lbrace n(\sig)b(t,\rho ) + f(t,\sig)\right\rbrace\phi(t,\sig)=0\end{aligned}\ ] ] where we denoted @xmath83 . [ remtestlip ] by approximating a lipschitz function by @xmath68 ones , it is possible to prove that the definition of weak solutions would be equivalent with test functions in @xmath84\times\bom)$ ] vanishing at time @xmath15 . let @xmath85 be the solution of the differential equation @xmath86 for each time @xmath51 , we define the entrance time @xmath87 and entrance point @xmath88 for a point @xmath89 : @xmath90 we consider the sets @xmath91 and @xmath92\times\bom;\;x\in\bom_1^t\},\;q_2:=\{(t , x)\in[0,t]\times \bom;\;x\in \bom_2^t\}.\ ] ] we also define @xmath93 and notice that @xmath94\times\d\om=\{(t , x);\;\tau^t(x)=0\},\text { and } \sigma_2=\{(t , x(t;0,\sig));\;0\leq t \leq t,\;\sig\in\d\om\}=\{(t , x);\;\tau^t(x)=0\}.\ ] ] see figure [ redresscaract ] for an illustration . we can now introduce the changes of variables that we will constantly use in the sequel . [ propchgtvar ] the maps @xmath95\times\bom & \rightarrow & q_2 \\ ( t , y ) & \mapsto & x(t;0,y ) \end{array}\ ] ] are bilipschitz . the inverse of @xmath96 is @xmath97 and the inverse of @xmath98 is @xmath99 with @xmath100 . denoting @xmath101 and @xmath102 , we have : @xmath103 we refer to the appendix for the proof of this result and to the figure [ redresscaract ] for an illustration . using these changes of variables we can write for a function @xmath1040,t[\times\om)$ ] @xmath105 we want to decompose the equation into two subequations : one for the contribution of the boundary term and one for the contribution of the initial condition since they are `` independent '' . defining @xmath106 we have , when the solution is regular : @xmath107 and the same for @xmath108 . it is thus natural to introduce the following equations where we denoted @xmath109 and we precise the definition of weak solutions to these equations . [ defsolrho1rho2 ] we say that a couple @xmath1100,t[\times \om)$ ] is a weak solution of the equations - if for all @xmath111 with @xmath112 we have : @xmath113 and for all @xmath114\times\bom)$ ] with @xmath115 we have @xmath116 if @xmath117 is a regular function which solves , then the weak formulation is satisfied since we have : @xmath118 we prove now the following theorem , establishing the equivalence between the problem and the problem - . [ proprhorho1rho2 ] let @xmath780,t[\times \om)$ ] be a weak solution of the equation . then @xmath119 given by is a weak solution of - . conversely , if @xmath117 and @xmath108 are weak solutions of and , then the function defined by @xmath120 is a weak solution of . + @xmath121 _ direct implication . _ let @xmath122 be a weak solution of the equation . we will prove that @xmath108 defined by solves . let @xmath123\times \bom)$ ] with @xmath115 . we define for @xmath124 @xmath125 and we intend to extend it in a lipschitz function of @xmath126\times\bom$ ] so that we can use it as a test function in the weak formulation for @xmath122 ( see remark [ remtestlip ] ) . we define , for @xmath127 , @xmath128 with @xmath129 being a truncature function in @xmath130 such that @xmath131 for @xmath132 . then @xmath133 and we set @xmath134 since @xmath87 and @xmath88 are lipschitz from proposition [ propchgtvar ] . we define then @xmath135 the function @xmath136 is lipschitz on @xmath137 , lipschitz on @xmath138 and @xmath139\times\bom)$ ] since @xmath140 . thus @xmath141\times\bom)$ ] with @xmath142 . using @xmath136 as a test function in , we have @xmath143dxdt + \int_0^t\int_{\d\om}\left\lbrace n(\sig ) b(t,\rho)+f(t,\sig)\right\rbrace\phi_1^\eps(t,\sig)dtd\sig \\ & + \int_{q_2}\rho[\d_t\phi_2 + g\cdot \nabla\phi_2]dxdt + \int_\om \rho^0(x)\phi_2(0,x)dx=0=i^1_\eps + i_2.\end{aligned}\ ] ] by doing the change of variables @xmath96 in the term @xmath144 and noticing that @xmath145 , we obtain @xmath146{}0.\ ] ] now doing the change of variables @xmath98 in the second term @xmath147 and noticing that @xmath148 gives the result . the equation on @xmath117 is proved in the same way . + @xmath121 _ reverse implication . _ let @xmath117 and @xmath108 be solutions of and respectively . define @xmath5 by , and consider a test function @xmath149\times \bom)$ ] with @xmath81 . then @xmath150 , with @xmath151 , thus @xmath152 is valid as a test function in the weak formulation of . in the same way @xmath153 with @xmath154 is valid as a test function for . thus we have @xmath155 doing the changes of variables gives the weak formulation of . this theorem simplifies the structure of the problem . in some sense , it formalizes the method of characteristics in the framework of weak solutions for our problem . the characteristics are straightened ( see figure [ redresscaract ] ) and the directional derivative along the field @xmath156 is transformed in only a time derivative . moreover , integrating the jacobians ( which contains the transformation of areas ) in the definitions of @xmath117 and @xmath108 , these functions are constant in time . the continuous analysis and discrete approximation of the problem is thus simplified . in order to obtain _ a priori _ properties on the solutions of the equation , we will use the theory of renormalized solutions first initiated by diperna - lions @xcite in the case of @xmath157 and further developed by boyer @xcite in the case of a bounded domain . let us first recall the result that we will use , which can be found in @xcite . we need to introduce the following measure on @xmath1580,t[\times \d\om$ ] : @xmath159 [ proprenormalisation ] let @xmath1600,t[\times\om)$ ] be a solution , in the distribution sense , to the equation : @xmath161 1 . the function @xmath122 lies in @xmath162;l^p(\om))$ ] , for any @xmath163 . furthermore , @xmath122 is continuous in time with values in @xmath164 weak-@xmath165 . 2 . there exists a function @xmath1660,t[\times \d\om;|d\mu_g|)$ ] such that for any @xmath167 , for any test function @xmath168\times \barre{\om})$ ] , and for any @xmath169\subset[0,t]$ ] , we have @xmath170 [ remvalabs ] + @xmath121 by approximating the function @xmath171 by @xmath68 functions , it is possible to show that the formula stands with @xmath172 . + @xmath121 the second point of the proposition implies in particular that @xmath173 has a trace which is @xmath174 . + @xmath121 in @xcite , this proposition is proved in the case of a much less regular field @xmath0 but with the technical assumption that @xmath175 , which is not the case here . though , the proof can be extended to our case . thanks to this result , we can prove the following proposition . [ propestimcont ] let @xmath780,t[\times \om)$ ] be a weak solution of the equation . the following estimates stand @xmath176 and @xmath177 + @xmath121 _ estimate in @xmath178 . _ let @xmath122 be a weak solution of the equation . then in particular it solves in the sense of distributions . thus the proposition [ proprenormalisation ] applies and gives a trace @xmath1660,t[\times\d\om;|d_{\mu g}|)$ ] . now , by using with @xmath179 and the definition of weak solutions to the equation we have that for all @xmath180 @xmath181 which gives @xmath182 in view of the remark [ remvalabs ] , we know that @xmath183 is also a weak solution to the equation , with initial data @xmath184 and boundary data @xmath185 . by integrating this equation on @xmath11 and using the divergence formula , we obtain in the distribution sense : @xmath186 and thus @xmath187 a gronwall lemma concludes . + @xmath121 _ estimate in @xmath4 . _ using the proposition [ proprhorho1rho2 ] , we have @xmath117 and @xmath108 solving and . by doing the changes of variables , using the definitions of @xmath117 and @xmath108 and the formulas , we see that @xmath1880,t[\times \d\om)}+||\trho_2(t,\cdot)||_{l^1(\om)},\quad \forall \,t>0\ ] ] @xmath1890,t[\times\om)}\leq ||\trho_1||_{l^\infty(\tq_1)}||g||_{l^\infty(\d\om)}e^{t||\div \ , g||_{\infty } } + ||\trho_2||_{l^\infty(]0,t[\times\om)}e^{t||\div \ , g||_{\infty}}\ ] ] but solving explicitely the equation , we have @xmath190 on the other hand , for @xmath108 we have @xmath1910,t[\times\om)}=||\trho_2(0)||_{l^\infty(\om)}=||\rho^0||_{l^\infty(\om)}$ ] . the expression shows that in the case of a zero boundary data @xmath192 , the trace @xmath193 has some extra regularity , namely it is @xmath162;l^1(\d\om))$ ] . [ propunicite ] if @xmath122 and @xmath194 are two weak solutions of the problem , then @xmath195 almost everywhere . in this section , we build a weak solution to the equation which , in view of the previous considerations , can be achieved by building a couple @xmath196 of solutions to the equations - ( recall proposition [ proprhorho1rho2 ] ) . we will achieve the existence by convergence of an approximation scheme to the problem - where the difficulty is restricted to the approximation of the boundary condition . then we establish an error estimate in the case of more regular data . in order to avoid heavy notations , we forget about the tilda when referring to the problem - . we place ourselves in the case where @xmath197 . let @xmath198 be a uniform subdivision of @xmath126 $ ] with @xmath199 . for the equation , we give ourself uniform subdivisions @xmath200 and @xmath201 , with @xmath202 . the scheme for the equation is then given by : @xmath203 that is , @xmath204 for all @xmath205 . + for the discretization of the equation , for each @xmath206 let @xmath207 with @xmath208 . let @xmath209 & \rightarrow & \d\om \\ s & \mapsto & \sig(s ) \end{array } $ ] be a parametrization of @xmath16 with @xmath210 a.e . , so that for @xmath211 we have @xmath212 . let @xmath213 be an uniform subdivision with @xmath214 . the scheme is given by @xmath215 with @xmath216 and @xmath217 notice that the schemes and are well - posed since the definition of @xmath218 involves values of @xmath219 only with @xmath220 . we denote by @xmath221 and define now the piecewise constant functions @xmath222 and @xmath223 on @xmath224 and @xmath225 by , for @xmath226 and @xmath227 @xmath228\tau_{i-1},\tau_i],\;s\in[s_j , s_{j+1}[\\ \rho_{1,h}(t,\tau,\sig(s))=0 & \text { for } t\in[t_k , t_{k+1}[,\,\tau \in ] t_k , t],\;s\in[s_j , s_{j+1 } [ \\ \rho_{2,h}(t , x,\tt)=\rho_2^k(l , m ) & \text { for } t\in[t_k , t_{k+1}[,x\in[x_{l},x_{l+1}[,\;\tt\in[\tt_m,\tt_{m+1}[. \end{array}\ ] ] see the figure [ schemadiscret ] for an illustration . notice that we have @xmath2290,t_k[\times\d\om)}=\sum_{i=1}^k\sum_{j=1}^{m}\left|\rho_1^k(i , j)\right|\delta t \delta \sig,\;||\rho_{2,h}(t_k,\cdot)||_{l^1(\om)}=\sum_{l , m=1}^m\abs{\rho_2^k(l , m)}(\delta x)^2.\ ] ] + @xmath121 we take the same discretization step in @xmath8 and @xmath9 for @xmath230 but it would work the same with two different steps . + @xmath121 for more regular data , we could take point values instead of . + @xmath121 it will be clear from the following that the scheme would converge the same regardless to the value that we give to @xmath231 . + we prove the equivalent of the proposition [ propestimcont ] in the discrete case . notice that there exists a constant @xmath232 such that @xmath233 and @xmath2340,t[;l^1(\d\om ) ) } + c_\sig \delta \sig:=||f||_h$ ] . [ propestimdisc ] let @xmath235 and @xmath236 being given by and respectively . then for all @xmath206 @xmath2370,t[\times\om)}=||\rho^0||_{l^\infty(\om)}\ ] ] @xmath2380,t_k[\times\d\om ) } \leq e^{t_k ||\beta||_{l^\infty}||n||_h}\left\lbrace||\rho^0||_{l^1(\om ) } + \frac{||f||_h}{||\beta||_{l^\infty}||n||_h}\right\rbrace,\ ] ] @xmath239 moreover , if @xmath240 then @xmath241 for all @xmath242 . the non - negativity of the scheme is straightforward from the definition . the estimate for @xmath223 follows directly from the scheme . for the @xmath178 estimate on @xmath222 we compute , using the scheme @xmath2430,t_k[\times\d\om ) } + \left|b^{k+1}(\rho_1^{k+1},\rho_2^{k+1})\right|\delta t||n||_h + \delta t||f||_h\end{aligned}\ ] ] now from the expression of @xmath244 @xmath245 thus we obtain @xmath246 using @xmath247 and @xmath248 ends the proof of the @xmath178 estimate . + for the @xmath4 estimate , we remark that @xmath249 [ existence ] under the assumptions , there exists @xmath250 and @xmath2510,t[\times\om)$ ] such that @xmath252 and @xmath253 for the weak-@xmath165 topology of @xmath4 . furthermore , @xmath196 is the unique weak solution of - . uniqueness of the solution is straightforward for the problem and follows from the @xmath178 estimate on @xmath254 which can be derived following the proof of the proposition [ propestimcont ] . the proof for the existence is rather classical and consists in passing to the limit in discrete weak formulations of and . from the previous proposition , we obtain that the families @xmath255 and @xmath256 are bounded in @xmath4 and thus there exist @xmath250 , @xmath2510,t[\times\om)$ ] and some subsequences @xmath257 and @xmath258 such that @xmath259 and @xmath260 for the weak-@xmath165 topology of @xmath4 . we have to prove now that @xmath196 is a weak solution of - . the uniqueness of solutions to the equation implies then by standard argument that the whole sequence converges . it remains to prove that @xmath196 solves - . + @xmath121__the function @xmath230 is a weak solution of . _ _ let @xmath261 be a test function for . we have @xmath262 where we denoted @xmath263 . using the scheme ( @xmath264 is constant in @xmath206 ) and @xmath265 since @xmath266 , we obtain @xmath267 { } - \int_\om \rho^0(y)\phi(0,y)dy\end{aligned}\ ] ] since @xmath268{l^1}\rho^0 $ ] . observing that the left hand side converges to @xmath269 gives the result . + @xmath121__the function @xmath254 is a weak solution of . _ _ let @xmath270 be a test function for . then the same calculation as above shows , with @xmath271 and using that @xmath272 as well as @xmath273 for @xmath274 and @xmath275 @xmath276 defining the following piecewise constant functions : @xmath277 and @xmath278 on @xmath279 , the previous equality reads @xmath280 we need the following lemma in order to conclude . we have @xmath2810,t[).\ ] ] we define the piecewise constant function @xmath282 as for @xmath283 and @xmath284 and @xmath285 for @xmath286 . let @xmath287 , then @xmath288 since we defined @xmath289 for @xmath290t_k , t]$ ] . thus , for @xmath2910,t[)$ ] we have @xmath292 and we obtain the result by using @xmath293 , @xmath294 , @xmath295{l^1}\beta$ ] , @xmath296 and noticing that the second term goes to zero in view of the @xmath4 bounds on @xmath222 ( proposition [ propestimdisc ] ) and @xmath61 . using the lemma as well as @xmath297 , @xmath298 and @xmath299{\c([0,t]\times\d\om)}\phi(t , t,\sig)$ ] , the previous calculations give @xmath300{}-\int_0^t\int_{\d\om}\left\lbrace n(\sig)\tb(t,\rho_1,\rho_2)+f(t,\sig)\right\rbrace \phi(t , t,\sig)d\sig dt.\ ] ] on the other hand the left hand side also goes to @xmath301 . this proves that @xmath254 verifies the definition [ defsolrho1rho2 ] and ends the proof . we establish now an error estimate for the approximation of the equations - . for this section , we make the following assumptions on the data : @xmath3020,t[\times \d\om)\ ] ] it can be noticed that in order to perform the weak convergence of the approximated solutions and establish theoretical existence to the continuous problem , we did not need to approximate the characteristics @xmath85 of the equation . in view of the error estimate though , we need to use another approximation of the data than . for @xmath303 we have to introduce an approximation @xmath304 of the characteristics given by a numerical integrator of the ode system - . then we define @xmath305 for @xmath306 and @xmath307 being two continuous functions on @xmath224 and @xmath1580,t[\times\om$ ] respectively , we define @xmath308\tau_{i-1},\tau_i],\;s \in [ s_j , s_{j+1}[\\ \p_1 g_1(t,\tau,\sig(s))=0 & \text { for } t\in[t_k , t_{k+1}[,\,\tau \in ] t_k , t],\;s\in[s_j , s_{j+1 } [ \\ \p_2 g_2(t , x,\tt)=g_2(t_k , x_l,\tt_m ) & \text { for } t\in[t_k , t_{k+1}[,\,x\in]x_{l},x_{l+1}],\;\tt \in [ \tt_m,\tt_{m+1 } [ \end{array}.\ ] ] [ lemprojection ] let @xmath3090,t[\times\om)$ ] . then there exists @xmath310 and @xmath311 such that @xmath312 we do nt give the proof of this lemma since it is classical . we define @xmath313 and @xmath314 the errors of the schemes , with @xmath119 solving the problem - . from the equation we have @xmath315 and thus , subtracting this to and denoting @xmath316 we obtain @xmath317 with @xmath318 hence the truncation error of the scheme @xmath319 comes only from the quadrature error coming from the approximation of the integral in @xmath320 . assume and that @xmath3210,t[\times\om)$ ] . then there exists @xmath322 such that @xmath323 we have @xmath324.\end{aligned}\ ] ] thus @xmath325(t_k,\tau,\sig ) -\left(\beta\circ x_1\right)\trho_1(t_k,\tau,\sig)\right|d\tau d\sig \\ & + \sum_{l , m=1}^{l}\int_{x_l}^{x_{l+1}}\int_{x_m}^{x_{m+1}}\left| \p_2\left[\left(\beta\circ x_2\right)\trho_2\right](t_k , y ) -\left(\beta\circ x_2\right)\trho_2(t_k , x,\tt)\right|dx d\tt + \left|\left|\left(\beta\circ x_1\right)\trho_1\right|\right|_{l^\infty } h\}.\end{aligned}\ ] ] using the lemma [ lemprojection ] and the @xmath178 a priori estimate of proposition [ propestimcont ] gives the result . [ remerrtronc ] in order to have a better order for the truncation error we could use a more sophisticated quadrature method like for instance the trapezoid method on @xmath11 for @xmath108 and on @xmath326 for @xmath117 ( completed by a left rectangle method on @xmath327 ) . adapting the previous proof shows that if the numerical integrator used for the characteristics has order larger than @xmath328 , then the truncation error would have order @xmath328 ( order of the trapezoid method ) . [ properrtrho1trho2 ] assume and that @xmath3290,t[\times \om)$ ] is a regular solution of - . let @xmath222 and @xmath223 solve and . then there exists some constants @xmath330 and @xmath331 such that @xmath312 in view of the lemma [ lemprojection ] , it is sufficient to prove the proposition with @xmath332 instead of @xmath333 ( with @xmath334 ) . for the second estimate , we notice that @xmath335 and the result follows from the definition of @xmath336 . for the first one , we have @xmath3370,t_k[)}=||e_{1,h}(t_k,\cdot)||_{l^1(]0,t_k[)}=\sum_{i=1}^k\sum_{j=1}^{m}\left|e_1^k(i , j)\right|\delta t \delta \sig.\ ] ] we can compute , using @xmath338 and conclude using a discrete gronwall lemma . + @xmath121 by looking more carefully at the propagation of errors in the proof , we see that if we set @xmath339 ( which is valid under ) , the error on @xmath108 comes only from the projection error . + @xmath121 if in addition , we follow the remark [ remerrtronc ] for the approximation of the data , then the error between @xmath222 and @xmath340 would be of order @xmath328 if we had used a trapezoid method for the integral term in @xmath320 . we explain now how we approximate the solution of from the approximation of the solutions of equations - given by the schemes , . we translate formula at the discrete level thanks to @xmath341 , @xmath342 given by and the solutions @xmath343 , @xmath344 of the schemes and : @xmath345 the jacobians of the changes of variables @xmath346 are approximated respectively by @xmath347 and @xmath348 , piecewise constant functions constructed similarly as in through @xmath349 and @xmath350 , where @xmath351 and @xmath352 are one - dimensional quadrature methods such that @xmath353 and @xmath354 . the errors of these quadrature methods are denoted by @xmath355 , @xmath356 and are assumed to be of order @xmath357 , @xmath358 : @xmath359 hence , we have @xmath360 we define the following meshes : @xmath361\tau_{i-1},\tau_i],\,s\in[s_j , s_{j+1}[\ } \\ v_2(k , l , m)=\{(t , x(t;0,(x_l,\tt_m)));\ ; t \in[t_k , t_{k+1}[,\,x\in[x_l , x_{l+1}[,\,\tt\in[\tt_m,\tt_{m+1}[\ } \end{array}\ ] ] and , for a function @xmath362\times\bom)$ ] @xmath363 [ remproj ] in the same way as the lemma [ lemprojection ] , there exists a constant @xmath364 such that for all function @xmath3650,t[\times \om)$ ] @xmath3660,t[\times \om)}\leq c_\p h.\ ] ] suppose that @xmath3670,t[\times\om)$ ] is a regular solution of the equation and let @xmath368 be defined by . then there exists a constant @xmath369 such that @xmath370}{\sup}\,||\rho_h(t,\cdot)-\rho(t,\cdot)||_{l^1(\om)}\leq c h.\ ] ] in view of the remark [ remproj ] , it is again sufficient to prove the proposition with @xmath371 instead of @xmath122 . let @xmath287 , then @xmath372 with @xmath373 ( @xmath334 ) . we do the proof only for @xmath254 since it is similar for @xmath230 . we also do nt write the dependency in @xmath374 in order to avoid heavy notations . to obtain the complete proof it suffices to add integrals with respect to @xmath375 in the following and @xmath374 in all the functions . doing the change of variables @xmath96 we have , noticing that @xmath376 @xmath377 now we have , using the definition @xmath378 thus , since @xmath379 from formula and the fact that @xmath380 , and using @xmath381 , there exists @xmath382 such that @xmath383 the last inequality comes from the lemma [ lemprojection ] since @xmath384 from the formula . using then the continuous and discrete _ a priori _ @xmath178 estimates and the proposition [ properrtrho1trho2 ] gives the result . in the case of less regularity on the solution , we still have @xmath3850,t[\times\om)$ ] . indeed , we write @xmath386 . then we use that for @xmath334 @xmath387{l^1}j_s^{-1}$ ] as well as @xmath388 with @xmath369 a constant . using the theorem [ existence ] for the convergence of @xmath341 and @xmath342 gives the result . in practical situations we are often only interested in the number of metastases and not in the density @xmath122 itself . notice that thanks to the formula @xmath389 , we do nt have to compute the jacobians @xmath390 , @xmath391 to get the number of metastases . yet , we still have to compute the characteristics since they are requested in the computation of the boundary condition ( see formula ) . since our equation is two - dimensional , the computational cost can be relatively high because of the integral term in the boundary , especially for large - time simulations . in order to take into account that the metastases are born with a vasculature very close to a given value @xmath58 , we examine replacing the function @xmath392 by a dirac measure . in @xcite , we demonstrate that if we take @xmath393}$ ] and let @xmath394 go to zero the solution of the problem , in the case of an autonomous velocity field @xmath0 and initial condition equal to zero , converges to the measure solution of a limit problem consisting in replacing @xmath395 by a dirac measure in @xmath396 . we use here these considerations to reduce the computational cost and simulate only along the characteristic passing through @xmath396 , that is to say the scheme with only one discretization point @xmath397 on @xmath16 and @xmath398 . moreover , we use a runge - kutta method of order 4 for the approximation of the characteristics and a trapezoid method for the approximation of the boundary condition . + the values of the parameters for the tumoral growth are taken from @xcite , where they were fitted to mice data . following @xcite and @xcite , we take @xmath399 and fix the value of @xmath63 arbitrarily . the values of the parameters ( without the treatment ) are gathered in the table [ valparam ] . the size (= volume ) is expressed in @xmath400 though until now it was thought as the number of cells . @xmath401 a very important issue for clinicians is to determine the number of metastases which are not visible with medical imaging techniques ( micro - metastases ) . having a model for the density of metastases structured in size allows us to compute the number of visible and non - visible metastases . we took as threshold for a metastasis to be visible a size of @xmath402 cells , that is @xmath403 by using the conversion @xmath404 cells . in the figure [ visiblevstotal ] , we plotted the result of a simulation showing both the total number of metastases as well as only the visible ones . we observe that at day @xmath405 the model predicts approximately one metastasis though it is not visible . at the end of the simulation , the total number of metastases is much bigger than the number of visible ones . + thus , an interesting application of the model would be to help designing a predictive tool for the total number of metastases present in the organism of the patient . in this perspective , we define a metastatic index as the integral of @xmath122 on @xmath11 : @xmath406 of course , this index depends on the values of the parameters , for example on the parameter @xmath63 , as shown in the table [ variation_m ] . the larger @xmath63 , the larger the metastatic index . @xmath407 in this table , we remark that at least for times less than @xmath408 days , it seems that the metastatic index is linear in @xmath63 . indeed , this can be explained by the fact that at the beginning , most of the metastases come from the primary tumor and not by the metastases themselves ( see figure [ primvsmeta].a ) . this means that the renewal term in the boundary condition of could be neglected for small times and that the solution of is close to the one of but then , integrating the equation on @xmath11 gives @xmath409 , where @xmath410 represents the primary tumor and solves the system - with initial condition @xmath411 . the figure [ primvsmeta].b shows that for larger times metastases emitted by the metastases themselves are more important than the ones emitted by the primary tumor . the metastatic index for large time is then not anymore linear in @xmath63 ( result not shown ) . .,width=188 ] , title="fig:",width=188 ] 0.9 cm , title="fig:",width=188 ] [ [ section ] ] we present various simulations of treatments , first involving an anti - angiogenic drug ( aa ) alone , in order to investigate the difference in effectiveness of various drugs regarding to their pharmacokinetic / pharmacodynamic parameters . the first result shown in figure [ droguesfolkman ] takes the three drugs which were used in @xcite where only the effect on tumor growth was investigated , and simulates the effect on the metastases . the three drugs are tnp-470 , endostatine and angiostatine and each drug is characterized by two parameters in the model : its efficacy @xmath76 and its clearance rate @xmath412 . these parameters were retrieved in @xcite by fitting the model to mice data . the administration protocols are the same for endostatine and angiostatine ( @xmath405 mg every day ) but for tnp-470 the drug is administered with a dose of @xmath413 mg every two days . we observe that tnp-470 seems to have the poorest efficacy due to its large clearance . as noticed in @xcite , the ratio @xmath414 should govern the efficacy of the drug and its value is @xmath415 for tnp-470 and @xmath416 for both endostatine and angiostatine . the model we developed is now able to simulate efficacy of the drugs on the metastatic evolution ( figure [ droguesfolkman].c ) . interestingly , the drug which seems to be more efficient regarding to the tumor size at the end of the simulation ( day @xmath408 ) , namely angiostatine , is not the one which gives the best result on the metastases . indeed , the lower efficacy of endostatine regarding to ultimate size is due to a relatively high clearance provoking a quite fast rebound of the angiogenic capacity once the treatment stops . but since the tumor size was lower for longer time , the metastatic evolution was better contained . this shows that the model could be a helpful tool for the clinician since the response to a treatment can differ from the primary tumor to metastases , but the clinician has no data about micro - metastases which are not visible with imagery techniques . [ [ section-1 ] ] one of our main postulate in the treatment of cancer is that for a given drug , the effect can vary regarding to the temporal administration protocol of the drug , due to the combination of the pharmacokinetic of the drug and the intrinsic dynamic of tumoral and metastatic growth . to investigate the effect of varying the administration schedule of the drug , we simulated various administration protocols for the same drug ( endostatine ) . the results are presented in figure [ endostatine ] . we gave the same dose and the same number of administrations of the drug but either uniformly distributed during 10 days ( endostatine 2 ) , concentrated in 5 days ( endostatine 1 ) or in 2 days and a half ( endostatine 3 ) . we observe that the tumor is better stabilized with a uniform administration of the drug ( endostatine 2 ) but the number of metastases is better reduced with the intermediate protocol ( endostatine 1 ) . it is interesting to notice that again if we look at the effects at the end of the simulation , the results are different for the tumor size and for the metastases . the two protocols endostatine 1 and endostatine 2 give the same size at the end , but not the same number of metastases . moreover , the best protocol regarding to minimization of the final number of metastases ( endostatine 1 ) is neither the one which provoked the largest regression of the tumor during the treatment ( endostatine 3 ) nor the one with the most stable tumor dynamic ( endostatine 2 ) . in the figure [ endostatinedose ] , we investigate the influence of the aa dose ( parameter @xmath417 ) on tumoral , vascular and metastatic evolution . we observe that the model is consistent since it exhibits a monotonous response to variation of the dose . ) 20 mg every day , tnp-470 ( @xmath418 ) 30 mg every two days and angiostatine ( @xmath419 ) 20 mg every day . a : tumor size . b : angiogenic capacity . c : number of metastasis.,title="fig:",width=226 ] 0.3 cm ) 20 mg every day , tnp-470 ( @xmath418 ) 30 mg every two days and angiostatine ( @xmath419 ) 20 mg every day . a : tumor size . b : angiogenic capacity . c : number of metastasis.,title="fig:",width=226 ] 0.3 cm ) 20 mg every day , tnp-470 ( @xmath418 ) 30 mg every two days and angiostatine ( @xmath419 ) 20 mg every day . a : tumor size . b : angiogenic capacity . c : number of metastasis.,title="fig:",width=453 ] 0.3 cm 0.3 cm 0.3 cm 0.3 cm an important question in clinical oncology is to determinate how to combine a cytotoxic drug ( ct ) that kills the proliferative cells and an anti - angiogenic ( aa ) drug which acts on the angiogenic process , either by blocking the angiogenic factors like vegf ( monoclonal antibodies , e.g. bevacizumab ) or by inhibiting the receptors to this molecule . the aa drugs are classified as part of the cytostatic drugs as they aim to stabilize the disease . for instance , in the treatment of breast cancer , patients which express the receptor her receive a combination of docetaxel ( ct ) and herceptine ( a tyrosine kinase inhibitor , aa ) . two questions are still open : which drug should come before the other and then what is the best temporal repartition for each drug ? here , we perform a brief _ in silico _ study of the first question . since we do nt have real parameters for the cytotoxic drug we fix arbitrarily the value of each parameter @xmath77 , @xmath420 and @xmath421 to @xmath46 , and perform simulations of the model to investigate combination of the ct and the aa . in the figure [ combinaisonctaa ] , we present the results of two simulations : one giving the aa before the ct ( fig . [ combinaisonctaa].a ) and the other one doing the opposite ( fig . [ combinaisonctaa].b ) . although in both cases the effect on the metastases is very good since the growth seems stopped ( fig . [ combinaisonctaa].d ) , it appears that the qualitative behaviors of the tumoral and metastatic responses are different regarding to the order of administration of the drugs ( fig . [ combinaisonctaa].c and [ combinaisonctaa].d ) . according to the model , it would be better to administrate first the ct in order to reduce the tumor burden and then use the aa to stabilize the disease . indeed , the number of metastasis at the end of the simulation is lower when the ct is applied first than in the opposite case . of course , this conclusion depends on the tumoral growth and drugs parameters but this simulation shows that the model is able to exhibit different responses regarding to the order of administration between ct and aa drugs . , @xmath422 and @xmath423 ) and a cytotoxic one ( ct ) . the parameters for the ct are : @xmath424 , @xmath425 , and @xmath426 . a. aa from day 5 to 10 then ct from day 10 to 15 , every day . tumor growth and angiogenic capacity . b. ct from day 5 to 10 then aa from day 10 to 15 , every day . tumor growth and angiogenic capacity . c : comparison between both combinations on the tumor growth . d : comparison between both combinations on the metastatic evolution.[combinaisonctaa],title="fig:",width=264 ] 0.9 cm , @xmath422 and @xmath423 ) and a cytotoxic one ( ct ) . the parameters for the ct are : @xmath424 , @xmath425 , and @xmath426 . a. aa from day 5 to 10 then ct from day 10 to 15 , every day . tumor growth and angiogenic capacity . b. ct from day 5 to 10 then aa from day 10 to 15 , every day . tumor growth and angiogenic capacity . c : comparison between both combinations on the tumor growth . d : comparison between both combinations on the metastatic evolution.[combinaisonctaa],title="fig:",width=264 ] 0.9 cm , @xmath422 and @xmath423 ) and a cytotoxic one ( ct ) . the parameters for the ct are : @xmath424 , @xmath425 , and @xmath426 . a. aa from day 5 to 10 then ct from day 10 to 15 , every day . tumor growth and angiogenic capacity . b. ct from day 5 to 10 then aa from day 10 to 15 , every day . tumor growth and angiogenic capacity . c : comparison between both combinations on the tumor growth . d : comparison between both combinations on the metastatic evolution.[combinaisonctaa],title="fig:",width=264 ] 0.9 cm , @xmath422 and @xmath423 ) and a cytotoxic one ( ct ) . the parameters for the ct are : @xmath424 , @xmath425 , and @xmath426 . a. aa from day 5 to 10 then ct from day 10 to 15 , every day . tumor growth and angiogenic capacity . b. ct from day 5 to 10 then aa from day 10 to 15 , every day . tumor growth and angiogenic capacity . c : comparison between both combinations on the tumor growth . d : comparison between both combinations on the metastatic evolution.[combinaisonctaa],title="fig:",width=264 ] 0.9 cm in this paper , we combined the models of @xcite and @xcite to obtain a model aiming at describing the effect of anti - angiogenic drugs on the metastatic growth . we established the well - posedness of the model and developed an efficient numerical scheme to perform simulations , which could be adapted to similar models in higher dimensions . the model can now be used in order to rationalize the temporal administration of the anti - angiogenic drugs . to achieve this , we have to implement the various pharmacokinetic models of the different aa drugs and then compare the _ in silico _ predictions to real patient data . + an important open problem in this direction is the mathematical parameter identifiability of the model , that is to say the inverse problem of uniqueness of the parameters resulting in a given observation . it is also important to develop efficient numerical methods able to achieve the parameter identification from the data . indeed , identifying the parameters @xmath63 and @xmath64 in a given patient could determine the metastatic aggressiveness of its cancer , through the metastatic index . this could lead to interesting clinical applications such as a refinement of the existing classifications like tnm or sbr , which deal only with the visible metastases . + as shown in @xcite , the metastatic response to aa treatment depends on the time schedule of the drug . the results of the simulations are encouraging in the perspective of using the model as a tool able to test various real temporal administration protocols of the drugs and to perform predictions of the mathematically optimized schedule for a given drug . moreover , aa are never used in a monotherapy but rather combined with a cytotoxic drug , and determining the best way to combine both drugs is still a clinical open question @xcite . as shown in the figure [ combinaisonctaa ] , the model could help in this direction , regarding both to tumor regression and metastatic evolution of the disease . we should also develop further the modeling in order to take into account for the competition effects between ct and aa . indeed , by reducing the vasculature aa drugs should induce worse supply of both drugs and on the contrary some arguments are expressed in favor of a normalization effect on the tumor vasculature by aa therapy @xcite , at least at the beginning of the treatment . these elements should be incorporated to the model _ via _ nonlinear terms involving the drugs concentrations in the equations - . the relative simplicity of the model ( @xmath427 parameters without treatment ) is a great advantage in view of concrete applications since we have to be able to fit the model to patients data in order to retrieve their parameters and then perform predictions about the optimal schedule . + a fundamental problem that we have to integrate in our model is the one of toxicities which have to be dynamically controlled to optimize the scheduling of the drug . in the case of ct and on the tumoral growth , a model dealing with hematological toxicities is used to drive phase i clinical trials @xcite . in our case , we also have to integrate a module to control the toxicity and address the resulting problem of optimization under constraints . + eventually , our model can be used to run _ in silico _ tests about the paradigm of metronomic chemotherapies which consists in delivering a cytotoxic drug at low doses and uniformly distributed in the treatment cycle rather than administrating the maximum tolerate dose ( mtd ) at the beginning of the cycle . indeed , these metronomic protocols seem to have a dynamical anti - angiogenic effect @xcite that can be integrated in the model of @xcite for the tumour growth and in our model for the effect on metastases . the result for the second map is classical . for the first one , we have to deal with irregular points of the boundary @xmath16 . we denote by @xmath428 the set of such points and set @xmath429 . in order to prove the result , it is sufficient to prove that the map @xmath4300,t[\times \d\om\setminus\chi & \rightarrow & \om_1^t\setminus\chi_t \\ ( \tau,\sig ) & \mapsto & x(t;\tau,\sig ) \end{array}\ ] ] is a diffeomorphism , that globally the map @xmath431\times\d\om\rightarrow \bom_1^t$ ] is bilipschitz and that its inverse is @xmath432 . for the first point , since we avoid the irregular points of the boundary by excluding the set @xmath428 , we have the @xmath433 regularity . it remains to prove that @xmath434 is one - to - one and onto , and that its inverse is @xmath433 . + @xmath121 _ the map @xmath435 is one - to - one and onto_. let @xmath51 and @xmath436 . we have @xmath437 . + for the injectivity , we remark that if we have @xmath438 with for instance @xmath439 , then @xmath440 which is prohibited by the assumption that @xmath441 . thus @xmath435 is one - to - one and we have , for @xmath442\times \d\om$ ] : @xmath443 which implies @xmath444 . thus , we have proven that the inverse of @xmath435 is @xmath432 . + @xmath121 _ the map @xmath435 is a diffeomorphism_. we will prove the formula for @xmath390 which will conclude the proof by using the local inversion theorem . we have @xmath445 , with @xmath446 for @xmath374 being a parametrization of @xmath16 and @xmath447 the derivative in @xmath448 of @xmath449 viewed as the flow on @xmath450 . we compute @xmath451 we compute now directly the value of @xmath452 . we define @xmath453 and now notice that we can write @xmath454 now when @xmath77 goes to zero @xmath455 since @xmath456 . finally , we have @xmath457 , thus @xmath458 and @xmath459 . solving the differential equation between times @xmath460 and @xmath18 and taking the absolute value then gives the formula . + @xmath121 _ globally , @xmath435 is bilipschitz . _ it is possible to show that @xmath461\times\d\om)}\leq e^{t|||dg|||_{l^\infty([0,t]\times\bom}}$ ] . on the other hand , using the formula @xmath462 and the fact that from @xmath463 is bounded on @xmath464 thanks to the assumption we have @xmath465 . thus @xmath435 and @xmath466 are lipschitz on @xmath467\times\d\om\setminus\chi$ ] and @xmath468 respectively , and they are both globally continuous on @xmath467\times\d\om$ ] and @xmath464 . hence they are globally lipschitz . f. billy , b. ribba , o. saut , h. morre - trouilhet , t. colin , d. bresch , j. boissel , e. grenier , j. flandrois . a pharmacologically based multiscale mathematical model of angiogenesis and its use in investigating the efficacy of a new cancer treatment strategy . _ journal of theoretical biology _ * 260 * ( 2009 ) 545 - 562 . j. ml ebos , c. r. lee , w. cruz - munoz , g. a. bjarnason , j. g. christensen and r. s. kerbel , accelerated metastasis after short - term treatment with a potent inhibitor of tumor angiogenesis . _ cancer cell _ , 2009 . p. hahnfeldt , d. panigraphy , j. folkman and l. hlatky , tumor development under angiogenic signaling : a dynamical theory of tumor growth , treatment , response and postvascular dormancy . _ cancer research _ * 59 * ( 1999 ) 4770 - 4775 . f. lignet , s. benzekry , f. billy , b. cajavec bernard , o. saut , m. tod , p. girard , g. freyer , e. grenier , t. colin and b. ribba , identifying optimal combinations of anti - angiogenesis drugs and chemotherapies using a theoretical model of vascular tumour growth . ( in preparation ) . m. paez - ribes , e. allen , j. hudock , t. takeda , h. okuyama , f. vinals , m. inoue , g. bergers , d. hanahan and o. casanovas , antiangiogenic therapy elicits malignant progression of tumors to increased local invasion and distant metastasis . _ cancer cell _ * 15 * ( 2009 ) 220 - 231 . g. j. riely et al . , randomized phase ii study of pulse erlotinib before or after carboplatin and paclitaxel in current or former smokers with advanced non - small - cell lung cancer . _ j. clin . ( 2009 jan 10 ) 264 - 270 . b. you , c. meille , d. barbolosi , b. tranchand , j. guitton , c. rioufol , a. iliadis and g. freyer , a mechanistic model predicting hematopoiesis and tumor growth to optimize docetaxel + epirubicin ( et ) administration in metastatic breast cancer ( mbc ) : phase i trial . _ j. clin . oncol.(meeting abstracts ) _ * 25 * ( 2007 ) . the author would like to express its gratitude to the following people for great support and helpful discussions : d. barbolosi , a. benabdallah , f. hubert and f. boyer . this work was partially supported by anr project memorex .

the propagation of ( massless ) scalar , electromagnetic and gravitational waves on fixed schwarzschild background spacetime is described by the general time - dependent regge - wheeler equation . we transform this wave equation to usual schwarzschild , eddington - finkelstein , painlev - gullstrand and kruskal - szekeres coordinates . in the first three cases , but not in the last one , it is possible to separate a harmonic time - dependence . then the resulting radial equations belong to the class of confluent heun equations , i.e. , we can identify one irregular and two regular singularities . using the generalized riemann scheme we collect properties of all the singular points and construct analytic ( local ) solutions in terms of the standard confluent heun function heunc , frobenius and asymptotic thom series . we study the eddington - finkelstein case in detail and obtain a solution that is regular at the black hole horizon . this solution satisfies causal boundary conditions , i.e. , it describes purely ingoing radiation at . to construct solutions on the entire open interval,\infty [ \,$ ] , we give an analytic continuation of local solutions around the horizon . black hole scattering and quasi - normal modes are briefly considered as possible applications and we use semi - analytically calculated graybody factors together with the damour - ruffini method to reconstruct the power spectrum of hawking radiation emitted by the black hole .

1503.08101

while many features of black hole spacetimes can be investigated in terms of the motion of ( massive and massless ) particles , wave propagation opens a perspective on additional phenomena such as interference effects , scattering of radiation at a black hole , its quasi - normal modes and black hole evaporation . scattering and interference patterns have proven to be very important tools in other branches of physics and yield observational properties of different physical systems . therefore , it is of great relevance to consider black - hole scattering as well and to describe the dynamics of fields in the presence of a black hole for different boundary and initial conditions . the central equation that describes the propagation of massless waves on schwarzschild background spacetime is the general ( time - dependent ) regge - wheeler equation which is a wave equation written in the usual schwarzschild time and the so - called tortoise coordinate . regge and wheeler found this equation for spin @xmath2 by investigating the stability of a schwarzschild black hole under gravitational perturbations in a linearized perturbation theory @xcite . to study electromagnetic wave propagation one can solve maxwell s equations on flat spacetime but consider a medium with specified constitutive relations @xcite . another possibility to derive the regge - wheeler equation for @xmath3 is to use the so - called debye equation , because the source - free maxwell equations reduce to this single scalar equation on a schwarzschild background @xcite . for scalar perturbations and their propagation the klein - gordon equation has to be employed and the regge - wheeler equation for @xmath4 follows from a straight - forward calculation @xcite . for each of the three different types of perturbations the respective regge - wheeler - type equation follows after an expansion of the fields into ( tensorial , vectorial and scalar ) spherical harmonics . assuming additionally a harmonic time dependence , i.e. , fourier expanding the solution of this wave equation , leads to a radial equation that is known as the stationary regge - wheeler equation . this radial equation is a schrdinger - type differential equation with a spin - dependent potential barrier . due to the form of this potential no exact solution is known as long as the tortoise coordinate is kept as the radial variable . changing the coordinate system and reconsidering the previous wave equations gives rise to new radial equations that admit analytic solutions in terms of series of special functions such as hypergeometric or coulomb wave functions , see e.g. ref . another very insightful method is to use the heun functions @xcite and the theoretical framework of singularity analysis to construct analytic ( local ) solutions to the radial equations . we will follow this strategy to a great extent in this article . we start with some necessary preparations in sec . [ sec : prep ] where we introduce the confluent heun equation and important notations concerning singularities of differential equations and local solutions around them . furthermore , we consider on a schwarzschild background a wave equation in the tortoise coordinate with an arbitrary potential term . this equation is transformed to usual schwarzschild coordinates . by considering a general coordinate transformation that keeps the angles unchanged we proceed to eddington - finkelstein , painlev - gullstrand and kruskal - szekeres coordinates . in the last part of this section we sketch how to derive the respective wave equations for the three different spin values . we combine the results to obtain the general regge - wheeler equation that has the form of the previously discussed wave equation in the tortoise coordinate and can , thus , be transformed to the respective wave equations in the other coordinate systems . in sec . [ sec : radialeqn1 ] we study the radial equation in the tortoise coordinate , which is the stationary regge - wheeler equation . a lot about the phenomenology can be learned from this simplest form of the radial equation . we describe asymptotic solutions that fulfill the causal boundary condition at the horizon , the so - called in - mode for scattering and the qnm - asymptotics . here and in the following , qnm stands for quasi - normal modes . since there is no known exact analytic solution of the stationary regge - wheeler equation beyond the asymptotic behavior , approximations or numerical methods have to be used to obtain the full solutions of this equation . afterwards , we derive the radial equation in the usual schwarzschild coordinate , which was already considered e.g. by leaver @xcite and fiziev @xcite . we briefly summarize their results and give properties as well as local solutions of this radial equation . all these solutions have only finite and non - sufficient domains of convergence and there is no regular solution at the black hole horizon that allows to penetrate this radius . in sec . [ sec : radialeqn2 ] we derive and discuss the radial equation in eddington - finkelstein coordinates in full detail , and we examine its properties and local solutions in terms of frobenius and thom series and using the confluent heun function @xmath5 . as far as we know , there is no previous work concerned with the analysis of the eddington - finkelstein radial equation to find analytic solutions of the perturbation equations on schwarzschild background , apart from a short remark on their asymptotic behavior at the horizon in ref . @xcite . to overcome the fact that the region where the local solutions converge is bounded , we show how an analytic continuation of the local frobenius solutions onto the entire interval @xmath60,\infty [ \,$ ] can be performed . in this way we construct an exact solution on this interval that is regular at the horizon and satisfies the causal boundary condition . in sec . [ sec : applications ] we examine some applications of our analytically continued solutions . we consider black hole scattering and briefly sketch how to calculate qnm - frequencies for the different perturbations . at the end of this section we use the method of damour and ruffini @xcite and our semi - analytically calculated graybody factors to reobtain the power spectrum of the hawking radiation emitted from a schwarzschild black hole . we consider an ordinary , second order , linear and homogeneous differential equation of the form @xmath7{}{z } + p_1(z ) \diff{}{z } + p_0(z ) \right ] y(z ) = 0 \label{eq : ode}\end{aligned}\ ] ] with rational coefficient functions @xmath8 , where @xmath9 shall be an element of the riemann sphere , i.e. , of the extended complex plane including @xmath10 . following @xcite , a point @xmath11 @xmath12 a singularity at the point @xmath13 @xmath14 if a differential equation possesses only regular singularities it is referred to as being fuchsian , otherwise it is called a confluent equation . the singly confluent heun equation ( che ) belongs to the remarkably large class of heun s differential equations . all members of this class can be derived from a single second order fuchsian differential equation of the form with four singularities , the general heun equation ( ghe ) , see e.g. @xcite . there are four different confluent cases which are obtained through ( consecutive ) confluence processes of regular singularities . in the case of the che , two regular singularities of the ghe , one of them located at infinity , merge and form an irregular singularity . one standard form of this che is given by , where the coefficient functions are [ eq : che ] @xmath15 and regular singularities are located at @xmath16 . by considering the transformation @xmath17 we find an irregular singularity at @xmath18 , i.e. , @xmath19 . the three singularities can be further characterized by their singular rank ( s - rank ) . the s - rank of a regular singularity is by definition always equal to one , so we only have to calculate the rank of the irregular singularity . for the equation in the form this s - rank is found to be two , where we follow the convention in ref . the generalized riemann scheme ( grs ) , in the form proposed by slavyanov and lay @xcite , is a very useful tool in the analysis of singular differential equations . it contains * s - ranks and locations of all singularities ( 1st and 2nd row ) , * indicial ( characteristic ) exponents of the regular singularities ( 3rd and 4th row ) and * characteristic exponents of second kind for the irregular singularity ( 3rd row and following ) . the corresponding grs for the confluent heun equation in the form is @xmath20 and yields all necessary information about properties of local solutions at the singularities of the differential equation : around both regular singularities local frobenius - type solutions @xcite can be constructed using the entries of the first and second column of the grs and we can construct thom - type solutions ( asymptotic series ) at the irregular singularity with the help of the third column . the two local frobenius solutions around the regular singularity at @xmath21 are @xmath22 and those around the other regular singularity at @xmath23 are given by @xmath24 the series expansion coefficients @xmath25 and @xmath26 can be obtained by inserting the respective solutions from above into the che and solving the resulting three term recurrence relations with chosen initial conditions . here and in what follows , we use the following notation : the superscript is either @xmath27 or @xmath28 to distinguish the two independent local solutions , while the corresponding singularity , at which the solution is constructed , is given in the argument after the semicolon . the local frobenius solutions @xmath29 are convergent within a circle in the complex plane , centered at the respective regular singularity with a radius that is the distance to the next neighboring singular point . hence , in each case the region of convergence is bounded by the unit circle centered at the considered singularity . we emphasize that at each regular singularity @xmath30 the solutions have the asymptotic behavior @xmath31 where the ( complex ) exponent @xmath32 is either @xmath33 or @xmath34 , depending on which of the two singularities is considered . another very useful canonical form of the che is obtained by choosing the coefficient functions in to be @xmath35 and , additionally , defining the two parameters @xmath36 and @xmath37 by @xmath38 this form will be called the maple - form of the che in the following , since it is implemented in the computer algebra system maple @xcite . the grs for the differential equation in the form then changes to @xmath39 and yields the relations between the parameter sets @xmath40 and @xmath41 . at a first glimpse , this looks like an unnecessary modification but it will prove very useful in the following . we define the standard confluent heun function @xmath42 in the domain @xmath43 as the first local frobenius solution at the regular singularity @xmath44 @xcite @xmath45 with the additional normalization @xmath46 & \left . \diff{\mathrm{heunc}(a , b , c , d , e , z)}{z } \right|_{z=0 } & & = \dfrac{b+(c - a)(b+1)+2e}{2(b+1 ) } \ , .\end{aligned}\ ] ] from we obtain the series coefficients @xmath47 as solutions of the recurrence relation @xmath48 depending on the five parameters @xmath49 and @xmath37 , where [ eq : che_recurrence ] @xmath50 we can read off the behavior for large @xmath51 , since the limit @xmath52 yields @xmath53 hence , we conclude that the standard solution @xmath42 is convergent and , thus , well defined only for @xmath54 as it should be for a local frobenius solution ; the radius of convergence is given by the distance to the next neighboring singularity . the major advantage is now that once the standard function @xmath5 is implemented together with the recurrence relation , the second frobenius solution at @xmath21 and even those two at the other regular singularity @xmath23 can be expressed using this single function @xcite . the second frobenius solution at @xmath21 can be represented as @xmath55 and the two frobenius solutions at @xmath23 are now given by @xmath56 we will need these representations to construct solutions to the radial part of wave equations on schwarzschild background spacetime in sections [ sec : radialeqn1 ] and [ sec : radialeqn2 ] . the thom solutions ( asymptotic series ) at the irregular singularity @xmath10 are constructed as @xcite @xmath57 and the coefficients @xmath58 and @xmath59 follow , again , from recursion relations that can be derived by inserting @xmath60 into the differential equation . for the entire article we choose natural units such that the speed of light , newton s gravitational constant , planck s constant and boltzmann s constant are set to unity , i.e. , @xmath61 . furthermore , all distances are measured in multiples of the schwarzschild radius @xmath62 . using this convention , the horizon is located at @xmath63 and the light sphere at @xmath64 , where @xmath65 is the usual schwarzschild radial coordinate ( i.e. , the areal radius ) . the schwarzschild metric in these units is then @xmath66 with @xmath67 being the standard metric on the unit two - sphere . since it will become very useful later , we start with an analysis of the wave equation @xmath68{}{t } - v(r ) \right ] \psi(t , r _ * ) = 0 \label{eq : waveeqn_general}\end{aligned}\ ] ] that involves the schwarzschild time coordinate @xmath69 and the tortoise coordinate [ eq : tortoisecoord ] @xmath70 here @xmath71 is some given spherically symmetric potential . for @xmath72 we obtain the free wave equation , which is solved by @xmath73 i.e. , by a superposition of in- and outgoing waves with arbitrary complex amplitudes @xmath74 and @xmath75 . in the following we transform the wave equation into the corresponding wave equations in usual schwarzschild , eddington - finkelstein ( ef ) , painlev - gullstrand ( pg ) and kruskal - szekeres ( ks ) coordinates . to obtain the wave equation in usual schwarzschild coordinates we perform the simple transformation @xmath76 that does not affect any of the other coordinates . using and the chain rule we rewrite @xmath77{}{{r _ * } } & = \dfrac{(r-1)}{r } \left [ \dfrac{(r-1)}{r } \diffp[2]{}{r } + \dfrac{1}{r^2 } \diffp{}{r } \right ] \end{aligned}\ ] ] and the wave equation becomes now @xmath78{}{r } + \dfrac{(r-1)}{r^3 } \diffp{}{r } \right . \\ - \left . \diffp[2]{}{t } - v(r ) \right ] \psi(t , r ) = 0 \ , . \label{eq : waveeqn_schwarzschild}\end{gathered}\ ] ] this equation contains a `` new '' first - derivative term and we notice that the values @xmath79 and @xmath63 , i.e. , the origin and the horizon in our coordinates , describe singularities of this differential equation . starting from we use a further coordinate transformation @xmath80 , where @xmath81 to consider additional coordinate systems . the angles @xmath82 and @xmath83 remain unchanged to keep the coordinates adapted to spherical symmetry . in the wave equation partial derivatives with respect to @xmath65 are meant that @xmath69 is kept fixed , and vice versa . in the new coordinates , we have to work with partial derivatives with respect to @xmath84 where @xmath85 is kept fixed , and vice versa . thus , we need the following transformation formulas for partial derivatives : @xmath86 & = : \dot{f}(r , t ) \diffp{}{v } + \dot{g}(r , t ) \diffp{}{u } \ , , \\[12pt ] \left . \diffp{}{r } \right|_{t=\text{const . } } & = \left . \diffp{v}{r } \diffp{}{v } \right|_{u=\text{const . } } + \left . \diffp{u}{r } \diffp{}{u } \right|_{v=\text{const . } } \notag \\[6pt ] & = : f'(r , t ) \diffp{}{v } + g'(r , t ) \diffp{}{u } \ , .\end{aligned}\ ] ] if , for the sake of readability , we omit the arguments of the functions @xmath87 and @xmath88 , the wave equation reads @xmath89 \diffp[2]{\psi(u , v)}{v } \notag \\ + & \left [ \left(\dfrac{r-1}{r}\right)^2 { g'}^2 - \dot{g}^2 \right ] \diffp[2]{\psi(u , v)}{u } \notag \\ + & \left [ 2\left ( \dfrac{r-1}{r}\right)^2 f ' g ' - 2 \dot{f } \dot{g } \right ] \diffp{\psi(u , v)}{u v } \notag \\ + & \left [ \left ( \dfrac{r-1}{r } \right)^2 f '' + \dfrac{(r-1)}{r^3 } f ' - \ddot{f } \right ] \diffp{\psi(u , v)}{v } \notag \\ + & \left [ \left ( \dfrac{r-1}{r } \right)^2 g '' + \dfrac{(r-1)}{r^3 } g ' - \ddot{g } \right ] \diffp{\psi(u , v)}{u } \notag \\ - & v(r ) \psi(u , v ) = 0 \ , , \label{eq : waveeqn_coordtrafo}\end{aligned}\ ] ] where @xmath65 has to be considered as a function of the new coordinates , i.e. , @xmath90 . now , we can introduce eddington - finkelstein , painlev - gullstrand and kruskal - szekeres coordinates by specifying the two functions @xmath91 and @xmath92 , respectively . for an overview of such regular coordinate systems for the schwarzschild spacetime the reader is referred to , e.g. , the work in @xcite and references therein . for ( ingoing ) eddington - finkelstein coordinates the proper coordinate transformation is given by [ eq : coordtrafo_ef ] @xmath93 the new time coordinate @xmath85 is constant on ingoing radial null geodesics . with the transformation above the wave equation reduces to the simpler form @xmath94{}{r } + 2 \dfrac{(r-1)}{r } \diffp{}{r v } \right . \\ + \dfrac{(r-1)}{r^3 } \diffp{}{r } -v(r ) \right ] \psi(v , r ) = 0 \ , . \label{eq : waveeqn_ef}\end{gathered}\ ] ] to consider painlev - gullstrand coordinates we have to use instead the slightly more complicated coordinate transformation [ eq : coordtrafo_pg ] @xmath95 calculating all necessary first and second derivatives we obtain @xmath78{}{r } - \dfrac{(r-1)}{r } \diffp[2]{}{v } + 2 \dfrac{(r-1)}{r^{3/2 } } \diffp{}{r v } \right . - \dfrac{(r-1)}{2 r^{5/2 } } \diffp{}{v } + \dfrac{(r-1)}{r^3 } \diffp{}{r } - v(r ) \right ] \psi(v , r ) = 0 \label{eq : waveeqn_pg}\end{gathered}\ ] ] as the corresponding wave equation in pg coordinates . its qualitative features are similar to those of the wave equation in ef coordinates . however , as the latter looks simpler we will concentrate in the following on the ef coordinates . the transformation to kruskal - szekeres coordinates ( in the outer region @xmath96 ) is achieved by [ eq : coordtrafoks ] @xmath97 and , thus , equation simplifies to @xmath98{}{u } - \diffp[2]{}{v } \right ) -4 \mathrm{e}^{-r } v(r ) \right ] \psi(u , v ) = 0 \ , . \label{eq : waveeqn_ks}\end{aligned}\ ] ] here , @xmath65 has to be considered as a function of @xmath84 and @xmath85 , implicitly given by the relation @xmath99 and we recognize the wave operator in double null coordinates @xmath100 . in the interior region @xmath101 @xmath84 and @xmath85 have to be interchanged and this corresponds to a change of signs in the wave operator . while in the four other coordinate systems the wave equation can be reduced with a separation ansatz to a stationary equation by splitting off a time factor , this is not true in ks coordinates . for this reason , we will not consider the ks coordinates in the rest of the paper . for each ( massless ) perturbation with any of the spin values @xmath102 a wave equation of the form can be derived . these wave equations are valid in the sense that the back reaction on the spacetime geometry is neglected and we do consider the fields as small perturbations on a fixed background spacetime . in the following , we sketch how to derive the respective wave equation on schwarzschild background for a chosen spin value and combine the results in the end to obtain the general regge - wheeler equation that contains the spin as a parameter . for ( massless ) scalar perturbations the wave equation is a result of the klein - gordon - equation @xmath103 \phi = 0 \ , , \label{eq : kleingordon}\end{aligned}\ ] ] where @xmath88 is the determinant of the schwarzschild metric . ( note that we use the signature convention ( -,+,+,+ ) . ) if we expand the field into scalar spherical harmonics @xmath104 according to @xcite @xmath105 and insert this ansatz into the klein - gordon equation , we notice that the remaining function @xmath106 has to fulfill the wave equation with a potential given by @xmath107 to describe the propagation of electromagnetic waves we consider the source - free maxwell equations @xmath108 owing to the conformal invariance of and we can use the metric @xmath109 which is conformally equivalent to the schwarzschild metric . following stephani @xcite , the source - free maxwell equations can be reduced to the debye - equation , which is a single differential equation for a scalar function called debye potential . for the background spacetime described by this debye equation is @xmath110{}{\vartheta } + \dfrac{1}{\sin^2 \vartheta } \diffp[2]{}{\varphi } \right ) \right . + \dfrac{\partial^2}{\partial r_*^2 } - \diffp[2]{}{t } \right ] \pi = 0 \ , , \label{eq : debyeeqn}\end{gathered}\ ] ] which is given in as well . to obtain the general solution of maxwell s equations we have to construct the four - potential of the electromagnetic field @xcite @xmath111 where @xmath112 and @xmath113 are two independent solutions of the debye - equation and @xmath114 . thereupon , we can obtain the components of the electromagnetic field as usual by @xmath115 since the debye potential @xmath112 in is a scalar function , we can expand it into scalar spherical harmonics according to @xmath116 inserting this ansatz into the debye equation , we notice that the function @xmath106 has to fulfill the wave equation with a potential given by @xmath117 in a linear perturbation approach , as considered by regge and wheeler @xcite , we write the total metric @xmath118 as the sum @xmath119 of a background metric @xmath120 and a small perturbation @xmath121 . we neglect all higher order terms in @xmath122 and its derivatives . a rather lengthy calculation , involving the expansion of the perturbation in scalar , vectorial and tensorial spherical harmonics and restricting to odd parity perturbations yields the wave equation with the potential @xmath123 as a result of the einstein vacuum field equation . the ( odd - parity ) components of the perturbation @xmath122 can be calculated in terms of the solutions to this wave equation . a similar treatment for the even parity perturbations leads to the zerilli equation @xcite which , however will not be considered in the present paper . combining the previous results for different spins we obtain the general time - dependent regge - wheeler equation @xmath124{}{r _ * } - \diffp[2]{}{t } - v_{s l}(r ) \right ] \psi_{s l}(t , r _ * ) = 0 \ , , \label{eq : rwe_time}\end{aligned}\ ] ] where the spin - dependent potential is now given by @xmath125 the differential equation is a wave equation of the form and can , thus , be transformed to the corresponding wave equations in schwarzschild , eddington - finkelstein , painlev - gullstrand and kruskal - szekeres coordinates using the results of section . a lot of the phenomenology can be understood by investigating the wave equation in the simplest form , i.e. , written in the tortoise coordinate . hence , we shall briefly recap the properties of this differential equation . assuming a harmonic time dependence @xmath126 in leads to the stationary form of the regge - wheeler equation ( rwe ) that is @xmath7{}{r _ * } + \omega^2 - v_{s l}(r ) \right ] r_{\omega s l}(r _ * ) = 0 \ , . \label{eq : rwe}\end{aligned}\ ] ] this radial equation is an ordinary , second order , linear and homogeneous differential equation in the form of a schrdinger equation with `` energy '' @xmath127 . the shape of the involved potential is shown in fig . for all three spin values and a partial wave index @xmath128 as an example . the regge - wheeler potential as a function of the tortoise coordinate @xmath129 for @xmath128 and spin @xmath4 ( dotted ) , @xmath3 ( dashed ) and @xmath2 ( solid).,scaledwidth=40.0% ] the maxima occur near the lightsphere radius @xmath64 ( @xmath130 ) . while for @xmath3 the maximum is always located exactly at that radius , for @xmath4 it is found slightly below and for @xmath2 slightly above , but in any case it coincides with the lightsphere radius in the limit @xmath131 . the value of the potential increases with the partial wave number ( angular momentum index ) . figure ( [ fig : rwepotential2 ] ) shows this @xmath132-dependence of the maximum for @xmath4 . the regge - wheeler potential as a function of the tortoise coordinate @xmath129 for @xmath4 and @xmath133 ( dotted ) , @xmath134 ( dashed ) and @xmath128 ( solid).,scaledwidth=40.0% ] the potential decreases exponentially near the horizon ( @xmath135 ) and behaves like @xmath136 for @xmath137 . hence , the asymptotic solutions in these limits are @xmath138 this is , together with the time - dependent factor @xmath139 , a superposition of in- and outgoing waves , where terms @xmath140 describe waves that propagate to smaller radii and terms @xmath141 describe waves propagating to larger radii with increasing time . to consider a black hole without quantum effects we have to assume causal boundary conditions to ensure that no waves emerge from the horizon . thus , the asymptotic solution that satisfies this causal boundary condition is the in - mode @xcite @xmath142 usually , the amplitude at the horizon is normalized such that @xmath143 and we can define the reflection and transmission coefficients @xcite @xmath144 the complex conjugate of gives the out - mode , @xmath145 . considering the wronskian @xmath146 , we can derive the relation @xmath147 that corresponds to flux conservation . to consider scattering at a black hole we have to find a solution of the rwe that satisfies the causal boundary condition , i.e. , that has the asymptotic behavior . from this solution we could read off the amplitudes @xmath148 and @xmath149 to obtain the reflection and transmission coefficients or the phase shifts @xmath150 @xmath151 which give rise to the ( total ) scattering and absorption cross sections of the black hole . to obtain quasi - normal modes ( qnm ) and calculate their frequencies one usually restricts to purely ingoing boundary conditions : no waves emerge from the horizon , but no waves are coming in from spatial infinity as well . hence , an initial perturbation radiates into the horizon and into spatial infinity and should decay in the cause of time in order to have a black hole that is stable against the considered perturbation . this leads , of course , to imaginary frequencies , where the real part describes an ordinary oscillation and the imaginary part ( with the proper sign ) assures the temporal decay . the proper asymptotic solution is then given by @xmath152 unfortunately , no exact solution of the rwe is known due to the involved potential . thus , to solve the scattering or quasi - normal mode problem the only possibilities seem to be either approximation methods ( e.g. , wkb @xcite , born @xcite , ... ) or numerical solutions . but yet another method is shown in the next sections where we derive exact analytic solutions , given in the form of convergent series , after changing the coordinate system and thus the underlying wave equation . as a result of section [ subsec : waveeq ] we know that the analog to the time - dependent rwe ( coordinates @xmath69 and @xmath129 ) in schwarzschild coordinates ( @xmath69 and @xmath65 ) is obtained by using equation together with the regge - wheeler potential . we can reduce this wave equation , again , to a radial equation with the ansatz @xmath153 as was done before to derive the stationary rwe . this leads to the radial equation @xmath154{}{r } + ( r-1 ) \diff{}{r } \right . + r^3 \omega^2 \\ \left . - ( r-1 ) \left ( l(l+1 ) + \dfrac{1-s^2}{r } \right ) \right ] r_{\omega s l}(r ) = 0 \ , , \label{eq : radialeqnschwarzschild}\end{gathered}\ ] ] which is discussed in the literature for example by leaver @xcite and by fiziev @xcite . leaver uses the ansatz @xmath155 to transform equation ( [ eq : radialeqnschwarzschild ] ) into a generalized spheroidal wave equation for @xmath156 , which is nothing but a special form of the confluent heun equation ( che ) , even though leaver never stated this nor did he use the term `` confluent heun equation '' in his work . fiziev on the other hand makes the ansatz @xmath157 and gives the confluent heun equation in the maple - form for the remaining function @xmath158 and its solutions in terms of the standard confluent heun function . it should be stated that in both cases the ansatz is not well defined at the horizon ( @xmath63 ) due to the term @xmath159 . in the following , we will briefly recall these known solutions since we can use them for a comparison of our results later . with the ansatz of fiziev the function @xmath158 has to satisfy a confluent heun equation in the maple - form with the five parameters @xmath160 the two regular singularities are located at the origin @xmath79 and at the horizon @xmath63 , while the irregular singularity is found at spatial infinity @xmath161 . we can give the grs of this equation using the general result ( [ eq : grs_che_mapleform ] ) to complement the work in @xcite and @xcite . this grs then reads @xmath162 thus , the two local frobenius solutions at the horizon can easily be constructed manually with the help of the above grs or given in terms of the standard confluent heun function as follows : @xmath163 now , we use again ( [ eq : ansatzfieziev ] ) and obtain the full solutions of the schwarzschild radial perturbation equation ( [ eq : radialeqnschwarzschild ] ) around the regular singularity at the horizon . these solutions are @xmath164 it is important to remark that both solutions are not well defined exactly at the horizon . while the modulus approaches one , the phase is not defined : in the complex plane infinitely many turns on the unit circle are performed when approaching the singularity at @xmath63 . thus , in these coordinates it is not possible to construct wavelike solutions that propagate through the horizon . furthermore , the representation of the solutions in terms of the @xmath5-function is only useful in the domain @xmath165 . the asymptotic thom solutions at the irregular singularity @xmath161 can be constructed , as shown in sec . [ sec : prep ] , using the characteristic exponents of the second kind in the third column of the grs @xmath166 thereupon , the full thom solutions of the schwarzschild radial equation at @xmath10 are @xmath167 which coincide with those given in @xcite . these series are convergent in the limit @xmath168 . for the expansion coefficients @xmath58 and @xmath59 recurrence relations can be derived and the normalizations @xmath169 and @xmath170 , i.e. , the amplitudes at spatial infinity remain free to choose . combined with the time - dependent factor @xmath171 the solutions @xmath172 and @xmath173 describe waves that propagate into the respective singularity , i.e. , waves going into the horizon and waves going into spatial infinity . the solutions @xmath174 and @xmath175 describe waves that emerge from the respective singular point . figure ( [ fig : inout_schwarzschild ] ) sketches these in- and outgoing properties as a mnemonic . in- and outgoing properties of the local solutions to the schwarzschild and eddington - finkelstein radial equations.,scaledwidth=35.0% ] to obtain the radial perturbation equation in eddington - finkelstein coordinates we use the wave equation together with the regge - wheeler potential . thereupon , we perform the ansatz @xmath176 to separate the eddington - finkelstein time coordinate @xmath85 . thus , the remaining function @xmath177 has to fulfill the differential equation @xmath178{}{r}+ \left ( \dfrac{1 - 2i \omega r^2}{r(r-1 ) } \right ) \diff{}{r } \right . \\ \left.- \left(\dfrac { l(l+1 ) + ( 1-s^2)/r}{r(r-1 ) } \right ) \right ] r_{\omega s l}(r ) = 0 \ , . \label{eq : radialef}\end{gathered}\ ] ] this radial equation exhibits two regular singularities which are located at @xmath79 and @xmath63 , since these points are poles of first order in the second term and poles of at most second order in the last term . by means of the substitution @xmath179 we obtain @xmath178{}{\zeta } + \left ( \dfrac{3\zeta^2 - 2\zeta-2i \omega}{\zeta^2(\zeta-1 ) } \right ) \diff{}{\zeta } \right . \\ \left . + \left ( \dfrac { l(l+1 ) + \zeta(1-s^2)}{\zeta^2(\zeta-1 ) } \right ) \right ] r_{\omega s l}(\zeta ) = 0\end{gathered}\ ] ] which yields a pole of second order at @xmath18 in the coefficient of the first derivative term . hence , we conclude that the original differential equation possesses an irregular singularity at @xmath180 . with two regular singularities at @xmath181 and the irregular singularity at @xmath161 the radial equation in eddington - finkelstein coordinates belongs to the class of singly confluent heun equations as well . it is indeed possible to find a substitution that transforms into a standard form of the che . one such possibility simply is @xmath182 and yields a che in the form with the set of parameter @xmath183 using the five parameters in we can immediately give the corresponding grs for the differential equation following the general result , @xmath184 this grs already reveals a lot of useful information : since @xmath185 can only take one of the integer values @xmath186 the two indicial exponents at the regular singularity @xmath79 differ only by an integer . hence , both local frobenius solutions constructed at this point are not linearly independent anymore @xcite . moreover , we can already see that it will be possible to construct a solution that is regular at the horizon . this is due to the indicial exponent @xmath187 in the second column of the grs and the transformation that is regular at the horizon @xmath63 , unlike in the schwarzschild coordinate case before . it is also possible to give the radial equation in ef coordinates in the maple form of the che . in this case the parameter for the confluent heun equation in the form become @xmath188 using these five parameters we can give the solutions around the regular singularity at the horizon in terms of the standard confluent heun function as follows : @xmath189 both solutions are independent and differ remarkably in their behavior at the horizon : while the first solution is well behaved and takes simply a constant value , the second solution @xmath190 performs infinitely many turns on the unit circle in the complex plane and has no well defined phase when approaching @xmath63 . figure ( [ fig : ef_frobenius ] ) shows the real part and the absolute value of both solutions for some arbitrarily chosen values of the parameter @xmath191 . the two local solutions around the horizon for @xmath192 : the modulus ( solid ) and the real part ( dotted ) are shown.,title="fig:",scaledwidth=40.0% ] the two local solutions around the horizon for @xmath192 : the modulus ( solid ) and the real part ( dotted ) are shown.,title="fig:",scaledwidth=40.0% ] around the irregular singularity at spatial infinity we can construct the thom solutions , using the characteristic exponents of second kind in the third column of the grs ( [ eq : grs_ef ] ) . the result is @xmath193 and the coefficients @xmath58 and @xmath59 can be calculated using recurrence relations obtained from the differential equation , again . we labeled all the solutions such that we get the same scheme ( [ fig : inout_schwarzschild ] ) as in the schwarzschild case before : the regular solution at the horizon @xmath194 describes waves that propagate into the horizon , while the other solution @xmath190 is associated with waves that emerge from the horizon and propagate to larger radii . at the irregular singularity at spatial infinity the two solutions @xmath195 describe in- and outgoing waves as well : the first solution @xmath173 describes waves that propagate into the singularity ( outgoing ) and the second solution @xmath175 is associated with waves that emerge from spatial infinity and propagate to smaller radii ( ingoing ) . it is quite unpleasant that the two local solutions @xmath196 in ( [ eq : ef_frobenius1 ] ) and ( [ eq : ef_frobenius2 ] ) have such a small radius of convergence which makes them useful only in the region @xmath197 . this is a consequence of the fact that the only known representation of the confluent heun function is the one in terms of a frobenius series . when we consider scattering of radiation at a black hole or when we are interested in its quasi - normal modes we have to impose boundary conditions at spatial infinity , or at least at a sufficiently large radius , and therefore the representation of the confluent heun function @xmath5 as a frobenius series is not appropriate . in this section we will work out a procedure that is related to a similar problem considered by jaff @xcite . he calculated the bound states and energy eigenvalues of the hydrogen molecule ion . actually , the same idea was applied by leaver @xcite to obtain a continued fraction expression for quasi - normal modes of black holes . lay @xcite and slavyanov and lay @xcite gave a general guide for applying the method of jaff to any kind of heun equation , including all the confluent cases , and they worked out what is called the central two point connection problem ( ctcp ) in great detail . the derivations in this section are based on the work of slavyanov and lay and we will put our radial equation in such a form that the results in can be used . the question of interest is now : is it possible to `` connect '' the local solutions at the horizon and those at spatial infinity and obtain expressions that are valid on a larger domain or cover even the entire positive real axis as the domain of convergence ? the basic idea is to `` glue '' frobenius- and thom - type solutions together . in this way we encounter a special case of the central two point connection problem . to solve this problem we will apply the following five step - procedure : 1 . shift the regular singularity and reorder the differential equation . 2 . perform an s - homotopic transformation of the dependent variable that involves the indicial exponent of the respective frobenius solution at the horizon and the characteristic thom exponents ( of order zero and one ) at spatial infinity . 3 . mbius - transform the independent variable such that the regular singularity remains unchanged at the origin , but the irregular singularity is moved to the boundary of the unit circle . 4 . solve the resulting differential equation using a series expansion around the origin that converges on the open unit circle . 5 . construct , thereupon , the total solution of the original radial equation by `` inverting '' the previous transformations . we use the transformation @xmath198 that maps the regular singularity from @xmath63 to @xmath21 , while the irregular singularity remains at @xmath10 . here it is important that no other singularity is located between those two on the positive real axis . this yields the differential equation @xmath178{}{z } + \left ( \dfrac{a_2}{z+z _ * } + \dfrac{a_1}{z } + g_0 \right ) \diff{}{z } \right . + \left ( \dfrac{c_2}{z+z _ * } + \dfrac{c_1}{z } \right ) \right ] y_{\omega s l}(z ) = 0\end{gathered}\ ] ] with the parameters @xmath199 and is done to match the conventions used in @xcite . we perform an s - homotopic transformation of the dependent variable @xmath200 that contains indicial exponents at @xmath21 and characteristic exponents of the second kind ( order zero and order one ) at @xmath10 . this transformation is given by @xmath201 where the exponents @xmath202 have the following meaning : * @xmath203 is the indicial exponent of the chosen frobenius solution at @xmath44 . thus , we can have either @xmath204 or @xmath205 according to the grs . * @xmath206 is the characteristic thom exponent of order one and depends on the chosen thom solution at @xmath10 . hence , we can have either @xmath207 or @xmath208 . * for large @xmath9 we get @xmath209 and we have to adjust the sum @xmath210 to be the characteristic exponent of order zero of the chosen thom solution . hence , the possible values for @xmath211 are @xmath212 or @xmath213 . the scheme in fig . ( [ fig : exponents ] ) shows which combination of the three exponents must be taken for the desired choice of frobenius and thom solutions to be connected . a scheme showing the proper choice of the exponents @xmath214 and @xmath211 related to local solutions for the connection problem.,scaledwidth=45.0% ] next , we apply the mbius transformation @xmath215 that maps the irregular singularity from @xmath10 to @xmath216 and leaves the regular singularity at @xmath217 . this yields a new differential equation for @xmath218 , that is @xmath219{\xi_{\omega s l}(x)}{x } + \sum_{k=1}^3 \omega_k x^k \ , \diff{\xi_{\omega s l}(x)}{x } \\ + \sum_{k=1}^2 \delta_k x^k \ , \xi_{\omega s l}(x ) = 0 \ , , \end{gathered}\ ] ] where the coefficients @xmath220 and @xmath221 are given by @xmath222 we use a series expansion ansatz for the unknown function @xmath223 @xmath224 and insert this ansatz into the differential equation . thus , we obtain a three term recurrence relation for the series coefficients @xmath225 @xmath226 where @xmath227 we have @xmath228 the solution ( [ eq : ctcp_ansatz ] ) with the recurrence relation ( [ eq : ctcp_recursion ] ) is absolutely convergent on the open unit circle around @xmath217 , i.e. , the radius of convergence is the distance between the two considered singularities at @xmath217 and @xmath216 . choose any @xmath229 with @xmath230 , then @xmath231 and convergence for @xmath230 is assured @xcite . however , uniform convergence will strongly depend on the properties of the coefficients @xmath232 , since @xmath233 = \lim_{k\to \infty } \left| \dfrac{\xi_{k+1}}{\xi_k } \right ] \ , .\end{aligned}\ ] ] hence , the solution is uniformly convergent if @xmath234 exists and is finite , which will depend on the values of @xmath235 and @xmath132 . the sum will be finite and , thus , uniform convergence will be assured if @xmath232 is a minimal solution of the recurrence relation ( [ eq : ctcp_recursion ] ) ( see , e.g. , * ? ? ? * ; * ? ? ? finally , we can construct the total solution to the radial equation in ef coordinates ( [ eq : radialef ] ) . we have to use @xmath236 to recover the original radius variable @xmath65 . thereupon , we obtain @xmath237 the exponents @xmath238 must be chosen according to the scheme in fig . ( [ fig : exponents ] ) and the requirements of the physical situation that is to be described . the solution ( [ eq : ctcp_solution ] ) is absolutely convergent on the open interval @xmath2390.5,\infty[$ ] owing to the absolute convergence of @xmath218 for @xmath240 and the mapping @xmath241 as explained in fig . ( [ fig : mapping ] ) . a scheme showing the mapping from @xmath229 to @xmath65 and the solutions @xmath177 and @xmath223 together with their domains of convergence.,scaledwidth=40.0% ] in the entire region where the frobenius solutions and the analytically continued solutions overlap , they do coincide . hence , we can interpret our solutions ( [ eq : ctcp_solution ] ) as new representations of the confluent heun function @xmath5 with larger domain of convergence . in the next section we show how to use these solutions for different physical applications . as an example , we consider the scattering of a massless scalar field at a schwarzschild black hole . numerical studies of such scattering processes can be found for example in @xcite and @xcite . in analogy to the usual scattering theory used in quantum mechanics and nuclear physics we consider an ingoing plane wave coming from spatial infinity as initial condition . such a plane scalar wave is in schwarzschild coordinates for @xmath168 given by @xcite @xmath242 this is a superposition of partial waves with angular momentum index @xmath132 . each partial wave will be reflected and/or absorbed by the black hole in a different manner and this will change the composition of the total wave field since different modes will be modified with different amplitudes that refer to scattering and absorption coefficients . this can be best understood by regarding the rwe ( [ eq : rwe ] ) and the involved potential . for a given frequency @xmath243 of the incoming wave the potential depends on the wave index @xmath132 as shown in fig . ( [ fig : rwepotential2 ] ) . as long as the energy @xmath127 of the incoming wave is larger than the potential peak for a given @xmath132-mode , this mode will be ( mostly ) absorbed by the black hole . when the energy becomes smaller than the potential peak for a fixed @xmath132-mode , this mode will be ( mostly ) reflected by the potential barrier . for the special case @xmath244 we expect the transmission and reflection probabilities to be @xmath245 . to derive the transmission and absorption probabilities we must use the solution ( [ eq : ctcp_solution ] ) with the exponents chosen to be @xmath246 to match the causal boundary condition , i.e. , to ensure purely ingoing waves at the horizon . the full wave field in ef coordinates is then given by @xmath247 we can , of course , transform this solution to usual schwarzschild coordinates simply by using @xmath248 . this might be done since the plane wave is given in schwarzschild coordinates as well . ( the other option is to give the plane wave in ef coordinates . ) thus , we have @xmath249 and we know that in the limit @xmath168 the asymptotic form must be @xmath250 now , we can proceed as in usual quantum mechanical scattering theory : at large distances the total wave field shall be the sum of the incoming plane wave and a back - scattered part due to reflection at the black hole @xmath251 and we define the complex valued phase shifts @xmath150 by @xmath252 we can show that this implies @xmath253 hence , we have to calculate the transmission and reflection amplitudes @xmath254 to obtain the phase shifts . we can do so by matching our solution ( [ eq : scattering_wavefield ] ) with the radial function given by to the asymptotic form ( [ eq : scattering_wavefield_asympt ] ) . even though the radial function in the form ( [ eq : scattering_radialfunction ] ) is not in general convergent in the limit @xmath168 for an arbitrary choice of the parameter @xmath235 and @xmath132 we can fit it to the asymptotic behavior on an interval for large values of @xmath65 and for fixed mode number @xmath132 and frequency @xmath243 . with this procedure we can calculate the phase shifts in a semi - analytical way for all desired combinations of the parameters . using the phase shifts or the reflection and transmission amplitudes we can calculate the transmission and reflection coefficients @xmath144 they describe the probability of being transmitted through the potential barrier , i.e. , being absorbed by the black hole and the probability of being reflected back to spatial infinity . the transmission coefficients @xmath255 are usually called graybody factors , since they are important when we derive the deviations from a thermal spectrum of hawking radiation . figure ( [ fig : graybody ] ) shows the calculated graybody factors as functions of the frequency for different wave modes . graybody factors as functions of the frequency for @xmath133 ( thick solid ) up to @xmath256 ( thin dashed).,scaledwidth=40.0% ] using the graybody factors we can , for example , calculate the total absorption cross section as a sum over all partial contributions @xmath257 in fig . ( [ fig : partialabsorption ] ) we show the partial absorption cross sections for different values of the wave mode starting from @xmath133 ( on the left ) to @xmath256 . the results are normalized by the horizon area . the figure further shows the total absorption cross section as a sum over all partial contributions . the result is normalized by the geometrical optics value of the cross section @xmath258 , which is recovered in the limit of high frequencies as it should be . for small frequencies @xmath259 the partial absorption cross section for the @xmath260 mode ( lowest mode ) approaches the horizon area of the black hole and therefore overcomes the problem shown in figure ( 7 ) in the phase - integral study by andersson @xcite . for higher frequencies our results match those given by andersson and our results in fig . ( [ fig : partialabsorption ] ) are in agreement with those shown in figures ( 2 ) and ( 3 ) in the early work by sanchez @xcite . top : partial absorption cross sections normalized by the horizon area @xmath261 for wave modes starting from @xmath133 ( left ) up to @xmath256 ; bottom : total absorption cross section as a sum over all contributions normalized by the geometrical optics value @xmath262.,scaledwidth=40.0% ] we can use the radial solutions ( [ eq : ctcp_solution ] ) to obtain the quasi - normal mode frequencies of a schwarzschild black hole for a spin @xmath185 perturbation . to ensure the proper qnm boundary conditions we have to choose , according to the scheme in fig . ( [ fig : exponents ] ) , the exponents to be @xmath263 and we further demand the solution to converge in the limit @xmath264 . the convergence property is fulfilled if the series coefficients @xmath225 are minimal solutions of the recurrence relation . this immediately yields the continued fraction equation that was given by leaver @xcite and the qnm frequencies @xmath265 can be calculated as complex roots of this equation . hence , we see that the method of leaver fits perfectly into our framework of continued local frobenius solutions . for details on quasi - normal mode frequencies the reader is referred to refs . @xcite and @xcite and to the living reviews @xcite . using the method of damour and ruffini @xcite we can calculate the spectrum of the hawking radiation emitted by a schwarzschild black hole . now , we have to consider that part of the solution to the radial equation which was always neglected up to now : we need to use the second solution @xmath266 that is not regular at the horizon . this solution is given by ( [ eq : ef_frobenius2 ] ) and describes purely outgoing waves at the horizon @xmath63 : this is the outgoing hawking radiation that is obtained here as the solution of a classical wave equation on curved background geometry . close to the horizon this solution behaves like @xmath267 following damour and ruffini , there exists a unique continuation of this solution for radii @xmath268 , which yields @xmath269 this continuation then describes an antiparticle of negative energy inside going down the black hole and a particle outside emerging from the horizon . the probability for transmission through the horizon is given by the squared modulus of the relative amplitude @xmath270 using feynman s method @xcite of scattering amplitudes one can deduce that this is also the probability for antiparticle - particle pair production just outside @xmath63 . in @xcite sannan showed how to get the mean number of particles that are radiated to spatial infinity in a given mode from the pair production probability @xmath271 . the result is @xmath272 where the @xmath273 is for fermionic particles and the @xmath274 must be used for bosons . the bosonic result is exactly hawking s result @xcite that was obtained using calculations involving the gravitational collapse . when we consider bosons , the spectrum is thermal except for the modification caused by the transmission probability ( or graybody factor ) @xmath255 , which causes an amplitude filtering of different modes . hence , the total spectrum is not really thermal anymore but is usually called graybody radiation ( instead of black - body radiation with thermal spectrum ) . from ( [ eq : hawkingradiationmeannumber ] ) we can derive a power spectrum of the emitted radiation . each massless boson takes away an energy of @xmath243 from the black hole and the change of energy per unit of time and frequency interval is minus the radiated power which is given by @xmath275 here , @xmath276 denotes additional degrees of freedom like the polarization of the particle @xcite . it should be mentioned that this radiation spectrum is valid for a time in which we consider the background geometry to remain unchanged , i.e. , the mass of the black hole is taken to be constant during this time interval . then , we may iterate and consider the mass to be decreased due to the evaporated energy and repeat the calculation of the radiated power with the decreased mass . in this way one can derive the mass of the black hole as a function of time and calculate the life time of black holes with given initial mass . power spectrum of hawking radiation emitted by a schwarzschild black hole for massless bosons with spin @xmath4 ( dashed ) , @xmath3 ( dotted ) , @xmath2 ( dash - dotted ) compared to a thermal spectrum for a cross section of @xmath258 ( solid).,scaledwidth=30.0% ] figure ( [ fig : hawking ] ) shows the obtained power spectrum according to ( [ eq : hawkingradiation_spectrum ] ) for massless bosons with spin @xmath277 . the results in fig . ( [ fig : hawking ] ) match previous results , for example those obtained by page @xcite using purely numerical calculations . the major contributions in the spectra in fig . ( [ fig : hawking ] ) are due to the lowest @xmath278 modes for each spin , respectively . the little `` bumps '' that we can see at higher frequencies are due to higher mode contributions . for a fixed spin the maxima of higher mode contributions are shifted to higher frequencies but their peak values decrease rapidly with increasing mode number . the regge - wheeler equation is usually written in terms of the tortoise coordinate . however , it can also be written in other coordinate systems , such as usual schwarzschild , eddington - finkelstein , painlev - gullstrand or kruskal - szekeres coordinates . while it is not possible to obtain regular solutions at the black - hole horizon in schwarzschild coordinates ( see the comment following eq . ) , we have shown that such solutions can be derived in ef coordinates . this is also possible in pg coordinates where , however , the expressions are somewhat more involved . these solutions are given in terms of the standard confluent heun function around the black hole horizon which is defined in terms of frobenius series . as these series converge on the unit sphere in the complex plane centered at @xmath0 , they can not be evaluated at larger radii . we have shown how this problem can be overcome by using an analytical continuation procedure . this gave us analytical solutions in terms of convergent series on the entire domain up to ( but not in general including ) infinity . based on these exact solutions we have re - obtained , in a semi - analytic fashion , several results on black - hole scattering , quasi - normal modes and hawking radiation that had been found numerically earlier . in a follow - up paper we consider a massive scalar field in ef coordinates which allows us to define purely ingoing solutions that penetrate the horizon . such a massive scalar field is a possible dark matter candidate and analytical solutions might be of interest in this area . in the same follow - up paper we further re - consider massless fields of spin @xmath4 , 1 or 2 but switch to the pg coordinates which are associated with infalling observers and therefore , at least in principle , with possible measurements . + the authors would like to thank domenico giulini , norman grlebeck and claus lmmerzahl for valuable discussions during the preparation of the manuscript . support from the deutsche forschungsgemeinschaft within the research training group 1620 `` models of gravity '' is gratefully aknowledged . the first author is financially supported by the german collaborative research center 1128 `` geo - q '' . 36ifxundefined [ 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 _ _ , fundamental theories of physics ( book 96 ) ( , ) @noop * * , ( ) http://stacks.iop.org/0264-9381/23/i=7/a=015 [ * * , ( ) ] @noop * * ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) , ed . , @noop _ _ ( , ) @noop _ _ ( , ) http://www.maplesoft.com/support/help/maple/view.aspx?path=heunc [ `` , '' ] ( ) http://eudml.org/doc/148221 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) link:\doibase 10.1140/epjc / s10052 - 011 - 1831-y [ * * , ( ) , 10.1140/epjc / s10052 - 011 - 1831-y ] @noop * * ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevlett.102.231103 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevd.18.1030 [ * * , ( ) ] link:\doibase 10.1103/physrevd.73.064030 [ * * , ( ) ] link:\doibase 10.1103/physrevd.69.044004 [ * * , ( ) ] @noop * * ( ) link:\doibase 10.12942/lrr-1999 - 2 [ * * ( ) , 10.12942/lrr-1999 - 2 ] @noop * * ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( )

the paper preposes a new scheme to promote the robustness of 3d structure and motion factorization from uncalibrated image sequences . first , an augmented affine factorization algorithm is proposed to circumvent the difficulty in image registration with imperfect data . then , an alternative weighted factorization algorithm is designed to handle the missing data and measurement uncertainties in the tracking matrix . finally , a robust structure and motion factorization scheme is proposed to deal with outlying and missing data . the novelty and main contribution of the paper are as follows : ( i ) the augmented factorization algorithm is a new addition to previous affine factorization family for both rigid and nonrigid objects ; ( ii ) it is demonstrated that image reprojection residuals are in general proportional to the error magnitude in the tracking data , and thus , the outliers can be detected directly from the distribution of image reprojection residuals , which are then used to estimate the uncertainty of inlying measurement ; ( iii ) the robust factorization scheme is proved empirically to be more efficient and more accurate than other robust algorithms ; and ( iv ) the proposed approach can be directly applied to nonrigid scenarios . extensive experiments on synthetic data and real images demonstrate the advantages of the proposed approach . computer vision , structure from motion , robust factorization , weighted factorization , reprojection residual , outlier detection .

1609.02638

from motion ( sfm ) is the process to find three - dimensional ( 3d ) structure and camera motion parameters from a set of 2d image sequences . the classical method for 3d reconstruction is stereo vision using two or three images @xcite . stereo vision , due to limited information , is sensitive to image noise . given a sequence of images , structure from motion is a powerful method to build a consistent 3d map with the knowledge of multiple - view geometry . over the past two to three decades , tremendous progress has been made in sfm @xcite@xcite@xcite@xcite@xcite , the results of this research have a wide range of potential applications , including robot navigation and obstacle avoidance , autonomous driving , video surveillance , and environment modeling . structure and motion factorization algorithm , pioneered by tomasi and kanade @xcite , is an effective approach for sfm . based on a bilinear formulation using singular value decomposition ( svd ) with low - rank approximation , the method decomposes image measurement directly into the 3d structure and the camera motion components , given a set of tracked features across the sequence . by dealing uniformly with the data from all images , the algorithm achieves a more robust and more accurate solution than stereo vision - based methods @xcite@xcite@xcite@xcite . a linear affine camera has been adopted by most research in sfm due to its simplicity @xcite . it was extended to a more accurate nonlinear perspective camera model in @xcite by incrementally performing the affine factorization of a scaled tracking matrix . triggs @xcite proposed a full projective factorization algorithm with projective depths recovered from epipolar geometry . the method was further studied and different iterative schemes were proposed to recover the projective depths by minimizing image reprojection errors @xcite . oliensis and hartley @xcite provided a complete theoretical convergence analysis for the iterative extensions . because full perspective model is computational intensive , a quasi - perspective model was proposed in @xcite as a trade - off of efficiency and accuracy . by assuming deformation constraints that the nonrigid 3d structure lies in a span of rigid bases , the factorization algorithm was extended to handle nonrigid deformation @xcite , where the shape bases , combination coefficients , and camera motions are solved simultaneously in a svd framwork . this idea has been extensively investigated and developed in @xcite@xcite@xcite@xcite . a manifold - learning framework was proposed in @xcite to relax the deformation assumption . @xcite proposed a sequential approach for nonrigid sfm . yan and pollefeys @xcite developed a similar factorization framework to recover the structure and motion of articulated objects . in a dual trajectory space , @xcite proposed a duality solution to this problem based on basis trajectories . most factorization algorithms are based on the svd decomposition of the tracking matrix composed by all features tracked across the sequence . in the presence of missing data , however , svd factorization could not be performed directly . different alternative factorization approaches have been proposed to handle incomplete data , such as power factorization @xcite , alternative factorization @xcite , and factor analysis @xcite . in practical application , the tracked features are usually corrupted by outliers or larger errors , in this case , most algorithms will degrade or even fail . the most popular strategies to handle outliers are random sample consensus ( ransac ) @xcite and its variations , least median of squares ( lmeds ) @xcite , and other similar framework based on hypothesis - and - test @xcite . most of these methods are computational intensive and only work well for two or three views . in recent years , some robust structure and motion factorization algorithms have been proposed @xcite@xcite@xcite@xcite . a scalar - weighted svd algorithm was proposed by aguitar and moura @xcite through minimizing weighted square errors . gruber and weiss @xcite enhanced the robustness to missing data and uncertainties using a factor analysis in an expectation maximization ( em ) framework . zelnik - manor _ @xcite defined a new type of motion consistency based on temporal consistency , and applied it to multi - body factorization with directional uncertainty . a gaussian mixture model with the em algorithm was introduced by zaharescu and horaud @xcite . @xcite proposed to correct the outliers with pseudo observations through iterations . ke and kanade @xcite dealt the outliers by minimizing a l1 norm of the reprojection errors . eriksson and hengel @xcite utilized the l1 norm to the wiberg algorithm to handle missing and outlying features . alternatively , okatani _ @xcite solved the problem by introducing a damping factor into the wiberg algorithm . more recently , yu _ @xcite presented a quadratic program formulation for robust multi - model fitting of geometric structures . @xcite proposed an adaptive kernel - scale weighted hypotheses to segment multiple - structure data with a large number of outliers . an alternating bilinear approach was proposed by paladini _ _ @xcite to solve nonrigid structure from motion by introducing a globally optimal projection step of the motion matrices onto the manifold of metric constraints . _ @xcite designed a spatial - and - temporal - weighted factorization algorithm to deal with significant noise in the measurement . @xcite developed an optimal approach based on branch - and - bound . in this paper , by exploring the fact that the reprojection residuals are in general proportional to measurement errors of the tracking data , we propose to handle the outlying data through a new viewpoint via the distribution of image reprojection residuals . the proposed approach is based on a new augmented factorization formulation , which circumvents the problem of image registration in the presence of missing data and outliers . an alternative weighted factorization algorithm is developed to handle the missing data and measurement uncertainties . finally , a robust factorization scheme is proposed to handle outlying and missing data in both rigid @xcite and nonrigid @xcite structure and motion recovery . in addition , the proposed scheme can be directly applied to handle nonrigid structure and motion factorization . the remainder of this paper is organized as follows . some background on affine factorization is offered in section [ sec : background ] . the augmented factorization algorithm is elaborated in section [ sec : rank4 ] . section [ sec : pf ] presents the alternative factorization algorithm for incomplete data . an outlier detection scheme and the robust factorization algorithm are discussed in section [ sec : outliers ] . section [ sec : nonrigid ] discusses the extension to nonrigid factorization . experimental evaluations and comparisons on synthetic and real images are described in sections [ sec : experiment1 ] and [ sec : experiment2 ] , respectively . finally , a short conclusion is drawn in section [ sec : conclusions ] . under perspective projection , a 3d point @xmath0^{t}$ ] is projected onto an image point @xmath1^{t}$ ] in frame @xmath2 according to the imaging equation @xmath3\mathbf{\tilde{x}}_j\ ] ] where @xmath4 is a non - zero scale factor ; @xmath5 and @xmath6 are the homogeneous form of @xmath7 and @xmath8 , respectively ; @xmath9 is a @xmath10 projection matrix of the @xmath2-th frame ; @xmath11 is the camera calibration matrix ; @xmath12 and @xmath13 are the corresponding rotation matrix and translation vector of the camera with respect to the world system . when the object is far away from the camera with relatively small depth variation , we may assume a simplified affine camera model as below to approximate the perspective projection . @xmath14 where @xmath15 is a @xmath16 affine projection matrix of the @xmath2-th camera ; @xmath17 is a two - dimensional translation term of the frame . under the affine assumption , the mapping from the space to the image becomes linear as the unknown depth scalar @xmath4 in ( [ eq : perspective_projection ] ) is eliminated in ( [ eq : affine_projection1 ] ) . consequently , the projection of all image points in the @xmath2-th frame can be denoted as @xmath18 = \mathbf{a}_{i}[\mathbf{{x}}_1 , \mathbf{{x}}_2 , \cdots , \mathbf{{x}}_n ] + \mathbf{c}_i\ ] ] where @xmath19 $ ] is the translation matrix of the frame @xmath2 . therefore , the imaging process of an image sequence can be formulated by stacking equation ( [ eq : ortho_i ] ) frame by frame . @xmath20}_{\mathbf{w}_{2m\times n } } = \underbrace{\left[{\begin{smallmatrix } \mathbf{a}_{1 } \\ \vdots \\ \mathbf{a}_{m } \\ \end{smallmatrix } } \right]}_{\mathbf{m}_{2m\times 3 } } \underbrace{\left[{\begin{smallmatrix } \mathbf { { x}}_1 , & \cdots , & \mathbf { { x}}_n \end{smallmatrix } } \right]}_{\mathbf{{s}}_{3\times n } } + \underbrace{\left[{\begin{smallmatrix } \mathbf{c}_{1 } \\ \vdots \\ \mathbf{c}_{m } \\ \end{smallmatrix } } \right ] } _ { \mathbf{c}_{2m\times n}}\ ] ] where @xmath21 is the frame number and @xmath22 is the number of features . it is easy to verify that @xmath17 in ( [ eq : affine_projection1 ] ) is the image of the centroid of all space points . thus , if we register all image points in each frame to the respective centroid and adopt relative image coordinates with respect to the centroid , the translation term will vanish , i.e. , @xmath23 and @xmath24 . consequently , the imaging process ( [ eq : affine_rigid_fact1 ] ) is written concisely as @xmath25 where @xmath26 is called the tracking matrix or measurement matrix which is composed by all tracked features . structure from motion is a reverse problem of the imaging process . suppose we have the tracking matrix @xmath26 , our purpose is to recover the motion matrix @xmath27 and the shape matrix @xmath28 . the tracking matrix is highly rank - deficient , as seen from the right side of ( [ eq : affine_rigid_fact1 ] ) , the rank of @xmath26 is at most 3 if the translation term @xmath29 vanishes . in practice , however , the rank of a real tracking matrix is definitely greater than 3 due to image noise and the affine approximation error . thus , we need to find a rank-3 approximation of the tracking matrix . a common practice is to perform svd decomposition on the tracking matrix and truncate it to rank 3 , then , the motion matrix @xmath27 and the shape matrix @xmath28 can be easily decomposed from the rank-3 approximation of @xmath26 . it is obvious that this decomposition is not unique since it is only defined up to a nonsingular linear transformation matrix @xmath30 as @xmath31 since we are interested in the structure and motion parameters in the euclidean space , for this purpose , we need to find a transformation matrix @xmath32 to upgrade the solution to the euclidean frame . the metric constraint on the motion matrix is usually adopted to recover this upgrading matrix @xmath32 @xcite@xcite . then , the euclidean structure is recovered from @xmath33 and the camera motion parameters of each frame are decomposed from @xmath34 . previous studies on affine factorization of rigid objects are based on the formulation of ( [ eq : sim_affine ] ) due to its simplicity . one necessary condition of the rank-3 affine factorization is that all image measurements should be registered to the corresponding centroid of each frame . however , when some tracked features are missing , or contaminated with outliers and significant noise , it is impossible to reliably retrieve the centroid . as will be shown in the experiments , the misalignment of the centroid will cause a significant error to the final solution . previous studies were either simply ignoring this issue or hallucinating the missing features with pseudo observations , which may lead to a biased solution . in this section , we propose an augmented factorization algorithm to circumvent this problem . let us rewrite the affine imaging process ( [ eq : affine_projection1 ] ) as below . @xmath35\mathbf{{\tilde{x}}}_j\ ] ] where the space point @xmath36 is denoted by a 4-dimensional homogeneous form as @xmath37^t$ ] . then , the projection for the entire sequence is formulated as follows . @xmath38}_{\mathbf{w}_{2m\times n } } = \underbrace{\left[{\begin{smallmatrix } \mathbf{a}_{1 } & | & \mathbf{c}_{1 } \\ \vdots & | & \vdots\\ \mathbf{a}_{m } & | & \mathbf{c}_{m } \\ \end{smallmatrix } } \right]}_{\mathbf{m}_{2m\times 4 } } \underbrace{\left[{\begin{smallmatrix } \mathbf { { \tilde{x}}}_1 , & \cdots , & \mathbf { { \tilde{x}}}_n \end{smallmatrix } } \right]}_{\mathbf{{s}}_{4\times n}}\ ] ] which can be written concisely as @xmath39 compared to the rank-3 factorization , the motion matrix in ( [ eq : affine_rigid_fact4 ] ) is augmented by an extra column , while the shape matrix is augmented by an extra row . as a result , the rank of the tracking matrix becomes four in this case . thus , given the tracking matrix , the factorization can be simply obtained via svd decomposition by imposing the rank-4 constraint . we call the equation ( [ eq : affine_rigid_fact4 ] ) augmented factorization . it is obvious that the expression ( [ eq : affine_rigid_fact4 ] ) is derived directly from the affine projection model ( [ eq : affine_projection1 ] ) , which does not require image registration with respect to the centroid . therefore , it is applicable to imperfect data with significant noise , missing entries , and outlying points . both factorization algorithms ( [ eq : sim_affine ] ) and ( [ eq : affine_rigid_fact4_concise ] ) can be equivalently denoted as the following minimization scheme . @xmath40 by enforcing different rank constraints , the residual errors corresponding to the algorithms ( [ eq : sim_affine ] ) and ( [ eq : affine_rigid_fact4_concise ] ) would be @xmath41 respectively , where @xmath42 are singular values of the tracking matrix in descending order , and @xmath43 . it is obvious that the error difference by the two algorithm is @xmath44 . for noise free data , if all image points are perfectly registered to the corresponding centroids of every views and the origin of the world system is set at the gravity center of the space points , i.e. , @xmath45 . then , the last column of the motion matrix in ( [ eq : affine_rigid_fact4 ] ) vanishes since @xmath46 , and the expression ( [ eq : affine_rigid_fact4_concise ] ) is equivalent to the rank-3 factorization ( [ eq : sim_affine ] ) . thus , ( [ eq : sim_affine ] ) is a special case of the augmented factorization after registration to the centroid . nonetheless , since the image centroid can not be accurately recovered due to the presence of outlying and missing data , the rank-@xmath47 algorithm will yield a big error as @xmath48 does not approach zero in this situation . suppose the rank-4 decomposition of ( [ eq : affine_rigid_fact4 ] ) yields a set of solutions @xmath49 . the decomposition is not unique since it is defined up to a nonsingular linear transformation @xmath50 as @xmath51 and @xmath52 . the recovery of the metric upgrading matrix , as discussed below , is different with that in the rank-3 factorization . the upgrading matrix @xmath32 is a @xmath53 nonsingular matrix which can be denoted as the following form . @xmath54\ ] ] where @xmath55 denotes the first three columns of @xmath32 , and @xmath56 is the fourth column , i.e. , the last column of @xmath32 . suppose @xmath57 is the @xmath2-th two - row submatrix of @xmath58 , then the upgraded motion matrix can be written as @xmath59 = \left [ \mathbf{a}_i | \mathbf{{c}}_i \right]\ ] ] the left @xmath60 submatrix of @xmath61 in ( [ eq : mi ] ) can be written as @xmath62\ ] ] where @xmath63 is the focal length of cameras , @xmath64 and @xmath65 are the first two rows of the camera rotation matrix . let us denote @xmath66 , then , @xmath67 is constrained from ( [ eq : mhl ] ) as @xmath68\ ] ] the above equation provides two independent constraints to @xmath67 , which is a @xmath69 positive semidefinite symmetric matrix with nine degree - of - freedom since it is defined up to a scale . thus , the matrix @xmath67 can be linearly solved via least squares given five or more images . furthermore , the submatrix @xmath55 can be decomposed from @xmath67 via extended cholesky decomposition as proved in @xcite . after recovering @xmath55 , the last column of the upgrading matrix is then determined straightforwardly . from the expression ( [ eq : mi ] ) , the projection equation ( [ eq : affine_projection4 ] ) can be written as @xmath70 it can be easily proved from ( [ eq : affine_projection4 m ] ) that the last column @xmath56 corresponds to the translation from the world coordinate system to the image system . under a given coordinate system , different values of @xmath56 will only result in a pure translation of the world system , however , it has no influence to the euclidean structure of the reconstructed object . thus , @xmath56 can be set freely as any 4-vector that is independent of the columns of @xmath55 so as to guarantee the nonsingularity of the resulted upgrading matrix . practically , @xmath56 may be constructed as follows . suppose the svd decomposition of @xmath55 is @xmath71 \left [ \begin { smallmatrix } \sigma_1 & 0 & 0 \\ 0 & \sigma_2 & 0 \\ 0 & 0 & \sigma_3\\ 0 & 0 & 0 \end{smallmatrix}\right ] [ \mathbf{v}_1,\mathbf{v}_2,\mathbf{v}_3]^t \nonumber\end{aligned}\ ] ] where @xmath72 and @xmath73 are two orthogonal matrices , and @xmath74 is a diagonal matrix of the three singular values . then , @xmath56 can be simply set as @xmath75 where @xmath76 is the second singular value , and @xmath77 is the last column of @xmath72 . the construction guarantees a good numerical stability in computing the inverse of @xmath32 , since the constructed matrix @xmath32 has the same condition number as @xmath55 . the above proposed augmented rank-4 affine factorization algorithm is summarized in algorithm [ alg : augmented ] . 1 . perform svd decomposition on the tracking matrix ; + 2 . obtain a set of rank-4 solutions of @xmath58 and @xmath78 ; + 3 . recover the metric upgrading matrix matrix @xmath32 ; + 4 . compute the euclidean solution from @xmath79 and @xmath80 . svd decomposition is a convenient technique for structure and motion factorization , however , svd only works when the tracking matrix is complete , i.e. , all features are tracked across the sequence . in practice , missing data are inevitable since some features may get lost during the process of tracking due to occlusion or other factors . researchers proposed different alternative factorization approaches to handle missing data @xcite . in this section , a two - step alternative and weighted factorization algorithms are introduced to handle missing data and image uncertainties . the essence of structure and motion factorization ( [ eq : affine_rigid_fact4 ] ) is equivalent to finding a set of rank-4 solutions @xmath27 and @xmath81 by minimizing the following frobenious norm . @xmath82 the basic idea of the two step factorization is to minimize the cost function ( [ eq : fact_fnormwhat ] ) over @xmath81 and @xmath27 alternatively until convergence , while leaving the other one fixed , i.e. , rcl f ( ) & = & _ -_f^2 [ eq : fact_pfw1 ] + f ( ) & = & _ -_f^2 [ eq : fact_pfw2 ] each cost function of the algorithm is indeed a convex function , and thus , a global minimum can be found . the algorithm converges fast if the tracking matrix is close to rank-4 , even with a random initialization . the idea has been adopted by several researches @xcite@xcite . different with the svd decomposition , the minimization process is carried out by least squares . let us rewrite the cost function ( [ eq : fact_pfw1 ] ) with respect to each feature as follows . @xmath83 where @xmath84 is the @xmath85-th column of @xmath26 , and @xmath86 is the @xmath85-th column of @xmath81 . thus , the least squares solution of @xmath81 is given as @xmath87 where @xmath88 is the moore - penrose pseudoinverse of matrix @xmath27 . the solution ( [ eq : lq_s2 ] ) can easily handle the missing data in the tracking matrix . for example , if some entries in @xmath84 is unavailable , one can simply delete those elements in @xmath84 and the corresponding columns in @xmath89 , or simply set those entries in @xmath84 to zeros , then , @xmath86 can still be solved from ( [ eq : lq_s2 ] ) via least squares . similarly , the second cost function ( [ eq : fact_pfw2 ] ) can be rewritten with respect to each frame as @xmath90 which yields the following least - square solution of the motion matrix @xmath27 . @xmath91 where the pseudoinverse @xmath92 , @xmath93 and @xmath94 denote the @xmath2-th row of the matrices @xmath27 and @xmath26 , respectively . in case of missing elements , one can simply set those entries in @xmath94 to zeros . the alternative algorithm is summarized in algorithm [ alg : alternative ] . measurement errors are inevitable in the process of feature detection and tracking . if prior knowledge about distribution of the errors is available , all elements of the approximation error can be weighted by taking account of the error distribution to increase the robustness and accuracy of the algorithm . the basic idea is to give each image measurement a weight according to its uncertainty . reliable features are assigned higher weights while unreliable features receive lower weights . the weighted factorization is formulated as follows . @xmath95 where @xmath96 denotes the hadamard product , which is an element - by - element multiplication of two matrices ; @xmath97 is an @xmath98 uncertainty matrix whose entries are weights derived from the confidence of the image measurements . the general weighted factorization could not be solved analytically using the singular value decomposition . many researchers have proposed different schemes to solve the problem @xcite@xcite@xcite . in this section , the solution of ( [ eq : fact_weightedall ] ) will be obtained using the alternative factorization algorithm by solving @xmath81 and @xmath27 alternatively as follows . rcl f ( ) & = & _ _ j_j(_j-_j)_f^2 [ eq : fact_pfwighted1 ] + f ( ) & = & _ _ i^t_i^t(_i^t-_i^t)_f^2 [ eq : fact_pfweighted2 ] where @xmath99 denotes the @xmath85-th column of @xmath100 and @xmath101 the @xmath2-th row of @xmath100 . the close - form solutions of the shape and motion matrices can be obtained by least squares . rcl _ j & = & ( ( _ j))^((_j)_j ) , j=1 , , n [ eq : fact_lswighted1 ] + _ i^t & = & ( _ i^t ( _ i^t ) ) ( ( _ i^t))^ , i=1 , , m [ eq : fact_lsweighted2 ] where @xmath102 denotes the pseudoinverse of a matrix , and @xmath103 stands for the diagonal matrix generated by a vector . equations ( [ eq : fact_lswighted1 ] ) and ( [ eq : fact_lsweighted2 ] ) yield the least - square solutions of @xmath81 and @xmath27 . same as the alternative factorization , when there are any missing data in the tracking matrix , one can simply set those entries in @xmath84 to zeros . the alternative weighted factorization algorithm is summarized in algorithm [ alg : weighted ] , where the motion matrix @xmath27 can be initialized randomly or using provious estimation , while the initial value of the weight matrix @xmath100 , as will be discussed in next section , is estimated from the reprojection residuals . based on the rank-4 factorization algorithm proposed in the foregoing sections . a fast and practical scheme for outlier detection is proposed in this section . outlying data are inevitable in the process of feature tracking . some most popular strategies in the computer vision field are based on the hypothesis - and - test scheme @xcite@xcite , which are computationally intensive . we will investigate the problem from a new viewpoint via the distribution of image reprojection residuals . both the svd - based and the alternative factorization - based algorithms yield a set of least - square solutions . the best fit model is obtained by minimizing the sum of the squared residuals between the observed data and the fitted values provided by the model . extensive experiments show that the least - square algorithms usually yield reasonable solutions even in the presence of certain amount of outliers and the reprojection residuals of these outlying data are usually outstandingly larger than those associated with inliers . suppose @xmath58 and @xmath78 are a solution set of the rank-4 factorization of a tracking matrix @xmath26 , the reprojection residuals can be computed by reprojecting the set of solutions back onto all images . let us define a residual matrix as follows . @xmath104_{2m\times n}\ ] ] where @xmath105\ ] ] is the residual of point @xmath106 in both directions . the reprojection error of the point is defined by the euclidean distance @xmath107 of the image point and its reprojection , and the reprojection error of the entire sequence is defined by an @xmath108 matrix @xmath109_{m\times n}\ ] ] fig.[fig : s1 ] shows an example of the distribution of the error matrix ( [ eq : error_matrix ] ) , where 40 images are generated from 100 random 3d space points via affine projection . the image resolution is @xmath110 , and the images are corrupted by gaussian noise and 10% outliers . the added noise level is 3 pixels , and the outliers are simulated by random noise whose level ( standard deviation of the noise ) is set at 15 pixels . the real added noise and outliers are illustrated by an image as shown in fig.[fig : s1 ] ( a ) , where the grayscale of each pixel corresponds to the inverse magnitude of the error on that point , the darker the pixel , the larger the error magnitude on that point . the distribution of the real added outliers is depicted as a binary image in fig.[fig : s1 ] ( c ) , which correspond to the darker points in fig.[fig : s1 ] ( a ) . ) ; ( c ) the distribution of the added outlying data ; ( d ) the outliers segmented from reprojection error by a single threshold ; ( e ) the distribution of false positive error given by the thresholding ; and ( f ) the false negative error given by the thresholding.,title="fig : " ] + ( a)(b ) + ) ; ( c ) the distribution of the added outlying data ; ( d ) the outliers segmented from reprojection error by a single threshold ; ( e ) the distribution of false positive error given by the thresholding ; and ( f ) the false negative error given by the thresholding.,title="fig : " ] + ( c)(d ) + ) ; ( c ) the distribution of the added outlying data ; ( d ) the outliers segmented from reprojection error by a single threshold ; ( e ) the distribution of false positive error given by the thresholding ; and ( f ) the false negative error given by the thresholding.,title="fig : " ] + ( e)(f ) + using the corrupted data , a set of motion and shape matrices were estimated by employing the rank-4 factorization algorithm and the error matrix was then computed . the distribution of the reprojection error ( [ eq : error_matrix ] ) is illustrated in fig.[fig : s1 ] ( b ) with each pixel corresponds to the reprojection error of that point . it is evident that the reprojection error and the real added noise have similar distribution . the points with large reprojection errors correspond to those with large noise levels . fig.[fig : s1 ] ( d ) shows the binary image of fig.[fig : s1 ] ( b ) by simply applying a global threshold to the reprojected residuals . it is obvious from the test that almost all outliers are successfully segmented by a single threshold . the distribution of false positive error ( the inlier points being classified as outliers by the given threshold ) and the false negative error ( the outliers not being detected by the thresholding ) are given in fig.[fig : s1 ] ( e ) and ( f ) , respectively . the false positive error is mainly caused by those inliers with large noise ( which should be treated as outliers ) , while the false negative error is caused by the outliers with small deviations ( which can actually be treated as inliers ) , these two types errors are related to the threshold , however , they will not make a big influence to the final solutions . inspired by this observation , an intuitive outlier detection and robust factorization scheme is proposed . the flowchart of the strategy is shown in fig.[fig : s2 ] , and the computation details is given in algorithm [ alg : robust ] . + 1 . normalize the tracking matrix via point - wise and image - wise rescalings , as in @xcite , to improve numerical stability ; 2 . perform rank-4 affine factorization on the tracking matrix to obtain a set of solutions of @xmath58 and @xmath78 ; 3 . estimate the reprojection residuals and determine a global threshold to segment the outliers from the distribution ; 4 . eliminate the outliers and recalculate the matrices @xmath58 and @xmath78 using the inlying data via algorithm [ alg : alternative ] ; 5 . estimate the uncertainty of each inlying feature from the distribution of the reprojection residuals ; 6 . refine the solutions by weighted factorization algorithm [ alg : weighted ] ; 7 . recover the metric upgrading matrix @xmath32 and upgrade the solutions to the euclidean space as @xmath51 and @xmath111 ; 8 . perform a global optimization via bundle adjustment . in algorithm [ alg : robust ] , steps 3 and 4 can be repeated for one more time to ensure a more refined inlying data and solutions . in practice , however , the repetition does not make much difference to the final results . during computation , the algorithms [ alg : alternative ] and [ alg : weighted ] are employed to handle the missing data and measurement uncertainties . concerning the initialization of the alternative algorithm , since an initial set of solutions have been obtained in the previous steps , these solutions can be used as initial values in the iteration so as to speed up the convergence of the algorithm . suppose the image noise is modeled by gaussian distribution , it can be verified that the reprojection residuals ( [ eq : residual_matrix ] ) also follow the same distribution as the image noise , while the reprojection errors ( [ eq : error_matrix ] ) follow @xmath112 distribution with codimention 2 . it is nature to assume that the noise at both coordinate directions in the image is independent and identically distributed ( i.i.d . ) . let @xmath113 be the @xmath114-vector formed by the residual matrix @xmath115 , then , @xmath113 should also be gaussian . fig.[fig : s3 ] shows an example , we add gaussian noise and outliers to a tracking matrix , as shown in fig.[fig : s3 ] ( a ) , then , we compute a set of solutions using the proposed augmented factorization and calculate the reprojection residuals , whose distribution is also gaussian , as shown in fig.[fig : s3 ] ( b ) . + ( a ) + + ( b ) + let @xmath116 and @xmath117 be the mean and standard deviation of @xmath113 , respectively , and register the residual vector @xmath113 with respect to its mean @xmath116 , then , the outlier threshold can be chosen as @xmath118 where @xmath119 is a parameter , which can be set from 3.0 to 5.0 ( the result is not sensitive to this value , and we choose 4.0 in our experiment ) . the points whose absolute values of the registered residuals in either direction , or the registered reprojection errors , are greater than @xmath120 will be classified as outliers , i.e. , rcl = & & \{_i , j ||u_ij-| > |v_ij-| > + & & ( ( u_ij-)^2+(v_ij-)^2)^ > } [ eq : outlier_se ] since the residual vector @xmath113 contains outliers , which have signifycant influence to the estimation of the mean and standard deviation ( std ) due to the large deviations of the outliers . in practice , the mean is estimated from the data that are less than the median value of @xmath121 . @xmath122 while the standard deviation is estimated from the median absolute deviation ( mad ) as @xmath123 since the mad is resistant to outliers . the above computation usually guarantees a robust estimation of the mean and the standard deviation . as for the weights in weighted factorization , most researchers estimate it from the uncertainty of each feature based on the information such as sharpness and intensity contrast around its neighborhood . some researchers modeled the errors isotropically with different variances @xcite@xcite ; others adopted directional distribution to describe the uncertainties @xcite@xcite . the uncertainty is usually estimated during the process of feature detection and tracking or given as prior information nonetheless , this information is unavailable in many practical applications . in our early study @xcite , it was shown that the uncertainty of each feature is generally proportional to the reprojection residual of that point . for example , from the structure and motion matrices computed at step 5 , the residuals of inlying data can be estimated . as depicted in fig.[fig : s4 ] , it is evident that the reprojection residuals have very similar distribution as the added noise . based on this observation , the weight of each point can be estimated from the residual value obtained from the data after eliminating the outliers . the points with higher residual values have larger uncertainties , and thus , lower weights are assigned . the weight of each point at each coordinate direction is determined as follows in a shape like normal distribution . @xmath124 where the standard deviation @xmath117 is estimated from equation ( [ eq : sigma ] ) , @xmath125 is the @xmath106-th element of the residual matrix ( [ eq : residual_matrix ] ) , and @xmath126 is a normalization scale . clearly , one point may have different weights at the two image directions based on the residual value at that direction . for the missing data and outliers , the corresponding weights are set as @xmath127 . direction.,title="fig : " ] + the preceding discussions are all based on the assumption that the scene is globally rigid or static . in case of nonrigid scenarios , we can follow bregler s assumption that represent the nonrigid structure as a linear combination of a set of rigid shape bases @xmath128 @xcite , i.e. , @xmath129 where @xmath130 is the combination weight ; and @xmath131 is the number of bases . under this assumption , the imaging process of one image can be modeled as @xmath132 = \mathbf{a}_{i}\mathbf{s}_i+[\mathbf{c}_{i } , \cdots , \mathbf{c}_{i}]\nonumber\\ & = & \left [ { \omega_{i1 } \mathbf{a}_{i } } , \cdots , { \omega_{ik } \mathbf{a}_{i } } \right ] \left[\begin{smallmatrix } \mathbf{b}_1 \\ \vdots \\ \mathbf{b}_k \\ \end{smallmatrix}\right]+ { [ \mathbf{c}_{i } , \cdots , \mathbf{c}_{i}]}\nonumber\end{aligned}\ ] ] similar to rigid factorization , if all image points in each frame are registered to the centroid and relative image coordinates are employed , the translation term vanishes , i.e. , @xmath23 . consequently , the nonrigid factorization under affine camera model is expressed as @xmath133}_{\mathbf{w}_{2m\times n } } = \underbrace{\left [ { { \begin{smallmatrix } { \omega_{11 } \mathbf{a}_1 } & \cdots & { \omega_{1k } \mathbf{a}_1 } \\ \vdots & \ddots & \vdots \\ { \omega_{m1 } \mathbf{a}_m } & \cdots & { \omega_{mk } \mathbf{a}_m } \\ \end{smallmatrix } } } \right]}_{\mathbf{m}_{2m\times 3k } } \underbrace{\left[\begin{smallmatrix } \mathbf{b}_1 \\ \vdots \\ \mathbf{b}_k \\ \end{smallmatrix}\right]}_{\mathbf{b}_{3k\times n}}\ ] ] it is obvious from ( [ eq : nonrigid_fact ] ) that the rank of the tracking matrix @xmath26 is at most @xmath134 . previous studies on nonrigid sfm are based on the rank-@xmath134 constraint due to its simplicity . however , the expression ( [ eq : nonrigid_fact ] ) is based on the same assumption as ( [ eq : sim_affine ] ) that all image measurements are registered to the corresponding centroid of each frame . obviously , the assumption is not valid in the presence of outlying and missing data . similar to the rigid case , we can employ a similar augmented formulation like ( [ eq : affine_rigid_fact4 ] ) to circumvent the registration step . by adopting homogeneous expression ( [ eq : affine_projection4 ] ) , the above nonrigid factorization can be expressed in the following augmented form . @xmath135}_{\mathbf{w}_{2m\times n } } = \underbrace{\left[{\begin{smallmatrix } { \omega_{11 } \mathbf{a}_{1 } } & \cdots & { \omega_{1k } \mathbf{a}_{1 } } & \mathbf{c}_{1 } \\ \vdots & \ddots & \vdots & \vdots\\ { \omega_{m1 } \mathbf{a}_{m } } & \cdots & { \omega_{mk } \mathbf{a}_{m } } & \mathbf{c}_{m } \\ \end{smallmatrix } } \right]}_{\mathbf{m}_{2m\times ( 3k+1 ) } } \underbrace{\left[\begin{smallmatrix } \mathbf{b}_1 \\ \vdots \\ \mathbf{b}_k \\ \mathbf{t}_i^t \end{smallmatrix}\right]}_{\mathbf{{b}}_{(3k+1)\times n}}\ ] ] it is obvious that the rank of the tracking matrix becomes @xmath136 in this case . given the tracking matrix , the shape and motion matrices can be easily obtained via svd decomposition by imposing rank-(@xmath136 ) constraint . the expression ( [ eq : affine_nonrigid_fact4 ] ) does not require any image registration , and thus , can directly work with outlying and missing data . based on the new formulation , the preceding proposed augmented factorization , alternative factorization , and weighted factorization algorithms can be directly extended to the nonrigid scenario . the only difference here with respect to the rigid case lies in the rank constraint applied to the tracking matrix . thus , a set of motion and structure matrices can be easily decomposed as shown in ( [ eq : affine_nonrigid_fact4 ] ) . obviously , the decomposition is not unique and we need to find a metric upgrading matrix @xmath137 to upgrade the solution to the euclidean space . then , the nonrigid structure and camera motion parameters can be factorized from @xmath27 and @xmath81 , respectively . a detailed discussion about this approach can be found in our early study @xcite . the proposed technique was validated and evaluated extensively on synthetic data and compared with other similiar algorithms in the literature . during the simulation , we produced 100 random space points within a cube of @xmath138 , and generated a sequence of 50 images from these points by affine projection . the following settings were used in the test : image resolution : @xmath110 pixel ; focal lengths : randomly varying from 500 to 550 ; rotation angles : randomly varying between @xmath139 and @xmath140 ; camera positions : randomly setting inside a sphere with a diameter of 40 ; average distance from the cameras to the object : 600 . these imaging conditions are very close to the assumption of affine projection . we compared the proposed rank-4 factorization algorithm with its rank-3 counterpart with respect to different image centroid displacements . during the test , different levels of gaussian white noise was added to the generated images and the centroid of each image was deviated with a displacement ; then , all imaged points were registered to the deviated centroid . this is a simulation of the situation when the centroid could not be reliably recovered due to missing and outlying data . using the misaligned data , we recover the motion and shape matrices using the svd factorization with the rank-4 and the rank-3 constraints , respectively ; then , reproject the solution back onto the images and calculate the reprojection residuals . in order to evaluate the performance of different algorithms , we calculate the difference between the ground truth and the corresponding back - projected features , we call the average of all these errors as the mean reprojection variance , which is defined as follows . @xmath141 where @xmath142 is the noise - free tracking matrix ; and @xmath58 and @xmath78 are the estimated motion and shape matrices , respectively . in order to obtain a statistically meaningful comparison , 100 independent tests were performed at each noise level . the mean reprojection variance with respect to different centroid displacements at different noise levels is shown in fig . [ fig : s4 ] , where the noise level is defined as the standard deviation of the gaussian noise . + ( a ) ( b ) + as shown in fig . [ fig : s4 ] , it is evident that the miscalculated centroid has no influence to the proposed rank-4 factorization algorithm , however , the centroid error has a huge impact to the performance of the rank-3 based algorithm . the test proves that the influence caused by the centroid errors is far more significant than that caused by the image noise . thus , the rank-4 affine factorization approach is a better choice in practice , especially in the presence of missing data , outliers , or large measurement errors . in this test , we evaluated and compared the performance of the proposed approach with respect to other robust factorization algorithms in terms of accuracy and computational complexity . we add gaussian noise to the above generated image sequence and vary the noise level from 1 to 5 pixels in steps of 1 pixel . in the mean time , 5% and 20% outliers were added to the tracking matrix , respectively . using the contaminated data , we recover the motion and shape matrices using the propose technique . the mean reprojection variance at different noise levels and outliers ratios is shown in fig . [ fig : s5 ] . as a comparison , two closely related algorithms in the literature were implemented as well , one is an outlier correction scheme proposed by huynh _ @xcite , the other one is proposed by ke and kanade @xcite based on @xmath143 minimization . + ( a ) ( b ) + the results in fig . [ fig : s5 ] were evaluated by 100 independent tests , where direct stands for the regular rank-4 factorization algorithm without outlier rejection . here the reprojection variance was estimated only from the original inlying data by eliminating the added outliers so as to provide a fair comparion of different approaches . it is obvious that the proposed scheme outperforms other algorithms in terms of accuracy . the direct factorization algorithm yields significantly large error due to the influence of outliers , and the error increases with the increase of outliers . the experiment also shows that all three robust algorithms are resilient to outliers , as can be seen in fig . [ fig : s5 ] , the ratio of outliers has little influence to the reprojection variance of the three robust algorithms . the complexity of different approaches was compared in terms of real computation time in the above test . all algorithms were implemented in matlab 2009 on a lenovo t500 laptop with 2,26ghz intel cpu . the frame number was varied from 50 to 300 in steps of 50 so as to generate different sizes of the tracking matrix , and 10% outliers were added in the data . table [ tab : time ] shows the real computation time of different algorithms . obviously , the complexity of the proposed scheme lies in between of @xcite and @xcite . this is because huynh s method does not include the weighted factorization , while the minimization of @xmath143 norm in @xcite is computationally more intensive than the alternative factorization algorithm . we evaluated the performance of the augmented nonrigid factorization algorithm . during the experiment , we generated a deformable space cube , which was composed of 21 evenly distributed rigid points on each side and three sets of dynamic points ( @xmath144 points ) on the adjacent surfaces of the cube that were moving outward . there are 252 space points in total as shown in fig.[fig : s6 ] . using the synthetic cube , we generated 100 images by the affine projection with randomly selected camera parameter . each image corresponds to a different 3d structure , and the image resolution is set at @xmath110 . + ( a ) ( b)(c)(d ) + + ( e ) ( f)(g)(h ) + 1@l * 6c@ frame no . & 50 & 100 & 150 & 200 & 250 & 300 + huynh & 1.19 & 2.35 & 3.68 & 6.41 & 10.45 & 12.69 + ke & 1.81 & 6.27 & 14.32 & 26.94 & 44.87 & 67.53 + proposed & 1.27 & 3.93 & 8.12 & 14.28 & 22.40 & 32.13 + [ tab : time ] for the above simulated image sequence , we added 3 pixels gaussian noise to the image sequence and also added 10% outliers to the tracking matrix . fig.[fig : s6 ] ( b)-(d ) show three noise and outlier corrupted images . using the contaminated data , all outliers were successfully removed and the motion and shape matrices were recovered by employing the proposed robust algorithm . the corresponding 3d dynamic structures recovered by the proposed approach are shown in fig.[fig : s6 ] ( f)-(h ) , respectively . it is evident that the deformable cubic structures are correctly retrieved by the proposed robust strategy . the method was extensively tested on a number of real image sequences . we will report the experimental results and evaluations on four real sequences in the paper . the first test is on a fountain base sequence captured at downtown san francisco . the sequence consists of 10 images and on average 5648 features were tracked across the sequence using the feature tracking system @xcite . since the images have a large portion of homogeneous and repetitive textures , it is thus hard to track accurately for this type of scene , though visually only a few features were obviously mismatched . in order to test the algorithm , an additional 5% outliers were added to the tracking data . all these features with disparities to the first image are shown in fig.[fig : r1 ] . using the proposed scheme , the outliers were successfully rejected . after rejecting the outliers , the weighted alternative algorithm was employed to recover the motion and structure matrices . finally , the solution was upgraded to the euclidean space . as shown in fig.[fig : r1 ] , the structure looks realistic and most details are correctly recovered . the histogram distribution of the reprojection residual matrix ( [ eq : residual_matrix ] ) with outliers is shown in fig.[fig : r2 ] ( a ) . the residuals are largely conform to the assumption of normal distribution . it can be seen from the distribution that the outliers can be obviously distinguished from inliers , the computed threshhold is shown in the figure . the histogram distribution of the residuals of the detected inlying data is shown in fig.[fig : r2 ] ( b ) . obviously , the residual error is reduced significantly after rejecting the outliers . as a quantitative evaluation , the final reprojection errors by different approaches are tabulated in table [ tab : reprojection_error ] , from which we can see that the proposed scheme yields the lowest reprojection error . + + + + ( a ) ( b ) + the second sequence is a corner of hearst gym at uc berkeley . there are 12 images in the sequence , and on average 1890 features were tracked in total . the correctly detected inlying features , together with about 5% outliers are shown in fig.[fig : r3 ] . using the proposed robust algorithm , we successfully recovered the euclidean structure of the scene , as shown in fig.[fig : r3 ] , all outliers are correctly detected and removed . as a comparison , the reprojection errors obtained using different algorithms are listed in table [ tab : reprojection_error ] , which shows that the proposed approach outperforms other robust algorithms . the third test is on a deformable dinosaur sequence from the literature @xcite . the sequence consists of 231 images with different movement and deformation of a dinosaur model . the image resolution is @xmath145 pixel and 49 features were tracked across the sequence . in order to test the robustness of the algorithm , an additional 8% outliers were added to the tracking data as shown in fig.[fig : nr1 ] . using the proposed approach , all outliers were successfully rejected , however , a few tracked features were also eliminated due to large tracking errors . the nonrigid extended approach was employed to recover the motion and structure matrices , and the solution was then upgraded to the euclidean space . fig.[fig : nr1 ] shows the reconstructed structure and wireframes . the vrml model is visually realistic and the deformation at different instant is correctly recovered , although the initial tracking data are not very reliable . + + + 1@l * 4c@ dataset & fountain & hearst & dinosaur & face + huynh & 0.736 & 0.742 & 0.926 & 0.697 + ke & 0.579 & 0.635 & 0.733 & 0.581 + proposed & 0.426 & 0.508 & 0.597 & 0.453 + [ tab : reprojection_error ] + + ( a ) ( b ) + the histogram distribution of the reprojection residual matrix ( [ eq : residual_matrix ] ) with outliers is shown in fig.[fig : nr2 ] ( a ) . the residuals are largely conform to the assumption of normal distribution . as can be seen from the histogram , the outliers are obviously distinguished from inliers . a threshhold is computed from the mean and std of the distribution , and the histogram of the residuals produced by the final solution after rejecting outliers is shown in fig.[fig : nr2 ] ( b ) , which shows a significant resuction on the residual errors . for comparison , we also extend the algorithms of huynh and ke to the nonrigid scenarios , and the reprojection errors by different algorithms are shown in table [ tab : reprojection_error ] . the proposed scheme yields the lowest reprojection error in this test . the last experiment is on a face sequence with different facial expressions . the dataset was downloaded from fgnet at http://www-prima.inrialpes.fr/fgnet/html/home.html . we selected 200 images from the sequence . the image resolution is @xmath146 with 68 automatically tracked feature points using the active appearance model ( aam ) . for test purpose , 8% outliers were added to the tracking data as shown in fig.[fig : nr3 ] . the proposed robust algorithm was used to recover the euclidean structure of the face . fig.[fig : nr3 ] shows the reconstructed vrml model with texture and the corresponding wireframes from different viewpoints . as demonstrated in the results , different facial expressions have been correctly recovered by the proposed approach . the reprojection errors obtained from different algorithms are tabulated in table [ tab : reprojection_error ] . as in other experiments , the proposed robust approach also yields the best performance . in this paper , we first proposed a new augmented factorization framework which has been proved to be more accurate than classical affine factorization , especially in the situation when the centroid of the imaged features could not be reliably recovered due to the presence of missing and outlying data . then , we presented an alternatively weighted factorization algorithm to handle incomplete tracking data and alleviate the influence of large image noise . finally , a robust factorization scheme was designed to deal with contaminated data with outliers and missing points . the proposed technique requires no prior information of the error distribution in the tracking data , and it can be directly extended to nonrigid factorization , which was rarely discussed in the literature . extensive evaluations on both synthetic and real datasets demonstrated the advantages of the proposed scheme over previous methods . a. agudo , l. agapito , b. calvo , and j. m. montiel , `` good vibrations : a modal analysis approach for sequential non - rigid structure from motion , '' in _ proceedings of the ieee conference on computer vision and pattern recognition _ , 2014 , pp . 15581565 . a. agudo , b. calvo , and j. montiel , `` finite element based sequential bayesian non - rigid structure from motion , '' in _ computer vision and pattern recognition ( cvpr ) , 2012 ieee conference on_.1em plus 0.5em minus 0.4emieee , 2012 , pp . . p. m. aguiar and j. m. moura , `` rank 1 weighted factorization for 3d structure recovery : algorithms and performance analysis , '' _ ieee transactions on pattern analysis and machine intelligence _ , vol . 25 , no . 9 , pp . 11341149 , 2003 . i. akhter , y. sheikh , s. khan , and t. kanade , `` trajectory space : a dual representation for nonrigid structure from motion , '' _ ieee transactions on pattern analysis and machine intelligence _ 33 , no . 7 , pp . 14421456 , 2011 . bazin , y. seo , r. hartley , and m. pollefeys , `` globally optimal inlier set maximization with unknown rotation and focal length , '' in _ european conference on computer vision_.1em plus 0.5em minus 0.4em springer , 2014 , pp . 803817 . c. bregler , a. hertzmann , and h. biermann , `` recovering non - rigid 3d shape from image streams , '' in _ computer vision and pattern recognition , 2000 . proceedings . ieee conference on _ , vol . 2.1em plus 0.5em minus 0.4emieee , 2000 , pp . 690696 . s. christy and r. horaud , `` euclidean shape and motion from multiple perspective views by affine iterations , '' _ ieee transactions on pattern analysis and machine intelligence _ , vol . 18 , no . 11 , pp . 10981104 , 1996 . m. a. fischler and r. c. bolles , `` random sample consensus : a paradigm for model fitting with applications to image analysis and automated cartography , '' _ communications of the acm _ , vol . 24 , no . 6 , pp . 381395 , 1981 . a. gruber and y. weiss , `` multibody factorization with uncertainty and missing data using the em algorithm , '' in _ computer vision and pattern recognition , 2004 . cvpr 2004 . proceedings of the 2004 ieee computer society conference on _ , vol . 1.1em plus 0.5em minus 0.4emieee , 2004 , pp . i707 . d. q. huynh , r. hartley , and a. heyden , `` outlier correction in image sequences for the affine camera , '' in _ computer vision , 2003 . proceedings . ninth ieee international conference on_.1em plus 0.5em minus 0.4em ieee , 2003 , pp . 585590 . q. ke and t. kanade , `` robust l 1 norm factorization in the presence of outliers and missing data by alternative convex programming , '' in _ 2005 ieee computer society conference on computer vision and pattern recognition ( cvpr05 ) _ , vol . 1.1em plus 0.5em minus 0.4emieee , 2005 , pp . 739746 . z. liu , p. monasse , and r. marlet , `` match selection and refinement for highly accurate two - view structure from motion , '' in _ european conference on computer vision_.1em plus 0.5em minus 0.4emspringer , 2014 , pp . 818833 . d. d. morris and t. kanade , `` a unified factorization algorithm for points , line segments and planes with uncertainty models , '' in _ computer vision , 1998 . sixth international conference on_.1em plus 0.5em minus 0.4emieee , 1998 , pp . 696702 . i. nurutdinova and a. fitzgibbon , `` towards pointless structure from motion : 3d reconstruction and camera parameters from general 3d curves , '' in _ proceedings of the ieee international conference on computer vision _ , 2015 , pp . 23632371 . t. okatani , t. yoshida , and k. deguchi , `` efficient algorithm for low - rank matrix factorization with missing components and performance comparison of latest algorithms , '' in _ 2011 international conference on computer vision_.1em plus 0.5em minus 0.4emieee , 2011 , pp . 842849 . j. oliensis and r. hartley , `` iterative extensions of the sturm / triggs algorithm : convergence and nonconvergence , '' _ ieee transactions on pattern analysis and machine intelligence _ , vol . 29 , no . 12 , pp . 22172233 , 2007 . m. paladini , a. del bue , j. xavier , l. agapito , m. stoi , and m. dodig , `` optimal metric projections for deformable and articulated structure - from - motion , '' _ international journal of computer vision _ , vol . 96 , no . 2 , pp . 252276 , 2012 . g. qian , r. chellappa , and q. zheng , `` bayesian algorithms for simultaneous structure from motion estimation of multiple independently moving objects , '' _ ieee transactions on image processing _ , vol . 14 , no . 1 , pp . 94109 , 2005 . b. resch , h. lensch , o. wang , m. pollefeys , and a. sorkine - hornung , `` scalable structure from motion for densely sampled videos , '' in _ proceedings of the ieee conference on computer vision and pattern recognition _ , 2015 , pp . 39363944 . d. scaramuzza , `` 1-point - ransac structure from motion for vehicle - mounted cameras by exploiting non - holonomic constraints , '' _ international journal of computer vision _ , 95 , no . 1 , pp . 7485 , 2011 . p. sturm and b. triggs , `` a factorization based algorithm for multi - image projective structure and motion , '' in _ european conference on computer vision_.1em plus 0.5em minus 0.4emspringer , 1996 , pp . 709720 . j. taylor , a. d. jepson , and k. n. kutulakos , `` non - rigid structure from locally - rigid motion , '' in _ computer vision and pattern recognition ( cvpr ) , 2010 ieee conference on_.1em plus 0.5em minus 0.4em ieee , 2010 , pp . 27612768 . l. torresani , a. hertzmann , and c. bregler , `` nonrigid structure - from - motion : estimating shape and motion with hierarchical priors , '' _ ieee transactions on pattern analysis and machine intelligence _ , vol . 30 , no . 5 , pp . 878892 , 2008 . b. triggs , `` factorization methods for projective structure and motion , '' in _ computer vision and pattern recognition , 1996 . proceedings cvpr96 , 1996 ieee computer society conference on_.1em plus 0.5em minus 0.4emieee , 1996 , pp . 845851 . g. wang and q. j. wu , `` perspective 3-d euclidean reconstruction with varying camera parameters , '' _ ieee transactions on circuits and systems for video technology _ , vol . 19 , no . 12 , pp . 17931803 , 2009 . g. wang , and q. j. wu , `` quasi - perspective projection model : theory and application to structure and motion factorization from uncalibrated image sequences , '' _ international journal of computer vision _ , vol . 87 , no . 3 , pp . 213234 , 2010 . g. wang , j. s. zelek , and q. j. wu , `` structure and motion recovery based on spatial - and - temporal - weighted factorization , '' _ ieee transactions on circuits and systems for video technology _ , vol . 22 , no . 11 , pp . 15901603 , 2012 . g. wang , j. s. zelek , and q. j. wu , `` robust structure from motion of nonrigid objects in the presence of outlying and missing data , '' in _ computer and robot vision ( crv ) , 2013 international conference on_.1em plus 0.5em minus 0.4emieee , 2013 , pp . 159166 . g. wang , j. s. zelek , q. j. wu , and r. bajcsy , `` robust rank-4 affine factorization for structure from motion , '' in _ applications of computer vision ( wacv ) , 2013 ieee workshop on_.1em plus 0.5em minus 0.4em ieee , 2013 , pp . 180185 . h. wang , t .- j . chin , and d. suter , `` simultaneously fitting and segmenting multiple - structure data with outliers , '' _ ieee transactions on pattern analysis and machine intelligence _ , vol . 34 , no . 6 , pp . 11771192 , 2012 . j. yu , t .- j . chin , and d. suter , `` a global optimization approach to robust multi - model fitting , '' in _ computer vision and pattern recognition ( cvpr ) , 2011 ieee conference on_.1em plus 0.5em minus 0.4em ieee , 2011 , pp . 20412048 .

we investigate the evolution of type ib / c supernova ( sn ib / c ) progenitors in close binary systems , using new evolutionary models that include the effects of rotation , with initial masses of 12 25 for the primary components , and of single helium stars with initial masses of 2.8 20 . we find that , despite the impact of tidal interaction on the rotation of primary stars , the amount of angular momentum retained in the core at the presupernova stage in different binary model sequences converge to a value similar to those found in previous single star models . this amount is large enough to produce millisecond pulsars , but too small to produce magnetars or long gamma - ray bursts . we employ the most up - to - date estimate for the wolf - rayet mass loss rate , and its implications for sn ib / c progenitors are discussed in detail . in terms of stellar structure , sn ib / c progenitors in binary systems at solar metallicity are predicted to have a wide range of final masses up to about 7 , with helium envelopes of . our results indicate that , if the lack of helium lines in the spectra of sne ic were due to small amounts of helium ( e.g. ) , the distribution of both initial and final masses of sn ic progenitors should be bimodal . furthermore , we find that a thin hydrogen layer ( ) is expected to be present in many sn ib progenitors at the presupernova stage . we show that the presence of hydrogen , together with a rather thick helium envelope , can lead to a significant expansion of some sn ib / c progenitors by the time of supernova explosion . this may have important consequences for the shock break - out and supernova light curve . we also argue that some sn progenitors with thin hydrogen layers produced via case ab / b transfer might be related to type iib supernova progenitors with relatively small radii of about .

1004.0843

it is generally believed that type ib and type ic supernovae result from core collapse events of naked helium stars . the helium stars are thought to be produced by the loss of the hydrogen envelope , via stellar winds mass loss from massive single stars or via mass transfers in close binary systems . according to recent stellar models adopting the most up - to - date stellar winds mass loss rates @xcite , the final masses ( @xmath5 ) of helium stars produced by mass - losing single stars appear to be too massive to produce typical sne ib / c ( i.e. , @xmath6 @xmath0 at solar metallicity ) . although the limiting mass for bh formation is not yet well determined , given their high binding energy , such massive progenitors of @xmath6 @xmath0 are likely to form black holes ( bhs ) , producing faint supernovae or no supernova at all ( cf . * ? ? ? although very bright sne ib / c like sn 1998bw could be produced from such massive helium stars if , for example , powered by rapid rotation ( e.g. , * ? ? ? * ; * ? ? ? * ) , such events are shown to be rare ( e.g. , * ? ? ? * ; * ? ? ? by contrast , helium stars with a wide range of masses ( @xmath7 ) can be made from @xmath8 @xmath0 primary components in close binary systems via the so - called case a / b mass transfer . many of them may end their life as bright sne ib / c leaving neutron stars as remnants , if their final masses are less than about 7 10 @xmath0 . population studies indeed show that close binary stars can produce a sufficient number of sne ibc to explain their observed rate , without the need of invoking single star progenitors ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? therefore , it is most likely that the majority of typical sne ibc are produced in binary systems . the observational evidence for the connection between sne ibc and long gamma - ray bursts ( grbs ) has particularly motivated many observational studies to better understand sne ibc since the last decade ( see * ? ? ? * for a review ) . theoretical stellar models of sne ibc progenitors are thus highly required nowadays . the most comprehensive studies on the detailed characteristics of sne ibc progenitors in binary systems were conducted by @xcite ( hereafter , wlw95 ) using mass - losing pure helium star models , and by @xcite ( hereafter , wl99 ) using self - consistent binary star models . although more recent theoretical studies on sne ibc progenitors in binary systems can be found in the literature , they have been focused on long grb progenitors or stellar populations , rather than on the detailed nature of typical sne ibc progenitors ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in this paper , we revisit the problem of sne ibc progenitors in close binary systems using both binary star and single helium star models up to the neon burning stage , with updated physics of two important ingredients . one is rotation , which was not considered in wlw95 and wl99 , and the other is the mass loss rate of wolf - rayet ( wr ) stars ( sect . [ sect : review ] ) . this paper is organized as follows . in sect . [ sect : review ] , we briefly review recent developments of stellar evolution models regarding the effects of rotation and the wr star mass loss rate , arguing for the need of updated physics in binary star models . our adopted physical assumptions and numerical method are discussed in sect . [ sect : method ] . in the following section ( sec . [ sect : rotation ] ) , using our binary star evolution models including the effect of rotation and the transport of angular momentum due to hydrodynamic instabilities and magnetic torques , we explore the role of tidal interaction and mass transfer in the redistribution of angular momentum in primary stars . in sec . [ sect : sn ] , the nature of sne ibc progenitors is investigated in terms of final masses , masses of helium and hydrogen layers , radii and mass loss rates at the presupernova stage , assuming these properties do not significantly change from neon burning to core collapse . for this purpose , we also present mass - losing single helium star models as a complement to our binary star models , given that the parameter space explored with our binary model sequences is limited . we conclude the paper by discussing observational implications of our results , in sect . [ sect : discussion ] . rotation has particular roles in the evolution massive stars as it changes the stellar structure , induces chemical mixing and enhances mass loss due to stellar winds @xcite . during the last decade , several authors have calculated massive star models up to the pre - supernova stage , considering the redistribution of angular momentum and chemical species due to rotationally induced hydrodynamic instabilities , such as eddington - sweet circulations , and the shear instability @xcite . although these models could explain some observational aspects such as surface abundances of cno elements of massive stars and wolf - rayet ( wr ) star populations at different metallicities ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , their adopted angular momentum transport mechanisms turned out to be too inefficient to explain the observed spin rates of stellar remnants @xcite . i.e. , their models predict nearly two orders of magnitude higher spin rates of white dwarfs and young neutron stars than the observed ones . these models also imply that almost all wr stars can retain enough angular momentum to produce grbs , either by magnetar or collapsar formation depending on the final mass . this gives a nearly 1000 times higher ratio of grbs to sne than the observationally implied value . on the other hand , @xcite suggested that magnetic torques resulting from dynamo actions in differentially rotating radiative layers ( the so - called spruit - tayler dynamo ) should be the dominant angular momentum transport mechanism compared to the pure hydrodynamic instabilities . recent magnetic models that adopt the spruit - tayler dynamo according to the prescription by @xcite are indeed more consistent with observations in terms of the spin rates of stellar remnants @xcite . magnetic models also better explain the fact that long grbs are rare events compared to normal core collapse supernovae . although the spruit - tayler dynamo mechanism is still subject to many uncertainties @xcite , these recent studies indicate that an efficient angular momentum transport mechanism comparable to what the spruit - tayler dynamo predicts is needed to understand the observations . the above discussion is based on single star models , and the role of rotation in the evolution of massive binary stars remained relatively unexplored . massive stars in close binary systems are supposed to experience an exchange of mass and angular momentum via mass transfer and tidal interaction , and thus the evolution of binary stars is more complex than that of single stars . @xcite and @xcite presented , for the first time , self - consistent calculations of massive binary star evolution models including many relevant effects of rotation and binary interactions : the change of the stellar structure due to the centrifugal force , transport of angular momentum and chemical species due to rotationally induced hydrodynamic instabilities , tidal interaction , transfer of mass from the primary , accretion of mass and angular momentum of the secondary , and resulting changes of the orbit . their work indicates that the secondary can be easily spun up by mass accretion even up to critical rotation , thus modulating the mass accretion efficiency by the interplay of the enhanced mass loss due to rotation from the spun - up secondary and the mass transfer from the primary . this effect provides important clues to better understand the evolutionary paths of some observed x - ray and wr star binary systems as discussed by @xcite and @xcite . their non - magnetic models also show that massive stars may end up with different core spin rates depending on the history of mass loss / gain during binary evolution , that could be related to the observational diversity of core - collapse supernovae . more recently , @xcite and @xcite included the spruit - tayler dynamo in their binary models , and discussed possible evolutionary paths of massive binary stars towards long grbs . in this paper , we present new evolutionary calculations of magnetic ( i.e. , the spruit - tayler dynamo is included ) massive binary stars for a large parameter space , focusing on the evolution of the primary stars to investigate the nature of typical sne ibc progenitors . wlw95 employed the mass - dependent mass loss rate of wr stars of @xcite for their helium star models , and wl99 used those of @xcite and @xcite . both studies concluded that the final masses of sne ibc progenitors in binary systems should converge to a limited range of 2.2 3.6 @xmath0 , even for an initial helium star mass of 20 @xmath0 . later developments of the wr wind theory considering clumpies pointed out , however , that the wr mass loss rates used by wlw95 and wl99 are significantly overestimated @xcite . for example , fig . [ fig : wrwind ] shows that , for helium stars on the zero age main sequence , the mass loss rate given by @xcite is almost an order - of - magnitude lower than that of hamann et al . used in wl99 . the most recent theoretical models of wr winds by @xcite and @xcite also give wr mass loss rates compatible to the nugis & lamers rate ( cf . figure 1 in * ? ? ? the reduced wr mass loss rates have been considered in many recent massive single star models ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , suggesting large final masses of wr stars at the presupernova stage ( @xmath9 @xmath0 at solar metallicity ) compared to @xmath10 @xmath0 found by @xcite . as argued in the introduction , this leads to the interesting conclusion that most core collapse events occurring in single wr stars should not produce bright supernovae , rendering the binary star channel even more important for sne ibc . surprisingly , the effect of the new wr mass loss rate on sne ibc progenitors in close binary systems has not been much discussed or largely overlooked . to our knowledge , @xcite are the only authors who addressed this issue in detail . although wl99 also presented some model sequences with a reduced wr mass loss rate , their adopted reduction factor was rather modest ( 2 times smaller than the hamann et al . rate ) . here we consider the hamman et al . wr mass loss rate reduced by factors of 5 and 10 , which reflects the most recent result as discussed above . we have used the same stellar evolution code as in @xcite , that follows the simultaneous evolution of the two stellar components of a binary system . the effect of the centrifugal force on stellar structure is considered following @xcite . the transport of angular momentum and chemical elements is treated as diffusion considering eddington sweet circulations , the shear instability , the goldreich - schubert - fricke instability and the spruit - tayler dynamo , as described in @xcite . we use @xmath11 for the semi - convection parameter @xcite , as in @xcite . the stellar winds mass loss rate is determined following @xcite with a metallicity scaling of @xmath12 @xcite for the main sequence phase . wr wind mass loss rates are computed according to hamman et al . with a correction factor @xmath13 , and with a metallicity dependence of @xmath14 @xcite : @xmath15 we use @xmath16 or 10 in most model sequences , which means the wr mass loss rate by hamman et al . is lowered 5 or 10 times , to consider the most recent estimates ( see fig . [ fig : wrwind ] ) . the enhancement of stellar winds mass loss due to the centrifugal force is considered by @xmath17 ~~,\ ] ] where @xmath18 with @xmath19 and @xmath20 is the eddington factor @xcite . here , to prevent a singularity that may occur as @xmath21 approaches @xmath22 , the mass loss rate is limited to @xmath23 , where @xmath24 is the thermal time scale of the star . the binary orbit is assumed to be circular , and the roche lobe radius is determined according to the approximation of @xcite . the mass loss rate of the roche - lobe filling component through the first lagrangian point is implicitly computed using the method given by @xcite . the equation of motion of a test particle is numerically solved to calculate the amount of angular momentum of the accreted matter if the transfered matter directly hits the secondary star , and the keplerian value is assumed otherwise @xcite . the change of the orbital period due to mass transfer and stellar wind mass loss is considered according to @xcite . we follow @xcite to determine the amount of the specific angular momentum carried away from the orbit by stellar winds . tidal synchronization is considered following @xcite ( see also * ? ? ? we assume a synchronization time scale according to @xcite who considered tidally driven meridional circulations as the main mechanism for tidal dissipation : @xmath25 where @xmath26 denotes the mass ratio and @xmath27 the orbital separation . this prescription gives a much shorter time scale than that given by @xcite . given that the physics of tidal dissipation is much debated in the literature @xcite , we introduce a parameter @xmath28 to investigate how an extremely fast / slow synchronization may influence the results . in most cases , however , we use @xmath29 . a few sequences are also computed with @xmath30 of @xcite for comparison : @xmath31 where @xmath32 is the moment of the star , and @xmath33 a constant measuring the coupling between the tidal potential and the gravity mode . using the data of table 1 in @xcite , we constructed a fitting formula for @xmath33 as the following : @xmath34 where @xmath35 is the radius of the convective core . note that both prescriptions by tassoul and zahn are not appropriate for a star with a convective envelope . however , the role of tidal synchronization is significant only on the main sequence , and not important in late evolutionary stages as discussed below . we computed 45 model sequences for initial masses of the primary star mostly from 12 to 25 @xmath36 at two different metallicities ( @xmath37 0.02 and 0.004 ) , for different mass ratios , initial orbital periods , and wr mass loss rates , as summarized in table . the initial rotational velocity at the equatorial surface of each star is set to be 20% of the keplerivan value . we could not calculate more massive systems because of a numerical difficulty encountered during the mass transfer phases , except for seq . 26 where a primary star of 60 @xmath0 is considered with a rather large wr mass loss rate ( i.e. , @xmath38 ) . the adopted initial orbital periods corresponds either to case a or to case b mass transfer . in the present study , we do not consider case c systems , but briefly discuss the possible outcomes of case c mass transfer in sect . [ sect : dischydrogen ] . the evolution of the primary stars is followed up to neon burning in most cases . we also present non - rotating single helium star models to discuss sne ibc progenitors in binary systems with initial masses larger than 25 @xmath0 , and also to compare them with binary star models ( sect . [ sect : sn ] ) . in this section , we focus our discussion on the evolution of primary stars and investigate whether binary evolution via case a or case b mass transfer could lead to diverse pre - collapse conditions of sne ibc in terms of the amount of core angular momentum . although the evolution of mass - accreting secondary stars is a matter of extreme interest as discussed in @xcite , @xcite and @xcite , it is beyond the scope of this paper . here , we first present some results including the spruit - tayler dynamo with our fiducial assumption on synchronization time ( i.e. , @xmath39 ) , showing that the final amount of angular momentum in the core of the primary star is not much affected by different histories of mass loss ( i.e. , case ab or case b ; sect . [ sect : fiducial ] ) . then , we discuss the influences of different assumptions on tidal synchronization and transport process of angular momentum ( sect . [ sect : nonfiducial ] ) . the evolution of the primary star in a close binary system is characterized by the rapid loss of mass due to roche - lobe overflow . as an example , the evolution of the primary star in seq . 14 is described in figs . [ fig : kippseq9 ] and [ fig : hrseq9 ] , where our fiducial value of @xmath39 is adopted , including the spruit - tayler dyanamo . the binary system initially consists of a @xmath40 star and a @xmath41 star in a 4 day orbit . mass transfer starts at @xmath42 , when the helium mass fraction in the hydrogen burning core has increased to 0.94 . the mass transfer rate rises up to @xmath43 , which roughly corresponds to @xmath44 where @xmath45 and @xmath46 denote the mass and the kelvin - helmoltz time scale of the primary star , respectively . the primary mass decreases to @xmath47 by the end of the case a transfer ( see fig . [ fig : chem ] ) . the second roche - lobe overflow begins at @xmath48 when the envelope of the primary star expands due to hydrogen shell burning during the helium core contraction phase ( case ab mass transfer ) . the primary star loses most of the hydrogen envelope as a result , exposing its helium core of @xmath49 having a small amount of hydrogen ( @xmath50 @xmath0 ) in the outermost layers , as shown in the third panel of fig . [ fig : chem ] . although the star remains compact ( @xmath51 ) during core helium burning , helium shell burning activated after core helium exhaustion leads to the expansion of the envelope up to @xmath52 ( see fig . [ fig : hrseq9 ] ) during core carbon burning . a case abb mass transfer does not occur , however , due to the large orbital separation ( @xmath53 ) at this stage , while it does occur in many other sequences . the final mass at the end of the calculation ( neon burning ) is @xmath54 . the mass of hydrogen decreases to 0.0015 @xmath0 at the end , and the remaning mass of helium is 1.49 @xmath0 , as shown in the last panel of fig . [ fig : chem ] . the star is likely to eventually explode as a type ib supernova given the rather thick helium envelope with a very thin hydrogen layer , but it might also appear as type iib if the supernova were found within several days after the explosion ( see sect . [ sect : hydrogen ] ) . the lower panel of fig . [ fig : hrseq9 ] shows that the mass transfer is not conservative . when the secondary star reaches critical rotation as a result of the accretion of angular momentum , the stellar wind mass loss rate increases so drastically as to prevent efficient mass accumulation ( see * ? ? ? * for a more detailed discussion on this effect ) . the ratio of the accreted mass in the secondary star to the transferred mass from the primary star is about 0.83 during the case a mass transfer , and 0.41 during the case b mass transfer . the previous calculations by wl99 and @xcite show that the non - conservative mass transfer leads to a shorter orbit than in the case of conservative mass transfer , in general . this effect should be kept in mind in the following discussion on the evolution of the binary orbit and its consequences . the evolution of rotation in the primary star is shown in figs . [ fig : totjsync ] , [ fig : omegasync ] and [ fig : jcoreseq9 ] . as shown in fig . [ fig : totjsync ] , the total angular momentum of the primary star rapidly decreases in the beginning as a result of tidal interaction , until the star is completely synchronized with the orbit when @xmath55 . [ fig : omegasync ] shows the evolution of the angular velocity inside the primary star from zams until it reaches complete synchronization . note that the star keeps rotating almost rigidly , even though the tidal interaction causes redistribution of angular momentum from the outermost layers of the star @xcite . this results from the very short time scale of the transport of angular momentum in the star ( about 300 years ) due mainly to convection and the spruit - tayler dynamo , in the convective and radiative layers respectively . later on , the angular velocity of the primary star still remains coupled with the orbital motion , until decoupling starts during case b mass transfer ( see below ) . interestingly the total angular momentum of the primary star gradually increases from the initial synchronization until the onset of case a mass transfer ( fig . [ fig : totjsync ] ) . this is because the star significantly expands , while the change of the orbital separation remains small durning this period , as shown in the third and last panels in fig . [ fig : jcoreseq9 ] . however , it rapidly decreases again during the case a and ab mass transfer phases , as explained below . [ fig : jcoreseq9 ] shows that angular momentum in the core of the primary star is mostly removed during the mass transfer phases ( case a and case ab ) . it should be noted , however , that the mechanism for braking the core is different for each case . during the case a mass transfer , the synchronization time scale according to eq . ( [ eq3 ] ) remains very short ( @xmath56 ) compared to the mass transfer time scale ( @xmath57 ) . the decrease of the core angular momentum during the case a mass transfer results from the synchronization that occurs even when the orbit is rapidly widened due to mass exchange ( fig . [ fig : jcoreseq9 ] ) . during the case ab mass transfer , the further increase of the orbital separation significantly weakens the role of synchronization for the redistribution of angular momentum . however , both the spruit - tayler dynamo and mass loss lead to rapid braking of the thermally contracting helium core . further significant core - braking by the spruit - tayler dynamo occurs during the co core contraction phase , and the mean specific angular momentum of the innermost @xmath58 becomes about @xmath59 at the neon burning phase . in seq . 20 , the initial masses of the stellar components are the same as in seq . 14 but the initial period is large enough that the first mass transfer occurs during the helium core contraction phase ( case b mass transfer ) . avoiding case a mass transfer , the primary star retains more angular momentum in the core at the end of main sequence than in seq . 14 however , the core loses more angular momentum during the helium core contraction phase than in seq . 14 , as stronger magnetic torques are exerted , mainly due to the more massive envelope ( fig . [ fig : caseb ] ) . at neon burning , @xmath60 is obtained , which is similar to that in seq . 14 . the effect of synchronization is negligible in seq . 15 , where @xmath61 is adopted . as shown in fig . [ fig : jcoreseq10 ] , a rather rapid decrease of @xmath62 occurs during and after the case a mass transfer , which is a combined effect of mass loss and magnetic torques : mass loss carries away angular momentum from the envelope , and magnetic torques brake the core rotation subsequently . the core is further slowed down during the helium core and co core contraction phases , resulting in @xmath63 at the neon burning phase . note that this value is very close to that in seq . we find that the result with zahn s prescription for synchronization ( seq . 17 ) is not much different from that of seq . 15 where synchronization is negligible . the synchronization time scale according to zahn is sensitive to the ratio of the convective core size to the stellar radius ( @xmath64 , see eqs . ( [ eq4 ] ) and ( [ eq5 ] ) ) . this ratio continuously decreases as the star evolves , and thus @xmath30 continuously increases to such an extent that the effect of synchronization can be ignored when case a mass transfer starts . when @xmath65 with tassoul s prescription is used ( seq . 16 ) , on the other hand , synchronization becomes important even after the case ab mass transfer phase , and the primary is further spun down by tidal interaction , giving @xmath66 at the neon burning phase . in seq . 18 where the spruit - tayler dynamo is not included , the core is spun down due to synchronization during the case a mass transfer phase ( fig . [ fig : jcoreseq10 ] ) as in the corresponding magnetic case ( seq . 14 ; fig . [ fig : jcoreseq9 ] ) . the spin - orbit coupling becomes significantly weakened as the orbit widens after the case a mass transfer phase . despite a significant amount of mass is lost via case ab mass transfer , the core in the primary star retains most of the remaining angular momentum in the following evolutionary stages . this is because the chemical gradient across the boundary between the helium core and the hydrogen envelope effectively prohibits the transport of angular momentum ( cf . * ? ? ? * ; * ? ? ? the core angular momentum at neon burning is thus about 10 times larger ( @xmath67 ) than in the corresponding magnetic case . as shown in the above examples , in the model sequences with the spruit - tayler dynamo , all of the primary stars retain similar amounts of angular momentum ( a few @xmath68 ) in the innermost @xmath58 at neon burning regardless of the detailed history of mass transfer , unless synchronization is extremely fast as in seq . 16 ( see table [ tab1 ] ) . according to the spruit - tayler dynamo , magnetic torques exerted to the core become stronger with a higher spin rate , a larger degree of differential rotation between the core and the envelope , and a heavier radiative envelope . therefore , although winds or roche - lobe overflows reduce the size of the hydrogen envelope and remove angular momentum from the star , this in turn weakens the torque exerted to the core , and vice versa . the remarkable convergence of @xmath69 to a few @xmath68 in our model sequences , even for different wind parameters and metallicities as shown in table [ tab1 ] , can be explained by this self - regulating nature of the spruit - tayler dynamo . this result indicates that not much diversity is expected in sne ibc progenitors produced via case a or case ab / b mass transfer , in terms of rotation : most of sne ibc of a similar progenitor mass may leave neutron stars with a similar spin rate . however , other types of binary interactions still may lead to various final rotation periods in sn progenitors ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * for such examples ) . in the literature , the detailed history of binary interactions such as mass transfer and tidal interaction is often neglected , and only the evolution of pure helium stars to discuss sne ibc progenitors in binary systems is investigated ( e.g. wlw95 ; @xcite ) . although this approach is very useful , our binary star models indicate that complications resulting from binary interactions have important consequences in sn ibc progenitors . to address this issue , here we also present evolutionary models of non - rotating single helium stars with initial masses of 2.8 20 @xmath0 at @xmath70 for comparison with binary star models , as presented in table [ tab2 ] . this also enables us to study sn ibc progenitors with initial masses higher than 25 @xmath0 , which are lacking in our binary model sequences . these helium star models were calculated up to neon burning , with the wr mass loss rate given in eq . ( [ eq1 ] ) with @xmath71 5 or 10 , as in the binary models . in the following , we focus our discussion mostly on the results with @xmath16 , while the influence of the wind parameter is briefly discussed . the final mass of the primary stars in a close binary systems is largely determined by the winds mass loss during core helium burning , if the initial mass ( @xmath72 ; i.e. , the mass right after case ab or case b mass transfer ) is significantly larger than about 3.0 @xmath0 . however , for a less massive helium star , the expansion of the helium envelope becomes so dramatic during co core contraction , and/or during core carbon burning , that it can lead to another mass transfer phase : case abb or case bb ( e.g. , compare the final radius of the 2.8 @xmath0 helium star model with those of more massive helium star models in table [ tab2 ] ) . the impact of this mass transfer phase becomes more important for a less massive helium star . in seqs . 1 , 2 and 4 where @xmath73 2.1 2.3 @xmath0 , the rapid loss of mass during case abb / bb transfer reduces the total mass of the primary stars to such an extent that they may not explode as supernovae , but die as white dwarfs . in seq . 3 , where @xmath74 @xmath0 , the carbon - oxygen core can grow beyond the chandrasekhar limit despite the significant loss of mass ( i.e. , about 1 @xmath0 ) during the case bb phase . the remaining helium mass in the envelop is only about 0.18 @xmath0 at core carbon exhaustion . for the other sequences at solar metallicity ( seqs . 5 , 8 , 9 , 10 , 11 , 13 , 21 ) where @xmath75 , the amounts of mass loss via case abb / bb transfer until core neon burning are rather moderate , varying from 0.02 to 0.1 @xmath0 . the wr wind mass loss rate has significant consequences in systems where @xmath76 3.0 @xmath0 . the previous study by wlw95 concluded that the final masses of single helium stars with initial masses of 3 20 @xmath0 should converge to @xmath77 2.2 3.5 @xmath0 . as expected from the lower wr mass loss rate adopted in this study , our helium star models give a significantly wider range of final masses ( @xmath5 ) at the given metallicity of 0.02 : @xmath78 2.9 7 @xmath0 ( 2.9 10 @xmath0 ) from helium stars of 3 20 @xmath0 for @xmath16 ( @xmath79 ; table [ tab2 ] ; cf . fig . [ fig : mimf ] ) . our binary star models also give generally larger final masses for sn ibc progenitors than those in wl99 : @xmath80 1.64 4.5 @xmath0 for @xmath81 from @xmath8213 25 @xmath0 , compared to @xmath80 2.0 3.4 @xmath0 in wl99 is determined by the detailed history of case abb / bb transfer , which depends on uncertain parameters such as the semi - convection efficiency . ] . figure [ fig : dmhe ] shows the amount of helium ( @xmath83 ) as a function of the final mass in the sn ibc progenitor models with @xmath84 at solar metallicity . in general , both binary and helium star models predict large amounts of helium in the envelope compared to the results of wlw95 and wl99 except for relatively low mass systems of @xmath85 @xmath0 . note that only rather massive progenitors with @xmath86 @xmath0 can be significantly helium deficient ( i.e. , @xmath87 ) . for the systems with @xmath88 @xmath0 , the role of case abb / bb mass transfer becomes important . in particular , the small amount of helium ( @xmath89 ) in seq . 3 shows that helium deficiency can also be achieved by the so - called case bb / abb mass transfer from relatively low mass helium cores ( @xmath90 ) . therefore , we expect two different classes for type ic progenitors at solar metallicity : one with @xmath91 @xmath0 and the other with @xmath92 5.5 @xmath0 ( see also * ? ? ? * ; * ? ? ? obviously , the upper limit of @xmath5 for potential type ic progenitors increases with a smaller mass loss rate : with @xmath93 , we have @xmath92 8.9 @xmath0 . in our binary star models , case ab or case b mass transfer does not completely remove hydrogen from the primary stars , as shown in fig . [ fig : chem ] . the resulting wr mass loss is therefore weaker than that from the corresponding pure helium star . this explains the fact that for a given wr mass loss rate , the binary star models with @xmath94 @xmath0 predict somewhat larger @xmath83 than the single helium star models do as shown in fig . [ fig : dmhe ] . more importantly , the binary star models show that , even at @xmath95 , small amounts of hydrogen can be retained up to the pre - supernova stage for a certain range of the final mass . [ fig : dmh ] ( see also table [ tab1 ] and fig . [ fig : chem ] ) shows the total mass of hydrogen in the primary stars at neon / oxygen burning phase . the time span from this stage to core collapse is supposed to be less than about 10 yr . the wr winds mass loss rate from the primary star models shown in the figure is about @xmath96 . some of the primary stars in this sample are still undergoing case abb or bb mass transfer ( seqs . 5 , 8 , 9 , 13 , & 21 and many of the smc models ; see table [ tab1 ] ) , but the mass transfer rate is only about @xmath97 . therefore , the amount of hydrogen at core collapse should remain close to the values given in the figure . fig [ fig : dmh ] indicates that the presence of hydrogen is only expected for @xmath98 \la 3.7 $ ] , at solar metallicity with @xmath99 . this results from two different reasons for different mass ranges . for @xmath100 , primary stars expand during the carbon burning phase to much larger radii than more massive ones do . the resulting mass loss rates via case abb or bb transfer thus become large enough to completely remove hydrogen from the primary stars by the time of core collapse . for @xmath101 , on the other hand , wr winds are rather strong and can remove hydrogen from the primary stars during the core helium burning phase . the upper limit of @xmath5 for the presence of hydrogen thus depends on the adopted mass loss rate . as shown in the figure , with @xmath93 , it increases to about 4.5 @xmath0 at solar metallicity . at smc metallicity , the stellar wind effect is less significant and even rather massive supernova progenitors of about 7 @xmath0 can retain fairly large amounts of hydrogen ( @xmath102 ) . we discuss implications of this result for supernova types in sect . [ sect : dischydrogen ] . relatively low mass helium stars usually experience a rapid expansion of the envelope during the core carbon burning phase . interestingly , our new models predict much larger radii at the pre - supernova stages than what the wlw95 models do , as shown in fig . [ fig : radius ] . the first reason for this difference is the updated opal opacity table by @xcite , which causes a strong iron bump at around @xmath103 . another reason is the smaller wr mass loss rate adopted in the present study , and the resultant thicker helium envelope in our models . the radius is further affected by the complication of binary interaction : the presence of a thin hydrogen layer in some of the binary models results in a more extended envelope than in the corresponding single helium star models . it is also worth noting that the radii of the binary star models at @xmath104 are smaller than those of the single helium star models . this is because the primary stars in these sequences are filling the roche - lobe , and the envelope can not expand beyond it . the metallicity effect is somewhat subtle . the hydrogen and helium layers in sn ibc progenitors become thicker due to the reduced mass loss rate for lower metallicity , while the opacity due to metals becomes smaller . as these two effects compensate each other , the radii of our sn ibc progenitor models at both smc and solar metallicities are found to be similar . finally , it should be noted that the radius depends on the final mass . in general , a less massive progenitor tends to have a larger radius at the presupernova stage . this might make mixing of nickel into the helium rich layer induced by the rayleigh - taylor instability during the supernova explosion more efficient for a less massive sn ibc progenitor , as implied by the simulations of @xcite and @xcite , resulting in more prominent helium lines in the spectra ( see sect . [ sect : discussion ] below ) . [ fig : mdot ] shows the mass loss rate due to wr winds at the central neon / oxygen burning phase from different sn ibc progenitor models . the mass loss rate ranges from @xmath105 to @xmath106 , depending on the final mass , at solar metallicity . the corresponding wind velocity ( i.e. , the escape velocity ) changes from @xmath107 to @xmath108 . the circumstellar interaction of such winds from primary stars in binary systems are supposed to be very complex because of the orbital motion and the wind - wind interaction with the secondary star . some of the primary star models shown in the figure are still undergoing case abb or bb mass transfer , which may add to the complexity . therefore , the nature of the circumstellar materials around sn ibc progenitors in a binary system may not follow the simple relation of @xmath109 that is expected for single wr star progenitors . we have presented new evolutionary models of massive close binary stars , considering tidal interaction , and transport of angular momentum and chemical species due to rotationally induced hydrodynamic instabilities and the spruit - tayler dynamo . we have investigated the redistribution of angular momentum in the primary star . although mass transfer and tidal interaction can significantly affect the evolution of the rotation velocity of the primary star on the main sequence , the amount of angular momentum retained in the core in the late evolutionary stages is rather insensitive to the previous history of such binary interactions because of the self - regulating nature of the spruit - tayler dynamo . we have also calculated non - rotating , mass - losing single helium star models and compared them with our primary star models in binary systems . our models adopt a much lower wr mass loss rate than in the previous studies by wlw95 and wl99 , and predict some new important properties of sn ibc progenitors accordingly . the following discussions are based on the models with @xmath16 , unless otherwise specified . 1 . the final masses of sn ibc progenitors in binary systems at @xmath110 @xmath111 are not limited to @xmath112 as predicted by wlw95 and wl99 , but a more wide range @xmath5 is expected ( i.e. , @xmath113 from @xmath114 @xmath0 ; see fig . [ fig : mimf ] ) . 2 . at @xmath115 @xmath111 , significant deficiency of helium ( @xmath116 @xmath0 ) is expected for @xmath117 and for @xmath118 ( fig . [ fig : dmhe ] ) . rather large amounts of helium up to 1.5 @xmath0 are expected for the other final mass range ( i.e. , @xmath119 5.5 @xmath0 ) . at @xmath115 @xmath120 , no such helium deficient sn progenitors are expected for the considered initial masses ( @xmath121 @xmath0 ) . a thin layer of hydrogen with @xmath122 @xmath0 is predicted for sn ibc progenitors with @xmath123 at @xmath124@xmath111 , and @xmath125 at @xmath115@xmath120 , respectively ( fig . [ fig : dmh ] ; table [ tab1 ] ) . 4 . most sn ibc progenitors with @xmath126 @xmath0 rapidly expand during core carbon burning , resulting in @xmath127 at the presuprnova stage . this is much larger than found in wlw95 ( fig . [ fig : radius ] ) . compact progenitors of @xmath128 are only expected for a relatively high mass ( @xmath129 @xmath0 at @xmath110 @xmath111 and @xmath130 @xmath0 at @xmath110 @xmath120 ; fig . [ fig : radius ] ) . our binary star models show that the mass transfer during helium core contraction ( case ab or case b ) in a close massive binary system can not remove the hydrogen envelope promptly enough to avoid the core braking due to the spruit - tayler dynamo during the helium core contraction phase . comparison of our binary star models with the single star models by @xcite and @xcite indicate that the amount of angular momentum retained in the core of the primary star at the presupernova stage should not be much different from those found in single star models if the spruit - tayler dynamo is adopted . i.e. , a specific angular momentum of a few @xmath68 in the innermost @xmath771.4 @xmath0 at the presupernova stage is expected in both single and binary stars . this value is smaller by one or two orders of magnitude than what is necessary to make a long gamma - ray bursts by magnetar or collapsar formation , or very energetic supernovae ( hypernovae ) powered by rapid rotation and strong magnetic fields ( e.g. * ? ? ? * ) , although it may suffice to produce millisecond pulsars @xcite . together with the work by @xcite , our results thus indicate that binary interactions with case ab / b mass transfers at @xmath131 may not particularly enhance the production of strongly rotation - powered events like long grbs or hypernovae . this is consistent with the observational evidence that such events are rare compared to normal core collapse events ( e.g. , * ? ? ? * ; * ? ? ? this also confirms the theoretical consensus that other evolutionary paths are needed to produce long grbs associated with sn ibc , such as the quasi - chemically homogeneous evolution of a metal poor star @xcite , tidal spin - up of a wr star in a very close binary system with a neutron star or black hole companion ( e.g. * ? ? ? * ; * ? ? ? * ) , or binary evolution with case c mass transfer with some specific conditions @xcite . larger radii of our sn ibc progenitor models than those previously found should have consequences in shock break - outs and bolometric light curves . for instance , a shock break - out from a larger envelope would be marked by a lower photosperic temperature . detailed comparison of numerical calculations with observational data may thus give strong constraints on sne ibc progenitor properties ( e.g. , * ? ? ? * ) . recent discovery of the x - ray outburst with sn 2008d by @xcite indeed suggests the usefulness of such a study for the probe of supernova progenitors ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , for which our new models would provide ideal input . we will address this issue in a forthcoming paper . although the weak signature or no evidence of helium in sne ic spectra may indicate the deficiency of helium in their progenitors , it is not well known how much helium can be hidden in the supernova spectra . it may also depend on the degree of mixing of nickel into helium rich layers @xcite . if we assume 0.5 @xmath0 , for instance , as the maximum amount of helium allowed for hiding helium lines in sn spectra , our models indicate that most sne ic progenitors at solar metallicity should belong to two distinct classes in terms of both zams and final masses , as summarized in fig . [ fig : mimf ] for the binary systems that undergo case b mass transfer . the same conclusion was also drawn by wl99 and @xcite . but pols & dewi considered different types of binary systems ( see below ) , and the finding of `` two mass classes '' for the final masses was not obvious in wl99 due to the very high wr rate adopted in their study , although it was clearly seen for the zams masses . if we assume that primary stars of @xmath132 produce low - mass - class sne ic via case bb mass transfer , about 62% of the sne ic from case b systems should belong to the high mass class and the rest ( @xmath7738% ) to the low mass class , at @xmath133 ( see fig . [ fig : mimf ] ) . it is important to note that the two classes are produced by different mechanisms . the high mass class of sne ic progenitors ( i.e. , @xmath134 @xmath0 in fig [ fig : mimf ] ) is a consequence of wr winds mass loss , while the low mass class results from case abb / bb mass transfer as discussed in sect . [ sect : masses ] . the zams mass range for the low mass class may not be much affected by metallicity , while it should be widened with increasing metallicity for the high mass class . this leads to the conclusion that the low and high mass classes would dominate at low and high metallicities respectively . it should also be noted that the final mass range for the low mass class may not change much for different metallicities , while it may decrease with increasing metallicity for the high mass class , due to the increasing wr winds mass loss rates , as implied by the result of wl99 . in the massive close binary systems considered in this paper , the primary star masses become much smaller than those of the secondary stars when case abb or case bb mass transfer begins . if the companion star mass were lower than the helium star in a close binary system , the mass transfer rate should become higher than in the systems of the present study . for example , if a helium star is located in a very short period binary system ( @xmath135 ) with a less massive companion ( e.g. a neutron star ) , mass transfer from the helium star may occur rapidly enough to make a helium - deficient carbon star , even for @xmath136 6.0 @xmath0 as shown by @xcite , @xcite and @xcite . the final masses of such sn ic progenitors may range from 1.5 @xmath0 to 3.0 @xmath0 . this scenario was also suggested by @xcite to explain the fast light curve of type ic sn 1994i . the zams mass of such a sn ic should be in the range of 12 20 @xmath0 @xcite . however , such close helium star plus neutron star systems are supposed to rarely form , and might not contribute much to the population of sne ic , compared to the systems considered in this study . our results should have several observational consequences . as mentioned above , the population of sne ic should be dominated by the high mass class for @xmath137 . they would have higher zams and final masses than those of typical sne ib progenitors ( cf . * ? ? ? * ; * ? ? ? given that the parameter space for the high mass class sne ic may become larger with a higher wr mass loss rate , the number ratio of sne ic to sne ib , and that of high mass class sne ic to low mass class sne ic should increase with increasing metallicity ( cf . * ; * ? ? ? * ; * ? ? ? the existence of the two mass classes of sne ic progenitors may be related to some aspects of the observational diversity of sne ic . for example , sne ic of the low mass class is likely to be characterized by rather fast declining light curves and low luminosities ( cf . * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , implying that the observed population of sne ic is likely to be baised to high - mass - class sne ic . we should also note the huge difference of the binding energy between the two classes . the binding energy of the envelope above 1.4 @xmath0 in the sn ic progenitor star model of seq . 3 ( @xmath138 @xmath0 ) is only about @xmath139 erg , while it should be one or two orders of magnitude higher for a sn ic progenitor with @xmath140 @xmath0 ( see table [ tab2 ] ) . as a consequence , the energetics of sne ic might be systematically different for the two different classes . on the other hand , the assumption of @xmath141 @xmath0 for sn ic progenitors leads to a ratio of type ic to ib supernova rate ( ic / ib ratio ) of about 0.4 from binary systems at solar metallicity . have @xmath142 @xmath0 with @xmath143 @xmath0 ( see table [ tab1 ] ) . this implies that the low - mass class sne ic would predominantly occur in binary systems at this metallicity . if we assume stars with @xmath144 would produce low - mass class sne ic as in the case of solar metallicity , the sn ic / sn ib ratio would be about 0.1 at smc metallicity . the exploration of the exact mass range for the low - mass class sne ic is a time - consuming task , and we plan to investigate this in near future . ] this appears in contradiction with recent observations that indicate rather a high ic / ib ratio of about 2.0 ( e.g. * ? ? ? this discrepancy would become even larger with @xmath145 , as implied by fig . [ fig : mimf2 ] . this raises a question on the nature of sn ic progenitors , and it should be kept in mind that we still do not fully understand what distinguishes sn ic progenitors from those of sn ib . a recent work by @xcite indicates that the mass fraction of helium in the outermost layers ( @xmath146 ) , rather than the total mass of helium , may be more relevant for the presence of helium lines in supernova spectra . specifically , it is shown that if helium is well mixed with co material such that @xmath146 becomes less than about 0.5 , helium lines are not seen in early time spectra , despite rather a large total amount of helium ( @xmath147 @xmath0 ) , if non - thermal excitation is absent . in our progenitor models , such a small @xmath146 is realized only for @xmath92 5.5 @xmath0 at solar metallicity ( with @xmath16 ) . this is not different from the above - discussed mass limit for having @xmath2 @xmath0 , implying that the initial mass range for sn ic progenitors would not change much even if we adopted @xmath146 as a criterion , at least for the high mass class . on the other hand , we have @xmath148 in the primary star of seq . 3 at carbon exhaustion while the total mass of helium is less than 0.2 @xmath0 . this imples that the initial mass range for the low mass class sn ic progenitors might be affected if the condition of @xmath149 for sn ic progenitors were applied . but @xcite did not yet calculate such a low mass sn progenitor model ( @xmath91 @xmath0 ) , and their analyses were limited to early times of supernovae . it remains to be an important subject of future work to systematically investigate which types of supernova progenitors would lead to the presence or absence of helium lines in the supernova spectra at different epochs , including the effect of non - thermal excitation . therefore , the above discussion on type ic progenitors based on the total amount of helium should only be considered indicative at this stage . it is interesting that , at @xmath133 , the presence of a thin hydrogen layer is only expected for a limited range of the initial / final mass of sn ib progenitors , as shown in figs [ fig : dmh ] , and [ fig : mimf ] . the detection of hydrogen absorption lines at high velocity has been indeed reported in many sne ib ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , in favor of our model prediction for the presence of a thin hydrogen layer in sne ib progenitors . this might provide a strong constraint for the progenitor masses of observed sne ib , in principle . note also that explosions of such helium stars with thin hydyrogen layers could be recognized as sn iib rather than ib , if hydrogen lines were detected short after supernova explosion , e.g. , as in the case of sn 2008ax @xcite and as recently discussed by @xcite and @xcite . the radii of these progenitor models range from @xmath150 to @xmath151 . they may corredpond to the compact category of sn iib progenitors , which is discussed in @xcite . the relatively low ejecta masses of such sne iib are consitent with our model predictions . on the other hand , case c mass transfer can also leave helium cores covered with small amounts of hydrogen envelope . as the life time of such stars made via case c mass transfer should be rather short , they can retain much more hydrogen ( @xmath152 @xmath0 ) , than what is predicted from our binary models with case ab / b mass transfer . such a star may eventually explode as a sn iib like sn 1993j ( e.g. * ? ? ? * ; * ? ? ? * ) , with a much extended envelope ( @xmath153 ) . therefore , the two categories of sn iib progenitors according to their sizes , which has been recently suggested by @xcite , may be understood within the framework of binary evolution ; sne iib of the compact type may be produced via case ab / b mass transfer ( especially at @xmath154 ) , and sne iib of the extended type via case c mass transfer . this work is supported by the doe scidac program ( doe de - fc02 - 06er41438 ) , the nsf grant ( nsf - arra ast-0909129 ) , and the nasa theory program ( nnx09ak36 g ) . we are grateful to luc dessart for useful discussions . cccccccccccccc 1 & 0.02 & 12 & 8 & 3.0 & 5 & b+bb & 57.9 & 1.40 & 1.21 & 0.17 & 0.0 & 0.33 & onemg wd + 2 & 0.02 & 12 & 11 & 4.0 & 5 & b+bb & 104.4 & 1.48 & 1.24 & 0.20 & 0.0 & 0.35 & onemg wd + 3 & 0.02 & 13 & 11 & 5.0 & 5 & b+bb & 123.3 & 1.64 & 1.43 & 0.18 & 0.0 & 0.22 & sn ic + 4 & 0.02 & 14 & 12 & 3.0 & 5 & a+ab+bb & 118.5 & 1.33 & 1.09 & 0.22 & 0.0 & @xmath155 & onemg wd + 5 & 0.02 & 14 & 12 & 5.0 & 5 & b+bb & 30.7 & 2.97 & 1.66 & 1.24 & 1.9(-4 ) & 0.25 & snib + 6 & 0.02 & 16 & 14 & 2.0 & 5 & a : contact & & & & & & & + 7 & 0.02 & 16 & 14 & 3.0 & 5 & a+ab+abb & 101.8 & 1.54 & 1.33 & 0.17 & 0.0 & 0.39 & onemg wd + 8 & 0.02 & 16 & 14 & 4.0 & 5 & b+bb & 26.2 & 3.66 & 2.05 & 1.47 & 4.5(-3 ) & 0.24 & snib + 9 & 0.02 & 16 & 14 & 5.0 & 5 & b+bb & 33.7 & 3.65 & 2.04 & 1.47 & @xmath156 & 0.25 & snib + 10 & 0.02 & 18 & 12 & 3.0 & 5 & a+ab+abb & 27.9 & 2.66 & 1.58 & 1.01 & @xmath157 & 0.26 & snib + 11 & 0.02 & 18 & 12 & 3.0 & 10 & a+ab+abb & 27.3 & 2.74 & 1.59 & 1.08 & @xmath157 & 0.26 & snib + 12 & 0.02 & 18 & 12 & 5.0 & 10 & b : contact & & & & & & & + 13 & 0.02 & 18 & 17 & 3.0 & 10 & a+ab+abb & 36.2 & 3.03 & 1.68 & 1.27 & @xmath158 & 0.25 & sn ib + 14 & 0.02 & 18 & 17 & 4.0 & 10 & a+ab & 29.7 & 3.79 & 2.14 & 1.49 & @xmath159 & 0.25 & sn ib + 15 & 0.02 & 18 & 17 & 4.0 & 10 & a+ab & 24.0 & 3.97 & 2.27 & 1.53 & @xmath160 & 0.26 & snib + 16 & 0.02 & 18 & 17 & 4.0 & 10 & a+ab & 50.0 & 3.80 & 2.16 & 1.50 & @xmath161 & 0.06 & snib + 17 & 0.02 & 18 & 17 & 4.0 & 10 & a+ab & 25.2 & 3.84 & 2.18 & 1.50 & @xmath162 & 0.26 & snib + 18 & 0.02 & 18 & 17 & 4.0 & 10 & a+ab & 30.6 & 3.73 & 2.14 & 1.43 & @xmath163 & 3.57 & snib + 19 & 0.02 & 18 & 17 & 5.0 & 3 & b & 33.1 & 3.73 & 2.33 & 1.23 & @xmath157 & 0.25 & snib + 20 & 0.02 & 18 & 17 & 5.0 & 5 & b & 32.4 & 4.04 & 2.45 & 1.4 & 0.00 & 0.33 & snib + 21 & 0.02 & 18 & 17 & 5.0 & 10 & b+bb & 31.5 & 4.41 & 2.51 & 1.68 & @xmath164 & 0.26 & snib + 22 & 0.02 & 18 & 17 & 6.0 & 10 & b & 39.3 & 4.39 & 2.56 & 1.62 & @xmath165 & 0.26 & snib + 23 & 0.02 & 25 & 19 & 6.0 & 10 & b : contact & & & & & & & + 24 & 0.02 & 25 & 24 & 2.0 & 10 & a : contact & & & & & & & + 25 & 0.02 & 25 & 24 & 3.0 & 3 & a+ab & 22.7 & 3.70 & 2.46 & 0.98 & 0.0 & 0.24 & snib + 26 & 0.02 & 25 & 24 & 3.0 & 5 & a+ab & 22.4 & 4.33 & 2.80 & 1.30 & 0.0 & 0.25 & snib + 27 & 0.02 & 25 & 24 & 3.0 & 10 & a+ab & 21.3 & 5.07 & 3.17 & 1.67 & 0.0 & 0.25 & snib + 28 & 0.02 & 25 & 24 & 3.0 & 10 & a+ab & 18.9 & 5.08 & 3.19 & 1.66 & 0.0 & 0.32 & snib + 29 & 0.02 & 25 & 24 & 4.0 & 5 & a + ab & 21.5 & 4.45 & 2.91 & 1.22 & 0.0 & 0.26 & snib + 30 & 0.02 & 25 & 24 & 6.0 & 10 & b & 27.4 & 6.49 & 4.45 & 1.63 & 0.0 & 0.39 & snib + 31 & 0.02 & 60 & 40 & 7.0 & 3 & a & 16.8 & 4.95 & 3.70 & 0.25 & 0 & 0.24 & snic + 32 & 0.004 & 16 & 12 & 3.0 & 5 & b+bb & 64.75 & 3.91 & 2.22 & 1.54 & @xmath166 & 0.24 & snib + 33 & 0.004 & 16 & 14 & 3.0 & 5 & b+bb & 19.6 & 3.90 & 2.21 & 1.53 & @xmath166 & 0.24 & snib + 34 & 0.004 & 16 & 14 & 5.0 & 5 & b+bb & 21.8 & 3.84 & 2.19 & 1.51 & @xmath167 & 0.24 & snib + 35 & 0.004 & 18 & 12 & 5.0 & 5 & b+bb & 14.4 & 4.64 & 2.76 & 1.68 & @xmath168 & 0.31 & snib + 36 & 0.004 & 18 & 12 & 8.0 & 5 & b+bb & 24.33 & 4.56 & 2.68 & 1.67 & @xmath169 & 0.33 & snib + 37 & 0.004 & 18 & 17 & 3.0 & 5 & a+ab & 18.4 & 4.42 & 2.55 & 1.67 & @xmath170 & 0.26 & snib + 38 & 0.004 & 18 & 17 & 3.0 & 10 & a+ab & 14.4 & 4.58 & 2.61 & 1.77 & @xmath171 & 0.26 & snib + 39 & 0.004 & 18 & 17 & 6.0 & 5 & b+bb & 27.0 & 4.57 & 2.71 & 1.65 & @xmath166 & 0.27 & snib + 40 & 0.004 & 25 & 12 & 3.0 & 5 & a : contact & & & & & & & + 41 & 0.004 & 25 & 12 & 6.0 & 5 & b : contact & & & & & & & + 42 & 0.004 & 25 & 19 & 3.0 & 5 & a+ab & 10.2 & 7.09 & 4.87 & 2.03 & @xmath172 & 0.32 & snib + 43 & 0.004 & 25 & 24 & 3.0 & 5 & a+ab & 13.0 & 7.31 & 5.05 & 2.07 & @xmath173 & 0.28 & snib + 44 & 0.004 & 25 & 24 & 6.0 & 5 & b : contact & & & & & & + 45 & 0.004 & 40 & 30 & 4.0 & 5 & a+ab & 9.31 & 12.0 & 9.42 & 1.24 & @xmath174 & 0.56 & bh cccccccc 5 & 2.8 & 2.73 & 1.51 & 1.16 & 163.54 & 5.08 & 0.01 + 5 & 3.0 & 2.91 & 1.60 & 1.22 & 33.40 & 1.79 & 0.03 + 5 & 4.0 & 3.72 & 2.15 & 1.43 & 8.01 & 0.73 & 0.18 + 5 & 6.0 & 4.23 & 2.77 & 1.19 & 4.19 & 0.38 & 0.25 + 5 & 8.0 & 4.95 & 3.40 & 0.85 & 3.07 & 0.30 & 0.29 + 5 & 10.0 & 5.49 & 3.93 & 0.54 & 1.59 & 0.22 & 0.47 + 5 & 15.0 & 6.44 & 4.85 & 0.31 & 0.59 & 0.18 & 0.55 + 5 & 20.0 & 7.09 & 5.44 & 0.22 & 0.48 & 0.16 & 0.66 + 10 & 3.0 & 2.95 & 1.60 & 1.27 & 34.96 & 1.84 & 0.03 + 10 & 4.0 & 3.84 & 2.20 & 1.49 & 6.94 & 0.70 & 0.15 + 10 & 6.0 & 4.93 & 3.14 & 1.55 & 4.29 & 0.51 & 0.19 + 10 & 8.0 & 6.06 & 4.13 & 1.41 & 2.55 & 0.35 & 0.33 + 10 & 10.0 & 7.01 & 4.96 & 1.12 & 1.94 & 0.25 & 0.55 + 10 & 15.0 & 8.89 & 6.80 & 0.47 & 0.54 & 0.20 & 0.82 + 10 & 20.0 & 10.33 & 8.21 & 0.31 & 0.40 & 0.16 & 1.17

it seems generic to have vacua with lower dimensionality than ours . we consider the possibility that the observable universe originated in a transition from one of these vacua . such a universe has anisotropic spatial curvature . this may be directly observable through its late - time effects on the cmb if the last period of slow - roll inflation was not too long . these affect the entire sky , leading to correlations which persist up to the highest cmb multipoles , thus allowing a conclusive detection above cosmic variance . further , this anisotropic curvature causes different dimensions to expand at different rates . this leads to other potentially observable signals including a quadrupolar anisotropy in the cmb which limits the size of the curvature . conversely , if isotropic curvature is observed it may be evidence that our parent vacuum was at least 3 + 1 dimensional . such signals could reveal our history of decompactification , providing evidence for the existence of vastly different vacua .

1003.0236

our current understanding of cosmology and high energy physics leaves many questions unanswered . one of the most fundamental of these questions is why our universe has three large dimensions . this may be tied to the more general question of the overall shape and structure of the universe . in fact , it is possible that our universe was not always three dimensional or that other places outside of our observable universe have a different dimensionality . there are surely long - lived vacua where one or more of our three dimensions are compactified , since this does not even rely on the presence of extra - dimensions and indeed happens in the standard model @xcite . eternal inflation can provide a means to populate these vacua , and naturally leads to a highly inhomogeneous universe on very long length scales . further , it seems likely that these lower - dimensional vacua are at least as numerous as three dimensional ones since there are generally more ways to compactify a greater number of spatial dimensions . if we do indeed have a huge landscape of vacua ( e.g. @xcite ) then it seems all the more reasonable that there should be vacua of all different dimensionalities and transitions between them ( see e.g. @xcite ) . we will ignore the subtle issues of the likelihood of populating those vacua ( the measure problem " ) . instead we will focus on the possibility of observing such regions of lower dimensionality since surely such a discovery would have a tremendous effect on our understanding of cosmology and fundamental physics . our compact dimensions are generically unstable to decompactification @xcite . thus it seems possible that the universe began with all the dimensions compact ( the starting point in @xcite for example ) . in this picture our current universe is one step in the chain towards decompactifying all dimensions . of course , eternal inflation may lead to a very complicated history of populating different vacua , but in any case , it seems reasonable to consider the possibility that we came from a lower dimensional ancestor " vacuum . we will assume that prior to our last period of slow - roll inflation our patch of the universe was born in a transition from a lower dimensional vacuum . our universe then underwent the normal period of slow - roll inflation . for our signals to be observable we will assume that there were not too many more than the minimal number of efolds of inflation necessary to explain the cmb sky . this may be reasonable because this is very near a catastrophic boundary : large scale structures such as galaxies would not form if inflation did not last long enough to dilute curvature sufficiently @xcite . since achieving slow - roll inflation is difficult and the longer it lasts the more tuned the potential often is , there may be a pressure to be close to this lower bound on the length of inflation . we will actually use the energy density in curvature , @xmath0 , in place of the number of efolds of inflation . the observational bound requires that @xmath1 today ( this corresponds to @xmath2 efolds for high scale inflation ) . the existence of galaxies requires @xmath3 today ( corresponding to @xmath4 efolds if we use the bound from @xcite ) . thus @xmath0 may be close to the observational bound today . other , similar arguments have also been made for a relatively large curvature today @xcite . most signals of the presence of other vacua , e.g. bubble collisions @xcite , also rely on this assumption . these signals have also mostly been explored assuming that the other vacua are all 3 + 1 dimensional . while an important first step , this seems like a serious oversimplification . we find interesting differences in the case that our parent vacuum was lower dimensional . in particular , our universe can be anisotropic , with different spatial curvatures in the different directions . this anisotropic curvature dilutes exponentially during inflation , making the universe appear very isotropic at early times . however , this curvature ( @xmath0 ) grows at late times , leading to several observable effects . this anisotropic curvature sources an anisotropy in the hubble expansion rate , since the different dimensions expand at different rates . the most interesting signal is an anisotropy in the normal cmb curvature measurement . the angular size of a standard ruler " now appears to depend on the orientation of that ruler . in the cmb this shows up as unexpected correlations between modes of all angular sizes . unlike the normal curvature measurement , this anisotropic curvature measurement is not degenerate with the scale factor expansion history and is thus easier to measure . this anistropic curvature also leads to a significant quadrupolar anisotropy in the cmb which constrains the size of @xmath0 . there are possibly other observables from 21 cm measurements , direct measurements of the hubble expansion ( e.g. from supernovae ) , or from searches looking for nontrivial topology of the universe . in this section we will compute the evolution of a universe that began with one or two of our three spatial dimensions compactified . we will consider the possibility that our universe began in a lower - dimensional vacuum . in particular we assume that just prior to our recent period of slow - roll inflation , the currently observable part of the universe ( our pocket universe " in landscape terminology ) was in a vacuum with only one or two large , uncompactified spatial dimensions . the other dimensions , including the one or two that will eventually become part of our three large spatial dimensions , are compactified and stable . the universe then tunnels , nucleating a bubble of our vacuum in which three spatial dimensions are uncompactified and thus free to grow with the cosmological expansion . we will consider starting from either a 1 + 1 or 2 + 1 dimensional vacuum . we will not consider the 0 + 1 dimensional case in great detail , as it is significantly different @xcite . however it is possible that it will have the same type of signatures as we discuss for the other two cases , depending on the details of the compactification manifold . consider first the case that the universe is initially 2 + 1 dimensional , and in the tunneling event one of the previously compactified spatial dimensions becomes decompactified , losing whatever forces were constraining it and becoming free to grow ( in the tunneling event it may also grow directly ) . we can think of this as a radion for that dimension which is initially trapped in a local minimum , tunneling to a section of its potential where it is free to roll . of course , the tunneling event may actually be due to a change in the fluxes wrapping the compact dimension , or in general to a change in whatever is stabilizing that dimension . the exact nature of this tunneling will not concern us since the further evolution of the universe is relatively insensitive to this . in all cases a bubble of the new vacuum is formed in the original 2 + 1 dimensional space . the bubble wall ( which is topologically an @xmath5 not an @xmath6 ) expands outward . the interior of this coleman - de luccia bubble @xcite is an infinite , open universe with negative spatial curvature ( see e.g. @xcite for this bubble in arbitrary dimensionality space - times ) . but this negative spatial curvature is only in two dimensions . the third , previously small , dimension may be topologically an @xmath5 or an interval , but in any case will not have spatial curvature . thus the metric after the tunneling inside the bubble is @xmath7 where @xmath8 is the coordinate of the previously compactified dimension and @xmath9 for negative spatial curvature in the @xmath10 plane . this is known as a bianchi iii spactime . if instead the universe is initially 1 + 1 dimensional and two spatial dimensions decompactify in the transition then the situation will be reversed . the single originally large dimension ( now labelled with coordinate @xmath8 ) will be flat but the other two dimensions may have curvature ( either positive or negative ) . for example , if they were compactified into an @xmath6 they would have positive curvature and so would be described by eqn . with @xmath11 , known as a kantowski - sachs spacetime . or if those two dimensions were a compact hyperbolic manifold , for example , they would be negatively curved with @xmath9 . in fact , generically compactifications do have curvature in the extra dimensions ( see for example @xcite ) . of course it is also possible that the two compact dimensions had zero spatial curvature . we will not consider this special case in great detail since it does not lead to most of our observable signals . we will thus assume that our universe begins with anisotropic spatial curvature , with metric as in eqn . . immediately after the tunneling event the universe is curvature dominated , though in this case of course the curvature is only in the @xmath10 plane . we assume the universe then goes through the usual period of slow - roll inflation , with a low number of efolds @xmath12 near the curvature bound . the equations of motion ( the frw equations " ) are : @xmath13 where the dot @xmath14 denotes @xmath15 , @xmath16 is the energy density , and @xmath17 and @xmath18 are the pressures in the @xmath19 and @xmath8 direction , i.e. the @xmath20 and @xmath21 components of the stress tensor @xmath22 . these can be rewritten in terms of the two hubble parameters @xmath23 and @xmath24 as @xmath25 at least in the case of tunneling from 2 + 1 to 3 + 1 dimensions , immediately after the tunneling event the universe is curvature dominated . in this case eqn . can be solved for @xmath26 directly . since this is just the usual isotropic frw equation , the solution is as usual @xmath27 , where @xmath28 is the bubble wall . actually , since we will assume the universe transitions to a period of slow - roll inflation after curvature dominance , we will assume there is a subdominant vacuum energy during the period of curvature dominance . this then gives a perturbutive solution accurate up to linear order in the vacuum energy @xmath29 of @xmath30 . then we can solve eqn . perturbatively for @xmath31 . there are several possible solutions but these are reduced because we will assume that immediately after the tunneling event @xmath32 . if we imagine the transition as a radion field tunneling through a potential barrier then we know that the radion generically starts from rest after the tunneling . with this boundary condition the solution to linear order in the vacuum energy is @xmath33 where @xmath34 is the initial value of @xmath35 . since the period of curvature dominance ends when @xmath36 , we see that roughly @xmath26 expands linearly while @xmath35 remains fixed during this period . thus the different expansion rates @xmath37 and @xmath38 remain very different during this period . @xmath37 is large while @xmath39 . the flat dimension will not begin growing rapidly until inflation begins . at that point though , it will be driven rapidly towards the same expansion rate as the other dimensions , @xmath40 , as we will now show . since our observed universe is approximately isotropic , we will only need to solve these equations in the limit of small @xmath41 . we will always work to linear order in @xmath42 . . - eqn . gives @xmath43 note that we have taken the pressure to be isotropic , @xmath44 , which is approximately true in all cases of interest to us . this is clearly true during inflation . during radiation dominance ( rd ) the radiation is in thermal equilibrium . since the reactions keeping it in equilibrium have rates much higher than the hubble scales during this time , the pressure is kept locally isotropic . during matter dominance ( md ) the pressure is zero to leading order . the sub - leading order piece due to the photons will also remain isotropic until after decoupling since the photons remain in equilibrium until this time . after decoupling the energy density in radiation is quite small compared to the matter density . further , this small pressure only develops anisotropy due to the differential expansion ( and hence redshifting ) between the @xmath19 and @xmath8 directions . thus the anisotropy in pressure is proportional to both @xmath42 and the small overall size of the pressure and is therefore negligible for us . the anisotropic spatial curvature in the metric is the only effect breaking isotropy in this universe and thus the only reason for a differential expansion rate @xmath42 . in fact , as we will see shortly , the differential expansion @xmath42 is proportional to @xmath0 , the curvature energy density , defined to be @xmath45 since @xmath0 grows during rd and md and it is @xmath46 today , it was quite small during the entire history of the universe after the period of curvature dominance ( to which we will return later ) . so we will treat both @xmath42 and @xmath0 as our small parameters and work to linear order in each . if we combine eqns . and we find an equation for @xmath38 which is true in the limit of small @xmath42 @xmath47 notice that this is exactly the same as the equation for @xmath37 ( eqn . ) but with the sign of the curvature term flipped . eqn . for @xmath37 is just the usual isotropic frw equation . thus @xmath48 behaves exactly as it would in the normal isotropic universe with a subleading curvature component and @xmath31 behaves as if it was the scale factor in a universe with an equal magnitude but opposite sign of curvature . can be solved easily because we only need the leading order behavior of @xmath26 and @xmath37 which are just the usual isotropic frw solutions as can be seen easily since eqns . and are just the usual frw equations . solving eqn . during the eras of interest and keeping only the inhomogeneous solutions yields @xmath49 as we will show later , the homogeneous solutions all die off as faster functions of time and are thus negligible . interestingly , this implies that @xmath42 is effectively independent of initial conditions . at every transition some of the homogeneous solution for @xmath42 is sourced , for example to make up the missing @xmath50 when transitioning from inflation to rd . but this homogeneous piece dies off faster , leaving only the inhomogeneous piece which is independent of the initial value of @xmath42 . to find the solutions for the scale factors @xmath48 and @xmath31 up to linear ( subleading ) order in the curvature , we solve eqns . , , and perturbatively in @xmath0 . the leading order behavior comes from the dominant energy density ( vacuum energy , radiation , or matter in our three eras ) . we will only need the solution during md so we can assume @xmath51 then . contains no @xmath35 s so it can be solved directly for @xmath48 . once we have the solution for @xmath48 up to linear order in @xmath0 we then plug in to eqn . to find @xmath31 also to linear order . the solutions during md to linear order in @xmath0 are @xmath52 where @xmath53 is an arbitrary , physically meaningless constant arising from the coordinate rescaling symmetry . thus this universe always has a differential expansion rate between the @xmath8 direction and the @xmath19-@xmath54 directions which is proportional to @xmath0 . the precise constant of proportionality depends only on the era ( inflation , rd , or md ) and not on initial conditions . further , the @xmath19-@xmath54 plane expands as in the usual isotropic frw universe , while the @xmath8 direction expands as if it was in that same universe except oppositely curved . during an initial period of curvature dominance the @xmath8 dimension remains constant while the other two dimensions expand , diluting curvature . during this period the expansion rates @xmath37 and @xmath38 are maximally different . then a period of slow - roll inflation takes over . during this period the expansion rates are driven exponentially close together . this difference in expansion rates is largest at the beginning of inflation , immediately after curvature dominance , when @xmath0 is still large . during inflation curvature dilutes exponentially as @xmath55 . so at the end of inflation the differential expansion rate is completely negligible @xmath56 . then during rd @xmath0 and hence also @xmath42 remain small , though growing as @xmath57 . during md @xmath0 and @xmath42 continue to grow @xmath58 , finally reaching their maximal value when the universe transitioned to vacuum energy dominance around redshift @xmath59 . since this final transition was so recent ( and the homogeneous solution for @xmath42 has not even had much time to die off yet ) we will approximate the universe as matter dominated until today . in this section , we discuss the late time observables of anisotropic curvature . we begin by computing its effects on standard rulers . these effects emerge due to the warping of null geodesics in the anisotropic background metric . null geodesics along different directions are warped differently by the curvature , leading to differences in the observed angular size of standard rulers in the sky . following this discussion , we compute the effect of anisotropic curvature on the cmb . the cmb is also affected by the warping of the null geodesics that propagate from the surface of last scattering to the current epoch . this warping affects the relation between the angle at which a cmb photon is observed today and the point at which it was emitted during recombination . in addition to this effect , the anisotropic metric discussed in section [ sec : metric ] also leads to differential hubble expansion . this leads to an anisotropic red shift in the universe , which causes a late time observer to see additional temperature anisotropies in the cmb . we conclude the section with a discussion of additional measurements that could be performed in upcoming experiments . in this section we present a calculation of the effect on standard rulers . while this is not a directly observable effect itself since we have no exact standard rulers in the sky , it does provide good intuition for the following calculation of the actual cmb observables in section [ subsec : cmb ] . further , many of the results of this section are used directly in that calculation . the spacetimes considered in this paper are curved and anisotropic . a canonical method to observe curvature is through the measurement of the angular sizes of standard rulers . curvature modifies the euclidean relationship between the measured angle and the linear size of the ruler . in a universe with anisotropic curvature , we expect this deviation from euclidean geometry to change with the angular position and orientation of the ruler . motivated by the use of baryon acoustic oscillations as cosmological standard rulers , we compute the present day angular size of standard rulers located at the surface of last scattering . this calculation gives intuition for the effects of anisotropic curvature on the cmb ( studied in detail in section [ subsec : cmb ] ) . to do so , we first determine the null geodesics that connect the surface of last scattering to a present day observer . the angle subtended between the two null geodesics that reach the end points of the standard ruler is then the angular size of the ruler . for simplicity , we assume that the universe was matter dominated throughout the period between recombination and the present epoch . we work with the metric @xmath60 which is produced when a @xmath61 dimensional universe is produced by tunneling from a @xmath62 dimensional vacuum . we restrict our attention to this scenario in order to facilitate concrete computation . however , our results can be applied to a wide class of scenarios that lead to anisotropic geometries . the metric describes a universe where two of the spatial dimensions ( parameterized by the coordinates @xmath63 in ) have negative curvature and grow with scale factor @xmath64 . the other dimension , parameterized by the coordinate @xmath8 in , grows with scale factor @xmath65 . the space - time geometry of such a universe can also be described using the metric with @xmath66 . these metrics are related by a coordinate transformation and they yield identical frw equations , and for the scale factors @xmath64 and @xmath65 . th is setup also describes anisotropic universes with positive curvature ( equation with @xmath11 ) . such a universe is described by the metric with the @xmath67 term replaced by @xmath68 . with this metric , the frw ( equations , and ) and null geodesic equations have the same parametric forms . our calculations also apply to this case , with the difference between the two cases being captured by the sign of the curvature term @xmath0 . an observer o ( see figure [ fig : geodesics ] ) at the present time receives photons from the surface of last scattering @xmath69 . this photon follows a null geodesic . in computing this null geodesic , we can assume without loss in generality that the point o lies at the origin of the coordinate system . with this choice , we focus on geodesics that lie along a direction of constant @xmath54 . these geodesics contain all the information required to describe our setup . the geodesics that connect the point o with the surface of last scattering have zero velocity along the @xmath54 direction . the @xmath70 symmetry in the @xmath63 plane then implies that @xmath54 remains constant during the subsequent evolution of the geodesic . using the metric , the null geodesic equations that describe the photon s trajectory @xmath71 are @xmath72 where the dots denote derivatives with respect to @xmath73 . with the boundary condition that the null geodesic reaches o at time @xmath74 , equation can be solved perturbatively to leading order in @xmath0 . the coordinates @xmath75 on @xmath69 from which the photon is emitted are @xmath76 in the above expression , @xmath77 is a parameter that governs the direction in the @xmath78 plane from which the photon is received at o and @xmath79 denotes the fractional energy density in curvature at the present time . the physical angle @xmath80 between the photon s trajectory and the @xmath8 axis , as measured by a local observer , is different from @xmath77 , and is given by @xmath81 for the geodesics computed in , the relation between the parameter @xmath77 and the physical angle @xmath82 observed at o is @xmath83 the @xmath84 corrections in the definition arise because the coordinates @xmath85 used to describe the metric are not locally flat . local coordinates @xmath86 can be constructed at any point @xmath87 of the space - time . these two sets of coordinates are related by @xmath88 the coordinate transformations in imply that operators constructed from global coordinates ( _ e.g. _ @xmath89 ) differ from the corresponding operator in the local inertial frame ( _ e.g. _ @xmath90 ) by quantities @xmath91 . the difference between these operators is suppressed by the ratio of the size of the local experiment over the hubble radius . these differences are negligible for any local experiment today . the angle defined by is therefore very close to the physical angle measured by a local experiment and we will use this definition for subsequent calculations . with the knowledge of the geodesics , we can calculate the angular size of a standard ruler of length @xmath92 at the time of recombination . since @xmath0 is very small during this time , the physical size of the ruler is independent of its location and orientation . first , consider a ruler oriented in the @xmath8 direction . this ruler lies between the co - ordinates @xmath93 and @xmath94 at the time @xmath95 of recombination . the length of this ruler is @xmath96 using and in , we find that the angular size @xmath97 subtended by a ruler of length @xmath92 at a local experiment o is @xmath98 a similar procedure can also be adopted to describe standard rulers that lie along the @xmath63 plane . the angular size of these rulers is given by the angle @xmath54 between the null geodesics that connect the ends of the ruler to the origin . following the above procedure , this angular size @xmath99 is @xmath100 the angular size of a standard ruler thus changes when its location and orientation are changed ( see figure [ fig : rulers ] ) . for a ruler located at @xmath101 ( _ i.e. _ @xmath102 ) the warp of the angle ( in equations and ) changes from @xmath103 for a ruler in the @xmath63 plane to @xmath104 for a ruler oriented in the @xmath8 direction . similarly , as a ruler oriented in the @xmath8 direction is moved from @xmath105 to @xmath106 , the angular warp factor changes from @xmath104 to @xmath107 . the reason for this change can be traced to the fact that for a ruler oriented in the @xmath8 direction , all of the angular warp occurs due to the effect of the curvature on the scale factor . @xmath64 and @xmath65 expand as though they have the same magnitude of the curvature but with opposite sig n. consequently , the angular warps along the two directions also have the same magnitude , but are of opposite sign . this angular dependence is an inevitable consequence of the anisotropic curvature @xmath0 endemic to this metric . we note that this measurement of the anisotropic curvature is relatively immune to degeneracies from the cosmological expansion history since the angular size changes depending upon the orientation of the ruler along every line of sight . we discuss how this measurement can be realized using cmb measurements in section [ subsec : statisticalanisotropy ] . the cmb offers a unique probe of the space - time geometry between the surface of last scattering and the current epoch . the spectral characteristics of the cmb photons at the time of last scattering are well determined . differences between this well determined spectrum and observations of the local flux of cmb photons arise during the propagation of the photons from recombination to the present epoch . these differences can be used to trace the space - time geometry since these photons travel along null geodesics of the geometry . in this section , we use the trajectories of cmb photons computed in sub section [ subsec : rulers ] to derive the spectrum of the cmb flux observed today . the cmb flux observed at o ( see figure [ fig : geodesics ] ) is @xmath108 where @xmath109 is the number of photons with energies between @xmath110 and @xmath111 received at o within a solid angle @xmath112 in an area @xmath113 during a time @xmath114 . the angle @xmath82 is defined as per since that definition corresponds to the physical angle that a local observer measures between the photon s trajectory and the @xmath8 axis . the photons that are received at o within this solid angle were emitted from the point p on the surface of last scattering @xmath69 ( see figure [ fig : geodesics ] ) . since the geometry of the universe ( equation ) is curved , the solid angle @xmath115 is different from the solid angle at o. the energy @xmath116 at which the photon is emitted is also different from the energy @xmath110 at which it is received owing to the expansion of the universe . furthermore , due to the differential expansion of the @xmath117 plane and the @xmath8 directi on , this energy shift is also a function of the solid angle . the photons received in the space - time volume @xmath118 are emitted from a volume @xmath119 . the ratio of these volume elements is proportional to the expansion of the universe . incorporating these effects , the flux can be expressed as @xmath120 or , in terms of the emission flux @xmath121 , @xmath122 to find the local flux , we have to relate the geometric and energy elements in at p to those at o. we begin with the angle @xmath82 . using the definition of @xmath80 and the solution for the geodesic , we solve for @xmath82 along the null geodesic and find that @xmath123 where @xmath82 and @xmath124 are the angles of the photon s trajectory at the observer s present location o and the point p ( see figure [ fig : geodesics ] ) on the surface of last scattering which is connected to o by the null geodesic . we have ignored contributions of order @xmath125 , the fractional energy in curvature at the time of recombination , in this solution . this is justified since @xmath126 . the angle @xmath54 is unaffected by the anisotropic curvature since there is an @xmath70 symmetry in the @xmath63 plane . consequently , @xmath127 . the volume elements are proportional to the expansion of the universe and are given by @xmath128 where @xmath129 and @xmath130 are the scale factors at the points p and o respectively . finally , we need to compute the relationship between the observed energy @xmath110 of the photon and the emission energy @xmath116 . the energy e observed by a local observer at some point along the photon s trajectory is given by @xmath131 where @xmath132 is an affine parameter along the photon trajectory . using the geodesic equations and the above expression , the present day energy @xmath110 is @xmath133 incidentally , this expression can also be arrived at by red shifting the momentum components of the photon along the radial and @xmath8 directions by @xmath134 respectively . we now have all the ingredients necessary to compute the present day flux @xmath135 given an initial flux @xmath121 at recombination . since @xmath0 prior to recombination is much smaller than @xmath136 , the cmb spectrum at recombination is identical to that of the usual frw universe . in particular , the cmb at p is a black - body at a temperature @xmath137 , with its spectrum , independent of angle , given by the planck distribution @xmath138 we define @xmath139 . this definition is motivated by the fact that cmb temperature should redshift roughly as the ratio of the scale factors of expansion . in this anisotropic universe , where two dimensions expand with scale factor @xmath26 and the other with scale factor @xmath35 , the quantity @xmath140 is roughly the mean expansion factor . using , , and in , we get @xmath141 where @xmath142 is the spherical harmonic with @xmath143 . it is well known that primordial density fluctuations lead to temperature anisotropies @xmath144 in the cmb . the temperature @xmath145 in inherits these anisotropies and is consequently a function of the angle @xmath146 in the sky . using this input , the distribution in describes a blackbody with a temperature @xmath147 at a given direction @xmath148 in the sky . note that the relation between the present day temperature @xmath149 and the temperature at recombination @xmath137 is warped both by the multiplicative factor ( the term in brackets ) in as well as the difference between the angles @xmath150 and @xmath151 . both these effects are proportional to @xmath79 and lead to effects in the cmb . in the following subsections , we highlight the key observables of this spectrum . the temperature @xmath152 is nearly uniform across the sky with an average temperature @xmath153 and primordial temperature fluctuations @xmath144 . substituting for @xmath152 in , we find that the anisotropic curvature leads to a quadrupole @xmath154 ( see equation ) in the cmb temperature . the source of this quadrupole is the differential expansion rate of the universe between the @xmath117 plane and the @xmath8 direction ( see equation ) , leading to differential red shifts along these directions . these differential red shifts lead to a quadrupolar warp of the average temperature of the surface of last scattering . unlike the primordial perturbations which are generated during inflation , this contribution to the quadrupole in the cmb arises from the late time emerge nce of the anisotropic curvature . fractionally , the additional power due to this effect is @xmath155 . current observations from the wmap mission constrains the quadrupolar temperature variation @xmath144 @xcite . naively , this constrains @xmath156 . however , the quadrupole that is observed in the sky is a sum of the quadrupole from the primordial density fluctuations and this additional contribution from the anisotropic curvature . it is then possible for these two contributions to cancel against each other leading to a smaller observed quadrupole . this cancellation requires a tuning between the primordial quadrupolar density perturbation and the anisotropic curvature contribution . additionally , this tuning can be successful only if the primordial quadrupolar perturbation is @xmath157 . the primordial density fluctuations are @xmath144 and it is difficult for the quadrupolar fluctuations to be much higher than this level . however , in a universe with a small number of e - foldings of inflation , the quadrupole is the mode that leaves the horizon at the very beginning of inflation and is therefore sensitive to physics in the primordial pre inflationary space - time . these phenomena are not constrained by inflationary physics and they could lead to additional power in the quadrupolar modes @xcite . it is therefore possible for the power in the primordial quadrupolar mode to be somewhat larger , leading to possible cancellation of the quadrupole from the late time anisotropic curvature . in fact , the measured quadrupole in our universe has significantly less power than expected from a conventional @xmath29cdm model @xcite . this anomaly may already be an indication of non - inflationary physics affecting th e quadrupole @xcite . there is also some uncertainty on the overall size of the quadrupole . for example , astrophysical uncertainties @xcite could potentially make the quadrupole in the cmb larger by a factor @xmath158 . owing to these uncertainties , it may be possible for @xmath79 to be as large as @xmath159 without running afoul of observational bounds . values of @xmath79 significantly larger than @xmath160 may also be possible . however , the additional tuning required to cancel the associated quadrupole may disfavor this possibility . it is interesting to note that anisotropic curvature is much more constrained than isotropic curvature . current cosmological measurements constrain the isotropic curvature contribution @xmath161 @xcite . however , anisotropic curvature leads to temperature anisotropies in the sky . since these anisotropies are well constrained by current measurements , the bounds on @xmath162 are more stringent ( for example , see @xcite ) . this bound is close to the cosmic variance limit on @xmath163 . consequently , there is an observational window of @xmath164 where the anisotropic curvature can be discovered . in this subsection we discuss the effects of anisotropic curvature on the power spectrum of the cmb . the warping of standard rulers by the anisotropic curvature ( see section [ subsec : rulers ] ) manifests itself in the cmb through these effects . at the present time , an observer o ( see figure [ fig : geodesics ] ) characterizes the cmb through the spectrum defined by @xmath165 where the present day temperature @xmath149 is defined in equation . the correlation functions @xmath166 of this spectrum contain all the information in the cmb . in a statistically isotropic universe , all non - diagonal correlators of the @xmath167 vanish . anisotropies mix different angular scales and will populate these non - diagonal correlators . we compute them in this section . @xmath149 inherits the density fluctuations at the time of recombination . since anisotropies were small prior to recombination , we will assume that the spectrum of density fluctuations at recombination is given by a statistically isotropic , gaussian distribution . the small anisotropies prior to recombination do alter this distribution and can give rise to additional observables @xcite . however , these corrections are proportional to the anisotropic curvature @xmath168 during recombination @xcite . since @xmath168 is smaller than the present day anisotropic curvature @xmath79 by a factor of @xmath169 , the experimental observables are dominated by the effects of the late time anisotropic curvature @xmath79 . in order to compute these late time effects , it is sufficient to assume that the spectrum of density fluctuations at recombination is statistically isotropic and gaussian . we will therefore make this assumption for the rest of the paper . our task is to start with this spectrum at recombination and compute the characteristics of the cmb spectrum observed by o. the anisotropic curvature warps the cmb spectrum at o in three ways . first , the photons from the surface of last scattering that reach o do not lie on a spherical surface ( see figure [ fig : geodesics ] ) . this warped surface @xmath69 is described by equation , where the deviations from sphericity are proportional to the late time curvature @xmath79 . second , the angle @xmath82 at which the photon is received at o is different from the co - ordinate angle @xmath170 on the surface of recombination at which this photon was originally emitted . third , the photon is red - shifted when it reaches o. this red - shift also depends upon the angle since the anisotropic curvature causes a differential hubble expansion leading to anisotropic red - shifts . we first determine the spectrum on @xmath69 , the surface from which photons at recombination reach o. @xmath69 can be described using spherical coordinates @xmath171 . @xmath172 is the physical distance at recombination between @xmath173 and a point @xmath174 on @xmath69 ( see figure [ fig : geodesics ] ) , @xmath170 is the polar angle between the @xmath8 axis and the unit vector at o that lies in the direction of @xmath174 and @xmath54 is the azimuthal angle . these flat space coordinates appropriately describe the recombination surface since the spatial curvature was very small during this period . in particular , the polar angle @xmath170 is given by @xmath175 while the physical distance @xmath172 ( using equation ) is @xmath176 the spectrum at @xmath69 can be characterized by @xmath177 where @xmath178 is the temperature at the recombination surface . the multiplicative factor @xmath179 in is introduced for convenience . it accounts for the red shift of the mean temperature from the era of recombination to the present time , but does not introduce additional correlations in the power spectrum . with this definition of @xmath180 , the correlation functions of the distributions and can be directly compared . after determining the correlators @xmath180 , we will incorporate the effects of the angular and energy warps to the spectrum . following @xcite , we express the temperature @xmath181 at any point @xmath182 on @xmath69 by the expansion @xmath183 the fourier components @xmath184 represent the power spectrum at recombination . since the anisotropic curvature is small in the era preceding recombination , the @xmath184 are drawn from a statistically isotropic , gaussian distribution . writing the term @xmath185 using spherical harmonics , we have @xmath186 where @xmath187 are the spherical bessel functions and @xmath188 are the spherical harmonics . using the expression for @xmath172 in equation , we expand @xmath189 for small @xmath79 . comparing this expansion with the definition of the @xmath180 in equation , we have @xmath190 the details of this expansion , including the definitions of the coefficients @xmath191 and @xmath192 can be found in appendices [ appendixbessel ] and [ appendixspherical ] . the @xmath188 in the above expression are all functions of the unit vector @xmath193 in the integrand . armed with the expression , we compute the correlators to first order in @xmath79 . each @xmath180 receives contributions from the spherical harmonics @xmath188 and @xmath194 . consequently , we expect non zero power in the auto correlation of each mode and correlation between modes separated by 2 units of angular momentum . these correlators are @xmath195 where the coefficients @xmath196 are @xmath197 numbers with a weak dependence on @xmath198 . all other correlators vanish . we relegate the details of this calculation to appendix [ appendixbessel ] . let us now relate the coefficients @xmath167 and @xmath180 . the present day temperature @xmath149 is given by . the relationship between @xmath170 and @xmath82 can be obtained from their respective definitions and . this relationship is given by @xmath199 owing to the @xmath200 symmetry in the @xmath63 plane , the angle @xmath54 is the same as the azimuthal angle @xmath201 used by o. we use the above relation to expand @xmath149 to leading order in @xmath79 , obtaining @xmath202 the second term in the above expression arises as a result of the differential red shift caused by the non - isotropic hubble expansion , while the third time arises due to the warp between the angles @xmath82 and @xmath170 ( as in equation ) . this expansion is valid for angular scales @xmath203 . using the spherical harmonic expansions for @xmath149 and @xmath178 in terms of @xmath167 and @xmath180 respectively , we find @xmath204 the coefficients @xmath205 in are obtained by combining the different spherical harmonics in . these coefficients are computed in the appendix [ appendixspherical ] . using the correlators of the @xmath180 ( see equation ) , we can compute the expectation values @xmath206 other correlation functions are unaffected by the anisotropic curvature @xmath79 . equation specifies that modes separated by 2 units of angular momentum @xmath198 are mixed while there is no mixing between modes of different @xmath207 . physically , this implies correlations between modes of different angular scales ( separated by two units of scale ) , but not of different orientation . the absence of mixing between modes of different orientation is due to the fact that the space - time preserves an @xmath200 symmetry in the @xmath63 plane . however , even though there is no correlation between modes of different @xmath207 , the power @xmath208 in a mode depends upon @xmath207 through the coefficients @xmath209 and @xmath210 . both these coefficients scale as @xmath211 ( see equations and ) . hence , we expect different amounts of power in the high @xmath207 mode versus th e low @xmath207 mode for a given @xmath198 . equipped with the knowledge of the correlators , we can perform tests of the statistical isotropy of the cmb . we follow the bipolar power spectrum analysis proposed by @xcite and adopt their notation ( note that the normalization convention adopted by @xcite is different from that used by the wmap team @xcite ) . in this analysis , one computes the correlator @xmath212 where the @xmath213 are clebsch gordan coefficients . in a statistically isotropic universe , these correlators are all zero except when @xmath214 and @xmath215 . in the present case , we use the correlators to compute the above statistic . for large @xmath198 , the only non - zero correlators are @xmath216 since the @xmath217 are smooth functions of @xmath198 , @xmath218 . the above correlators then scale as @xmath219 we note that these correlators are non - zero for all angular scales . this is precisely because the late time warp caused by the anisotropic curvature affects all the modes in the cmb . consequently , this is a statistically robust test of anisotropy . furthermore , this test of anisotropic curvature is immune to degeneracies from the expansion history of the universe that plague the measurement of isotropic curvature . indeed , in an isotropic universe , irrespective of the cosmological expansion history , this statistic would be zero . this is similar to the effect discussed in section [ subsec : rulers ] on standard rulers . in both cases , the anisotropic curvature affects measurements along every line of sight , breaking degeneracies with the cosmological expansion history . the similarity between these two effects is not surprising since the statistic captures the effect of the angular warp of the cmb by the anisotropic curvature ( the third term in ) . minimum variance estimators obtained from the cmb temperature / polarization for a power asymmetry of this type and the observability are calculated in @xcite . statistical analyses of the sort discussed in this section have been performed with the wmap data @xcite . in a universe with anisotropic curvature , these statistical tests can lead to quadrupolar dependence of the two point function . the expected answer for the statistic has power only in the @xmath220 and @xmath221 modes . furthermore , since these correlators are proportional to @xmath217 , the effect shows a bump around the first acoustic peak . interestingly , the two point quadrupolar anomaly in the wmap data shows similar characteristics with power only in the @xmath220 and @xmath221 modes , which peaks around the first acoustic peak . this anomaly could be explained in our scenario if the anisotropic curvature @xmath222 . however , such a large anisotropic curvature is heavily constrained by the absence of a correspondingly large quadrupole in the cmb ( see section [ subsec : cmbquadrupole ] ) . while this anomaly may be due to other systematic effects @xcite , similar searches could be performed with upcoming cmb experiments . it is conceivable that these experiments could discover correlations from anisotropic curvatures @xmath223 , as allowed by the size of the cmb quadrupole . we have so far considered only signals arising from the geometry of the universe , but observable signals may also arise from the topology . the normal eternal inflation picture makes it appear that space should be very large or infinite in all directions @xcite . if our observable universe nucleated as a bubble from ( 3 + 1 dimensional ) false vacuum inflation then it will appear as an infinite , open universe . however , in our picture it is natural that the observable universe could have one or two compact dimensions , even though it came from an eternally inflating space . interestingly , the size of these compact dimensions may be close to the hubble scale today because the period of slow - roll inflation was not too long . in the case of a 2 + 1 dimensional parent vacuum , the topology of the spatial dimensions of the observable universe would be @xmath224 . since the curvature is all in the @xmath225 and not the @xmath5 , the curvature radius of the universe and the topology scale ( in this case the radius of the @xmath5 ) are disconnected . thus , even though the curvature radius today is restricted to be @xmath226 times longer than the hubble scale , the size of the compact dimension can be smaller than the hubble scale . in fact , we expect that slow - roll inflation began when the curvature radius was around the hubble scale of inflation . thus , for the @xmath5 to be around the hubble scale today it would have needed to be about @xmath227 times smaller than the hubble size at the beginning of inflation . for high scale inflation this is near the gut scale , a very believable initial size for that dimension . this scenario is interestingly different from the compact topologies often considered , for which an isotropic geometry ( @xmath228 , @xmath229 or @xmath230 ) is usually assumed ( though see @xcite ) . any compact topology necessarily introduces a global anisotropy , but in our scenario even the local geometry of the universe is anisotropic . this allows the curvature radius and the topology scale to be different by orders of magnitude . thus it is reasonable that in our picture we may also have the circles in the sky " signal of compact topology @xcite . current limits from the wmap data require the topology scale to be greater than 24 gpc @xcite . this limit can be improved by further searching , especially with data from the planck satellite , to close to the @xmath231 gpc diameter of our observable universe . if discovered in conjunction with anisotropic curvature this would provide a dramatic further piece of evidence that we originated in a lower dimensional vacuum . further the directions should be correlated . if the parent vacuum was 2 + 1 dimensional then we expect the circles in the sky to be in the previously compact direction ( the @xmath5 ) while the curvature is in the other two dimensions . on the other hand , if the parent vacuum was 1 + 1 dimensional then it seems possible that both the signals of curvature and the compact topology would be in the same two dimensions , with the third dimension appearing flat and infinite . thus seeing both the anisotropic curvature and signals of the compact topology may provide another handle for determining the dimensionality of our parent vacuum . the cmb is a precise tool to measure cosmological parameters . however , it is a two dimensional snapshot of the universe at a given instant in time . additional information can be obtained through three dimensional probes of the universe . several experiments that yield three dimensional data are currently being planned . these include 21 cm tomography experiments and galaxy surveys . a complete study of the effects of anisotropic curvature in these experiments is beyond the scope of this work . in this section , we briefly mention some possible tests of this scenario in these upcoming experiments . a three dimensional map of the universe can be used to distinguish anisotropic curvature from fluctuations in the matter density . anisotropic curvature does not lead to inhomogeneities in the matter distribution . consequently , measurements of the large scale matter density can be used to distinguish between these two situations . such measurements may be possible using upcoming 21 cm experiments and high redshift surveys , for example lsst . lsst should be sensitive to isotropic curvatures down to @xmath232 with objects identified out to redshift @xmath233 @xcite . since the dominant effect of anisotropic curvature occurs at late times , lsst should be a good way to probe our signals . additionally , 21 cm experiments may also be sensitive to isotropic curvatures @xmath223 @xcite , and so may offer a very precise test of anisotropic curvature . the curvature anisotropy also gives rise to a differential hubble expansion rate @xmath234 ( see section [ sec : metric ] ) , which contributes to the quadrupole in the cmb ( see section [ subsec : cmb ] ) . this effect will also be visible in direct measurements of the hubble parameter . current experimental constraints on this effect are at the level of a few percent @xcite and are likely to become better than @xmath161 in future experiments @xcite . a universe produced as a result of bubble nucleation from an ancestor vacuum which has two large dimensions and one small , compact dimension is endowed with anisotropic curvature @xmath0 . such an anisotropic universe is also produced in the case when our 3 + 1 dimensional universe emerges from a transition from a 1 + 1 dimensional vacuum . in this case , depending upon the curvature of the compact dimensions , the resulting universe can have either positive or negative curvature along two dimensions , with the other remaining flat . the geometry of the equal time slices of the daughter universe are such that two of the directions are curved while the other dimension is flat . immediately after the tunneling event , the energy density of the universe is dominated by this anisotropic curvature @xmath0 . this curvature drives the curved directions to expand differently from the flat direction , resulting in differential hubble expansion @xmath42 between them . the expansion of the universe dilutes @xmath0 until it becomes small enough to allow slow roll inflation . at this time , the universe undergoes a period of inflation during which the curvature @xmath0 and the differential hubble expansion @xmath42 are exponentially diluted . however , during the epochs of radiation and matter domination , the curvature red shifts less strongly than either the radiation or the matter density . consequently , the fractional energy density @xmath0 in curvature grows with time during these epochs . this late time emergence of an anisotropic curvature @xmath0 also drives a late time differential hubble expansion @xmath42 in the universe . these late time , anisotropic warps of the space - time geometry are all proportional to the current fractional energy density in curvature , @xmath79 . they can be observed in the present epoch if inflation does not last much longer than the minimum number of efolds required to achieve a sufficiently flat universe ( @xmath235 efolds for high scale inflation ) . anisotropic curvature leads to the warping of the angular size of standard rulers . this warping is a function of both the angle and orientation of the ruler in the sky . consequently , this effect is immune to degeneracies from the expansion history of the universe since it affects rulers that are along the same line of sight but oriented differently . the cmb is also warped by the anisotropic curvature . in addition to the geometric warping , the differential hubble expansion @xmath42 also preferentially red shifts the energies of the cmb photons . this energy shift differentially changes the monopole temperature of the cmb giving rise to a quadrupole in the cmb . furthermore , since the anisotropic curvature is a late time effect , it affects all the modes that can be seen in the cmb . consequently , this effect leads to statistical anisotropy on all angular scales . this effect is different from other signatures of anisotropy considered in the literature @xcite . previous work has concentrated on the correlations that are produced due to the initial anisotropy in the universe at the beginning of inflation . since these modes are roughly stretched to the hubble size today , these initial anisotropies only affect the largest modes in the sky and are hence low l effects in the cmb . the late time anisotropy however warps the entire sky and leads to statistically robust correlations on all angular scales . the anisotropies in the pre - inflationary vacuum can however lead to other interesting signatures , for example in the gravitational wave spectrum @xcite . these signatures are an independent check of this scenario . anisotropies that affect all angular scales have also been previously considered @xcite . these required violations of rotational invariance during inflation and the anisotropy emerges directly in the primordial density perturbations . in our case , the density perturbations are isotropic and the anisotropy observed today is a result of a late time warp of the space - time . anisotropic curvature is already more stringently constrained than isotropic curvature . while isotropic curvature is bounded to be @xmath161 , it is difficult for anisotropic curvature to be much larger than @xmath160 without running afoul of current data , in particular , the size of the cmb quadrupole . since the measurement of curvature is ultimately limited by cosmic variance @xmath144 , there is a window between @xmath236 that can be probed by upcoming experiments , including planck . future cosmological measurements like the 21 cm experiments will significantly improve bounds on the curvature of the universe . a discovery of isotropic curvature would be evidence suggesting that our ancestor vacuum had at least three large space dimensions . on the other hand , a discovery of anisotropic curvature would be strong evidence for the lower dimensionality of our parent vacuum . the anisotropy produced from such a transition has a very specific form due to the symmetries of the transition . it leads to correlations only amongst certain modes in the cmb ( for example , only @xmath220 and @xmath221 ) . this distinguishes it from a generic anisotropic @xmath61 dimensional pre - inflationary vacuum which will generically have power in all modes . in these scenarios , it is also natural for the universe to have non - trivial topology . the existence of a non - trivial topological scale within our observable universe can be searched for using the classic circles in the sky " signal . if both the non - trivial topology and anisotropic curvature can be discovered , implying a period of inflation very close to the catastrophic boundary , it would be powerful evidence for a lower dimensional ancestor vacuum . a discovery of these effects would establish the existence of vacua vastly different from our own standard model vacuum , lending observational credence to the landscape . we would like to thank savas dimopoulos , sergei dubovsky , ben freivogel , steve kahn , john march - russell , stephen shenker , leonard susskind , and kirsten wickelgren for useful discussions and the dalitz institute at oxford for hospitality . s.r . was supported by the doe office of nuclear physics under grant de - fg02 - 94er40818 . is also supported by nsf grant phy-0600465 . while this work was in progress we became aware of interesting work by another group working on different signals of a similar general framework @xcite . the temperature @xmath181 at any point @xmath182 on the surface of last scattering @xmath69 ( see figure [ fig : geodesics ] ) can be expressed using spherical harmonics ( see equation ) @xmath237 expanding the bessel junctions @xmath187 in around @xmath238 to linear order in @xmath79 , we get @xmath239 where the coefficient @xmath191 is @xmath240 this expansion is valid for @xmath241 . the spherical harmonic @xmath242 in multiplies @xmath243 in the expansion . these harmonics can be combined , yielding @xmath244 the definitions of the @xmath192 are given in appendix [ appendixspherical ] . using , the coefficient @xmath180 of @xmath245 in is the expression in equation . with this information , we can compute the correlations amongst the @xmath180 . imposing the requirement that @xmath184 are drawn from a statistically isotropic , gaussian distribution @xcite , the two point function @xmath246 to linear order in @xmath79 is @xmath247 where @xmath248 is the two point function @xmath249 of the temperature anisotropies @xmath250 ( as defined in @xcite ) . we have again scaled out the piece that accounts for the red shift between the era of recombination and the present epoch . the first term in the integrand is the usual contribution @xmath217 to the power in the @xmath198 mode . the second term , proportional to @xmath79 , arises from the anisotropic curvature . to compute this term , we substitute for @xmath191 ( from ) in . the resulting integral has the form @xmath251 the second term in the above integrand is @xmath252 . for the first term , @xmath253 we use the fact that the @xmath187 satisfy the identity @xmath254 this implies @xmath255 physically , since there is roughly similar amounts of power in all the @xmath217 , we expect @xmath256 . this implies @xmath257 where @xmath258 is an order one coefficient . @xmath196 can be computed by integrating . this calculation requires explicit use of the two point function @xmath248 at recombination and is beyond the scope of this paper . using in and , we have @xmath259 a similar calculation can be performed for the other correlators of the @xmath180 . in the expression for @xmath180 , each @xmath180 receives contributions from the spherical harmonics @xmath188 and @xmath194 . consequently , we expect non trivial correlations only between modes separated by 2 units of angular momentum . this correlator is @xmath260 with this knowledge , we can compute the correlators of the @xmath167 . equation expresses the temperature @xmath149 ( characterized by @xmath167 ) observed today in terms of the temperature @xmath178 ( characterized by @xmath180 ) at recombination . writing @xmath178 in terms of the @xmath180 in , we get , @xmath261 the spherical harmonics in are all functions of @xmath148 . the products of the spherical harmonics @xmath262 and @xmath263 can be expressed as a linear combination of the @xmath188 . the term @xmath262 can be expressed as the combination , whilst @xmath263 is expressed as @xmath264 the coefficients @xmath265 are defined in appendix [ appendixspherical ] . using and in , we get the expression for the @xmath167 in terms of the @xmath180 . the @xmath167 in are expressed as a linear combination of @xmath180 and @xmath266 . consequently , to linear order in @xmath79 , we expect power in the modes @xmath267 and @xmath268 . using , the correlators can be computed . we now give the exact answers for the measures of statistical anisotropy computed approximately in eqn . in this appendix , we give the definitions of the coefficients @xmath192 , @xmath265 and @xmath205 . the coefficients @xmath205 ( see equation ) are defined by the addition of the effects from the energy warp of the cmb by the anisotropic hubble expansion and the warp of the angle @xmath82 at which the photon is observed and the angle @xmath170 at which it was emitted at the surface of last scattering . these evaluate to @xmath274 n. arkani - hamed , s. dubovsky , a. nicolis and g. villadoro , jhep * 0706 * , 078 ( 2007 ) [ arxiv : hep - th/0703067 ] . r. bousso and j. polchinski , jhep * 0006 * , 006 ( 2000 ) [ arxiv : hep - th/0004134 ] . c. krishnan , s. paban and m. zanic , jhep * 0505 * , 045 ( 2005 ) [ arxiv : hep - th/0503025 ] . s. m. carroll , m. c. johnson and l. randall , jhep * 0911 * , 094 ( 2009 ) [ arxiv:0904.3115 [ hep - th ] ] . j. j. blanco - pillado , d. schwartz - perlov and a. vilenkin , jcap * 0912 * , 006 ( 2009 ) [ arxiv:0904.3106 [ hep - th ] ] . j. j. blanco - pillado , d. schwartz - perlov and a. vilenkin , arxiv:0912.4082 [ hep - th ] . z. c. wu , phys . b * 585 * , 6 ( 2004 ) [ arxiv : hep - th/0309178 ] . s. b. giddings and r. c. myers , phys . rev . d * 70 * , 046005 ( 2004 ) [ arxiv : hep - th/0404220 ] . r. h. brandenberger and c. vafa , nucl . b * 316 * , 391 ( 1989 ) . b. greene , d. kabat and s. marnerides , arxiv:0908.0955 [ hep - th ] . b. freivogel , m. kleban , m. rodriguez martinez and l. susskind , jhep * 0603 * , 039 ( 2006 ) [ arxiv : hep - th/0505232 ] . j. garriga , t. tanaka and a. vilenkin , phys . d * 60 * , 023501 ( 1999 ) [ arxiv : astro - ph/9803268 ] . a. vilenkin and s. winitzki , phys . d * 55 * , 548 ( 1997 ) [ arxiv : astro - ph/9605191 ] . b. bozek , a. j. albrecht and d. phillips , phys . d * 80 * , 023527 ( 2009 ) [ arxiv:0902.1171 [ astro-ph.co ] ] . a. aguirre , m. c. johnson and a. shomer , phys . d * 76 * , 063509 ( 2007 ) [ arxiv:0704.3473 [ hep - th ] ] . s. chang , m. kleban and t. s. levi , jcap * 0804 * , 034 ( 2008 ) [ arxiv:0712.2261 [ hep - th ] ] . s. chang , m. kleban and t. s. levi , jcap * 0904 * , 025 ( 2009 ) [ arxiv:0810.5128 [ hep - th ] ] . s. r. coleman and f. de luccia , phys . d * 21 * , 3305 ( 1980 ) . r. bousso and b. freivogel , phys . d * 73 * , 083507 ( 2006 ) [ arxiv : hep - th/0511084 ] . e. silverstein , arxiv : hep - th/0405068 . d. larson _ et al . _ , arxiv:1001.4635 [ astro-ph.co ] . a. fialkov , n. itzhaki and e. d. kovetz , jcap * 1002 * , 004 ( 2010 ) [ arxiv:0911.2100 [ astro-ph.co ] ] . c. l. bennett _ et al . _ , arxiv:1001.4758 [ astro-ph.co ] . c. l. francis and j. a. peacock , arxiv:0909.2495 [ astro-ph.co ] . l. r. abramo and t. s. pereira , arxiv:1002.3173 [ astro-ph.co ] . a. hajian and t. souradeep , astrophys . j. * 597 * , l5 ( 2003 ) [ arxiv : astro - ph/0308001 ] . y. mao , m. tegmark , m. mcquinn , m. zaldarriaga and o. zahn , phys . rev . d * 78 * , 023529 ( 2008 ) [ arxiv:0802.1710 [ astro - ph ] ] . m. tegmark , a. de oliveira - costa and a. hamilton , phys . d * 68 * , 123523 ( 2003 ) [ arxiv : astro - ph/0302496 ] . b. mota , g. i. gomero , m. j. reboucas and r. tavakol , class . * 21 * , 3361 ( 2004 ) [ arxiv : astro - ph/0309371 ] . n. j. cornish , d. n. spergel and g. d. starkman , arxiv : gr - qc/9602039 . n. j. cornish , d. n. spergel , g. d. starkman and e. komatsu , phys . lett . * 92 * , 201302 ( 2004 ) [ arxiv : astro - ph/0310233 ] . z. ivezic , j. a. tyson , r. allsman , j. andrew , r. angel and f. t. l. collaboration , arxiv:0805.2366 [ astro - ph ] . v. c. rubin , annals of the new york academy of sciences * 302 * , 408 a. e. gumrukcuoglu , l. kofman and m. peloso , phys . rev . d * 78 * , 103525 ( 2008 ) [ arxiv:0807.1335 [ astro - ph ] ] . a. e. gumrukcuoglu , c. r. contaldi and m. peloso , jcap * 0711 * , 005 ( 2007 ) [ arxiv:0707.4179 [ astro - ph ] ] . l. ackerman , s. m. carroll and m. b. wise , phys . d * 75 * , 083502 ( 2007 ) [ erratum - ibid . d * 80 * , 069901 ( 2009 ) ] [ arxiv : astro - ph/0701357 ] . c. g. boehmer and d. f. mota , phys . b * 663 * , 168 ( 2008 ) [ arxiv:0710.2003 [ astro - ph ] ] . a. r. pullen and m. kamionkowski , phys . d * 76 * , 103529 ( 2007 ) [ arxiv:0709.1144 [ astro - ph ] ] . j. j. blanco - pillado and m. p. salem , arxiv:1003.0663 [ hep - th ] . m. demianski and a. g. doroshkevich , phys . d * 75 * , 123517 ( 2007 ) [ arxiv : astro - ph/0702381 ] . j. adamek , d. campo and j. c. niemeyer , arxiv:1003.3204 [ hep - th ] .

we have discovered 12 new molecular hydrogen emission - line objects ( mhos ) in the vicinity of the candidate massive young stellar object , in addition to five that were previously known . 2.12-m / h 2.25- m flux ratios indicate another region dominated by fluorescence from a photo - dissociation region ( pdr ) , and one region that displays an anomalously low h 2.12-m / h 2.25- m flux ratio ( ) and coincides with a previously reported deeply embedded source ( des ) . continuum observations at 3 mm reveal five dense cores ; the brightest core is coincident with the des . the next brightest cores are both associated with cm continuum emission . one of these is coincident with the _ iras _ source ; the other lies at the centroid of a compact outflow defined by bipolar mhos . the brighter of these bipolar mhos exhibits [ ] emission and both mhos are associated with choh maser emission observed at 95 ghz and 44 ghz . masses and column densities of all five cores are consistent with theoretical predictions for massive star formation . although it is impossible to associate all mhos with driving sources in this region , it is evident that there are several outflows along different position angles , and some unambiguous associations can be made . we discuss implications of observed h 2.12-/h 2.25- and [ ] 1.64-/h 2.12- flux ratios and compare the estimated total h luminosity with the bolometric luminosity of the region . we conclude that the outflows are driven by massive young stellar objects embedded in cores that are likely to be in different evolutionary stages . = 1

1211.3440

outflows play a critical role in the formation of isolated low - mass stars throughout the embedded accretion phases of evolution . they are thought to be driven by magnetic stresses in accretion disks that eject some of the inflowing matter and carry off angular momentum ( e.g. , shu et al . 1994 ; pudritz et al . 2006 ; matsakos et al . in contrast , the role of outflows in the formation of stars @xmath5 8 m@xmath6 remains elusive , because these stars are rare , form in tight clusters along with low- and intermediate - mass stars that may be in different stages of evolution , lie in heavily obscured regions often at distances greater than several kiloparsecs , and may be evolving on dynamical time scales less than a few thousand years ( e.g. , zinnecker & yorke 2007 ) . large surveys with single - antenna microwave telescopes indicate that energetic molecular outflows are present in massive star - forming regions ( e.g. , zhang et al . 2005 ) , but it is not generally possible to link individual outflows with driving sources , even with the current generation of millimeter - wave interferometers . the situation is somewhat better in the near - infrared , where the h@xmath0 1 - 0 s(1 ) line at 2.12 @xmath4 is a particularly powerful tracer of shocks in molecular outflows ( davis et al . 2010 and references therein ; smith 2012 ) . the current generation of sensitive wide - field near infrared imagers enables large surveys of massive star - forming regions to be conducted at arcsecond or sub - arcsecond resolution in reasonable periods of observing time . using the apache point observatory ( apo ) 3.5-m telescope , wolf - chase et al . ( 2012 , in preparation ) have conducted a narrow - band near - infrared survey for shocked emission in 28 massive star - forming regions that exhibit energetic molecular outflows ( zhang et al . 2005 ) . in wolf - chase ( 2012 , hereafter wss12 ) , we presented near - infrared results for mol 160 ( iras 23385 + 6053 ) , along with 3-mm continuum and line observations obtained with the combined array for research in millimeter - wave astronomy ( carma ) . in this paper , we present narrowband , @xmath7 1@xmath8-resolution near - infrared apo images and @xmath7 2@xmath8-resolution carma 3-mm continuum and line observations for mol 121 ( _ iras _ 20188 + 3928 ) . mol 121 has a spectral energy distribution ( sed ) resembling an ultra - compact ( uc ) region and is associated with radio continuum emission ( wood & churchwell 1989 ; molinari et al . varricatt et al . ( 2010 ) included it in a nir imaging survey of 50 regions containing intermediate- and high - mass yso outflow candidates . they present an overview of background observations of this object in the appendix to their paper ( a37 ) and note that prior distance estimates to mol 121 span an order of magnitude ( 0.3 - 3.91 kpc ) . mol 121 ( msx6c designation g077.4622 + 01.7600 ) is categorized in the red msx survey ( rms ) as class /yso , and is among the rms objects that have uniquely constrained kinematic distances and good sed fits ( mottram et al . 2011 , table 1 ) . in this paper , we assume the rms catalog distance of 1.7 kpc and bolometric luminosity of 8.2@xmath9 l@xmath6 . we obtained narrowband , near - infrared images of mol 121 using the near - infrared camera and fabry - perot spectrometer ( nicfps ) on the astrophysical research consortium ( arc ) 3.5-m telescope at the apo in sunspot , nm on ut 2008 oct 13 . we used the following filters and total integration times : h@xmath0 2.12-@xmath4 ( 1740 s ) , h@xmath0 2.25-@xmath4 ( 1740 s ) , and [ ] 1.64-@xmath4 ( 1800 s ) . we also used two narrowband filters to allow for continuum subtraction in the final images : an h@xmath0-continuum filter centered on 2.13 @xmath4 ( 1740 s ) and an [ ] -continuum filter centered on 1.65 @xmath4 ( 1800 s ) . our basic data acquisition , reduction , and calibration methods are described in wss12 . thirteen isolated stars having the best 2mass photometry flags were chosen in the narrow - band line images as well as in the corresponding narrow - band continuum images . we performed aperture photometry on these stars using the daophot package in iraf with an aperture radius of 12 pixels , @xmath74 times the fwhm , since the seeing was @xmath71@xmath8 in each image . the continuum images were scaled using a scaling factor and then subtracted from the narrow - band line images so that on average the stars disappear . the scaling factor was chosen so that the fractional residual fluxes of the stars within the aperture are as close to zero as possible in the resulting continuum - subtracted image . this procedure worked well except for the brightest stars in the field , which show small residual artifacts . we flux calibrated our data using equations ( 1 ) & ( 2 ) from wss12 with @xmath10 nm for the [ ] 1.64-@xmath4 filter ( apache point observatory 2010 ) . since the h@xmath11 sources can have irregular morphologies , we performed irregular aperture photometry using the highly flexible kang software version 1.3 . the apertures were defined using polygons . for each aperture in each image , we drew ten background regions . these background regions were chosen so they sample the background as close to the aperture as possible , and were roughly of the same size as the apertures . we then used the average background - subtracted flux within each aperture for subsequent calculations . the uncertainties in tables 1 & 2 correspond to @xmath12 , and are the combinations of three categories of uncertainty ; calibration , subtraction , and aperture . the calibration uncertainty incorporates the 2mass catalog k@xmath13 magnitude uncertainties , our instrumental stellar magnitude uncertainties and the standard deviation for the zero - point correction . the subtraction uncertainty is estimated using the standard deviation from zero for the fractional residual fluxes in the continuum subtraction procedure . the aperture uncertainty is the combination of photon noise within the aperture and background regions and the standard deviation from the ten background regions used for each aperture . we observed a field around mol 121 approximately 1.5@xmath14 across with carma on ut 2010 march 11 . the interferometer was in the c configuration with baselines ranging from 26 - 370 meters . we simultaneously observed the continuum and the ch@xmath3oh j=8@xmath157 line at 95.169 ghz . two spectral bands were configured for maximum continuum bandwidth , covering 468 mhz each , for a total of 1.872 ghz ( including both sidebands ) , on either side of a central frequency of 93.585 ghz . the band observing ch@xmath3oh was in the upper side band with a 31-mhz bandwidth and 63 channels , resulting in a velocity resolution of 1.538 km s@xmath16 . the passband calibrator was 1927@xmath17730 , the flux calibrator was mwc349 , and the phase calibrator was 2007@xmath17404 . the total integration time on source was 4.92 hours . the size ( fwhm ) of the synthesized beam was 1.967@xmath181.516@xmath8 . we reduced and calibrated the data as described in wss12 . figures 1 - 3 present our narrow - band nir images . our h@xmath0 2.12-@xmath4 , h@xmath0 2.25-@xmath4 , [ ] , and continuum - subtracted line images are shown in figure 1 . figure 2 displays the continuum - subtracted h@xmath0 2.12-@xmath4 image with the regions used to identify molecular hydrogen emission - line objects ( mhos ) , compact emission - line sources thought to be associated with outflows . two other emission features that are not mhos are also identified : fluorescent gas in a photo - dissociation region ( pdr ) and a region where the h@xmath0 2.12 @xmath4/2.25 @xmath4 flux ratio is less than one ( marked as ` anomaly ' ) . the anomalous h@xmath0 emission is discussed further in 4.1 . for the mhos , we follow the numbering scheme of the on - line catalog of these objects , hosted by the joint astronomy centre in hawaii ( davis et al . 2010 ) . in all , we identify 12 new mhos in addition to five that have been catalogued previously ( see figure 1 : bottom left panel , figure 2 , & table 1 ) . the feature originally identified as mho 867 by varricatt et al . ( 2010 ) actually encompasses both the pdr and the bow - shaped feature we associate with the mho . although the _ iras _ source appears to be offset at the eastern edge of the pdr in the h@xmath0 2.12-@xmath4 and [ ] images , it is centrally located in the h@xmath0 2.25-@xmath4 images ( figures 1 : top and bottom middle panels ) and the 3-color wide - field infrared survey explorer ( wise : wright et al . 2010 ) image presented in figure 4 . the presence of shocked gas to the west of the _ iras _ source may explain the enhancement of h@xmath0 2.12-@xmath4 and [ ] emission in this direction . table 1 lists the mho designations ( column 1 ) , areas used in the aperture photometry ( column 2 ) , positions of the peak emission ( columns 3 & 4 ) , h@xmath0 2.12-@xmath4 line fluxes ( column 5 ) , h@xmath0 2.25-@xmath4 line fluxes ( column 6 ) , and h@xmath0 2.12-@xmath4/h@xmath0 2.25-@xmath4 flux ratios ( column 7 ) . columns 8@xmath1910 present results for fluxes integrated over 2-pixel circular apertures centered on 2.25-@xmath4 peaks . where the h@xmath0 2.25-@xmath4 line flux is below 3 times the noise ( 3@xmath20 ) within an aperture , the h@xmath0 2.25-@xmath4 line is regarded as undetected and no value is given in columns 6 or 9 . in these cases , lower limits to the h@xmath0 2.12-@xmath4/h@xmath0 2.25-@xmath4 flux ratios are calculated using the 3@xmath20 value for that aperture as an upper limit for the h@xmath0 2.25-@xmath4 line flux . the difference in extinction at 2.12 @xmath4 and 2.25 @xmath4 is very small , only a tenth of a magnitude ( assuming a@xmath21 ) , yielding a corresponding brightness ratio of @xmath71.1 , which has a negligible effect on the flux ratios . the expected h@xmath0 2.12-@xmath4/2.25-@xmath4 line ratio for uv excitation ( as in pdrs ) is @xmath7 1.9 . it is significantly higher for h@xmath0 excited by shocks , with a ratio of @xmath7 3 - 20 for different types of shocks ( e.g. , black & van dishoeck 1987 ; smith 1994 ; gredel & dalgarno 1995 ; smith 1995 ; smith et al . 2003 ) . two types of shocks are observed in outflows from star - forming regions . continuous ( or c - type ) shocks occur when the shock is magnetically cushioned by a relatively high transverse magnetic field , such that the shock thickness is broad and deceleration of gas is relatively gradual , resulting in lower excitation and temperatures than in jump ( or j - type ) shocks . in the latter , the gas undergoes a sharp increase in temperature within a narrow zone , where molecules are dissociated ( e.g. , barsony et al . 2010 ; smith 2012 ) . the observed line ratios for objects listed as mhos are consistent with shock excitation , but not for the pdr and anomaly . since the high flux ratios for many of the mhos may reflect the fact that 2.25-@xmath4 emission is not detected over the full extent of the apertures used for the 2.12-@xmath4 emission ( see table 1 , columns 5 - 7 ) , we also computed flux ratios using a circular aperture of radius 2 pixels centered on the 2.25-@xmath4 emission peaks ( see table 1 , columns 8 - 10 ) . a comparison of columns 10 and 7 indicates that although h@xmath0 2.12-@xmath4/2.25-@xmath4 flux ratios are typically lower in the 2-pixel radius apertures , the ratios are still consistent with shock excitation . the [ ] 1.64-@xmath4 line probes fast , dissociative shocks with velocities @xmath5 50 km s@xmath16 . when it is observed in jets from low- and intermediate - luminosity ysos , it is typically localized in compact spots , representing only a small fraction of the region emitting in h@xmath0 , and its presence does not appear to correlate strictly with the evolutionary stage of the driving source ( caratti o garatti et al . fewer [ ] observations have been reported for high - luminosity objects , in part due to the greater extinction at 1.64 @xmath4 . curiously , caratti o garatti et al . ( 2008 ) found that the only [ ] emission toward the h@xmath0 jet associated with the massive young stellar object ( myso ) _ iras _ 20126@xmath174104 is close to the source . table 2 lists the three [ ] emission - line features we detected . the first two features are associated with mho 864 , and the third feature is associated with , and largely overlaps , mho 867 , the bright bow- or ` bean'-shaped feature that opens toward the _ iras _ source . in mho 864 , the two [ ] emission - line features lie along the inferred jet axis of the outflow . one is located at the apex of the mho and the other is close to core d , which presumably harbors the driving source of the outflow ( see figure 2 and 4.2.4 ) . in mho 867 , the morphology of the [ ] emission is similar to the mho , but over a smaller region . to investigate excitation conditions in these mhos , we computed line ratios in 2-pixel radius apertures along the major and minor axes of mho 864 , and along and transverse to the bow - shaped structure of mho 867 . placement of these apertures is indicated in figure 3 and emission line flux ratios are presented in table 3 . for the apertures with h@xmath0 2.25-@xmath4 detections , 2.12 @xmath4/2.25 @xmath4 flux ratios range from 6.8 - 14.5 . 1.64-@xmath4/2.12-@xmath4 flux ratios range from 0.03 - 0.9 for mho 864 apertures and 0.5 - 1.9 for mho 867 apertures . we interpret these results in 4.1 . the terms `` green fuzzies '' and `` extended green objects '' ( egos ) have been applied to regions of extended 4.5-@xmath4 emission that are often associated with shocked h@xmath0 emission from outflows ( e.g. , cyganowski et al . the nomenclature reflects the use of green as the representative color for the infrared array camera ( irac : fazio et al . 2004 ) 4.5-@xmath4 band in spitzer space telescope 3-color images produced from galactic legacy infrared mid - plane survey extraordinaire observations ( glimpse : benjamin et al . 2003 ; churchwell et al . figure 4 uses a similar 3-color scheme to represent the wise 3.4 , 4.6 , & 12-@xmath4 bands . here , objects that are bright in the 4.6-@xmath4 band appear to be green . mhos 864 , 865 , & 866 are the only mhos that are clearly prominent in the 4.6-@xmath4 band . mhos 864 & 865 have the highest h@xmath0 2.12-@xmath4 fluxes in mol 121 . on the other hand , there are several patches of faint 4.6-@xmath4 emission that do not correspond to mhos . we note that this result is similar to that of lee et al . ( 2012 ) , who imaged 34 spitzer glimpse egos with the united kingdom infrared telescope ( ukirt ) and noted that shocked molecular hydrogen emission is typically more extended than egos and that egos occasionally trace scattered continuum light ( see also barsony et al . this suggests that although the 4.5-@xmath4 & 4.6-@xmath4 bands may be useful in identifying features associated with bright outflows , they are not always dominated by shocked outflow emission . our carma 3-mm continuum data reveal five cores in the mol 121 region . figure 5 shows contours for the 3-mm continuum emission overlaid on the relevant portion of our continuum - subtracted h@xmath0 2.12-@xmath4 image . table 4 lists physical properties of the cores . assuming a distance of 1.7 kpc , deconvolved core diameters , @xmath22 , range from 790 au - 3500 au . the cores are discussed individually in 4.2 . assuming the dust emission is optically thin at 3 mm , the mass of a core is calculated in the following manner : @xmath23 here , @xmath24 is the distance to the source , @xmath25 is the total flux density , @xmath26 is the gas - to - dust ratio , @xmath27 is the planck function at dust temperature @xmath28 , and @xmath29 is the dust opacity per gram of gas ( e.g. , enoch et al . we use @xmath29 = 0.178 @xmath30 g@xmath16 , the value obtained by extrapolating from table 1 in ossenkopf & henning ( 1994 ) using a power law ( @xmath31 ) with @xmath32 = 1.8 , and assuming dust grains with thin ice mantles coagulated for 10@xmath33 yr at a gas density of 10@xmath34 @xmath35 . typically , dust temperatures in protostellar cores are assumed to be @xmath7 20 k ( e.g. , rosolowsky et al . 2010 ; lee et al . 2011 ; chen et al . 2012 ) , but dust temperatures in pdrs and some massive protostellar cores are closer to 30 k ( e.g. , molinari et al . 2008 ; anderson et al . 2012 ) , and enoch et al . ( 2007 ) note that plausible values might vary from 5@xmath1930 k. the sed of _ iras _ 20188@xmath173928 has been fit with a cold dust component of 39 k and warm dust component of 100 k ( mccutcheon et al . 1995 ) . most of the cores detected by carma are probably colder than this , so we have opted to calculated masses for t@xmath36 = 20 k & 40 k ( table 3 , columns 7 & 8) . for t@xmath36 = 40 k , derived masses are @xmath7 2.1 times lower than for t@xmath36 = 20 k. we calculate the column density from the peak flux density @xmath37 in a similar manner to enoch et al . ( 2007 ) : @xmath38 here , @xmath26 , @xmath29 , and @xmath27 , are the same as in equation ( 1 ) , @xmath39 is the size of carma s main beam at 3 mm , and @xmath40 is the mass of molecular hydrogen . columns 9 & 10 in table 3 show the results of this calculation for both dust temperatures used in the mass calculation . theoretical work predicts a mass column density threshold @xmath41 1 g @xmath42 for massive star formation ( krumholz & mckee 2008 ; krumholz et al . the mass column density is the column density @xmath43 multiplied by @xmath40 and a factor of 1.36 to account for elements heavier than hydrogen ( simon et al . regions where intermediate mass stars are thought to be forming typically have mass column densities @xmath7 0.1 - 0.5 g @xmath42 ( e.g. , arvidsson et al . 2010 ; wolf - chase et al . 2003 ; wolf - chase , walker , & lada 1995 ) . all of the carma cores have mass column densities significantly @xmath5 1 g @xmath42 , the smallest value being @xmath7 6 g @xmath42 for core e at t@xmath36 = 40 k. we have discovered three 95-ghz ch@xmath3oh masers in the mol 121 region , coincident with mhos 864 , 865 , and 892 . they are all unresolved by carma s 2@xmath8 resolution , and each appear in two contiguous velocity channels . there are very slight ( @xmath21@xmath8 ) shifts in position of the masers between channels , and that , combined with our relatively coarse 1.538 km s@xmath16 channel width , means that our masers may each be multiple , tightly clustered masers , but without better spatial and spectral resolution we will set aside this possibility . for each maser , we measured the total flux and estimated the central velocity , and these results are presented in table 5 . one 44-ghz class i ch@xmath3oh maser was previously discovered in the mol 121 region ( s. kurtz , private communication ) , at the same location as the brightest 95-ghz maser and with a velocity of 2.3 km s@xmath16 . the channel width of the 44-ghz data was 0.166 km s@xmath16 , accounting for the discrepancy with our estimated velocity . the 44-ghz entries in table 5 were identified from vla data provided by kurtz and are discussed in 4.4 . lorenzetti et al . ( 2002 ) found that [ ] 1.64-@xmath4/h@xmath0 2.12-@xmath4 flux ratios for knots associated with a jet from irs 8 in the vela molecular ridge ranged from @xmath7 0.1 - 1.1 ( uncorrected for extinction ) . since the [ ] and h@xmath0 emitting regions are unsurprisingly physically distinct , they computed flux ratios based on two sets of zones , defined by the [ ] peak emission and the h@xmath0 peak emission . they argue that these high ratios require the presence of a fast j - shock component , which produces [ ] 1.64-@xmath4/h@xmath0 2.12-@xmath4 flux ratios in the range 0.5 - 2 for a wide range of parameters ( hollenbach & mckee 1989 ) . given that the effects of extinction make our observed [ ] 1.64-@xmath4/h@xmath0 2.12-@xmath4 flux ratios lower limits , this suggests the presence of j - shocks in both mhos 867 & 864 ; however , the high observed h@xmath0 2.12-@xmath4/2.25-@xmath4 ratios across all of the apertures with 2.25-@xmath4 detections ( 6.8 - 14.5 ) in table 3 suggest the presence of c - shocks as well . for the high excitation temperatures associated with j - shocks , one expects h@xmath0 2.12-@xmath4/2.25-@xmath4 @xmath7 3 @xmath19 4 , and for the lower - excitation c - shocks , @xmath7 6 @xmath19 20 . above 20 , the excitation temperature is @xmath2 1000 k , and observable emission is not expected ( smith 1994 , 1995 ; smith et al . in fact , both types of shocks may be found in jets from ysos . typically , the molecular emission arises from the flanks of a bow shock , and the atomic or ionized component originates close to the apex ( smith 2012 ) . neither fluorescence nor shocks can explain a h@xmath0 2.12-@xmath4/2.25-@xmath4 flux ratio @xmath2 1 , which is the case for the `` anomaly '' listed in table 1 and indicated on figure 2 . we hand - checked each image for possible saturation effects near the bright _ iras _ source . our combination of narrow - band filters and short exposure times ( 30 s ) kept even the brightest sources below 30k adu , well within the linear range of the detector ( 3% linear to 54.4k adu ) and far below its full - well capacity of more than 100k adu . therefore , we conclude that this effect is real . to the best of our knowledge , this is the first time an observed h@xmath0 2.12-@xmath4/2.25-@xmath4 flux ratio @xmath2 1 has been reported . although high - resolution nir spectroscopy is essential to determining the nature of this emission , we note that dove et al . ( 1987 ) found that certain circumstances could produce population inversions in para - hydrogen transitions . they suggested future work to include the ortho - hydrogen transitions and more completely model the processes occurring in shock - heated clouds . another intriguing possibility is that the anomalous emission comes from the immediate front of a fast dissociative shock ( t@xmath4110,000 k ) , where statistical equilibrium is not reached ( m. smith , private communication ) . if statistical equilibrium is not achieved , the 2.12-@xmath4/2.25-@xmath4 ratio should be the ratio of the transition probabilities . the transition probabilities for the h@xmath0 2.12-@xmath4 1 - 0 s(1 ) and h@xmath0 2.25-@xmath4 2 - 1 s(1 ) transitions are 3.47@xmath44 s@xmath16 and 4.98@xmath44 s@xmath16 , respectively ( turner et al . 1977 ) . in this case , the expected ratio would be about 0.7 . although this is still somewhat larger than we calculate in table 1 , the ratio is affected by choice of aperture size . the anomaly is bright in both line and continuum nir images ( see fig . 1 ) , and it is coincident with a deeply embedded infrared source ( des ) identified from polarimetric patterns in the scattered nir emission ( yao et al . the des is also coincident with core c , the brightest 3-mm core in the mol 121 region ( see 4.2.3 ) . in this section , we discuss each of the five cores we detected with carma , previous related observations , and the possible association of cores with outflows , individually . there is direct evidence of massive star formation occurring in three of the five cores ( a , c , & d ) . the remaining two cores ( b & e ) may be starless . in the absence of significant external heating , these cores are likely to have temperatures @xmath45 20 k , in which case the derived masses and mass column densities suggest that they may be sites of future massive star formation . core a is associated with the _ iras _ source and several signposts of massive star formation , including a compact region observed at 3.6 cm and 6 cm ( jenness et al . 1995 ; molinari et al . 1998 ) , a pdr detected in our h@xmath0 2.12-@xmath4 , 2.25-@xmath4 , and [ ] 1.64-@xmath4 observations ( figs . 1 & 2 ) and evident in the wise image ( fig . 4 ) , and h@xmath0o maser emission ( palla et al . 1991 ; brand et al . 1994 ; anglada et al . 1997 ) . furthermore , core a lies at or near the continuum peak identified through previous sub - millimeter and millimeter observations ( fir 2 - jenness et al . 1995 ; mccutcheon et al . 1995 ; curran & chrysostomou 2007 ; motte et al . core a may be associated with a @xmath7e - w outflow defined by several mhos . in particular , the bow - shaped mho 867 opens toward this source , as does mho 938 further to the west ( see figures 2 & 5 ) . mho 867 also exhibits strong [ ] emission . there are several mho groupings toward the east , which do not align in any obvious way ( mhos 939@xmath19942 ) . core b has not been detected previously with millimeter or sub - millimeter observations , nor is it associated with any obvious signposts of star formation such as masers , radio continuum , or obvious outflow activity . it is located at the northern end of the bright pdr , coincident with an apparent dip in the nir emission , which suggests high extinction towards this core . it may be prestellar in nature . if this is true , its temperature may be lower than we assumed , and thus its mass may be larger . core c has not been detected previously with millimeter or sub - millimeter continuum observations , even though it has the largest integrated flux of the cores we detected with carma . it does , however , coincide with the location of the des reported by yao et al . ( 2000 ) , as well the peak of the anomalous h@xmath0 emission . although motte et al . ( 2007 ) mapped the entire cygnus x molecular cloud complex in 1.2-mm continuum at iram , they detected only one dense core with their 11@xmath8 beam within the region we identify as a pdr associated with _ iras _ 20188@xmath173928 . our cores a , b , & c all lie within this region . little et al . ( 1988 ) first detected a bipolar molecular outflow in the vicinity of _ iras _ 20188@xmath173928 from co and hco@xmath46 observations . the molecular outflow is aligned roughly n - s , with the redshifted gas to the north . using a distance of 4 kpc , they noted the outflow properties were similar to other massive outflows such as cep a and ngc 2071 . zhang et al . ( 2005 ) identified an outflow with the same orientation from co 2@xmath151 observations at a similar resolution to the observations of little et al . they derived significantly smaller outflow mass and energetic parameters ; however , they used a distance of only 0.3 kpc , which is much smaller than the most recent estimate of the distance ( mottram et al . zhang et al . ( 2005 ) noted that the outflow appears to be centered @xmath76@xmath8 to the north of the _ iras _ source , coincident with the des identified by yao et al . ( 2000 ) ; however , given their @xmath730@xmath8 beam and spatial sampling , it is impossible to identify the driving source(s ) . the k@xmath47 image of yao et al . ( 2000 ) shows nir reflection nebulosity oriented generally n - s in the vicinity of the _ iras _ source ; however , the polarization vectors exhibit a different pattern in a nw - se direction about the des ( coincident with the bar - shaped extension of the anomaly evident in figs . 2 & 5 ) . yao et al . ( 2000 ) suggest this may indicate the direction of a newly - burst outflow from the des . the association of core c with the des identified by yao et al . ( 2000 ) , strong millimeter flux , and corresponding lack of cm - continuum emission , make this object an excellent candidate myso in a pre - uc phase of evolution . if this is the case , anomalous h@xmath0 emission might be a signpost for the early outflow phase of a myso , but clearly high - resolution nir spectroscopy is necessary to search for any such evolutionary trends . core d is associated with a h@xmath0o maser and faint continuum emission at 3.6 cm and 6 cm , indicative of the presence of a uc region ( jenness et al . 1995 ; molinari et al . 1998 ) . using the ukt14 receiver at the jcmt , jenness et al . ( 1995 ) detected a submillimeter source ( fir 1 ) at a position between our cores d & e from 450 @xmath4 & 800 @xmath4 @xmath7 18@xmath8-resolution observations . mhos 864 & 865 , the brightest mhos in the region ( with the exception of mho 867 , which may be contaminated with fluorescent emission from the pdr ) , exhibit bipolar morphology in a ne - sw direction about core d. both mhos are associated with ch@xmath3oh maser emission at 95 ghz & 44 ghz . we have also detected [ ] emission at two positions within mho 864 . several mhos lie near or along a slightly different position angle from the direction defined by mhos 864 & 865 : to the ne , mhos 863 , 891 & 892 , and to the sw , mho 937 . mho 892 is also coincident with ch@xmath3oh maser emission at 95 ghz & 44 ghz . given the presence of both radio continuum emission and one or more well - collimated h@xmath0 outflow , this object may be close to transitioning to a stage where an expanding uc region destroys the myso(s) accretion disk . the bright mho 866 lies directly to the nw of core d , but its association is unclear . the position of core e coincides with the position of a dense core detected by motte et al . ( 2007 ) in 11@xmath8-resolution 1.2-mm continuum observations using the iram 30-m telescope . for an assumed distance of 1.7 kpc and temperature of 20 k , motte et al . ( 2007 ) derive a core mass of 40 m@xmath6 and density , n(h@xmath0 ) , of 3.5@xmath48 @xmath35 . for the same temperature , we derive a core mass of 12.8 m@xmath6 . the difference can be explained if the carma observations resolve out some of the large - scale structure revealed in sub - millimeter jcmt maps ( mccutcheon et al . 1995 ; jenness et al . 1995 ; curran & chrysostomou 2007 ) . mhos 893@xmath19897 lie in the vicinity of core e , but it is not possible to associate any of these with a driving source presently . wss12 noted that the myso , mol 160 , exhibits a very high l@xmath49/l@xmath50 ratio , falling approximately a factor of two above the relationship that was determined for low - mass ysos ( caratti o garatti et al . 2006 ) and extended to mysos based on results for iras 20126@xmath174104 ( caratti o garatti et al . 2008 ) . although it is not generally possible to link individual outflows with driving sources in massive star - forming regions , and caution must be exercised in attempting to draw conclusions about the nature of myso outflows from the l@xmath49/l@xmath50 relationship alone ( wss12 ) , it is instructive to compare the total h@xmath0 luminosity from all detected outflows to the total bolometric luminosity of the mol 121 region . if the h@xmath0 flows are produced by low- or intermediate - mass ysos , one would expect the h@xmath0 emission to be under - luminous with respect to the bolometric luminosity , which comes primarily from the _ iras _ source . there are many uncertainties in estimating the total h@xmath0 luminosity , including the distance to mol 121/_iras _ 20188@xmath173928 , the extinction toward individual h@xmath0 knots , and the ratio of total h@xmath0 to h@xmath51 luminosity . nevertheless it is possible to place reasonable limits on the assumptions . without spectra of the mhos , we can only estimate the effects of extinction . although the extinction is very large in the dense cores , most of the mhos are located in the more extended cloud . it seems reasonable to assume the column density derived for the parsec - scale nh@xmath3 clump in this region by anglada et al . ( 1997 ) , n(h@xmath0 ) = 5.5@xmath52 @xmath42 . using the relation a@xmath53/n(h@xmath0 ) @xmath54 5.3 @xmath55 @xmath30 mag yields an estimate of the visual extinction a@xmath53 @xmath54 29 mag . further assuming an optical parameter for dense clouds of r@xmath53 = 5 , yields a@xmath51 @xmath54 3.6 mag ( allen 2000 ) . the luminosity of the h@xmath0 2.12 @xmath4 emission is given by : @xmath56 f@xmath51 = 4.59 @xmath57 10@xmath58 w m@xmath59 , summed over all the mhos listed in table 1 ( including the pdr and anomaly increases f@xmath51 by a factor of @xmath7 2.5 . ) depending upon the temperature of the shocked h@xmath0 , the 2.12-@xmath4 luminosity is typically @xmath7 5 - 10% of the total ro - vibrational h@xmath0 emission , l@xmath49 ( e.g. , caratti o garatti 2006 , 2008 ) . this yields l@xmath49 @xmath7 11.7 - 23.3 l@xmath6 for the combined mho luminosity . the bolometric luminosity of _ iras _ 20188@xmath173928 , calculated from the far - infrared sed and assuming d = 1.7 kpc , is 8.2@xmath5710@xmath60 l@xmath6 ( mottram et al . these results place mol 121 a factor of 1.5 to three above the l@xmath49/l@xmath50 relationship determined by caratti o garatti et al . ( 2006 , 2008 ) . this is very similar to the result obtained by wss12 for mol 160 , and strongly suggests that the h@xmath0 flows are driven by massive ysos . class i ch@xmath3oh masers are well - correlated with molecular outflows in massive star - forming regions ( cyganowski et al . 2009 and references therein ; fontani et al . 2010 ) and with the presence of compact 3-mm continuum emission ( schnee & carpenter 2009 ) , suggesting that these masers are signposts of an early stage in the evolution of a myso before an expanding uc region has destroyed the accretion disk . the strongest ch@xmath3oh maser detection ( m1 ) in mol 121 is associated with mho 864 ( figure 5 ) . it is by far the brightest at 95 ghz and was the only previously detected maser at 44 ghz ( s. kurtz , private communication ) . mho 865 and the associated maser ( m3 ) are positioned exactly opposite core d , while mho 892 and its associated maser ( m2 ) lie along a slightly different position angle , though they also appear to be associated with core d. the masers all have low velocities , close to the v@xmath61 of the ambient cloud , which is reported to be 1.5 km s@xmath16 by zhang et al . ( 2005 ) and listed as 2.1 km s@xmath16 in the rms catalog ( mottram et al . this is consistent with the suggestion that ch@xmath3oh masers arise in systemic gas in outflow cavity walls ( debuizer et al . 2009 ) . wss12 suggested that the 95-ghz/44-ghz ratio might prove to be a useful diagnostic for the shortest - lived ( e.g. , very early ) phases of massive star formation . in order to estimate the flux ratios of our masers , we binned the 44-ghz data ( provided by s. kurtz ) to approximate our wider velocity channels . this summed data lowered the noise and revealed 44-ghz maser emission at the location of all three of our masers , not just the one known previously . we were then able to measure and compare the fluxes at 44 ghz and 95 ghz . these fluxes are all presented in table 5 . all three masers are brighter at 95 ghz than at 44 ghz , in contrast to the statistical result reported by fontani et al . ( 2010 ) , who concluded that the 95-ghz line is intrinsically fainter based on their detection rates . m1 and m3 , which are associated with the same outflow angle , have remarkably similar flux ratios ( 3.73 and 3.69 , respectively ) . m2 , which is associated with mho 892 , has a 95-ghz to 44-ghz flux ratio of 6.63 , dramatically higher than the other two . mho 892 is much more compact than mhos 864 & 865 , though we do not know if this is related in any way . we can tentatively order the dense cores in mol 121 by evolutionary stage , based on their continuum and spectral - line properties . ( 1 ) core a is associated with the _ iras _ source , a compact pdr , and bright cm - continuum emission , suggesting it is the oldest site of massive star formation in mol 121 . ( 2 ) core d , with its associated faint cm - continuum emission , luminous h@xmath0 outflow , ch@xmath3oh masers , and emission may harbor one or more massive objects that has not yet destroyed its accretion disk . ( 3 ) core c , with its des , likely contains a myso in a pre - uc region phase of evolution . ( 4 ) b & e appear to be massive starless or prestellar cores . we note that cores b & c also lie near the periphery of the pdr , which suggests that mol 121 would be an excellent candidate for follow - up studies to explore the possibility of star formation triggered by the expanding region associated with _ iras _ 20188@xmath173928 . 1 . using nicfps on the arc 3.5-m telescope at apo , we have discovered 12 new mhos in the vicinity of mol 121/_iras _ 20188@xmath173928 , bringing the total known mhos in this region to 17 . we have been able to separate one of the five previously discovered mhos from a pdr and a region marked by an anomalous h@xmath0 2.12-@xmath4/2.25-@xmath4 flux ratio @xmath2 1 . to the best of our knowledge , this is the first time an observation of a h@xmath0 2.12-@xmath4/2.25-@xmath4 flux ratio @xmath2 1 has been reported . we have detected [ ] 1.64-@xmath4 emission toward mhos 864 & 867 . [ ] 1.64-@xmath4/h@xmath0 2.12-@xmath4 and h@xmath0 2.12-@xmath4/2.25-@xmath4 flux ratios measured along and across these mhos indicate the presence of both j - shock and c - shock components . carma 3-mm continuum and line observations have revealed the presence of five compact cores ( a , b , c , d , & e ) and three ch@xmath3oh masers ( m1 , m2 , & m3 ) in the mol 121/_iras _ 20188@xmath173928 region . the physical properties of the cores indicate that cores a , c , & d are presently forming massive stars , while cores b & e may be the sites of future massive star formation . masers m1 , m2 , & m3 lie toward mhos 864 , 892 , & 865 ; all of which are associated with outflow activity from one or more sources embedded in core d. the 95-ghz masers all have 44-ghz counterparts , but they are all brighter at 95 ghz . the two masers ( m1 & m3 ) associated with bipolar mhos 864 & 865 have strikingly similar ratios of @xmath7 3.7 , while m2 has a higher ratio of @xmath7 6.6 . the significance of this is presently unknown , but suggests that future studies of maser line ratios are warranted , and may hold important implications for probing physical conditions in massive star - forming regions . the positions and orientations of the mhos relative to the cores suggest the presence of multiple outflows originating from at least two of the cores . a comparison of the total h@xmath0 luminosity of the mhos to the total bolometric luminosity of this region strongly suggests the outflows are driven by massive ysos . we tentatively age - order the cores from most to least evolved based on the different signposts of star - formation activity . ( 1 ) core a is associated with the _ iras _ source , a compact pdr , and bright cm - continuum emission , suggesting it is the oldest site of massive star formation in mol 121 . ( 2 ) core d , with its associated faint cm - continuum emission , luminous well - collimated h@xmath0 outflow(s ) , ch@xmath3oh masers , and emission may harbor one or more massive objects that has not yet destroyed its accretion disk . ( 3 ) core c is not associated with cm - continuum emission , but it coincides with a des , the anomalous h@xmath0 2.12-@xmath4/2.25-@xmath4 emission , and possible outflow . it likely harbors a myso in a pre - uc region phase of evolution . ( 4 ) while cores b & e are massive , they also appear to be starless or prestellar . this research is based on observations obtained with the apache point observatory 3.5-meter telescope , which is owned and operated by the astrophysical research consortium . we particularly thank the 3.5-meter observing specialists for their assistance in acquiring the data ; al harper , rs s advisor during acquisition of the carma observations ; and michael medford , who performed the original nicfps data reduction for mol 121 . support for carma construction was derived from the gordon and betty moore foundation , the kenneth t. and eileen l. norris foundation , the james s. mcdonnell foundation , the associates of the california institute of technology , the university of chicago , the states of california , illinois , and maryland , and the national science foundation . ongoing carma development and operations are supported by the national science foundation under a cooperative agreement , and by the carma partner universities . special thanks go to stan kurtz , for supplying us with his vla ch@xmath3oh 44 ghz maser data and to michael smith , for useful discussions on shock physics and the possible origin of the anomalous h@xmath0 emission reported in this paper . this research has made use of saoimage ds9 , developed by the smithsonian astrophysical observatory . we acknowledge use of the nasa / ipac infrared science archive , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . this publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation ; and the wide - field infrared survey explorer , which is a joint project of the university of california , los angeles , and the jet propulsion laboratory / california institute of technology , funded by the national aeronautics and space administration . gw - c was funded in part through nasa s illinois space grant consortium , and the authors gratefully acknowledge support from the brinson foundation grant in aid of astrophysics research at the adler planetarium . ka thanks jamie riggs for help in developing the near - infrared continuum subtraction procedure . we also thank the anonymous referee for providing helpful and timely comments to improve this paper . anderson , l. d. , zavagno , a. , deharveng , l. , et al . 2012 , a&a , 542 , 10 anglada , g. , sepulveda , i. , & gomez , j. f. 1997 , a&as , 121 , 255 apache point observatory 2010 , nicfps ( sunspot , nm : apo ) , http://www.apo.nmsu.edu/arc35m/instruments/nicfps/ arvidsson , k. , kerton , c. r. , alexander , m. j. , kobulnicky , h. a. , & uzpen , b. 2010 , aj , 140 , 462 barsony , m. , wolf - chase , g. a. , ciardi , d. r. , & olinger , j. 2010 , apj , 720 , 64 benjamin , r. a. , churchwell , e. , babler , b. l. , et al . 2003 , pasp , 115 , 953 black , j. h. & van dishoeck , e. f. 1987 , iau symposium 115 , 139 brand , j. , cesaroni , r. , caselli , p. , et al . 1994 , a&as , 103 , 541 caratti o garatti , a. , froebrich , d. , eislffel , j. , giannini , t. , & nisini , b. 2008 , a&a , 485 , 137 caratti o garatti , a. , giannini , t. , nisini , b. & lorenzetti , d. 2006 , a&a , 449 , 1077 chen , c - c . , williams , j. p. , & pandian , j. d. 2012 , apj , 752 , 102 churchwell , e. , babler , b. l. , meade , m. r. , et al . 2009 , pasp , 121 , 213 curran , r. l. , & chrysostomou , a. 2007 , mnras , 382 , 699 cyganowski , c. j. , brogan , c. l. , hunter , t. r. , & churchwell , e. 2009 , apj , 702 , 1615 cyganowski , c. j. , whitney , b. a. , holden , e. , et al . 2008 , aj , 136 , 2391 davis , c. j. , gell , r. , khanzadyan , t. , smith , m. d. , & jenness , t. 2010 , a&a , 511 , a24 de buizer , j. m. , redman , r. o. , longmore , s. n. , caswell , j. , & feldman , p. a. 2009 , a&a , 493 , 127 dove , j. e. , rusk , a. c. m. , cribb , p. h. , & martin , p. g. 1987 , apj , 318 , 379 enoch , m. l. , glenn , j. , evans , n. j. ii , et al . 2007 , apj , 666 , 982 fazio , g. g. , hora , j. l. , allen , l. e. , et al . 2004 , apjs , 154 , 10 fontani , f. , cesaroni , r. , & furuya , r. s. 2010 , a&a , 517 , a56 gredel , r. , & dalgarno , a. 1995 , apj , 446 , 852 hollenbach , d. , & mckee , c. f. 1989 , apj , 342 , 306 jenness , t. , scott , p. f. , & padman , r. 1995 , mnras , 276 , 1024 krumholz , m. r. , cunningham , a. j. , klein , r. i. , & mckee , c. f. 2010 , apj , 713 , 1120 krumholz , m. r. , & mckee , c. f. 2008 , nature , 451 , 1082 lee , h - t , takami , m. , duan , h - y , et al . 2012 , apjs , 200 , 2 lee , k. i. , looney , l. w. , klein , r. , & wong , s. 2011 , mnras , 415 , 2790 little , l. t. , bergman , p. , cunningham , c. t. , et al . 1988 , a&a , 205 , 129 lorenzetti , d. , giannini , t. , vitali , f. , massi , f. , & nisini , b. 2002 , apj , 564 , 839 matsakos , t. , massaglia , s. , trussoni , e. , et al . 2009 , a&a , 502 , 217 mccutcheon , w. h. , sato , t. , purton , c. r. , matthews , h. e. , & dewdney , p. e. 1995 , aj , 110 , 1762 molinari , s. , brand , j. , cesaroni , r. , palla , f. , & palumbo , g. g. c. 1998 , a&a , 336 , 339 molinari , s. , faustini , f. , testi , l. , et al . 2008 , a&a , 487 , 1119 motte , f. , bontemps , s. , schilke , p. , et al . 2007 , a&a , 476 , 1243 mottram , j. c. , hoare , m. g. , urquhart , j. s. , et al . 2011 , a&a , 525 , a149 ossenkopf , v. & henning , th . 1994 , a&a , 291 , 943 palla , f. , brand , j. , cesaroni , r. , comoretto , g. , & felli , m. 1991 , a&a , 246 , 249 pudritz , r. e. , rogers , c. s. , & ouyed , r. 2006 , mnras , 365 , 1131 rosolowsky , e. , dunham , m. k. , ginsburg , a. , et al . 2010 , apjs , 188 , 123 schnee , s. , & carpenter , j. m. 2009 , apj , 698 , 1456 shu , f. , najita , j. , ostriker , e. , et al . 1994 , apj , 429 , 781 simon , r. , jackson , j. m. , clemens , d. p. , bania , t. m. , & heyer , m. h. 2001 , apj , 551 , 747 smith , m. d. 1994 , a&a , 289 , 256 smith , m. d. 1995 , a&a , 296 , 789 smith , m. d. , khanzadyan , t. , & davis , c. j. 2003 , mnras , 339 , 524 smith , m. d. 2012 , astrophysical jets and beams ( cambridge : cambridge university press ) , pp . 116 - 117 turner , j. , kirby - docken , k. , & dalgarno , a. 1977 , apjs , 35 , 281 varricatt , w. p. , davis , c. j. , ramsay , s. , & todd , s. p. 2010 , mnras , 404 , 661 wolf - chase , g. , smutko , m. , sherman , r. , harper , d. a. , & medford , m. 2012 , apj , 745 , 116 ( wss12 ) wolf - chase , g. , moriarty - schieven , g. , fich , m. , & barsony , m. 2003 , mnras , 344 , 809 wolf - chase , g. a. , walker , c. k. , & lada , c. j. 1995 , apj , 442 , 197 wood , d. o. s. , & churchwell , e. 1989 , apj , 340 , 265 wright , e. l. , eisenhardt , p. r. m. , mainzer , a. k. , et al . 2010 , aj , 140 , 1868 yao , y. , ishii , m. , nagata , t. , nakaya , h. , & sato , s. 2000 , apj , 542 , 392 zhang , q. , hunter , t. r. , brand , j. , et al . 2005 , apj , 625 , 864 zinnecker , h. , & yorke , h. w. 2007 , ara&a , 45 , 481 rrcccccccc [ table1 ] 863 & 8.944 & 20 20 41.5 & + 39 38 32 & 27.4@xmath622.1 & 1.3@xmath620.3 & 20.4@xmath624.3 & 4.49@xmath620.34 & 0.40@xmath620.05 & 11.2@xmath621.5 + 864 & 26.627 & 20 20 39.6 & + 39 38 16 & 158.3@xmath6213.1 & 17.4@xmath622.7 & 9.1@xmath621.6 & 10.20@xmath620.77 & 0.92@xmath620.10 & 11.0@xmath621.4 + 865 & 19.225 & 20 20 39.0 & + 39 38 09 & 99.4@xmath628.5 & 19.2@xmath622.1 & 5.2@xmath620.7 & 9.28@xmath620.70 & 2.09@xmath620.16 & 4.4@xmath620.5 + 866 & 6.648 & 20 20 39.0 & + 39 38 16 & 33.0@xmath622.7 & 2.6@xmath620.5 & 12.7@xmath622.7 & 2.23@xmath620.20 & 0.46@xmath620.08 & 4.9@xmath621.0 + 867 & 7.179 & 20 20 38.1 & + 39 37 52 & 42.2@xmath623.2 & 2.1@xmath620.4 & 20.1@xmath624.5 & 3.03@xmath620.27 & 0.33@xmath620.10 & 9.1@xmath622.9 + 891 & 7.676 & 20 20 42.5 & + 39 38 40 & 14.9@xmath621.1 & 0.7@xmath620.2 & 21.0@xmath626.6 & 2.59@xmath620.20 & 0.18@xmath620.04 & 14.6@xmath623.3 + 892 & 6.186 & 20 20 40.5 & + 39 38 18 & 9.0@xmath620.9 & & @xmath57.5 & 1.79@xmath620.14 & & @xmath5 10.1 + 893 & 2.724 & 20 20 38.7 & + 39 38 27 & 2.3@xmath620.3 & & @xmath55.6 & 0.39@xmath620.05 & & @xmath5 2.6 + 894 & 2.707 & 20 20 38.2 & + 39 38 26 & 5.4@xmath620.4 & & @xmath516.7 & 2.01@xmath620.15 & & @xmath5 14.0 + 895 & 3.941 & 20 20 37.5 & + 39 38 25 & 3.2@xmath620.4 & & @xmath55.4 & 0.51@xmath620.06 & & @xmath5 3.5 + 896 & 1.268 & 20 20 38.2 & + 39 38 30 & 1.7@xmath620.2 & 0.3@xmath620.1 & 5.4@xmath621.4 & 0.68@xmath620.06 & 0.18@xmath620.05 & 3.9@xmath621.2 + 937 & 3.187 & 20 20 37.9 & + 39 38 02 & 4.2@xmath620.6 & & @xmath57.0 & 0.70@xmath620.12 & & @xmath5 2.4 + 938 & 4.078 & 20 20 34.9 & + 39 37 58 & 6.8@xmath620.5 & & @xmath516.1 & 1.61@xmath620.13 & 0.13@xmath620.04 & 12.7@xmath623.8 + 939 & 11.360 & 20 20 41.1 & + 39 38 02 & 19.0@xmath622.8 & & @xmath52.0 & 1.13@xmath620.11 & & @xmath5 7.3 + 940 & 11.669 & 20 20 43.6 & + 39 37 52 & 13.1@xmath621.0 & & @xmath515.2 & 1.75@xmath620.14 & & @xmath5 14.7 + 941 & 8.430 & 20 20 43.4 & + 39 37 41 & 11.1@xmath620.9 & 0.9@xmath620.2 & 12.0@xmath623.0 & 1.34@xmath620.11 & 0.12@xmath620.04 & 11.6@xmath624.3 + 942 & 7.642 & 20 20 44.4 & + 39 37 40 & 9.5@xmath620.7 & & @xmath516.4 & 0.60@xmath620.07 & & @xmath5 5.0 + anomaly & 28.718 & & & 124.9@xmath628.5 & 269.0@xmath6216.6 & 0.46@xmath620.04 & & & + pdr & 95.886 & & & 552.5@xmath6237.8 & 356.9@xmath6223.7 & 1.5@xmath620.1 & & & + cccccc [ table1 ] 864 & 1 & 20 20 39.8 & + 39 38 23 & 0.9@xmath620.2 & @xmath55.9 + & 2 & 20 20 39.8 & + 39 38 21 & 0.8@xmath620.1 & 7.0@xmath621.8 + & 3 & 20 20 39.7 & + 39 38 19 & 0.1@xmath620.03 & 6.8@xmath621.3 + & 4 & 20 20 39.7 & + 39 38 17 & @xmath20.03 & 13.8@xmath622.8 + & 5 & 20 20 39.6 & + 39 38 15 & @xmath20.03 & 14.5@xmath624.5 + & 6 & 20 20 39.6 & + 39 38 14 & 0.4@xmath620.1 & 12.8@xmath624.3 + & 7 & 20 20 39.8 & + 39 38 17 & @xmath20.04 & @xmath57.9 + & 8 & 20 20 40.0 & + 39 38 17 & @xmath20.3 & @xmath51.4 + & 9 & 20 20 39.6 & + 39 38 19 & @xmath20.03 & @xmath57.1 + & 10 & 20 20 39.4 & + 39 38 19 & & @xmath50.7 + 867 & 11 & 20 20 38.2 & + 39 37 55 & 1.2@xmath620.3 & @xmath54.0 + & 12 & 20 20 38.1 & + 39 37 53 & 0.5@xmath620.1 & @xmath57.2 + & 13 & 20 20 38.1 & + 39 37 52 & 0.6@xmath620.1 & 10.4@xmath623.6 + & 14 & 20 20 38.2 & + 39 37 50 & 0.5@xmath620.1 & @xmath56.5 + & 15 & 20 20 38.3 & + 39 37 53 & 1.2@xmath620.2 & @xmath54.5 + & 16 & 20 20 38.2 & + 39 37 51 & 1.8@xmath620.3 & @xmath55.6 + & 17 & 20 20 38.4 & + 39 37 52 & 1.9@xmath620.5 & @xmath52.3 + ccccccccccc [ table1 ] a & 20 20 39.262 & + 39 37 51.018 & 1.36@xmath570.31 & 13.6 & 17.4 & 29.9 & 14.1 & 6.00 & 2.83 + b & 20 20 38.865 & + 39 37 56.635 & 2.60@xmath570.08 & 7.3 & 12.0 & 20.6 & 9.74 & 3.22 & 1.52 + c & 20 20 39.325 & + 39 37 57.770 & 1.16@xmath570.45 & 15.0 & 18.7 & 32.1 & 15.2 & 6.62 & 3.12 + d & 20 20 39.330 & + 39 38 11.770 & 2.87@xmath571.50 & 7.3 & 17.4 & 29.9 & 14.1 & 3.22 & 1.52 + e & 20 20 37.935 & + 39 38 20.393 & 2.83@xmath570.61 & 6.4 & 10.5 & 18.1 & 8.55 & 2.82 & 1.33 + cccccccc [ table 2 ] m1 & 864 & 20 20 39.8 & 39 38 15 & 5.29 & 1.8 & 1.42 & 3.73 + m2 & 892 & 20 20 40.6 & 39 38 18 & 1.79 & 0.5 & 0.27 & 6.63 + m3 & 865 & 20 20 39.0 & 39 38 07 & 1.18 & 0.1 & 0.32 & 3.69 +

distributed gossip algorithm has been studied in literature for practical implementation of the distributed consensus algorithm as a fundamental algorithm for the purpose of in - network collaborative processing . this paper focuses on optimizing the convergence rate of the gossip algorithm for both classical and quantum networks . a novel model of the gossip algorithm with non - uniform clock distribution is proposed which can reach the optimal convergence rate of the continuous - time consensus algorithm . it is described that how the non - uniform clock distribution is achievable by modifying the rate of the poisson process modeling the clock of the gossip algorithm . the minimization problem for optimizing the asymptotic convergence rate of the proposed gossip algorithm and its corresponding semidefinite programming formulation is addressed analytically . it is shown that the optimal results obtained for uniform clock distribution are suboptimal compared to those of the non - uniform one and for non - uniform distribution the optimal answer is not unique i.e. there is more than one set of probabilities that can achieve the optimal convergence rate . based on the optimal continuous - time consensus algorithm and the detailed balance property , an effective method of obtaining one of these optimal answers is proposed . regarding quantum gossip algorithm , by expanding the density matrix in terms of the generalized gell - mann matrices , the evolution equation of the quantum gossip algorithm is transformed to the state update equation of the classical gossip algorithm . by defining the quantum gossip operator , the original optimization problem is formulated as a convex optimization problem , which can be addressed by semidefinite programming . * _ index terms _ * quantum networks , distributed gossip algorithm , optimal convergence rate

1512.03551

many manmade systems of the present day are networks of a large number of cooperating standalone units . examples of such systems include power grids , transportation and water distribution networks , telephone networks , the internet , the global financial network and social networks @xcite . the appealing features of networked systems are their natural flexibility , reliability and robustness to failure , easy re - configuration , and low cost . distributed consensus algorithm @xcite is one of the fundamental algorithms for the purpose of in - network collaborative processing in the context of coordinated control of networked systems modeled as networks of autonomous agents . using distributed consensus algorithm , agents achieve a stable and reliable opinion regarding a phenomenon , through the local exchange of information with their neighbors . distributed consensus algorithm has found applications in numerous fields including vehicle formation @xcite , multiagent collaboration and flocking @xcite , data fusion and tracking @xcite , distributed inference @xcite and load balancing @xcite . however , there are major challenges in the implementation of distributed consensus algorithm to enable agents to exchange their state with all their neighbors in every round of the algorithm . distributed gossip algorithm @xcite emerges as a mean for implementing the distributed consensus algorithm . there is an extensive body of literature on consensus and gossip algorithms , see the surveys @xcite , and references therein . one of the pioneering works on the gossip algorithms is @xcite where the authors provide the bound on the running time of the gossip algorithm to converge to the global average with certain accuracy . @xcite analyzed the effects of quantized communication on the gossip algorithm . @xcite enhanced the convergence rate of the gossip algorithm by exploiting the geographic information of the network . in @xcite accelerated gossip algorithms are proposed where local memory is exploited by installing shift - registers at each agent . @xcite studied the effect of node mobility on the convergence rate of the gossip algorithm . gossip algorithm has been employed for different applications . @xcite developed several distributed gossip - based algorithms to solve convex optimization problems defined over networks . @xcite introduced a gossip - based algorithm performing distributed kalman filtering for networked systems . @xcite employed gossip algorithm for distributed estimation in static distributed random fields . @xcite studied different deterministic gossiping protocols for avoiding deadlocks and achieving consensus under different assumptions . in the conventional gossip algorithm @xcite , each agent updates its state with one of its neighbors in random when its clock ticks . the clock ticks are at the times of a poisson process . the probability that each agent selects one of its neighbors is referred to as the transition probabilities and the probability that each agent starts the state updating process is the clock distribution probabilities . in the gossip algorithm introduced in @xcite , the clocks of agents follow the poisson processes with the same rate . this results in uniform clock distribution , which in turn leaves the transition probabilities as the only optimization parameters . the main motivation behind the analysis presented in this paper is to propose and optimize a gossip algorithm that can achieve the convergence rate of the optimal continuous - time consensus algorithm over the same network topology . to do so , in the gossip algorithm proposed in this paper , the clock distribution probabilities are considered as variables in the optimization problem . this results in a more general framework for optimizing the gossip algorithm where the case of uniform clock distribution @xcite is a subset of this problem . first we introduce the gossip algorithm with non - uniform clock distribution over classical networks . then we explain how the poisson process is the most suitable for the clock tick model of the gossip algorithm , and subsequently how non - uniform clock distribution can be achieved by modifying the rate of the poison process according to the clock distribution probabilities . after formulating the asymptotic convergence rate of the gossip algorithm , the minimization problem and its corresponding semidefinite programming problem for optimizing the asymptotic convergence rate of the gossip algorithm over classical networks is presented . we have provided the closed - form formulas for the probabilities and the convergence rate of the gossip algorithm for both cases of uniform and non - uniform clock distributions over different network topologies . this is done to show that the optimal results obtained for uniform clock distribution are suboptimal compared to those of the non - uniform clock distribution . for the case of non - uniform clock distribution , the optimal answer is not necessarily unique i.e. there is more than one set of transition and clock distribution probabilities that can achieve the fastest convergence rate . this is due to the fact that the number of variables in the semidefinite programming formulation of the problem is more than the number of equations . we have shown that for gossip algorithm with non - uniform clock distribution , one of the optimal answers is achievable by using the optimal results of the continuous - time consensus algorithm @xcite and ensuring that the transition and clock distribution probabilities satisfy the detailed balance property . the concept of consensus and gossip algorithms over quantum networks is a relatively new topic . the initial activities in this field have been initiated by the recent progress in the field of quantum information science and quantum distributed computing @xcite . this development is driven by the fact that centralized and big quantum computers with numerous qubits are expensive and difficult to build and maintain . an alternative to such systems is a network of smaller quantum computers which is analyzed as networks of quantum nodes . consensus and gossip problems are brought to the quantum domain in @xcite where they are reinterpreted as a symmetrization problem . authors in @xcite derive the general conditions for convergence of the consensus algorithm for quantum networks . authors in @xcite took a different approach and employed the induced graphs of the quantum interaction graph . by doing so , they could relate the consensus problem over a network of n - qubit to its equivalent classical consensus problem , and established the necessary and sufficient conditions for the exponential and asymptotic convergence rate of the quantum consensus . in @xcite , the convergence rate of the continuous - time quantum consensus is optimized , and it is shown that the optimal convergence rate is independent of the value of @xmath0 in qudits . they have shown that the induced graphs are the schreier graphs and using the young tabloids ; they have categorized all possible induced graphs . furthermore , by extending the proof of the aldous conjecture @xcite to all possible induced graphs , they have reduced the optimization of the continuous - time quantum consensus to optimizing the second smallest eigenvalue of the laplacian of the induced graph corresponding to partition @xmath1 . for a wide range of topologies , they have analytically addressed the semidefinite programming formulation of the reduced optimization problem . @xcite studies the optimization of the discrete - time model of the quantum consensus over a quantum network with @xmath2 qudits . in contrast to the results obtained for the continuous time model @xcite , in @xcite it is shown that the convergence rate of the consensus algorithm depends on the value of @xmath0 is qudits . exploiting the specht module representation of partitions of @xmath2 , in @xcite it is shown that the laplacian matrix corresponding to partition @xmath3 includes the corresponding spectrum of all other partitions . based on this result and the generalization of the aldous conjecture to all partitions of integer @xmath2 , they have shown that the problem of optimizing the convergence rate of the discrete - time quantum consensus reduces to optimizing the second largest eigenvalue modulus ( slem ) of the weight matrix . by providing the semidefinite programming formulation of the resultant problem , they have addressed it analytically . the analysis presented in this paper aim at optimizing the convergence rate of the quantum gossip algorithm to that of its quantum consensus state . by expanding the density matrix in terms of the generalized gell - mann matrices , we transform the evolution equation of the quantum gossip algorithm to the state update equation of the classical gossip algorithm . after defining the quantum gossip operator , we have formulated the original optimization problem as a convex optimization problem . we have shown that for different topologies , the resultant convex optimization problem can be solved using semidefinite programming and linear programming . in section [ sec : classicalgossip ] our model for the gossip algorithm with non - uniform clock distribution is introduced . followed by derivation of the optimization problem for optimizing the convergence rate of the gossip algorithm and the semidefinite programming formulation of the obtained optimization problem . section [ sec : quantumgossipalgorithm ] presents the quantum gossip algorithm and explains how it can be modeled as a classical gossip algorithm . formulation of the convex optimization problem for optimizing the convergence rate of the quantum gossip algorithm is also presented in this section . in section [ sec : optimizationsectioncomplete ] analytical optimization of the gossip algorithm along with closed - form expressions for the optimal probabilities and convergence rate for a range of topologies have been presented . section [ sec : conclusion ] concludes the paper . in our model for the gossip algorithm , once the clock in agent @xmath4 ticks , it selects one of its neighbors , ( agent @xmath5 ) at random with probability @xmath6 and communicates with the selected neighboring agent . upon communication , two agents update their states @xmath7 to the average of their current values @xmath8 . this can be described by the following equation , @xmath9 where @xmath10/2 . \end{aligned}\ ] ] @xmath11 is the identity matrix and @xmath12 is a column vector where its @xmath4-th element is @xmath13 and the rest is zero . we refer to each one of the matrices @xmath14 as an averaging matrix . defining @xmath15 as the permutation operator that exchanges the indices @xmath4 and @xmath5 as below , @xmath16 the averaging matrix @xmath14 can be written as @xmath17 @xmath15 can be represented in the matrix form as below , @xmath18 the probabilities @xmath6 form a square matrix @xmath19 that has the same sparsity pattern as the adjacency matrix of the graph @xmath20 . in other words , @xmath21 if @xmath22 . also all of its elements are nonnegative and each row of the matrix @xmath19 should sum up to one , i.e. @xmath23 . this is due to the fact that when the clock in agent @xmath4 ticks , it has to communicate with one of its neighbors . based on the update equation ( [ eq : gossip1 ] ) the state of agents at @xmath24-th time interval @xmath25 can be related to the initial state of agents @xmath26 as @xmath27 , where @xmath28 . each one of matrices @xmath29 is an averaging matrix similar to @xmath14 in ( [ eq : gossip2 ] ) . the product @xmath30 should converge to @xmath31 , in order to guarantee the convergence of @xmath32 to the average value or consensus state , in other words , @xmath33 @xmath34 is a column vector with @xmath2 elements , all equal to one . since the iteration ( [ eq : gossip1 ] ) computes the average of states of agents then the sum of values of agents should be preserved , i.e. @xmath35 . by substituting ( [ eq : gossip1 ] ) we have @xmath36 this means that the vector of all ones @xmath37 is the left eigenvector of @xmath38 corresponding to eigenvalue one . in addition , the vector of averages @xmath39 must serve as the fixed point of iteration ( [ eq : gossip1 ] ) i.e. @xmath40 , or in other words , the vector of all ones @xmath37 is the right eigenvector of @xmath38 corresponding to eigenvalue one . thus it can be deduced that @xmath34 is the eigenvector corresponding to eigenvalue one for all averaging matrices that can serve as @xmath38 in iteration ( [ eq : gossip1 ] ) . let @xmath41 be the probability that the clock in agent @xmath4 ticks at @xmath24-th time interval and @xmath42 be the probability that agent @xmath4 communicates with agent @xmath5 given that its clock has ticked , then the probability that @xmath38 is equal to the averaging matrix @xmath14 is @xmath43 and the matrices @xmath38 are drawn independent and identically distributed ( i.i.d . ) from the set of possible averaging matrices . defining the gossip operator @xmath44 as the mean of @xmath38 we have @xmath45 in the literature @xcite the probability that the clock in vertex @xmath4 ticks at @xmath24-th time interval ( @xmath41 ) is assumed to be @xmath46 ( i.e. uniform clock distribution ) by default . but here we have relaxed this constraint to enable a more general feasible region for the optimization problem and as a result a faster convergence rate for the gossip algorithm . uniform clock distribution is obtained by assuming that the clock at each vertex ticks according to a poisson process where the poisson processes in vertices have the same rate . thus , the inter - tick times at each vertex are rate @xmath13 exponentials , independent across vertices and over time . this is equivalent to a single clock ticking according to a rate @xmath2 poisson process . the main reason that makes the poisson process most suitable for the clock tick model of the gossip algorithm is the memoryless property of the poisson process . consider a clock that ticks according to a poisson process with rate @xmath47 , we define the random variable @xmath48 as the waiting time from time @xmath49 until the next clock tick . then for any @xmath50 , the random variable @xmath48 has the same exponential distribution with mean @xmath51 , i.e. @xmath52 , @xmath53 , independent of time @xmath49 . the memoryless property of poisson process states that at each point in time , the waiting time until the next clock tick has the same exponential distribution as the original inter - tick time , regardless of how long ago the last clock tick occurred . the poisson process is the only renewal process having this memoryless property . how much time is elapsed since the last clock tick gives no information about how long to wait until the next clock tick . another conclusion from this remarkable property of the poisson process is that the number of clock ticks in any time interval of length @xmath54 has a poisson distribution with mean @xmath55 , independent of the beginning point of the time interval . the following properties can be intuitively concluded from the memoryless property of the poisson process . * independent increments : the numbers of clock ticks occurring in disjoint intervals of time are independent . * stationary increments : the number of clock ticks occurring in a given time interval depends only on the length of the interval . * the probability of one clock ticking occurring in a time interval of length @xmath56 is @xmath57 for @xmath58 . * the probability of two or more clock ticks occurring in a time interval of length @xmath56 is @xmath59 for @xmath58 . the last property states that the probability of two or more clock ticks in a very small time interval of length @xmath56 is negligibly small compared to @xmath56 itself as @xmath58 . the clock in each vertex ticks according to an independent poisson process with rate @xmath60 , where @xmath61 is the number of clock ticks from @xmath4-th vertex up to time @xmath49 . in ( * theorem 1.1.3 ) it is shown that the merged process @xmath62 for @xmath50 is a poisson process with rate @xmath63 . denoting by @xmath64 the time interval between the @xmath65-th and @xmath24-th clock tick in the merged poisson process and letting @xmath66 if the @xmath24-th clock tick in the merged poisson process is the clock tick from @xmath4-th vertex , then the probability that the clock in vertex @xmath4 ticks at @xmath24-th time interval @xmath67 is @xmath68 which is independent of @xmath49 . another perspective on ( [ eq : clockdistribution341 ] ) is by considering the merged poisson process @xmath69 . suppose that each clock tick in the merged poisson process is the clock tick from @xmath4-th vertex with respective probabilities @xmath70 , independently of the clocks in all other vertices . let @xmath61 be the number of clock tick from @xmath4-th vertex up to time @xmath49 . then @xmath71 for @xmath72 are @xmath2 independent poisson processes having respective rates @xmath73 . the remarkable result ( [ eq : clockdistribution341 ] ) states that the next clock tick is from the @xmath4-th vertex with probability @xmath74 regardless of how long it takes until the next arrival . based on the merging and the memoryless property of the poisson process , it is apparent that the clock distribution of the gossip algorithm can be modified by setting the rate of the poisson process in each vertex according to ( [ eq : clockdistribution341 ] ) . in the final optimization stage we have provided the optimal answers for both cases of uniform and non - uniform clock and it is shown that the uniform case is suboptimal compared to the non - uniform one . from ( [ eq : gossip4 ] ) it can be deduced that in expectation convergence of @xmath30 to @xmath31 is equivalent to convergence of @xmath75 to @xmath31 . in @xcite the following conditions have been stated as the necessary conditions for convergence of @xmath75 to @xmath31 under the assumption of uniform selection of vertices . @xmath77 here @xmath78 refers to the spectral radius of a matrix . it is straightforward to show that the same results holds true for the case of non - uniform selection of vertices . in other words , assuming non - uniform selection of vertices , the conditions ( [ eq : gossip5 ] ) and ( [ eq : gossip7 ] ) are the necessary conditions for convergence of @xmath75 to @xmath31 . the matrix @xmath79 satisfies conditions ( [ eq : gossip5 ] ) . this is independent of the distribution @xmath80 that each one of averaging matrices is selected as @xmath38 . this is due to the fact that @xmath79 is the expected value of averaging matrices ( [ eq : gossip2 ] ) which all of them satisfy both conditions ( [ eq : gossip5 ] ) . satisfying condition ( [ eq : gossip7 ] ) relies solely on the probabilities @xmath41 and @xmath42 . considering the averaging matrix ( [ eq : gossip2 ] ) , the gossip operator @xmath44 can be written as @xmath81 in terms of the permutation operator @xmath82 the gossip operator @xmath44 can be written as @xmath83 . the last equality is obtained from the fact that @xmath84 . introducing @xmath85 as @xmath86 the gossip operator @xmath79 can be written in terms of laplacian operator @xmath87 as @xmath88 with the laplacian operator @xmath87 defined as @xmath89 to analyze the convergence rate of the gossip algorithm as described above , first we have to define the averaging time for the gossip algorithm . averaging time [ averagingtimegossip ] + for any @xmath90 , the @xmath91-averaging time of the randomized gossip algorithm ( denoted by @xmath92 is defined as follows : @xmath93 @xmath34 is the vector of all ones and @xmath94 refers to the @xmath95 norm of the vector @xmath96 . in ( * theorem 3 ) it is shown that the averaging time @xmath97 monotonically increases by second largest eigenvalue of @xmath98 . here the matrices @xmath14 are the averaging matrices defined in ( [ eq : gossip2 ] ) . thus optimizing the convergence rate of the gossip algorithm is equivalent to minimizing the second largest eigenvalue of the matrix @xmath99 . this minimization problem can be formulated as follows : @xmath100 this is a convex optimization problem since the second largest eigenvalue of a doubly stochastic matrix is a convex function on the set of symmetric matrices . this optimization problem can be reformulated as the following semidefinite programming problem @xcite . @xmath101 @xmath102 means that the matrix @xmath103 is a positive semidefinite matrix . we refer to problem ( [ eq : classicalgossipsdp ] ) as the fastest classical gossip algorithm ( fcga ) problem . an automorphism of the graph @xmath104 is a permutation @xmath105 of @xmath106 such that @xmath107 if and only if @xmath108 , the set of all such permutations , with composition as the group operation , is called the automorphism group of the graph and denoted by @xmath109 . for a vertex @xmath110 , the set of all images @xmath111 , as @xmath105 varies through a subgroup @xmath112 , is called the orbit of @xmath4 under the action of @xmath113 . the vertex set @xmath106 can be written as disjoint union of distinct orbits . in @xcite , it has been shown that the optimal probabilities on the edges within an orbit are equal . we consider a quantum network as a composite ( or multipartite ) quantum system with @xmath2 qudits . assuming @xmath114 as the d - dimensional hilbert space over @xmath115 , then the state space of the quantum network is within the hilbert space @xmath116 . the state of the quantum system is described by its density matrix @xmath117 . this matrix is positive hermitian and its trace is one @xmath118 . the network is associated with an underlying graph @xmath119 , where @xmath120 is the set of indices for the @xmath2 qudits , and each element in @xmath121 is an unordered pair of two distinct qudits , denoted as @xmath122 with @xmath123 . permutation group @xmath124 acts in a natural way on @xmath106 by mapping @xmath106 onto itself . for each permutation @xmath125 we associate unitary operator @xmath126 over @xmath127 , as below @xmath128 where @xmath129 is an operator in @xmath114 for all @xmath130 . a special case of permutations is the swapping permutation or transposition where @xmath131 , @xmath132 and @xmath133 for all @xmath110 and @xmath134 we denote the swapping permutation between the qudits indices @xmath5 and @xmath24 by @xmath135 and the corresponding swapping operator by @xmath136 . in ( * appendix a ) the swapping operator @xmath136 has been expressed as linear combination of the cartesian product of gell - mann matrices . in the model of quantum network presented here , every vertex represents a particle . employing the quantum gossip interaction introduced in @xcite , on each iteration of the gossip algorithm , one particle ( say @xmath5-th particle ) is selected with probability @xmath137 . then this particle selects one of its neighbours for updating its state . this neighbouring particle is selected with probability @xmath138 according to the underlying graph of the quantum network , and the updating procedure on the states of @xmath5-th and @xmath24-th particles can be written in terms of the evolution of the density matrix as below , @xmath139 where its resultant quantum consensus state is as below @xmath140 a necessary condition for the quantum gossip algorithm to reach its quantum consensus state ( i.e. @xmath141 ) is that the underlying graph of the quantum network should be connected . the analysis presented in this paper aim at optimizing the convergence rate of the quantum gossip algorithm to its quantum consensus state . to this aim , first we expand the density matrix @xmath117 as the linear combination of the generalized gell - mann matrices ( * ? ? ? * appendix a ) as below , @xmath142 where @xmath2 is the number of particles and @xmath143 denotes the cartesian product and @xmath47 matrices are the generalized gell - mann matrices ( * ? ? ? * appendix a ) . due to hermity of density matrix , its coefficients of expansion @xmath144 are real numbers and because of unit trace of @xmath145 we have @xmath146 . based on the decomposition of @xmath145 ( [ eq : decompositiondensitygeneral ] ) , its permutations can be written as @xmath147 note that in ( [ eq : decompositiondensitypermutation ] ) due to permutation operators , the place of indices @xmath148 and @xmath149 in the index of parameter @xmath145 are interchanged . substituting the density matrix @xmath145 from ( [ eq : decompositiondensitygeneral ] ) and its permutation ( [ eq : decompositiondensitypermutation ] ) in ( [ eq : lindblad2 ] ) and considering the independence of the matrices @xmath150 , we can conclude the following for the evolution of the density matrix ( [ eq : lindblad2 ] ) , @xmath151 for all @xmath152 . following the same procedure , the tensor component of the consensus state of the quantum gossip algorithm ( [ eq : qcmefinalconsensus ] ) can be written as below @xmath153 and for the connected underlying graph , the quantum gossip algorithm reaches quantum consensus , componentwise as @xmath154 . defining @xmath155 as a column vector of length @xmath156 with components @xmath157 , the evolution of the density matrix ( [ eq : lindblad2 ] ) can be written as blow , @xmath158 where @xmath136 is the swapping operator ( * ? ? ? * appendix a ) , provided that @xmath0 is replaced with @xmath159 which in turn results in gell - mann matrices of size @xmath160 . comparing the set of equations in ( [ eq : densityequation1 ] ) with those of the classical gossip in ( [ eq : gossip1 ] ) we can see that the quantum gossip ( [ eq : quantumstateupdate ] ) is transformed into the classical gossip ( [ eq : gossip1 ] ) with @xmath161 tensor component @xmath162 as the agents states . thus , similar to the analysis performed in section [ sec : classicalgossip ] for the classical gossip algorithm , it is straight forward to see that the convergence rate of the quantum gossip algorithm is governed by the second largest eigenvalue of the quantum gossip operator @xmath163 defined as @xmath164 the quantum gossip operator @xmath163 is obtained by taking the average over all possible swapping ( according to the underlying graph ) where the probability that a particular pair @xmath165 is swapped at time @xmath49 is @xmath166 . the quantum gossip operator @xmath167 can be written in terms of quantum laplacian operator @xmath168 as @xmath169 , where @xmath168 can be written as @xmath170 . based on ( [ eq : meanaveragingmatricesquantum ] ) and the analysis presented in section [ sec : classicalgossip ] for the classical gossip algorithm , it can be concluded that optimizing the convergence rate of the quantum gossip algorithm is equivalent to minimizing the second largest eigenvalue of the quantum gossip operator @xmath167 . this minimization problem can be formulated as following convex optimization problem , @xmath171 ( [ eq : quantumgossipconvexformulation ] ) is referred as the fastest quantum gossip ( fqg ) problem . the underlying graph of classical gossip algorithm corresponding to ( [ eq : quantumstateupdate ] ) is a cluster of connected components where each connected graph component corresponds to a given partition of @xmath2 into @xmath172 integers , namely @xmath173 , where @xmath174 and @xmath175 for @xmath176 is the number of indices in @xmath177 with equal values . the connected graph components are referred to as the induced graphs . the induced graphs originate from the tensor components @xmath162 that can be transformed into each other by permuting their indices and thus updating their states to their average value according to ( [ eq : densityequation1 ] ) . by establishing the intertwining relation between the induced graphs in @xcite , the results of aldous s conjecture is generalized to the set of all induced graphs and it is shown that the second eigenvalue of the laplacian matrices of all the induced graphs are the same . therefore for optimizing the convergence rate of the corresponding classical gossip algorithm ( [ eq : quantumgossipconvexformulation ] ) , it is enough to optimize the problem for the induced graph corresponding to partition @xmath178 which is same as the underlying graph of the quantum network . in @xcite similar results regarding generalization of aldous s conjecture has been achieved by exploiting the specht module representation of partitions of @xmath2 . in this section we address the optimization of the fastest classical gossip algorithm ( fcga ) problem ( [ eq : classicalgossipsdp ] ) over different network topologies . in the first two subsection , we represent the optimal results for all possible topologies with @xmath179 vertices which are connected and non - isomorphic and the topologies which the fcga problem ( [ eq : classicalgossipsdp ] ) can be solved using linear programming . in the third and fourth subsections , the optimization of the fcga problem ( [ eq : classicalgossipsdp ] ) is addressed using semidefinite programming ( sdp ) for uniform and non - uniform clock distributions , respectively . in the analysis presented in this section , the complete solution procedure is provided only for the symmetric star topology . for the rest of topologies we have provided the final results i.e. the optimal probabilities and the optimal value of the second largest eigenvalue of the gossip operator @xmath44 . the notations and derivations of the semidefinite programming used in this section are adopted from ( * ? ? ? * appendix b ) . here we provide the optimal results for all possible topologies with @xmath179 vertices which are connected and non - isomorphic . in table [ tab : n4slemoptimal ] , all possible connected topologies with @xmath180 vertices ( as depicted in figure [ fig : n4graphs ] ) are listed along with their optimal transition and clock distribution probabilities along with the optimal value of the second largest eigenvalue of the gossip operator @xmath181 . note that for all topologies listed in table [ tab : n4slemoptimal ] , the optimal value of the second largest eigenvalue of the gossip operator @xmath182 obtained for uniform clock distribution is same as that of the non - uniform clock distribution . vertices which are non - isomorphic . ( a ) path graph , ( b ) star graph , ( c ) lollipop graph , ( d ) cycle graph , ( e ) paw graph , ( f ) complete graph.,width=604 ] clock distribution + probabilities & transition probabilities + path & @xmath183 & @xmath184 + @xmath185 & @xmath186 + @xmath187 + @xmath188 + star & @xmath189 & @xmath190 + @xmath191 & @xmath192 + @xmath193 + lollipop & @xmath194 & @xmath184 + @xmath195 + @xmath196 & @xmath197 + @xmath198 , + @xmath199 + @xmath200 + cycle & @xmath201 & @xmath185 + @xmath202 & @xmath203 + for @xmath204 + @xmath205 + paw & @xmath201 & @xmath185 + @xmath202 & @xmath203 + for @xmath204 + @xmath205 + @xmath206 + complete graph & @xmath207 & @xmath208 for @xmath209 & @xmath210 for @xmath211 + in this subsection , we provide the optimal results for a number of topologies where the fcga problem ( [ eq : classicalgossipsdp ] ) can be solved using linear programming @xcite . note that the results reported in this subsection ( other than that of wheel topology ) are globally optimal and they are obtained for uniform clock distribution . in other words , optimizing the fcga problem ( [ eq : classicalgossipsdp ] ) with non - uniform clock distribution over topologies ( other than wheel topology ) discussed in this subsection would result in the same convergence rate obtained from solving the fcga problem ( [ eq : classicalgossipsdp ] ) with uniform clock distribution . in case of the wheel topology non - uniform clock distribution would result in the global optimal answer . this topology is obtained from cartesian product of @xmath212 edge - transitive graphs . the laplacian operator @xmath213 for the whole graph can be written as below , @xmath214 where @xmath215 is the laplacian operator of @xmath4-th graph . @xmath216 is the identity matrix with size @xmath217 and @xmath217 is the number of vertices in the @xmath5-th graph . due to edge - transitivity , all edges of each edge - transitive graph have the same optimal transition probabilities . thus the laplacian operator @xmath218 for each one of the graphs can be written as @xmath219 in terms of its unweighted laplacian matrix @xmath220 of the @xmath4-th graph . using this relation the laplacian operator for the whole graph can be derived as below , @xmath221 we denote the eigenvalues of the @xmath4-th unweighted laplacian matrix @xmath222 in their sorted order by @xmath223 where @xmath224 varies from @xmath13 to @xmath225 . using this notation the eigenvalues of the laplacian operator of the whole graph can be written as below , @xmath226 where @xmath224 for @xmath227 varies from @xmath13 to @xmath225 . based on the derivation in ( [ eq : cartesianlaplacianeigenvalue ] ) and considering the fact that the first eigenvalue of each unweighted laplacian matrix @xmath222 is zero ( i.e. @xmath228 for @xmath229 ) , the second smallest eigenvalue of the laplacian operator of the whole graph can be written as @xmath230 . using this result the optimization problem for the fcga problem can be written as below , @xmath231 where @xmath232 and @xmath233 is the number of edges in the @xmath5-th edge - transitive graph . for the optimal answer we have @xmath234 from this relation we can conclude the following for the optimal value of the second largest eigenvalue of the gossip operator @xmath181 and the transition probabilities @xmath235 @xmath236 where @xmath237 . in ( [ eq : cartesianlpoptimalprobability ] ) , we have used the notation @xmath238 to refer to each vertex in the topology , where @xmath239 varies from @xmath13 to @xmath225 . using this notation , @xmath240 is the transition probability from vertex @xmath238 to vertex @xmath241 . as an example consider the cartesian product of two complete graphs each with @xmath242 and @xmath243 vertices . let @xmath244 and @xmath245 be defined over the edges of each one of the complete graphs , then for the optimal results we have [ eq : cartesiancompletegraphoptimalanswer ] @xmath246 an obvious example for the cartesian product of two complete graphs is the cartesian product of @xmath247 and @xmath248 , known as the prism graph . the optimal value of @xmath181 is @xmath249 and for the optimal transition probabilities we have @xmath250 and @xmath251 . a complete graph with @xmath2 vertices is a topology where all vertices are connected to each other . due to the symmetry of the topology all edges have the same transition probability @xmath252 , and its laplacian operator @xmath213 can be written as @xmath253 , where @xmath254 is a square matrix with all elements equal to one and @xmath255 can be written as @xmath256 in terms of the transition probability @xmath257 . the eigenvalues of the laplacian matrix for a complete graph are @xmath258 , @xmath259 . considering the constraint on the summation of the transition probabilities in the fcga problem ( [ eq : classicalgossipsdp ] ) , for the complete graph topology , the optimal value of the second largest eigenvalue of the gossip operator @xmath44 is @xmath260 and the transition probabilities is @xmath261 . in this topology @xmath2 vertices are connected in form of a cycle . a cycle graph with four vertices in depicted in figure [ fig : n4graphs](d ) . cycle topology is edge transitive and thus the optimal transition probability @xmath252 on all edges is the same . the optimal value of @xmath181 is @xmath262 and the optimal transition probability is @xmath263 . in this topology , all @xmath264 vertices of a circle are connected to one central vertex . this topology has @xmath265 vertices . due to symmetry of this topology the transition probabilities can be categorized in three groups . first and second groups are the probabilities on the edges connected to the central vertex . these probabilities are denoted by @xmath266 and @xmath267 . third group is the probability on the edges connecting the @xmath264 vertices in the circle to each other . this probability is denoted by @xmath268 . for @xmath269 , the optimal transition probabilities are @xmath270 , @xmath271 , @xmath272 , and the optimal value of the second largest eigenvalue of the gossip operator is @xmath273 note that the results above for @xmath269 , are obtained for both uniform and non - uniform clock distribution . for @xmath274 , the uniform clock distribution results in a optimal answer different than that of the non - uniform clock distribution where the result of obtained from uniform clock distribution is suboptimal . for @xmath274 and uniform clock distribution , the optimal transition probabilities are as below , @xmath275 and the optimal value of the second largest eigenvalue of the gossip operator is equal to @xmath276 for @xmath274 and non - uniform clock distribution , the optimal transition probabilities are as in ( [ eq : gossiptransitionprobability1174 ] ) , and the optimal value of the second largest eigenvalue of the gossip operator is equal to @xmath277 , and the clock distribution probabilities are @xmath278 , @xmath279 . in this subsection we address the optimization of the second largest eigenvalue of the fcga problem ( [ eq : classicalgossipsdp ] ) with the assumption that the clock distribution of vertices is uniform ( i.e. @xmath280 ) . the complete solution procedure is provided only for the symmetric star topology . for the rest of topologies we have provided the final results i.e. the optimal transition probabilities and the optimal value of the second largest eigenvalue of the gossip operator @xmath281 . in symmetric star topology @xmath264 path branches of length @xmath24 ( each with @xmath24 vertices ) are connected to a central vertex . a symmetric star graph with parameters @xmath282 and @xmath283 is depicted in figure [ fig : gossipfiguresdp](a ) . branches of length @xmath282 . ( b ) complete - cored symmetric star topology with @xmath283 branches of length @xmath282 . ( c ) the two coupled complete graphs topology with parameters @xmath284 , @xmath285 and @xmath286 . ( d ) a palm graph with parameters @xmath287 and @xmath282 . ( e ) the graph of ccs star graph with two types of branches with @xmath283 branches of length @xmath288 and @xmath289 . ( f ) the lollipop graph with parameters @xmath287 and @xmath282.,width=604 ] we model the symmeitric star topology by graph @xmath290 with vertex set @xmath291 and edge set @xmath292 . this graph has @xmath293 vertices in total . we associate with each vertex @xmath294 a column vector @xmath295 , for @xmath296 . @xmath297 and @xmath298 are column vectors with @xmath299 and @xmath300 elements , respectively , where their @xmath301-th and @xmath302-th elements are equal to one , respectively and the rest is zero . the gossip operator @xmath79for the symmetric star graph can be written as below , @xmath303 here @xmath304 is defined as @xmath305 and @xmath306 . note that @xmath307 is associated with the edges connecting vertex @xmath294 to vertex @xmath308 . due to symmetry of graph , the probabilities over edges with same distance from the central vertex are the same . therefore , we have only @xmath309 variables , namely @xmath310,for @xmath311 and @xmath312 for @xmath313 . note that @xmath314 , due to symmetry of graph and @xmath315 , since @xmath316-th vertex for @xmath317 is only connected to @xmath318-th vertex . due to the symmetry of symmetric star graph towards its central vertex , we can state that the automorphism group of symmetric star graph @xmath319 is isomorphic to permutation of branches . the orbits of @xmath319 acting on the vertices are @xmath320 , in order to apply stratification method , we define the new basis as below @xmath321 with @xmath322 . on new basis @xmath323 , the gossip operator @xmath79 converts into smaller block diagonal matrices , namely @xmath324 and @xmath325 as below , @xmath326 } , \end{gathered } \ ] ] @xmath327 } , \end{gathered}\ ] ] note that @xmath325 is a submatrix of @xmath324 . according to cauchy interlacing theorem @xcite , for the eigenvalues of @xmath328 and @xmath329 we have @xmath330 . from these relations we can conclude that the second largest eigenvalue of @xmath79 is included in @xmath328 and thus the original optimization problem reduces to the following minimization problem , @xmath331 @xmath212 is the number of edges ( starting from the central vertex ) which their transition probabilities have not reached the boundaries of feasible region . in other words they can take values between one and zero . the transition probabilities over these edges are @xmath332 and @xmath310 for @xmath333 . for the rest of the edges , the transition probabilities on the edges towards the center of graph are one , i.e. @xmath334 for @xmath335 , and the transition probabilities in the opposite direction are zero , i.e. @xmath336 for @xmath335 . note that we have @xmath337 variables in the optimization problem . the values of variables @xmath338 can be written as below , [ eq : gossip42 ] @xmath339 the problem parameters @xmath340 of the standard semidefinite programming ( * ? ? ? * appendix b ) are as following @xmath341 , \end{gathered } \ ] ] @xmath342 , \\ \text{for } \quad i=1,\ldots , m , \end{gathered } \ ] ] @xmath343 , \\ \text{for } \quad i=1,\ldots , m-1 , \end{gathered } \ ] ] @xmath344 , \quad \boldsymbol{f}_{0 } = \left [ \begin{array}{cc } { -\boldsymbol{i}_{k \times k } } & { 0 } \\ { 0 } & { \boldsymbol{i}_{m \times m } } \end{array } \right ] . \end{gathered } \ ] ] here @xmath345 for @xmath346 is a column vector with @xmath24 elements , where its @xmath4-th element is equal to one and the rest is zero . @xmath347 for @xmath227 is a column vector with @xmath212 elements , where its @xmath4-th element is equal to one and the rest is zero . @xmath348 is a vector with length @xmath349 with @xmath13 at index @xmath349 and zero elsewhere . minimization variable @xmath350 can be defined as @xmath351 @xmath352 @xmath353 , @xmath354 , @xmath355 , @xmath356 , @xmath354 , @xmath357 , @xmath358^t$ ] . for the dual problem we choose the dual variable @xmath359 as @xmath360 , where @xmath96 is a column vector defined as @xmath361^t $ ] . from the constraints of dual problem @xmath362=\boldsymbol{c}_{i})$ ] we obtain the following relations , @xmath363 , and @xmath364 , for @xmath365 , and @xmath366 , for @xmath365 , and @xmath367 and @xmath368 from these equations , we can conclude the following relations , @xmath369 from complementary slackness condition ( @xmath370 where @xmath371 and @xmath372 are the optimal primal dual feasible points ) we have @xmath373 . this in turn results in the following relations for optimal values of primal feasible point @xmath374 and dual feasible point @xmath375 , [ eq : gossip13 ] @xmath376 by substituting ( [ eq : gossip12 ] ) in ( [ eq : gossip13a ] ) to ( [ eq : gossip13d ] ) we get the following equations , [ eq : gossip14 ] @xmath377 defining @xmath378 , and @xmath379 , the equations ( [ eq : gossip14a ] ) to ( [ eq : gossip14c ] ) can be written as @xmath380 for @xmath227 , or equivalently , @xmath381 in order to satisfy ( [ eq : gossip111 ] ) either @xmath382 should be set to zero or @xmath383 @xmath384 @xmath385 , where the former is not acceptable , since setting @xmath386 for @xmath387 will result in all elements of the vector @xmath96 to be zero . thus the following can be concluded from ( [ eq : gossip111 ] ) @xmath388 substituting @xmath389 and ( [ eq : gossip16 ] ) in ( [ eq : gossip15 ] ) we get @xmath390 from equations ( [ eq : gossip15 ] ) and ( [ eq : gossip16 ] ) for @xmath391 , we have @xmath392 on the other hand , from ( [ eq : gossip17 ] ) for @xmath393 we have @xmath394 comparing the equations ( [ eq : gossip18 ] ) and ( [ eq : gossip19 ] ) we can conclude the following @xmath395 and thus for @xmath396 we have @xmath397 substituting ( [ eq : gossip21 ] ) in ( [ eq : gossip14d ] ) we obtain the following @xmath398 from equations ( [ eq : gossip13e ] ) and ( [ eq : gossip13f ] ) we obtain the following inductive equations [ eq : gossip23 ] @xmath399 where ( [ eq : gossip23a ] ) holds for @xmath400 . using the recursive equations ( [ eq : gossip23 ] ) we can calculate @xmath401 for @xmath402 in terms of @xmath403 as below , @xmath404 where @xmath405 is a polynomial of order @xmath4 in terms of x. by substituting @xmath406 and @xmath407 in ( [ eq : gossip22 ] ) , we get the following polynomial , @xmath408 we refer to this polynomial as the final polynomial . this is a polynomial of order @xmath409 in terms of @xmath410 . all roots of this polynomial are in interval @xmath411 where @xmath54 is obtained from the largest root @xmath412 of the final polynomial ( [ eq : gossip25 ] ) according to the formula @xmath413 . after finding the optimal value of @xmath410 and correspondingly @xmath54 , the optimal value of probabilities can be obtained by substituting the optimal value of @xmath410 in formulas ( [ eq : gossip16 ] ) , ( [ eq : gossip17 ] ) , ( [ eq : gossip20 ] ) . the optimal results presented above are obtained for a given @xmath212 . the correct value of @xmath212 is the largest value where the resultant optimal probabilities and the second largest eigenvalue @xmath414 are within the feasible region ( i.e. between zero and one ) . to find this value of @xmath212 , first one should set the value of @xmath212 to its minimum value i.e. zero and then calculate the optimal probabilities and @xmath54 . if the obtained values are acceptable and they are within the feasible region ( i.e. between zero and one ) then it should increase the value of @xmath212 and redo the calculation . this process continues until the smallest value of @xmath212 that results in optimal probabilities outside of feasible region is obtained . in algorithm [ mfindingalgorithm ] we have provided the pseudo code for this process . + + @xmath415 set @xmath212 equal to @xmath4 and calculate the optimal value of @xmath54 and probabilities using final polynomial ( [ eq : gossip25 ] ) and ( [ eq : gossip16 ] ) , ( [ eq : gossip17 ] ) , ( [ eq : gossip20 ] ) [ marker724 ] + increase the value of @xmath4 by one and + set the optimal value of @xmath212 to @xmath416 and exit for the rest of topologies presented in this section the main procedure for finding the optimal value of probabilities and the second largest eigenvalue @xmath414 is more or less the same as algorithm [ mfindingalgorithm ] . the main difference is on the final polynomial and the formulas determining the optimal probabilities in terms of the second largest eigenvalue @xmath414 or equivalently @xmath410 . in tables [ tab : gossiptabless1 ] and [ tab : gossiptabless2 ] we have provided the optimal value of @xmath212 and @xmath54 ( second largest eigenvalue ) for different symmetric star topologies . the results in these tables are presented in order to analyse the optimal value of @xmath212 and @xmath54 in terms of topological parameters of symmetric star . [ tab : gossiptabless1 ] .the optimal value of @xmath212 in terms of number @xmath417 and length @xmath418 of branches for a symmetric star topology . [ cols="^,^,^,^,^,^,^",options="header " , ] similar to symmetric star topology , in ccs star topology , increasing either the length or the number of branches degrades the convergence rate of the randomized gossip algorithm . in this topology , two complete graphs each with @xmath419 and @xmath420 vertices respectively , share @xmath243 vertices . in figure [ fig : gossipfiguresdp](c ) two coupled complete graphs with parameters @xmath284 , @xmath285 and @xmath286 is depicted . this topology has @xmath421 vertices . due to the symmetry of the complete graphs , the transition probabilities can be divided into seven groups . @xmath422 is the transition probability on edges connecting the @xmath242 vertices on the top complete graph to each other . @xmath423 is the transition probability on the edges connecting the @xmath242 vertices on the top complete graph to the @xmath243 vertices in the middle and @xmath424 is the transition probability on the same edges as in @xmath423 but with reverse direction . @xmath425 is the probability on edges connecting the @xmath243 vertices in the middle to each other . similarly the transition probabilities @xmath266 , @xmath267 and @xmath268 are defined for the transition probabilities on the edges of the complete graph on the down side of the topology . for the symmetric case where @xmath426 the optimal transition probabilities and @xmath181 are obtained for two following categories . for @xmath427 , the optimal probabilities are @xmath428 , @xmath429 , @xmath430 , @xmath431 , and the optimal value of the second largest eigenvalue of the gossip operator @xmath432 is @xmath433 . for @xmath434 , the transition probabilities are @xmath435 , @xmath429 , @xmath436 , and the optimal value of the second largest eigenvalue of the gossip operator @xmath432 is @xmath437 . it is apparent that for the last symmetric case where @xmath434 the whole topology reduces to a 3-partite graph . in this section we address the optimization of the fcga problem ( [ eq : classicalgossipsdp ] ) with the assumption that the clock distribution of vertices is non - uniform ( i.e. @xmath438 ) . the symmetric star topology with parameters @xmath264 and @xmath24 is introduced in subsection [ sec : symmetricstar ] . as explained in subsection [ sec : symmetricstar ] , all edges with the same distance from the central vertex have the same optimal transition probability . for non - uniform selection of vertices , the optimal value of the second largest eigenvalue of the gossip operator is @xmath439 and the optimal transition probabilities are @xmath440 , @xmath441 , @xmath442 , @xmath443 for @xmath444 . the optimal clock distribution probabilities of vertices are @xmath445 , @xmath446 for @xmath447 . for nonzero @xmath448 , the optimal clock distribution probability @xmath449 is @xmath450 and the rest of the optimal clock distribution probabilities are same as the case of @xmath445 . a special case of the symmetric star topology is the case of @xmath451 , where this topology reduces to a path topology with odd number of vertices ( i.e. @xmath452 ) . for this case , the optimal value of the second largest eigenvalue of the gossip operator is @xmath453 and the optimal transition probabilities are @xmath454 , @xmath441 , @xmath442 and @xmath443 for @xmath444 . the optimal clock distribution probabilities are @xmath455 and @xmath456 for @xmath447 . for nonzero @xmath457 , the optimal clock distribution probability @xmath449 is @xmath458 and the rest of the optimal clock distribution probabilities are same as the case of @xmath445 . the complete - cored symmetric ( ccs ) star topology with parameters @xmath459 is introduced in subsection [ sec : ccsoptimizationuniform ] . as explained in subsection [ sec : symmetricstar ] , all edges ( vertices ) with the same distance from the central core have the same optimal transition ( clock distribution ) probabilities . for non - uniform clock distribution , the optimal transition probabilities are @xmath460 , @xmath461 and @xmath462 for @xmath463 . the optimal clock distribution probabilities are @xmath464 and @xmath465 for @xmath466 . the optimal value of the second largest eigenvalue of the gossip operator is @xmath467 . a special case of the ccs star topology is the path topology with even number of vertices which is obtained for @xmath451 . for the path topology with @xmath468 vertices , the optimal clock distribution probabilities are @xmath469 and @xmath470 for @xmath447 and the second largest eigenvalue of the gossip operator is @xmath471 . the complete - cored symmetric ( ccs ) star with two types of branches is identified with parameters @xmath472 . this topology is a ccs star topology where two types of tails ( each with @xmath473 and @xmath474 edges ) are connected to each vertex in the complete core . a ccs star graph with two types of branches with parameters @xmath283 , @xmath288 and @xmath475 is depicted in figure [ fig : gossipfiguresdp](e ) . for non - uniform clock distribution , the optimal transition probabilities are @xmath476 , @xmath461 , @xmath477 for @xmath478 , and @xmath479 , @xmath480 for @xmath481 . the optimal second smallest largest eigenvalue of the gossip operator is @xmath482 with @xmath483 and @xmath484 . the optimal transition probabilities are @xmath485 , @xmath486 for @xmath487 and @xmath488 for @xmath489 . a palm topology with parameters @xmath459 consists of a path graph with @xmath24 vertices connected to the central vertex of a star graph with @xmath264 branches as shown in figure [ fig : gossipfiguresdp](d ) for parameters @xmath287 and @xmath282 . this topology has @xmath490 vertices and @xmath491 edges . due to the symmetry of the topology , all edges connected to the central vertex of the star graph have the same transition probabilities ( denoted by @xmath492 and @xmath423 ) except the edge connecting the path graph to the central vertex and the vertices on the star graph ( other than the central vertex ) have the same clock distribution probability . by assigning the transition probabilities as in @xmath461 , @xmath477 for @xmath447 and @xmath493 , @xmath494 for @xmath495 , the optimal answer varies , depending on the values of the parameters @xmath264 and @xmath24 . for @xmath496 the optimal value of the second largest eigenvalue of the gossip operator is @xmath497 , and the optimal clock distribution probabilities are @xmath498 , @xmath499 , for @xmath500 and @xmath501 , for @xmath447 . for @xmath502 the optimal value of the second largest eigenvalue of the gossip operator is @xmath503 , and the optimal clock distribution probabilities are @xmath498 , @xmath504 for @xmath505 and @xmath506 , for @xmath463 . this topology is obtained by connecting a path graph ( with @xmath24 vertices ) to one of the vertices in a complete graph with @xmath299 vertices . by bridging vertex , we refer to the vertex in complete graph that is connected to the path graph . considering the symmetry of the complete graph the transitive probabilities in the complete graph can be categorized into two groups . the first group is those connecting the vertices in the complete graph other than the bridging vertex . we denote the transitive probability on these edges by @xmath422 . the second group is the edges from the bridging vertex to other vertices in the complete graph and vice versa , where the transitive probability over these edges are denoted by @xmath424 and @xmath423 , respectively . the transitive probabilities on the edges of the tail are denoted by @xmath507 and @xmath508 for @xmath509 . the lollipop topology is depicted in figure [ fig : gossipfiguresdp](f ) along with the probabilities assigned to the edges . for the case that @xmath510 the optimal value of the transition probability @xmath422 is zero and the lollipop topology reduces to palm topology where the optimal answer for this topology is provided in section [ sec : palmgraph ] . for the case of @xmath511 , by selecting the transition probabilities as @xmath461 , @xmath477 for @xmath466 and @xmath512 , @xmath513 , @xmath514 , the optimal value of the second largest eigenvalue of the gossip operator is obtained as @xmath515 where @xmath516 is @xmath517 , and the optimal transition clock distribution probabilities are @xmath518 , @xmath519 , for @xmath520 and @xmath521 for @xmath522 . for every underlying topology , there is a set of clock distribution and transition probabilities for the global optimal solution of the gossip algorithm . a straightforward method to obtain these probabilities is based on the sdp formulation of the continuous time consensus algorithm @xcite equivalent to that of that of the gossip algorithm . the equivalent sdp formulation of gossip algorithm can be obtained from that of the continuous time consensus algorithm by setting the upper limit on the total amount of weights @xmath523 to @xmath524 and defining the clock distribution and transition probabilities @xmath525 in terms of the weights of the continuous time consensus algorithm as below , [ eq : detailedbalanced ] @xmath526 where @xmath527 is the set of neighbours of @xmath4-th vertex in the underlying graph of the network . it is obvious that @xmath528 . also for variable @xmath304 we have @xmath529 and since @xmath530 then @xmath531 . by choosing the clock distribution and transition probabilities as in ( [ eq : detailedbalanced ] ) , the following can be concluded for the second largest eigenvalue of the gossip operator @xmath44 and the second smallest eigenvalue of the laplacian matrix in the continuous - time consensus problem , @xmath532 for the probabilities ( [ eq : detailedbalanced ] ) , it can be shown that @xmath533 , therefore , the selection of the probabilities in ( [ eq : detailedbalanced ] ) is said to have the detailed balance property @xcite . this is due to the fact that for all pairs of states , @xmath534 ( the rate at which transitions occur from state @xmath4 to state @xmath5 ) balances @xmath535 ( the rate at which transitions occur from @xmath5 to state @xmath4 ) . detailed balance is intimately related to time reversibility in markov chain . consider a continuous - time markov chain @xmath536 with unique equilibrium distribution @xmath537 and transition probabilities @xmath538 . the continuous - time markov chain @xmath536 is said to satisfy detailed balance property if the following holds , @xmath539 in other words , the long - run average number of transitions from state k to state @xmath4 per time unit is equal to the long - run average number of transitions from state @xmath4 to state @xmath5 per time unit for all @xmath540 . to explain the time reversibility , consider the stationary markov chain @xmath536 , i.e. @xmath541 , for all @xmath50 . it is shown @xcite that the condition ( [ eq:3151 ] ) is satisfied if and only if the stationary markov chain @xmath536 has the property that for all @xmath542 and all @xmath543 , @xmath544 is distributed as @xmath545 for all @xmath546 . in other words , the process reversed in time has the same probabilistic structure as the original process when the process has reached statistical equilibrium . a markov process with this property is said to be time reversible . we have optimized the convergence rate of the gossip algorithm for both classical and quantum networks . in contrast to previous models of the gossip algorithm in literature , in this paper we have proposed a novel model of the gossip algorithm with non - uniform clock distribution . this is done to propose and optimize a model for gossip algorithm that can achieve the convergence rate of the optimal continuous - time consensus algorithm over the same network topology . by modeling the clock tick model of the gossip algorithm as a poisson process , we have explained how non - uniform clock distribution can be achieved by modifying the rate of the poison process according to the clock distribution probabilities . deriving the minimization problem for optimizing the asymptotic convergence rate of the proposed gossip algorithm , we have analytically addressed its corresponding semidefinite programming formulation for different topologies and obtained closed - form formulas for the probabilities and the convergence rate of the gossip algorithm for both cases of uniform and non - uniform clock distributions . based on the results provided , it can be concluded that the optimal results obtained for uniform clock distribution are suboptimal compared to those of the non - uniform one . since the number of variables in the semidefinite programming formulation of the problem is more than the number of equations , the optimal answer for non - uniform clock distribution is not unique . in other words , there is more than one set of transition and clock distribution probabilities that can achieve the fastest convergence rate . based on the optimal results of continuous - time consensus algorithm @xcite , a method for achieving one of the optimal results for the case of non - uniform clock distribution is proposed . the proposed method ensures that the transition and clock distribution probabilities satisfy detailed balance property . in the case of the gossip algorithm for quantum networks , by expanding the density matrix in terms of the generalized gell - mann matrices , we transform the evolution equation of the quantum gossip algorithm to the state update equation of the classical gossip algorithm . defining the gossip operator for quantum gossip algorithm , we have formulated the original optimization problem as a convex optimization problem , which can be solved using semidefinite and linear programming . w. ren and r. w. beard , _ distributed consensus in multi - vehicle cooperative control : theory and applications _ , communications and control engineering.1em plus 0.5em minus 0.4emlondon : springer , 2008 . [ online ] . available : http://opac.inria.fr/record=b1131049 j. tsitsiklis , `` problems in decentralized decision making and computation , '' _ ph.d . dissertation , dept . massachusetts inst . technol . , cambridge , ma , nov . _ , 1984 . j. tsitsiklis , d. bertsekas , and m. athans , `` distributed asynchronous deterministic and stochastic gradient optimization algorithms , '' _ ieee trans . 31 , no . 9 , pp . 803812 , 1986 . j. duchi , a. agarwal , and m. wainwright , `` dual averaging for distributed optimization : convergence analysis and network scaling , '' _ automatic control , ieee transactions on _ , vol . 57 , no . 3 , pp . 592606 , march 2012 . , `` convergence rate analysis of distributed gossip ( linear parameter ) estimation : fundamental limits and tradeoffs , '' _ selected topics in signal processing , ieee journal of _ , vol . 5 , no . 4 , pp . 674690 , 2011 . a. broadbent and a. tapp , `` can quantum mechanics help distributed computing ? '' _ sigact news _ , vol . 39 , no . 3 , pp . 6776 , sep . [ online ] . available : http://doi.acm.org/10.1145/1412700.1412717 h. buhrman and h. rohrig , `` distributed quantum computing , '' in _ international symposium on mathematical foundations of computer science ( mfcs ) , lncs 2747 _ , 2003 , pp . v. s. denchev and g. pandurangan , `` distributed quantum computing : a new frontier in distributed systems or science fiction ? '' _ sigact news _ , vol . 39 , no . 3 , pp . 7795 , sep . 2008 . [ online ] . available : http://doi.acm.org/10.1145/1412700.1412718 l. mazzarella , a. sarlette , and f. ticozzi , `` a new perspective on gossip iterations : from symmetrization to quantum consensus , '' in _ ieee 52nd annual conference on decision and control ( cdc ) _ , dec 2013 , pp . 250255 . g. shi , d. dong , i. petersen , and k. johansson , `` reaching a quantum consensus : master equations that generate symmetrization and synchronization , '' _ ieee transactions on automatic control _ , pp . 114 , 2015 .

finite eccentricities in mass - transferring eccentric binary systems can be explained by taking into account mass - loss and mass - transfer processes that often occur in these systems . these processes can be treated as perturbations to the general two - body problem . the time - evolution equations for the semi - major axis and the eccentricity derived from perturbative methods are in general phase - dependent . the osculating semi - major axis and eccentricity change over the orbital timescale and they are not easy to implement in binary evolution codes like mesa . however , the secular orbital element evolution equations can be simplified averaging over the rapidly varying true anomalies . in this paper , we derive the secular time - evolution equations for the semi - major axis and the eccentricity for various mass - loss / transfer processes using either the adiabatic approximation or the assumption of delta - function mass - loss / transfer at periastron . we begin with the cases of isotropic and anisotropic wind mass - loss . we continue with conservative and non - conservative non - isotropic mass ejection / accretion ( including rlof ) for both point - masses and extended bodies . we conclude with the case of phase - dependent mass accretion . comparison of the derived equations with similar work in the literature is included and explanation of the existing discrepancies is provided .

1603.06593

the evolution of stellar binaries can be affected by a variety of physical phenomena , including mass - loss and mass - transfer processes . such processes affect the orbital evolution of these systems and thus consist an important part of the binary evolution theory . in dosopoulou @xmath0 kalogera 2016 ( from now on paper i ) , we treated different cases of mass - loss / transfer as perturbations to the general two - body problem . in that paper , we described the general perturbation method that can be used to calculate the orbital evolution of the binary and using this method we derived the general phase - dependent time - evolution equations for the semi - major axis and the eccentricity . here , assuming that the binary is slowly evolving we orbit - average the phase - dependent time - evolution equations and derive the secular time - evolution equations for the semi - major axis and the eccentricity . the secular evolution equations are derived both in the adiabatic regime as well as under the assumption of delta - function perturbation . we consider two bodies of masses @xmath1 and @xmath2 in an eccentric binary system of total mass @xmath3 . we refer to these as the donor and the accretor , respectively . the donor @xmath1 is losing mass with a rate @xmath4 and the accretor is accreting mass with a rate @xmath5 so that the total mass - loss rate is given by @xmath6 . for non - conservative mass - transfer , only a fraction @xmath7 of the lost mass is accreted . this means that in this case we have @xmath8 , where @xmath9 . for the general case of non - conservative mass - transfer , we derived in paper i the following phase - dependent time - evolution equations of the osculating semi - major axis @xmath10 and eccentricity @xmath11 in the @xmath12 reference system [ see figure @xmath13 here and equations @xmath14 and @xmath15 in paper i ] @xmath16\\ \nonumber & -2ha\left [ \frac{e^{2}+2e\cos{\nu}+1}{1-e^{2}}\right]\\ & -(\zeta+1)\left [ \frac{e^{2}+2e\cos{\nu}+1}{1-e^{2}}\right]\frac{\dot{m}}{m}\label{agen}\\ \begin{split } \left(\frac{de}{dt}\right)_{non - con}&=\frac{(1-e^{2})^{1/2}}{na}\left[b_{r}\sin{\nu } \right . \\ & \left . + b_{\tau}\frac{2\cos{\nu}+e(1+\cos^{2}\nu)}{1+e\cos{\nu}}\right]\\ & -2h\left[e+\cos{\nu}\right]-(\zeta+1)\left[e+\cos{\nu}\right]\frac{\dot{m}}{m } \label{egen } \end{split}\end{aligned}\ ] ] where the subscripts @xmath17 and @xmath18 refer to the radial and tangential components respectively , @xmath19 is the true anomaly and @xmath20 is the fraction of the total orbital angular momentum removed from the system . the other parameters are defined by @xmath21 and @xmath22 @xmath23 in equations @xmath24 , @xmath25 and @xmath26 are the absolute velocities of the ejected and accreted mass respectively and @xmath27 and @xmath27 are the positions of the ejection and accretion points relative to the bodies centers - of - mass . furthermore , @xmath28 and @xmath29 are the orbital velocities of the two bodies with respect to the inertial frame and @xmath30 is the orbital angular frequency . in the case the bodies in the binary system can be treated as point - masses we can simplify equations @xmath31 and @xmath32 and rewrite them as [ see equations @xmath33 and @xmath34 in paper i ] @xmath35\\ & -(\zeta+1)\left [ \frac{e^{2}+2e\cos{\nu}+1}{1-e^{2}}\right]\frac{\dot{m}}{m}\label{agenpoint}\\ \begin{split } \left(\frac{de}{dt}\right)_{non - con}^{point}&=\frac{(1-e^{2})^{1/2}}{na}\left[c_{r}\sin{\nu } \right . \\ & \left . + c_{\tau}\frac{2\cos{\nu}+e(1+\cos^{2}\nu)}{1+e\cos{\nu}}\right]\\ & -(\zeta+1)\left[e+\cos{\nu}\right]\frac{\dot{m}}{m}.\label{egenpoint } \end{split}\end{aligned}\ ] ] where we make use of the definition @xmath36 and the subscripts @xmath17 and @xmath18 refer to the radial and tangential components , respectively . equations @xmath31 and @xmath32 are in general phase - dependent and indicate that the semi - major axis and the eccentricity change over orbital timescales undergoing periodic oscillations . this makes these equations difficult to implement in binary evolution codes like startrack @xcite , bse @xcite or mesa @xcite . this difficulty emerges from the fact that the semi - major axis and eccentricity change on an orbital timescale @xmath37 while in the stellar evolution codes mentioned before the time - step is limited by the thermal timescale @xmath38 , where @xmath39 . this means that these codes can not keep track of the evolutionary changes in the orbital elements using equations @xmath31 and @xmath32 . however , equations @xmath31 and @xmath32 can be simplified if they are orbit - averaged - a procedure that can smooth out the aforementioned oscillations . when the assumption of slowly evolving systems can be made , the orbit - averaged equations provide information about the secular evolution of the orbital elements and from now on we are going to refer to these equations as _ secular time - evolution equations_. the secular time - evolution equations can then be implemented in the stellar evolution codes mentioned before . in this paper , we perform the orbit - averaging process either under the adiabatic approximation or considering delta - function mass - loss / transfer at periastron . under either of these assumptions , the derived secular time - evolution equations for the semi - major axis and the eccentricity are simplified further and can be used to efficiently model interacting eccentric binary systems . this paper is organized as follows . in section 2 we present the orbit - averaging process and describe as well as comment on the adiabatic approximation and the assumption of delta - function perturbation . in section 3 we discuss briefly the importance of the reference frame choice and derive for two different reference frames the secular orbital element time - evolution equations both in the adiabatic regime and for delta - function perturbation under the additional assumption of a phase - independent perturbation . in section 4 we derive the secular time - evolution equations for the semi - major axis and the eccentricity both in the adiabatic regime and under the assumption of delta - function mass - loss / transfer at periastron for the case of isotropic wind mass - loss . in section 5 we apply the results of section 3 in the case of anisotropic wind mass - loss and discuss qualitatively the effect of some simple wind structures on the binary orbital evolution , including jets . in section 6 we study the non - isotropic ejection and accretion in a binary for the cases of both conservative and non - conservative mass - transfer ( including roche - lobe - overflow ( rlof ) ) assuming extended bodies . in section 7 we derive the secular orbital element time - evolution equations for some commonly used characteristic examples of mass - loss / transfer under the additional point - mass approximation . section 8 deals with the case of a phase - dependent mass accretion rate ( bondi - hoyle scenario ) and in section 9 we summarize our results . the time - evolving orbital elements of a perturbed two - body system undergo physical oscillations due to the periodicity of the orbit [ see equations @xmath31 and @xmath32 ] . these oscillations can be smoothed out adopting orbit - averaging techniques that provide information about the secular evolution of the system . following lagrange planetary equations ( which refer to osculating orbital elements ) for a general perturbation [ see equations @xmath41 in paper i ] the time - evolution equation for the true anomaly @xmath19 is given by @xmath42 where @xmath43 is the argument of periapsis and @xmath44 is the longitude of the ascending node . for a general perturbation , the osculating orbit is precessing relative to the unperturbed orbit [ see equations @xmath45 in paper i ] and the true anomaly time - evolution deviates from the first unperturbed term in equation @xmath46 . the additional terms come from the fact that , while time @xmath47 is measured from a fixed moment in time , @xmath19 is measured from the periapsis , which changes through both precessions @xmath48 and @xmath49 . for perturbations induced by mass - loss / transfer processes we study in this paper we proved in section 3 in paper i [ see equation @xmath50 in paper i ] that @xmath51 . assuming in addition that in first order approximation the precession of the argument of periapsis @xmath43 due to mass - loss / transfer is small ( i.e. , @xmath52 ) then to first order we can compute the secular evolution of the orbital elements neglecting the second and third term in equation @xmath46 and using instead the substitution @xmath53 when we integrate the equations @xmath54 and @xmath55 over @xmath19 from @xmath56 to @xmath57 . in order to incorporate higher - order effects in the perturbation equations we have to proceed one step further and include all the terms in equation @xmath46 when applying the orbit - averaging process . in this paper , we derive first - order secular time - evolution equations where the general orbit - averaging rule is defined by the orbit - averaging integral @xmath58 where we made use only of the first term in equation @xmath46 and the quantity to be orbit - averaged is depicted with the three dots and is generally phase - dependent . in most cases , this makes the above integral tedious and not easy to resolve into a simple analytical form . however , there are cases that lead to significant simplifications to the time - evolution equations for the secular orbital elements . two commonly used simplifying approximations are the _ adiabatic approximation _ and the assumption of a _ delta - function perturbation _ that we will describe in detail in the following two subsections . in the next sections , we orbit - average the various phase - dependent time - evolution equations derived in paper i under both of these approximations . to proceed with orbit - averaging equations @xmath31 and @xmath32 we must make some assumptions about the perturbation timescale . if the perturbation timescale is much longer compared to the orbital timescale , the natural oscillations of the orbital elements can be smoothed out assuming the perturbation is nearly constant along the orbit . keeping the perturbation an adiabatic variant " of the system , this assumption is called _ the adiabatic approximation _ and the regime where the adiabatic approximation is valid is known as the _ adiabatic regime_. the system leaves this regime when the perturbation timescale becomes comparable to the orbital period . at this point adiabaticity is broken and the orbit - averaging method no longer applies ( e.g. , * ? ? ? * ) . for mass - transferring binary systems the adiabatic regime is valid by definition when the mass - loss rates involved in a problem are constants or under a reasonable approximation when the body s orbital period is much smaller than its mass - loss timescale . mathematically , this means that the mass - loss is so slow that both the body s mass - loss rate and its mass can be assumed constant over one orbital period , i.e. , we can take factors like @xmath59 out of the averaging integral @xmath60 assuming they are constant along the orbit . under certain conditions , the effect of a perturbation on a two - body system can have its peak at specific orbital phases . in this case the perturbation force can be approximated like delta - function perturbation centered at specific points in the orbit . in the case of mass - loss / transfer in eccentric orbits , the evolution of the mass - transfer rate has a gaussian - like behavior , with a maximum value occurring at periastron . thus , it is reasonable to assume that mass - loss will mostly occur near periastron , since this is the point in an eccentric orbit where the component stars are closest to each other . because of this , it is logical to examine the case of delta - function mass - loss / transfer at periastron . in this case , we can assume a mass - loss rate @xmath61 averaged over @xmath57 such that @xmath62 where @xmath3 is the mass of the mass - losing / accreting star and @xmath63 the relevant mass - loss / accretion rate . the function @xmath64 is the delta - function as a function of the true anomaly @xmath19 centered at the periastron position where @xmath65 . the time - evolution equations for the osculating orbital elements in their vector form [ see equations @xmath66-@xmath67 in paper i ] are independent of the reference frame choice . decomposing the perturbing force into its components in different coordinate systems is helpful and leads to different sets of equations for the semi - major axis and the eccentricity in each system . in section 3 of paper i , we presented the form of the phase - dependent time - evolution equations for the osculating orbital elements in three different reference systems [ see figure 1 here and section 3 of paper i for a detailed description of the reference systems ] . , while hatch shaped angles refer to the inclined by angle @xmath68 orbital plane . in the latter , two reference systems are defined , @xmath69 and @xmath70 , rotated by angle @xmath19 relative to the former.,scaledwidth=50.0% ] if we assume that the perturbation @xmath71 is nearly constant ( i.e. , phase - independent ) along the orbit , then under the adiabatic approximation equations @xmath72 and @xmath73 of paper i orbit - averaged become @xmath74 while equations @xmath75 and @xmath76 of paper i yield @xmath77 equations @xmath78 and @xmath79 describe the secular evolution of the osculating semi - major axis and eccentricity for a phase - independent perturbation under the adiabatic approximation and in the reference system @xmath80 shown in figure 1 . equations @xmath81 and @xmath82 describe the same but in the reference system @xmath83 shown also in figure 1 . in the case of isotropic wind mass - loss , there is spherical symmetry and thus the quantity @xmath84 defined by equation @xmath85 is zero , i.e. , @xmath86 . since there is no mass - transferred we can set @xmath87 . we assume that the mass lost does not remove any extra angular momentum other than the one lost due to the total mass change , i.e. , @xmath88 . substituting @xmath89 into equations @xmath31 and @xmath32 and performing the averaging rule @xmath60 under the adiabatic approximation ( @xmath59 constant along the orbit ) leads to _ the secular time - evolution equations for the osculating semi - major axis and eccentricity in the adiabatic regime in the case of isotropic wind mass - loss _ ( e.g. , * ? ? ? * ) @xmath90 where @xmath3 is the total mass , @xmath63 is the mass - loss rate and the subscript @xmath91 stands for isotropic . equations @xmath92 and @xmath93 indicate that an isotropic wind mass - loss leads to an increase in the secular semi - major axis while keeping the secular eccentricity unchanged . in other words , under adiabatic isotropic wind mass - loss an initially circular orbit will remain on average circular while an initially eccentric remain will retain on average its initial eccentricity . we mention here that @xcite considered the case in which only one body is losing mass isotropically . following equation @xmath94 in their paper , they assume for this case a perturbation of the form @xmath95 instead of @xmath96 . however , the latter is the correct perturbation for the system in the case only one component is losing mass in an isotropic way [ see section 4 in paper i ] . the former perturbation refers to case of anisotropic ejection of mass with an absolute velocity of zero [ see section 6 here ] . this important difference leads to factor of @xmath97 as well as different denominator discrepancy in their equations @xmath98 , where @xmath99 . the semi - major axis secular evolution equation is different compared to our equation @xmath92 due to both the different denominator and the factor of @xmath97 while the eccentricity secular evolution remains in both zero . in paper i we used the generalized lagrange gauge @xmath100 [ see subsection 2.1.2 of paper i ] to derive the phase - dependent time - evolution equations for the _ contact orbital elements _ [ see equations @xmath101 and @xmath102 in paper i ] . if we orbit - average these equations under the adiabatic approximation we see that the secular evolution of these elements is zero which is different from what equations @xmath92 and @xmath93 indicate for the _ osculating _ semi - major axis and eccentricity . however , we remind here that unlike the osculating orbital elements the contact orbital elements have no physical meaning but are sometimes useful mathematical tools [ see see subsection 2.1.2 of paper i for a more detailed discussion ] . in other words , this means that the fact that the contact orbital elements do not evolve secularly has no physical interpretation . under the assumption of isotropic delta - function mass - loss at periastron , the orbit - averaging procedure @xmath60 on equations @xmath31 and @xmath32 ( after the substitution @xmath89 ) yields @xmath103 equations @xmath104 and @xmath105 indicate that isotropic wind mass - loss at periastron leads to a secular increases in the both the semi - major axis and eccentricity . we note here that throughout the whole process no assumption is made for low eccentricities , so the above equations are valid for any value of eccentricity . besides , equations @xmath104 and @xmath105 show that the circular case is equivalent to the adiabatic isotropic wind mass - loss case [ see equations @xmath92 and @xmath93 ] . this can be seen by plugging in @xmath107 to equations @xmath104 and @xmath105 . for completeness , we mention here that for a perturbing force linearly proportional to the orbital velocity , i.e. , @xmath108 ( e.g. , adiabatic isotropic wind mass - loss studied here ) the time - evolution of @xmath43 is @xmath109 which means that on average the change in @xmath43 is zero . in paper i , we derived the perturbing acceleration in the case of anisotropic wind mass - loss from the donor of mass @xmath110 which is given by @xmath111 . for the sake of simplicity , we choose the reference system @xmath112 [ see figure @xmath13 ] . substituting the aforementioned perturbing acceleration @xmath71 to the adiabatic equations @xmath78 and @xmath79 leads to the following _ secular time - evolution equations for the osculating semi - major axis and the eccentricity in the adiabatic regime for the case of anisotropic wind mass - loss _ @xmath114 where the components of the perturbing force @xmath115 in the inertial reference system @xmath116 [ see figure @xmath13 ] are given by @xmath117 and the transformation @xmath118 is done as follows using the matrix @xmath119 defined by equations @xmath120 in paper i @xmath121 it is obvious from equation @xmath122 that for a phase - independent perturbing force and under the adiabatic approximation , there is no secular evolution of the osculating semi - major axis due to phase - independent anisotropic wind mass - loss [ see also @xcite ] . on the other hand , according to equation @xmath123 a secular change will be induced to the eccentricity if @xmath124 . the validity of this relation depends on the form and structure of the anisotropic wind . in the next two subsections we discuss briefly and qualitatively the effect of different kind of anisotropic wind structures ( including jets ) on the secular eccentricity . equation @xmath125 indicates that the only case in which we can have @xmath126 is when @xmath127 . under these conditions the secular eccentricity will remain unchanged . we will discuss qualitatively different types of anisotropic wind mass - loss , separating the cases of latitudinal and longitudinal dependence . we assume that the mass - flux rate we can be decoupled in its space and time dependence as @xmath128 , where @xmath129 . for no longitudinal dependence we have @xmath130 . equations @xmath131 show that in this case @xmath132 always . 1 . the final form of @xmath133 depends on the functional form of the latitudinal dependence @xmath134 . for an even function @xmath134 we have @xmath135 . this means that if the wind has only latitudinal dependence and is _ symmetric about the equator _ , there is no additional secular effect on the eccentricity . any asymmetry in the northern to southern hemisphere wind can be described by the relation @xmath136 , where @xmath137 is a constant parameter that measures the asymmetry between the two hemispheres . this relation leads to @xmath138 since @xmath139 . based on equation @xmath123 we get @xmath140 which means that the secular eccentricity changes . the actual sign and type of the secular change in eccentricity depends on the form of the wind latitudinal dependence @xmath134 . in the case of only longitudinal dependence , @xmath141 and @xmath142 . under this assumption , there is no way to simultaneously have @xmath127 . more specifically , regardless of whether the mass is lost symmetrically from the western and eastern hemisphere or not , we have a secular change in the eccentricity that will depend on the form of the wind longitudinal dependence @xmath143 . for general longitudinal and latitudinal dependence the secular eccentricity changes , while the form of the wind space dependence expressed through the functions @xmath134 and @xmath143 determines the sign and type of the eccentricity secular time - evolution equation . in thin bipolar jets , the mass - loss flux rate does not depend on the longitude @xmath144 . based on what we showed in the previous subsection 5.1 in this case we have @xmath132 . furthermore , it is reasonable to assume that in thin bipolar jets both the mass - loss flux rate and the wind velocity are constant and uniform . if the bipolar jets are also symmetric about the equator of the donor then we have @xmath145 , where @xmath146 and @xmath147 . this leads to @xmath135 as well and the secular eccentricity remains constant . asymmetric thin bipolar jets introduce a secular eccentricity evolution proportional to the quantity @xmath148 . a jet oriented in any other random direction in @xmath149 , breaks the symmetry of the ejection and the secular eccentricity changes accordingly . in all the following sections we choose to decompose our perturbations in the reference system @xmath12 [ see figure 1 ] . when all the mass ejected from the donor is accreted by the companion , there is no change in the total mass and orbital angular momentum of the system . for the so - called conservative case , this leads to @xmath151 in equations @xmath31 and @xmath32 . _ the secular time - evolution equations @xmath31 and @xmath32 for the case of conservative mass - transfer _ ( i.e. , @xmath151 ) in the _ adiabatic regime _ for extended bodies lead to @xmath152 - 2 ha \end{split}\label{conss1}\\ \begin{split } \left\langle\frac{de}{dt}\right\rangle_{con}^{ad}&=\frac{(1-e^{2})^{1/2}}{na}\left [ \left\langle b_{r}\sin{\nu}\right\rangle \right . \\ & \left . + \left\langle b_{\tau}\frac{2\cos{\nu}+e(1+\cos^{2}\nu)}{1+e\cos{\nu}}\right\rangle\right ] \end{split}\label{conss2}\end{aligned}\ ] ] while for a _ delta - function mass - loss / transfer at periastron _ we have @xmath153 -2ha(1-e^{2})^{1/2 } \end{split}\label{conss3}\\ \begin{split } \left\langle\frac{de}{dt}\right\rangle_{con}^{\delta}&=\frac{(1-e^{2})^{1/2}}{na}\left[\left\langle b_{r}\sin{\nu}\right\rangle\right . \\ & \left . + \left\langle b_{\tau}\frac{2\cos{\nu}+e(1+\cos^{2}\nu)}{1+e\cos{\nu}}\right\rangle \right ] \\ & -2h(1-e^{2})^{1/2}(1-e ) . \end{split}\label{conss4}\end{aligned}\ ] ] we remind here that first terms in square brackets on the rhs of equations @xmath154 depend through the quantity @xmath84 defined in equation @xmath85 on the ejection and accretion points and associated velocities . the second terms on the rhs emerge from the mass - transferred from the donor to the accretor and are proportional through the quantity @xmath155 defined in equation @xmath156 to the mass - loss / transfer rate @xmath157 . in paper i we used the generalized lagrange gauge @xmath100 [ see section 2.1.2 of paper i ] to derive the phase - dependent time - evolution equations for the contact elements [ see equations @xmath158 and @xmath159 in paper i ] . either under the adiabatic approximation or the delta - function mass - loss / transfer at periastron , equations @xmath158 and @xmath159 in paper i give for the secular contact semi - major axis and eccentricity @xmath160 \end{split}\label{contact1}\\ \begin{split } \left\langle\frac{de_{con}}{dt}\right\rangle_{con}^{ad,\delta}&=\frac{(1-e^{2})^{1/2}}{na}\left[\left\langle b_{r}\sin{\nu}\right\rangle \right . \\ & \left . + \left\langle b_{\tau}\frac{2\cos{\nu}+e(1+\cos^{2}\nu)}{1+e\cos{\nu}}\right\rangle \right].\label{contact2 } \end{split}\end{aligned}\ ] ] from equations @xmath161 and @xmath162 we see that the secular evolution for contact and osculating eccentricity under the adiabatic approximation is the same while from @xmath163 and @xmath164 the secular evolution of contact and osculating semi - major axis is different . under delta - function mass - loss / transfer at periastron equations @xmath163 and @xmath165 as well as @xmath161 and @xmath166 show that both the secular evolution of contact and osculating eccentricity and semi - major axis are different . we remind here again that the contact orbital elements have no physical meaning but still remain useful mathematical tools . in the case of roche - lobe - overflow ( rlof ) the mass is lost from the donor through the lagrangian point @xmath167 . this means that for rlof conservative mass - transfer we have @xmath168 , where @xmath169 is the location of the lagrangian point at periastron with respect to the center of mass of the donor and @xmath170 is the radius vector of the accretor s surface with respect to its center of mass . @xcite assumed rlof conservative delta - function mass - loss / transfer at periastron so that @xmath171 . although not explicitly stated but implied by their equation @xmath172 they assumed so - called stationary particles which have zero absolute velocities , i.e. , @xmath85 @xmath173 . under this assumption equation @xmath85 becomes @xmath174 from equations @xmath165 and @xmath166 we see that for delta - function mass - loss / transfer since @xmath175 at periastron and both @xmath176 and @xmath170 are not phase - dependent quantities , the only term that depends on @xmath84 and survives in equations @xmath165 and @xmath166 is the one including the tangential component of @xmath84 i.e. , @xmath177 . substituting equations @xmath156 and @xmath178 into @xmath179 and @xmath180 and after some algebra we are led to the following _ secular time - evolution equations for the osculating semi - major axis and eccentricity in the case of conservative rlof _ @xmath181\\ & -2\dot{m}_{2}\left(\frac{1}{m_{2}}-\frac{1}{m_{1}}\right)a(1-e^{2})^{1/2 } \label{deltarloa}\\ \nonumber \left\langle\frac{de}{dt}\right\rangle_{con}^{\delta , rl}&= -\frac{\dot{m}_{2}}{m_{1}}(1-e^{2})^{1/2}\left [ q\frac{r_{a_{2}}}{a}\cos{\phi_{p}}+\frac{r_{a_{1}}}{a}\right]\\ & -2\dot{m}_{2}\left(\frac{1}{m_{2}}-\frac{1}{m_{1}}\right)(1-e^{2})^{1/2}(1-e)\label{deltarloe}\end{aligned}\ ] ] where @xmath182 is the mass ratio and @xmath183 is the angle between the unit vector @xmath184 and the vector from the center of mass of the accretor to the accretion point @xmath185 at periastron . equations @xmath186 and @xmath187 are the same as equations @xmath188 and @xmath189 of @xcite . @xcite compared the timescale for the evolution of the eccentricity due to tides and rlof and found that , in contrast to tides which always act to circularize the orbit , mass - transfer may either increase or decrease the eccentricity , over timescales ranging from a few myr to a hubble time . furthermore , the timescale over which mass - transfer acts to increase the eccentricity may be shorter than the tidal timescale which acts to decrease it . for completeness , we note here that equations @xmath186 and @xmath187 reduce to the point - mass for stationary particles equations @xmath190 and @xmath191 described in section 7.4 by setting @xmath192 . thus , the first two terms in the rhs of equations @xmath186 and @xmath187 emerge from the fact that we took into account the physical extent of the ejection and accretion point , which introduces additional perturbations in our problem . these perturbations can be seen by looking into terms dependent on @xmath193 in equation @xmath85 . when a fraction of the mass ejected from the donor is accreted by the companion and the residual mass is lost from the system , there is change in the total mass and orbital angular momentum of the system . for this so - called non - conservative case , @xmath195 and in the most general case @xmath196 too in equations @xmath31 and @xmath32 . we mention here that from now on in all cases with @xmath196 when we perform the orbit - average in the adiabatic regime we assume that the parameter @xmath20 is changing adiabatically so that we can take it out from the orbit - averaging integral @xmath60 . _ the secular time - evolution equations @xmath31 and @xmath32 for the case of non - conservative mass - transfer _ ( i.e. , @xmath151 ) in the _ adiabatic regime _ for extended bodies lead to @xmath197 - 2 ha \\&-(\zeta+1)\frac{\dot{m}}{m } \end{split}\label{cons1}\\ \begin{split } \left\langle\frac{de}{dt}\right\rangle_{non - con}^{ad}=&\frac{(1-e^{2})^{1/2}}{na}\left [ \left\langle b_{r}\sin{\nu}\right\rangle \right . . + \left\langle b_{\tau}\frac{2\cos{\nu}+e(1+\cos^{2}\nu)}{1+e\cos{\nu}}\right\rangle\right ] \end{split}\label{cons2}\end{aligned}\ ] ] while for _ delta - function mass - transfer at periastron _ we have @xmath198 -2ha(1-e^{2})^{1/2 } \\&-(\zeta+1)\frac{\dot{m}}{m}(1-e^{2})^{1/2 } \end{split}\label{cons3}\\ \begin{split } \left\langle\frac{de}{dt}\right\rangle_{non - con}^{\delta}=&\frac{(1-e^{2})^{1/2}}{na}\left[\left\langle b_{r}\sin{\nu}\right\rangle\right . \\ & \left . + \left\langle b_{\tau}\frac{2\cos{\nu}+e(1+\cos^{2}\nu)}{1+e\cos{\nu}}\right\rangle \right ] \\ & -2h(1-e^{2})^{1/2}(1-e ) \\&-(\zeta+1)\frac{\dot{m}}{m}(1-e^{2})^{1/2}(1-e ) . \end{split}\label{cons4}\end{aligned}\ ] ] comparing equations @xmath154 with @xmath199 we see an additional third term arising on the rhs of the above equations . this term emerges from the non - conservative nature of the mass - transfer which carries away from the system both mass and orbital angular momentum . the first component of this term is proportional to the parameter @xmath20 describing the loss in the system s orbital angular momentum while the second component describes the change in the total system mass and is proportional to total mass - loss rate @xmath63 . the additional effect induced by an extra angular momentum removal from the system can be seen in equations @xmath200 and @xmath201 or @xmath179 and @xmath180 . assuming an adiabatically changing @xmath196 this additional removal of angular momentum enhances the increase rate of the secular semi - major axis and has no effect in the secular eccentricity . within the same considerations the limiting case @xmath202 does not induce any secular change in eccentricity but speeds up to the maximum level possible from this effect the secular expansion of the orbit . in the case of delta - function mass - transfer at periastron @xmath196 amplifies the increase rate of the semi - major axis while depending on the system under study speeds up the increase or decrease of eccentricity . similarly to the adiabatic case , the limiting case @xmath202 maximizes the aforementioned effects . for non - conservative delta - function rlof we assume that only a fraction @xmath7 of the lost mass is accreted such that @xmath203 . following a similar procedure with the one described in subsection 6.1.1 , equations @xmath31 and @xmath32 give the following for the _ secular time - evolution equations for the osculating semi - major axis and eccentricity in the case of non - conservative rlof _ -2\frac{\dot{m}_{2}}{m_{1}}a(1-e^{2})^{1/2}\left[\gamma ( q-1 ) \right.\\ & \left . + ( 1-\gamma)\left(\zeta+\frac{1}{2}\right)\frac{q}{1+q}\right ] \end{split } \label{rlononcon1 } \\ \begin{split } \left\langle\frac{de}{dt}\right\rangle_{non - con}^{\delta , rl}&= -\frac{\dot{m}_{2}}{m_{1}}(1-e^{2})^{1/2}\left [ \gamma q\frac{r_{a_{2}}}{a}\cos{\phi_{p}}\right.\\ & \left.+\gamma \frac{r_{a_{1}}}{a}\right ] -2\frac{\dot{m}_{2}}{m_{1}}(1-e^{2})^{1/2}(1-e)\left[\gamma ( q-1 ) \right.\\ & \left . + ( 1-\gamma)\left(\zeta+\frac{1}{2}\right)\frac{q}{1+q}\right ] . \end{split}\label{rlononcon2}\end{aligned}\ ] ] equations @xmath205 and @xmath205 do not change qualitatively the results for the mass - transfer timescale and the mass - transfer effect on the secular semi - major axis and eccentricity we mentioned in subsection 6.1.1 for the case of conservative mass - transfer . these equations are also derived in @xcite . however , in that study it was assumed that mass is lost only from the accretor and not from the system as a whole as we assume here and is described in detail in section 6.2 in paper i. this difference leads to the factor @xmath206 in our equations instead of @xmath207 in equations @xmath66 in @xmath208 in @xcite . the same reason also leads to the missing @xmath7 factor in the first term in the rhs of equation @xmath66 and the second term in the rhs of equation @xmath208 in @xcite . for the case of both binary components treated as point - masses , we have derived the form of the perturbing force [ see equations @xmath209-@xmath210 in paper i ] as well as the phase - dependent time - evolution equations @xmath211 and @xmath212 for the semi - major axis and the eccentricity , respectively . in the following sections , we apply these equations in some commonly used characteristic examples of mass - loss and mass - transfer under the point - mass approximation . furthermore , we compare our equations with equations derived other studies that have considered applications similar to the ones presented here . we assume that the mass ejected from the donor is always in the radial direction pointing towards the accretor while none of this matter is accreted . in this case , we have @xmath213 and @xmath214 . the quantity @xmath215 defined by equation @xmath216 then becomes @xmath217 from which in this case we have @xmath218 and @xmath219 in equations @xmath211 and @xmath212 . + we introduce a parameter @xmath220 which is a measure of the ejection velocity as a fraction of the orbital velocity at periastron and is defined by @xmath221 the _ secular point - mass time - evolution equations @xmath211 and @xmath212 for the semi - major axis and the eccentricity in the adiabatic regime for radial ejection with no accretion _ then become @xmath222 from equations @xmath223 and @xmath224 we see that a radial ejection from the donor with no accretion is equivalent to the case of isotropic wind mass - loss , which was studied in section 4 and is governed by equations @xmath92 and @xmath225 . equations @xmath223 and @xmath224 are consistent with the work of @xcite . in this case the mass is ejected in the tangential direction while the assumption of no accretion remains the same as in the section 7.1 . this means that @xmath213 and @xmath226 . in the last relation , for @xmath227 , the plus sign refers to ejection in the direction opposite to the direction of motion of the body . we can then write the quantity @xmath215 defined in equation @xmath216 as @xmath228 from which in this case we have @xmath229 and @xmath230 in equations @xmath211 and @xmath212 . we introduce here the parameter @xmath231 defined in the same way as in equation @xmath232 by @xmath233 the _ secular point - mass time - evolution equations @xmath211 and @xmath212 for the semi - major axis and the eccentricity in the adiabatic regime for tangential ejection with no accretion _ then become @xmath234 where we remind here that the plus sign refers to ejection in the direction of motion of the body ( front side ) and the minus sign to the opposite ( rear side ) . equation @xmath235 seems to have a singularity for circular orbits ( @xmath107 ) . however , as we describe below equation @xmath235 reduces to simpler forms with no singularity for either rear and frontal ejection . specifically , for rear ejection , tangential but opposite to the orbital motion , we have the minus sign and equations @xmath236 and @xmath235 simplify to @xmath237 equations @xmath238 and @xmath239 indicate that _ an ejection of mass opposite to the motion of the body direction leads to an increase in the secular semi - major axis and a decrease in the secular eccentricity of the orbit . _ for frontal ejection , tangential but along the orbital motion , we have the opposite result since in this case @xmath240 which shows that _ an ejection in the same to the motion of the body direction leads to a secular semi - major axis decrease and a secular eccentricity increase_. equations @xmath241 for the circular case of @xmath107 predict a zero secular evolution for the eccentricity which means that an initially - circular orbit will remain circular under the assumption of only tangential adiabatic mass - loss with no accretion in the case of point masses . we note here that both quantitatively and qualitatively we agree with @xcite , who used a different formalism to derive equations similar to our equations @xmath241 . it is interesting to see at this point what happens in the rear ejection case for delta - function mass - loss at periastron . in this case we have @xmath242 which depicts that _ both the secular semi - major axis and eccentricity increase for rear delta - function mass - loss at periastron_. if the same ejection is to take place at apastron , one can prove that this will lead to an increase in the secular semi - major axis and a decrease to the secular eccentricity . since the case of frontal ejection differs only by a sign compared to the rear ejection case , _ frontal delta - function mass - loss at periastron will lead to a semi - major axis and eccentricity secular decrease _ , while delta - function mass - loss at apastron will decrease the secular semi - major axis and increase the secular eccentricity . we have shown in section 7.1 that a radial ejection of mass is in the adiabatic regime secularly equivalent to the isotropic wind mass - loss case studied in section 4 . following the same logic the same is true for any radially accreted matter . for our case here we assume the donor is losing mass either isotropically in a wind or with a radial ejection pointing always towards the accretor . part of this mass is accreted radially to the companion and the rest of it is tangentially lost from it with a constant absolute velocity @xmath243 . what introduces perturbing forces in the scenario described above is the tangential mass - loss and we assume for it to always happen with a velocity opposite to the tangential motion of the accreting body . we note here that under these assumptions , the relative velocity of the tangentially lost mass to the center of mass of the companion is phase - dependent . based on the above assumptions , we have @xmath244 and @xmath245 . because the quantity @xmath215 defined by equation @xmath216 is now phase - dependent , the tangential wind mass - loss introduces an effective @xmath246 . we note here that as we mentioned in section 7.1 a radial ejection is in adiabatic regime secularly equivalent to the isotropic wind mass - loss and does not introduce any reaction force . hence , the perturbing force defined by equation @xmath209 in paper i , is only due to the tangential mass - loss and can be written as @xmath247 from equation @xmath248 we have for the value of @xmath215 and @xmath249 @xmath250 based on equations @xmath251 and @xmath252 we have in equations @xmath211 and @xmath212 that @xmath229 , @xmath253 and @xmath246 . similarly to the previous subsections , we introduce here the parameter @xmath254 such that @xmath255 we find that apart from the known secular equations for isotropic wind mass - loss / radial ejection in the adiabatic regime @xmath256 the secular time - evolution equations @xmath211 and @xmath212 in the case studied here become @xmath257 where @xmath258 . in this case , equations @xmath259 and @xmath260 indicate that the secular semi - major axis decreases while the secular eccentricity increases . we note here that equations @xmath261 and @xmath262 are included in equations @xmath259 and @xmath260 . we mention here that we agree with the results by . in that study based on the work of @xcite made some corrections and re - derived equations @xmath259 and @xmath260 . derived similar equations to our equations @xmath259 and @xmath260 by making specific assumptions for the radial and tangential components of the wind velocity when the latter is accreted . as it can be seen from the general equations @xmath263 and @xmath32 , the final expressions for the time - evolution of @xmath10 and @xmath11 depend strongly on the specific assumptions made about wind velocity when it is ejected or is accreted thus , comparison between the time - evolution equations for the semi - major axis and eccentricity between previous work is strongly driven by the different assumptions made separately in each of these works . in many previous work , the ejected or accreted mass particles were assumed to not have an absolute velocity on their own , . this assumption of the so - called _ stationary point particles _ leads to the additional relation @xmath173 , so that from equation @xmath85 we have for this case @xmath86 . under these assumptions , equations @xmath31 and @xmath32 lead to the following _ secular time - evolution equations for the semi - major axis and eccentricity in the adiabatic regime for stationary point particles and conservative mass - transfer _ @xmath264 while for _ delta - function conservative mass - transfer at periastron _ we have @xmath265 equations @xmath266 show that , in the case of stationary particles , the effect of mass - transfer on the secular semi - major axis and eccentricity depends on the mass ratio @xmath267 . in the adiabatic regime , the secular eccentricity does not change while the secular semi - major axis decreases for @xmath268 and increases for @xmath269 . in the case of delta - function mass - transfer at periastron , both the secular semi - major axis and the eccentricity decrease for @xmath268 and increase for @xmath269 . we mention here that the same perturbation for conservative mass - transfer and stationary particles , which is equal to @xmath270 is derived in @xcite [ see equation @xmath156 in here and equation @xmath271 in page @xmath272 in @xcite ] . however , for the case of delta - function mass - transfer at periastron @xcite derives the following secular time - evolution equation for the eccentricity @xmath273 which has clearly a very different functional dependence on @xmath11 compared to our equation @xmath191 . formula @xmath274 has also been used in previous work , . the reason for this discrepancy is that in these studies there is an error in the computation of the orbit - averaging integral @xmath60 for a delta - function conservative mass - transfer in the case of stationary particles . in more detail made a substitution @xmath275 ( @xmath276 at periastron ) in equation @xmath32 instead of performing the complete averaging integral @xmath60 with delta - function mass - transfer at periastron . based on the bondi - hoyle accretion scenario @xcite the accretion rate from a stellar wind can be written as @xmath277 where @xmath278 is the bondi - hoyle accretion efficiency , @xmath279 is the wind sound speed and @xmath280 is the is the relative velocity of the wind with respect to the accreting star . an assumption commonly being made ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) is that the density of the wind @xmath281 in the vicinity of the accretor follows the steady spherically symmetric profile @xmath282 , where @xmath4 is the wind - mass loss rate and @xmath283 is the wind velocity magnitude . assuming that the wind flow is supersonic ( @xmath284 ) the accretion rate @xmath5 for a radial wind ( with respect to the orbit ) of speed @xmath285 is given by @xmath286 where the efficiency factor for bondi - hoyle accretion @xmath278 is in the range @xmath287 in the bondi - hoyle model ( e.g. , * ? ? ? * ; * ? ? ? * ) , although it may be as low as @xmath288 in some specific cases ( e.g. , * ? ? ? * ; * ? ? ? for fast wind we have @xmath289 and equation @xmath290 reduces to @xmath291 where all the constants have been summed up in @xmath292 to indicate dimensionality and @xmath292 is defined by @xmath293 for slow wind we have @xmath294 and equation @xmath290 reduces to @xmath295 where again @xmath296 is a constant indicating dimensionality defined by @xmath297 . both cases are non - conservative mass - transfer since only the fraction given by equation @xmath290 is accreted by the companion . this fraction is phase dependent and in the case of fast wind is given by @xmath298 , which is defined as @xmath299 and is given by @xmath300 we then have @xmath301 . assuming point stationary masses ( i.e. , @xmath86 ) and no extra angular momentum removal ( i.e. , @xmath88 ) , the total perturbing force for the phase - dependent partial fast wind accretion can be written as @xmath302 where the subscript @xmath303 stands for _ fast wind _ and for comparison we kept the same definition as in @xcite @xmath304 equations @xmath31 and @xmath32 then give for the _ secular semi - major axis and eccentricity evolution due fast wind bondi - hoyle ( bh ) partial accretion in the adiabatic regime _ the following equations @xmath305 where we defined for comparison with @xcite the quantity @xmath61 as @xmath306 we mention here that in page @xmath307 and equation @xmath308 in @xcite , there is an error and the factor @xmath309 in the denominator should be @xmath310 as in our equation @xmath311 . after some algebra we can see that the sign of the right - hand - side of equation @xmath312 is determined by the sign of the quantity @xmath313 with a single zero solution in the range @xmath314 for @xmath315 . for @xmath316 the right - hand - side of equations @xmath317 and @xmath312 is positive and thus for a mass - ratio @xmath267 in the range @xmath316 , both the secular semi - major axis and eccentricity increase . for a mass - ratio @xmath318 the right - hand - side of equation @xmath312 is negative and the secular eccentricity always decreases while the sign of the right - hand - side of equation @xmath317 can not be determined since it involves many and different parameters this means that for a mass - ratio @xmath318 the semi - major secular evolution depends on many parameters including the orbital elements , the bondi - hoyle accretion efficiency parameter @xmath278 , the masses and the fast wind absolute velocity @xmath319 . the general case of slow wind is described by equation @xmath320 and leads to a elliptical integral that demands numerical integration . in the case that the wind velocity @xmath283 is of the same order like the orbital velocity @xmath321 , equation @xmath290 must be employed . this case needs numerical manipulation . the same is true for the any non - radial ( with respect to the orbit ) wind . however , assuming that the orbital velocity @xmath322 is always perpendicular to the radial wind which is always true in the circular case ( bondi - hoyle circular approximation ) , then equation @xmath290 simplifies to @xmath323\dot{m}_{1}\label{circbondi}\ ] ] where we have assumed a constant velocity wind , a bondi - hoyle factor @xmath278 and we defined @xmath324 where the subscript @xmath325 is for reminding the circular approximation made here . under these approximations equation @xmath326 reduces to an equation similar to the case of fast wind equation @xmath327 but with @xmath292 defined by @xmath328\label{r0}\ ] ] using @xmath292 defined in equation @xmath329 equations @xmath330 can be used accordingly to describe the bondi - hoyle accretion under the circular approximation . , @xcite , @xcite and made use of equation @xmath326 but they first substituted @xmath331 by its average @xmath332 $ ] . with this approach , they forced the mass - transfer rate to be phase - independent , so performing the averaging integral @xmath60 in equations @xmath31 and @xmath32 is oversimplified in their analyses . the right treatment involves orbit - averaging the full expression including the phase - dependent quantity @xmath298 given by equation @xmath333 . moreover , in equation @xmath334 in the last terms in the rhs of equations @xmath31 and @xmath32 are missing . these terms come from the change in total mass and orbital angular momentum of the system and can not be neglected . in this paper , we discussed the effect of various and different types of mass - loss and mass - transfer processes on the orbital evolution of eccentric binary systems . we made use of the general formulation and the phase - dependent time - evolution equations of orbital elements presented in paper i. based on them , we derived the secular time - evolution equations for the semi - major axis and the eccentricity under either the adiabatic approximation or the assumption of delta - function mass - loss / transfer at periastron . we first discussed the cases of isotropic and anisotropic wind mass - loss . we then studied the non - isotropic ejection and accretion in a binary ( including rlof ) in the conservative and non - conservative cases and for both point - masses and extended bodies . we presented specific examples of mass - loss / transfer under the point - mass approximation and compared them with similar work in the literature . we concluded discussing the case of phase - dependent mass accretion under the bondi - hoyle scenario . + + the main results of the paper are summarized below : * we studied the case of isotropic wind mass - loss . we proved that under the adiabatic approximation the secular osculating semi - major axis increases , while the secular osculating eccentricity remains constant . on the other hand , for delta - function isotropic mass - loss at periastron there is an increase in both the secular osculating semi - major axis and eccentricity . * we studied the case of anisotropic wind mass - loss . we showed that when the wind mass - loss is phase - independent the secular osculating semi - major axis is not affected . for wind mass - loss that is longitudinal independent and symmetrical about the equator , the secular osculating eccentricity remains constant . however , any asymmetry about the equator or any existing longitudinal dependence introduces changes in the secular eccentricity that depend on the specific spatial structure of the wind on both latitude and longitude . symmetric thin bipolar jets leave the secular eccentricity unchanged while any other asymmetrical or randomly directed jets induce secular eccentricity evolution . * we continued with the non - isotropic ejection and accretion in the case of point - masses . in this case : 1 . a radial ejection from the donor with no accretion is secularly equivalent to the isotropic wind mass - loss case mentioned above . 2 . an ejection of mass in the opposite to the motion of the body direction leads to an increase in the secular semi - major axis and a decrease in the secular eccentricity of the orbit . an ejection in the same to the motion of the body direction leads to a semi - major axis secular decrease and an eccentricity secular increase . 3 . for rear ejection and delta - function mass - transfer at periastron , both the secular semi - major axis and eccentricity increase . if the case of apastron , this will lead to an increase in the secular semi - major axis and a decrease to the secular eccentricity . for frontal ejection , delta - function mass - transfer at periastron will lead to a secular semi - major axis and eccentricity decrease , while delta - function mass - transfer at apastron will decrease the secular semi - major axis and increase the secular eccentricity . 4 . in the case of isotropic / radial wind mass - loss from the donor and radial partial wind accretion and tangential mass - loss from the accretor , the secular semi - major axis decreases while the secular eccentricity increases . * for the case of conservative or non - conservative rlof , in contrast to tides which always act to circularize the orbit , mass - transfer may either increase or decrease the secular eccentricity , over timescales ranging from a few myr to a hubble time . * for stationary particles , the secular semi - major axis and eccentricity time - evolution depends on the mass ratio @xmath182 . for adiabatic mass - loss rate , the secular eccentricity does not change while the secular semi - major axis decreases for @xmath268 and increases for @xmath269 . in the case of delta - function mass - loss at periastron , both the secular semi - major axis and eccentricity decrease for @xmath268 and increase for @xmath269 . * for phase - dependent mass - transfer assuming point and stationary particles , applying the general bondi - hoyle accretion scenario requires numerical integration . however , the cases of fast wind or radial wind under the circular approximation are analytically integrable . for @xmath316 both the secular semi - major axis and eccentricity increase . for a mass - ratio @xmath318 the secular eccentricity always decreases while the semi - major secular evolution depends on many parameters including the orbital elements , the bondi - hoyle accretion efficiency parameter @xmath278 , the masses and the fast wind absolute velocity @xmath319 . the secular time - evolution equations for the semi - major axis and the eccentricity we presented here for various and different cases of mass - loss / transfer could be used to model interacting eccentric binaries in star and binary evolution codes like startrack @xcite , bse @xcite or mesa @xcite . in a future paper , we will attempt to relax the assumption of tidal locking in the case of extended bodies in the binary and study the effect of mass - loss and mass - transfer on the time - evolution of the binary orbital elements as well as the binary components spins . assuming strong coupling between the spin and orbital angular momentum , we will derive secular time - evolution equations for the orbital elements and spins for the specific examples of mass - loss and mass - transfer studied in this paper including new applications . we want to thank our colleagues dimitri vears , jeremy sepinsky and lorenzo iorio who provided insight and expertise that greatly assisted the research . we are thankful to their comments that greatly improved the manuscript during its composition and revising stages . belczynski , k. , kalogera , v. , rasio , f. a. , et al . 2008 , , 174 , 223 - 260 bondi , h. , & hoyle , f. 1944 , , 104 , 273 boffin , h. m. j. , & jorissen , a. 1988 , , 205 , 155 bonai marinovi , a. a. , glebbeek , e. , & pols , o. r. 2008 , , 480 , 797 davis , p. j. , siess , l. , & deschamps , r. 2013 , , 556 , a4 duncan , m. j. , & lissauer , j. j. 1998 , , 134 , 303 eggleton , p. 2006 , evolutionary processes in binary and multiple stars , by peter eggleton , pp . . isbn 0521855578 . cambridge , uk : cambridge university press , 2006 . , gurfil , p. , & belyanin , s. 2008 , advances in space research , 42 , 1313 hadjidemetriou , j. d. 1963 , , 2 , 440 huang , s. s. 1956 , , 61 , 49 hurley , j. r. , tout , c. a. , & pols , o. r. 2002 , , 329 , 897 jura , m. 2008 , , 135 , 1785 karakas , a. i. , tout , c. a. , & lattanzio , j. c. 2000 , , 316 , 689 lajoie , c .- p . , & sills , a. 2011 , , 726 , 67 passy , j .- c . , mac low , m .- , & de marco , o. 2012 , , 759 , l30 paxton , b. , bildsten , l. , dotter , a. , et al . 2011 , , 192 , 3 paxton , b. , cantiello , m. , arras , p. , et al . 2013 , , 208 , 4 paxton , b. , marchant , p. , schwab , j. , et al . 2015 , , 220 , 15 shima , e. , matsuda , t. , takeda , h. , & sawada , k. 1985 , , 217 , 367 sepinsky , j. f. , willems , b. , kalogera , v. , & rasio , f. a. 2007 , , 667 , 1170 sepinsky , j. f. , willems , b. , kalogera , v. , & rasio , f. a. 2009 , , 702 , 1387 soker , n. 2000 , , 357 , 557 spiegel , d. s. , & madhusudhan , n. 2012 , , 756 , 132 theuns , t. , boffin , h. m. j. , & jorissen , a. 1996 , , 280 , 1264 veras , d. , wyatt , m. c. , veras , d. , hadjidemetriou , j. d. , & tout , c. a. 2013 , , 435 , 2416 veras , d. 2016 , royal society open science , 3 , 150571 villaver , e. , & livio , m. 2009 , , 705 , l81 vos , j. , stensen , r. h. , marchant , p. , & van winckel , h. 2015 , , 579 , a49

we have performed _ hst _ imaging of a sample of 23 high - redshift ( ) active galactic nuclei , drawn from the combo-17survey . the sample contains moderately luminous quasars ( ) . the data are part of the gems imaging survey that provides high resolution optical images obtained with the advanced camera for surveys in two bands ( f606w and f850lp ) , sampling the rest - frame uv flux of the targets . to deblend the agn images into nuclear and resolved ( host galaxy ) components we use a psf subtraction technique that is strictly conservative with respect to the flux of the host galaxy . we resolve the host galaxies in both filter bands in 9 of the 23 agn , whereas the remaining 14 objects are considered non - detections , with upper limits of less than 5 % of the nuclear flux . however , when we coadd the unresolved agn images into a single high signal - to - noise composite image we find again an unambiguously resolved host galaxy . the recovered host galaxies have apparent magnitudes of and with rest - frame uv colours in the range . the rest - frame absolute magnitudes at 200 nm are . the photometric properties of the composite host are consistent with the individual resolved host galaxies . we find that the uv colors of all host galaxies are substantially bluer than expected from an old population of stars with formation redshift , independent of the assumed metallicities . these uv colours and luminosities range up to the values found for lyman - break galaxies ( lbgs ) at . our results suggest either a recent starburst , of e.g. a few per cent of the total stellar mass and 100 myrs before observation , with mass - fraction and age strongly degenerate , or the possibility that the detected uv emission may be due to young stars forming continuously . for the latter case we estimate star formation rates of typically ( uncorrected for internal dust attenuation ) , which again lies in the range of rates implied from the uv flux of lbgs . our results agree with the recent discovery of enhanced blue stellar light in agn hosts at lower redshifts .

astro-ph0403462

around redshifts of @xmath103 , luminous quasars were orders of magnitude more numerous than today . although the physics of how active galactic nuclei evolve is still not understood , several links between galaxy and quasar evolution have emerged over recent years . the observational confirmation of supermassive black holes in the nuclei of all galaxies with a substantial bulge component ( e.g. * ? ? ? * ) makes every such galaxy a potential agn host . the strong evolution of the agn space density could therefore be related to the availability of accretion fuel in the host galaxies , or to the frequency of agn triggering events . gravitational interaction and major or minor merging of galaxies have long been suggested as important factors in driving nuclear activity in galaxies . confirming any of these as the dominant process has proved difficult , mainly because the morphological characteristics found for relatively nearby agn host galaxies are so diverse . furthermore , the properties of the hosts in the ` heyday ' of quasars ( @xmath11 ) are still elusive , a consequence of the substantial observational difficulties . the contrast between the bright nuclear point sources and the surrounding galaxy increases dramatically beyond @xmath12 as a result from both surface brightness dimming and waveband shifts towards the rest frame uv . the last years have seen numerous attempts to resolve the host galaxies of high - redshift quasars . owing to the observational challenges of detecting distant host galaxies the observational effort for each object is large , and the observed samples have consequently been very small , of the order of @xmath13 per target group . while radio - loud quasars appear to be very extended and have been resolved out to @xmath14 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , this is not the case for the large majority of radio - quiet quasars . at high redshifts two constraints dominate observational studies of host galaxies : on one side , very good seeing conditions are required to maximize the spatial contrast of the compact nuclear source compared to the extended host galaxy . on the other , large telescope apertures are preferrential to trace faint quasar hosts to as far away from the nucleus as possible . thus significant progress had to wait for 8m - class telescopes at very good sites with active optics systems with a very high light collecting power but atmospheric seeing limitations and for the _ hst _ and its high space - based sensitivity , combined with unprecedented spatial resolution , but limited size that might miss light from faint outer structures of the hosts . some host galaxies of radio - quiet quasars at @xmath15 have now been resolved both in the near infrared @xcite and in the optical domains @xcite , showing these objects to be moderately luminous , corresponding to present - day @xmath16 or slightly brighter . however , host galaxy colours have been unavailable , precluding estimates of the mass - to - light ratio ( @xmath17 ) . thus , without colours the observed luminosities , and their evolution with redshift , can not be mapped to the mass evolution if young stars contribute a major fraction of agn host s light . this is important as several high - luminosity quasars at @xmath11 appear to be located in very uv - luminous host galaxies @xcite . also , at low redshifts there is a link between nuclear activity and enhanced global star formation in the host galaxies . @xcite reported that sdss spectra of local seyfert 2 galaxies show a significant contribution from young stellar populations , and that this trend is strongly correlated with nuclear luminosity . in a multicolour study of low-@xmath18 qso hosts @xcite as well as at intermediate redshifts ( * ? ? ? * see below ) we found that hosts of elliptical morphology can be significantly bluer than the bulk of inactive ellipticals . these results indicate that in the recent past the star formation activity in galaxies hosting an agn may be different from normal galaxies . the details are far from understood . clearly more information is required to investigate the relation of starformation and agn activity , their common cause or causal order and the evolution of these properties with redshift . the new generation of wide - field imaging mosaics obtained with the hubble space telescope ( _ hst _ ) , especially in the conjunction with deep agn surveys , has opened a new observational avenue towards agn host galaxy studies . here we present first results on agn within the gems project @xcite , the largest _ hst _ colour mosaic to date . in the present paper we investigate the presence of rest - frame ultraviolet light in a substantial sample of @xmath19 agn , all with nuclear luminosities near @xmath20 . in a companion paper @xcite we study rest - frame colours and morphological properties of a sample of intermediate - redshift ( @xmath21 ) agn . the paper is organised as follows . we first describe the sample selection and properties together with a summary of the observational data ( sect . [ sec : data ] ) . we then comment on the decomposition of the nuclear and galaxy contribution , including a brief summary of the extensive simulations that we use to estimate measurement errors ( sect . [ sec : analysis ] ) . in sect . [ sec : results ] we present the measured host galaxy magnitudes and describe our treatment of non - detections . we move on to discuss the results in sect . [ sec : discussion ] , followed by our conclusions in sect . [ sec : conclusions ] . we use @xmath22kms@xmath23mpc@xmath23 , @xmath24 and @xmath25 throughout this paper . all quoted magnitudes are zeropointed to the ab system with zp@xmath26 and zp@xmath27 . gems , galaxy evolution from morphologies and seds @xcite is a large imaging survey in two bands ( f606w and f850lp ) with the advanced camera for surveys ( acs ) aboard _ hst_. centered on the chandra deep field south ( cdfs ) , it covers an area of @xmath28 ( 78 acs fields ) . each acs field was integrated for @xmath2913min exposures per filter ( one orbit ) , dithered by @xmath30 between exposures . the individual images were then combined , corrected for the acs geometric distortion , and at the same time rebinned to a finer pixel grid of @xmath31 , achieving approximate nyquist - sampling of the psf . the image combination also removed artefacts such as cosmic ray hits and satellite trails . resulting point source limiting magnitudes are @xmath32 ( @xmath33 ) and @xmath34 ( @xmath33 ) . in its central @xmath81/5 , gems incorporates the epoch 1 data from the goods@xcite project , that are similarly deep as the other gemsfields . further details of the data reduction procedure will be given in a forthcoming paper ( caldwell et al . 2004 , in prep . ) . the area covered by gems coincides with one of four fields covered by the combo-17 survey @xcite , which produced a low - resolution spectrophotometric data base ( based on photometry in 17 filters ) for about 10000 galaxies and 60 type 1 agn brighter than @xmath35 ( vega zeropoint ) in the cdfs area @xcite . the large number of filters permitted simultaneous assignment of accurate seds and redshifts for both galaxies and type 1 agn . galaxies and agn are classified by matching an sed template library to the set of 17 photometric points . the agn sed is composed of a range of continuum spectra with added broad emission lines ( all details are given in * ? ? ? type 2 agn as well as very low luminosity agn are invariably classified as galaxies . photometric redshifts are very reliable , with an rms scatter of @xmath36 for galaxies ( at @xmath37 ) and @xmath38 for agn at all redshifts . in this paper we address specific combo-17 sources just by their running identifiers ; the full combo-17 list of classifications in the cdfs will be made available in the future ( wolf et al . 2004 , in prep . ) . combo-17 redshifts and @xmath39 magnitudes of the agn in the sample . the dashed line marks the expected @xmath39 band magnitude for an agn of @xmath40 at the corresponding redshifts , assuming a typical quasar spectrum @xcite , the dotted lines correspond to @xmath41 and @xmath42 for reference . , width=321 ] the combo-17-selected agn in the gems field range over redshifts from @xmath43 up to @xmath44 . in this study we investigate the high-@xmath18 part of the agn distribution . our sample contains all agn brighter than @xmath45 in the redshift range @xmath46 that show a meaningful counterpart in the gems images . this excludes three objects with a primary classification as agn by combo-17 that were apparently low redshift ( @xmath47 ) emission line galaxies . at these redshifts , all light detected in the gems bands will originate in the rest frame ultraviolet . in fact , the lower redshift limit has been imposed to ensure that the long wavelength cutoff of the f850lp filter is still located below the balmer jump for all objects . the upper redshift boundary , on the other hand , was set to avoid contamination from possible extended ly@xmath48 emission in the f606w filter . the resulting sample contains 23 agns , three of which are positioned in the overlap region of two tiles so two separate images exist for these . one object ( combo-17 05696 ) shows inconsistent photometry between the two tiles due to variability over the 111 days between the two integrations . another object ( combo-17 19965 ) was classified in combo-17 with a redshift of @xmath49 that had to be revised following spectroscopy observations ( g. worseck , private communication ) and now entered this sample with @xmath50 . table [ tab : sample ] gives an overview over the sample properties , and figure [ z_r ] shows the distribution in @xmath39 and @xmath18 . the average absolute magnitudes for these objects which we call intermediately luminous quasars place them close to the canonical division of @xmath51 between seyfert galaxies ( low - luminosity ) and qsos ( high - luminosity ) . several x - ray sources in the cdfs have already been studied by _ hst _ with the wfpc2 camera @xcite , most of them faint agn , but their sample is completely disjoint from the combo-17 agn selection for this redshift range . one object falls into our redshift range but has @xmath52 , outside our selection limits . cccccccc i d & tile & ra ( 2000 ) & dec ( 2000 ) & @xmath18 & @xmath39 ( vega ) & f606w & f850lp + 12325 & 11 & 033301.7 & 275819 & 1.843 & 20.38 & 20.13 & 19.65 + 19965 & 23 & 033145.2 & 275436 & 1.90 & 19.96 & 20.59 & 20.29 + 30792 & 82 & 033243.3 & 274914 & 1.929 & 22.36 & 21.60 & 22.06 + 02006 & 04 & 033232.0 & 280310 & 1.966 & 19.76 & 19.59 & 18.98 + 04809 & 08 & 033136.3 & 280150 & 1.988 & 22.31 & 21.18 & 20.91 + 06817 & 09 & 033127.8 & 280051 & 1.988 & 21.59 & 21.61 & 20.86 + 18324 & 19 & 033300.9 & 275522 & 1.990 & 22.58 & 22.00 & 21.33 + 05498 & 01 & 033316.1 & 280131 & 2.075 & 22.29 & 22.91 & 22.28 + 11941 & 10 & 033326.3 & 275830 & 2.172 & 20.51 & 20.80 & 20.40 + 62127 & 62 & 033136.7 & 273446 & 2.175 & 23.57 & 24.91 & 24.37 + 51835 & 55 & 033140.1 & 273917 & 2.179 & 22.92 & 23.01 & 22.41 + 00784 & 05 & 033227.1 & 280336 & 2.282 & 23.29 & 23.28 & 22.80 + 36120 & 39 & 033149.4 & 274634 & 2.306 & 22.37 & 22.70 & 22.23 + 05696 & 02 & 033321.8 & 280121 & 2.386 & 23.06 & 22.73 & 22.32 + 05696 & 03 & @xmath53 & @xmath53 & @xmath53 & @xmath53 & 23.14 & 22.68 + 07671 & 07 & 033151.8 & 280026 & 2.436 & 22.35 & 22.35 & 22.24 + 07671 & 15 & @xmath53 & @xmath53 & @xmath53 & @xmath53 & 22.36 & 22.22 + 06735 & 02 & 033306.3 & 280056 & 2.444 & 21.98 & 22.14 & 21.88 + 01387 & 08 & 033144.0 & 280320 & 2.503 & 23.24 & 24.05 & 23.17 + 33644 & 31 & 033259.9 & 274748 & 2.538 & 21.87 & 21.28 & 21.17 + 11922 & 11 & 033309.1 & 275827 & 2.539 & 22.25 & 22.65 & 21.93 + 16621 & 19 & 033309.7 & 275614 & 2.540 & 19.98 & 20.41 & 20.06 + 15396 & 21 & 033216.2 & 275644 & 2.682 & 22.64 & 22.69 & 22.41 + 33630 & 33 & 033140.1 & 274746 & 2.719 & 21.74 & 22.21 & 21.98 + 42882 & 45 & 033201.6 & 274328 & 2.719 & 23.18 & 23.89 & 23.48 + 42882 & 95 & @xmath53 & @xmath53 & @xmath53 & @xmath53 & 24.04 & 23.54 + even though space based , acs shows a non - negligible background from stray light . in the reduction process already a global , outlier clipped median background was subtracted ( caldwell et al . 2004 , in prep . ) . as the deblending of nuclear and host galaxy component with two - dimensional modelling is sensitive to background sources , we applied an extra procedure to remove net residuals in the local background . this included an iterative masking of all objects in the field and the determination of the local background from the object - free regions . for each square of @xmath54 pixels an average from the unmasked pixels was computed , with a subsequent bilinear interpolation between these values to yield a background estimate for the whole field . after background adjustment , small subimages of @xmath55 pixels were extracted around all agns , corresponding to a field of view of @xmath56 . this field size contains @xmath57 % of the acs psf flux . since the agn are not strongly resolved this fraction also applies to the agn flux . the data reduction procedure kept record of the individual pixel weights throughout the process of reduction and combination . this information was then used in combination with the shot noise derived from pixel count rates to construct variance images which were later used in the error budget calculations . compared to ground based telescopes , the point spread function ( psf ) of acs is very stable in time . however , coma , astigmatism and defocus from surface height variations of the two ccds lead to variations over the field of view that need to be taken into account . the variations are much weaker than for the wfpc2 camera but remain non - negligible . also the ` breathing ' of _ hst _ changes the focal length which leads also to a small time variability in the psf . in figure [ fig : psf ] we show the mean acs psf compiled from @xmath8500 stars in the extensive gems area and the variation of the psf over the fov . + the variations were also investigated by @xcite from the crowded field of 47 tucanae . such an analysis was not possible for individual gems fields due to the small number of stars ( @xmath810 ) per field . thus a simultaneous characterisation of the spatial and temporal variations was not possible . however , since the spatial variations dominate ( further details of our investigations of the acs psf variations will be given in a dedicated technical paper , jahnke et al . 2004 , in prep . ) we used the large number of unsaturated stars in the gems area to construct an empirical psf individually for each object . at a given position we combined the nearest @xmath835 undisturbed stars to create a position - specific psf estimate . in this way we average over time , but only stars from a radius @xmath58 were used , and psf shape errors due to _ spatial _ variation were minimized , while a very high s / n was achieved for each of these psf estimates . finally , the subimage of each agn and its connected psf were registered to a common centroid . from our psf analysis we found that while coherent large - scale variations were essentially absent within each stack of @xmath835 psf stars , there was still considerable mismatch between the individual stars , in particular in the central pixel regions . as such mismatched pixels could be spuriously assigned to a host galaxy , we took the variations within each psf stack to derive rms frames describing an inherent psf uncertainty ; these were then also included in the variance images , artificially reducing the weight in the inner pixel regions . psf rms errors per pixel range from up to 30% in individual pixels inside 1 fwhm to 515 % inside @xmath59 , and to generally below 5% outside . for luminous agn at high redshifts , separating the galaxy image from the nuclear point source is a daunting task . even with a very good knowledge of the psf , the problem is still that the relative scalings of galaxy and agn are not known _ a priori_. in fact , one can not even be certain that the host galaxy is detectable at all and not swamped by the central point source . we therefore started out with the well - established technique of simple psf subtraction . in each case , the psf was scaled to the central flux of the agn , integrated inside a circular aperture of 4 pixel ( 012 ) diameter centered on the nucleus . this radius encircles approximately 34% of the total energy of a point source . a smaller radius ( e.g. 1 pixel ) would become too sensitive to shot noise and psf mismatch , while larger radii contain a significantly higher fraction of the total flux ( 50%/60%/70% at 3/4/5 pixel radius ) and thus would make a detection of any host galaxy component successively harder . this procedure somewhat oversubtracts the nuclear component by an amount corresponding to the underlying host galaxy contribution inside the encircled region . however , this method yields strictly conservative estimates the host galaxy flux , i.e. always unserestimating it . we used extensive simulations to determine correction factors for this oversubtraction ( see section [ sec : errors ] ) . the peak - scaled psf subtraction method has the advantage of being independent of any assumption about the host galaxy morphology . as a second method we employed the modelling package galfit in version 1.7a @xcite that allows the simultaneous fitting of several two - dimensional components to an image , convolved with a given psf . we describe our application of galfit to quasar images in @xcite in detail . in total we ran galfit in three configurations , always fitting two components , which were the point - source nucleus and either an exponential disk @xcite , a @xmath60 de vaucouleurs spheroid @xcite , or a @xcite model with the srsic index as a free parameter . however , simulations ( sect . [ sec : errors ] ) showed that with the present data , the galfit version used was operating near its limit , due to the very high contrast between nuclei and host galaxies . in the context of this paper we thus used galfit only as a cross - check on the peak subtraction method . to determine the limits for detecting host galaxies we constructed a sample of 200 randomly selected unsaturated stars , 100 in each observed band , to mimic unresolved quasars . this way we could investigate how our nucleus - removal techniques responded to a undetectable host galaxy , and we could set limits on the size and shape of expected residuals , thus lower flux limits for detectable host galaxies . for these stars the psfs were created in exactly the same way as for the agns . the object itself was always excluded in the psf production thus each test star and its psf were fully independent . this set of simulated ` naked quasars ' showed that in 88% ( 97% ) of all cases , any residuals which could be taken as spurious ( g)host galaxy detections had fluxes of less than 5% ( 10% ) of the total object flux . from this we adopted the condition that a real detection should show a residual flux after peak subtraction of at least 5% of the total flux , corresponding to a maximum nucleus - to - host ratio of 20 . because of the systematic oversubtraction inherent in the procedure corrections for the flux have to be applied ( see next section ) . the final decision if a host galaxy is resolved is based on this criterion . in addition we visually inspected whether the detected flux indeed came from a host galaxy or whether other , unmasked structures were present , using the peak subtracted images and radial profile . if this could be ruled out we classified a host galaxy as detected . while the mere detection of an agn host galaxy can be achieved with comparably little effort , the determination of flux error bars and systematic offsets is much more complicated . we performed extensive simulations of artificial quasar images , composed from empirical psfs and host galaxy models plus artificial noise matching the actual flux and noise distribution in real images . these simulations are described in detail by @xcite . we applied the peak subtraction nucleus removal as well as galfit to a set of @xmath82000 quasar images created in this way . comparing input and output parameter values yielded mean magnitude offsets as well as statistical errors for the individual host galaxy magnitudes ( fig . [ fig : corrs_and_errors_z ] ) . the simulations give reliability regions and error bars . the left panel in figure [ fig : corrs_and_errors_z ] shows which magnitudes are recovered for a given synthetic host galaxy . since the input set covers a large range of different morphological configurations , scale lengths , nucleus - to - host ratios , etc . , the recovered values will scatter . close to the detection limit , the scatter and the corrections grow rapidly as a function of magnitude ; additionally , the ability to differentiate between different morphological types will generally be lost . the combination of these effects is reflected in the spread of the output of the simulations . this measured spread is a direct estimate for the uncertainties of the total flux ( right panel in figure [ fig : corrs_and_errors_z ] ) . from these simulations we adopt approximate regions in brightness where host galaxy magnitudes can be reliably determined , with correction of 0.25 to a maximum of 0.6 mag . these regions go down to @xmath61 , @xmath62 for the peak subtraction method and the present data . outside the corrections and errors increase . in three cases , marked in table [ tab : results_vz ] with ` ? ' in column @xmath63 , the observed magnitudes extend to outside these regions ; here we continue using the derived corrections for these three objects , but the so derived host galaxies are more uncertain and their magnitudes should be taken with care . notice that the f850lp band data are substantially shallower than the f606w band , mainly a consequence of the acs detector sensitivity . using the above criteria for detecting a residual host galaxy , we find nine of the 23 host galaxies to be resolved in both bands , although some lie close to the sensitivity limit . one object formally fell above the 5% level in one band but not the other ; combo-17 33630 at @xmath64 might be marginally resolved in f850lp and shows a structure at 1distance that might be a tidal arm or a foreground object . with this object lying at the highest redshift of the sample we do not consider this a clear detection . as mentioned , for three further cases in the f850lp band the host galaxies are very faint ( marked with a ` ? ' in the @xmath63 column in tab . [ tab : results_vz ] ) . while their flux is above 5% of the total , their raw magnitudes fell 0.81.0 mag outside the reliability region where corrections and associated errors are still small . this low s / n is also reflected in the radial profiles ( see appendix [ sec : appendix ] ) . as described above , tests with field stars show that 12% of all objects ( @xmath83 objects ) might show spurious ` host galaxies ' at the 5% flux level , and 3% ( 0 or 1 objects ) at the 10% flux level . in f606w five of our objects fall with their host fluxes between these two values . in the f850lp - band these are four which include the three uncertain ones from above . according to statistics 13 of these might be spurious detections . however , including or excluding these more uncertain data points in the following analysis does not have an influence on the conclusions drawn . for each object the host galaxy flux was determined by simple aperture photometry after subtraction of the scaled psf , excluding resolved companion objects . the radius of the aperture was matched the used image size . all extracted magnitudes are collected in table [ tab : results_vz ] , and shown together with the upper limit for the unresolved objects in the top panels of figure [ fig : errorsimpeak ] . the extracted host galaxy images and radial surface brightness profiles are shown in fig . [ fig : allimages ] in the appendix . to illustrate the behaviour of true point sources , we included a selection of 24 field stars , 12 each in the f606w- and f850lp - band ( figures [ fig : stars_v ] and [ fig : stars_z ] ) , that were subjected to the same psf determination and peak subtraction as described in section [ sec : detection ] . we plot the same profiles as for the agn . this selection is random apart from the fact that the 24 stars were observed on 24 different tiles . for most of the stars there is no systematic positive residual flux visible , as expected . the few that do show positive fluxes form the spurious detection statistic described above . while with deeper images or at lower redshift @xcite the morphological appearance of the host galaxies can be determined , this was generally not possible for the present data . apart from the extentions in combo-17 00784 , apparent residual structure visible in the colour images shown in the appendix is dominated by the psf subtraction procedure . this includes elongations and apparent off - centering that is due to limited centering precision of the order of 0.050.1 pixel ( see e.g. combo-17 05696 from tile 2 ) . thus we adopted the host galaxy morphological class in all cases as ` undecided ' , and then applied the systematic corrections for oversubtraction provided by the simulations ( fig . [ fig : corrs_and_errors_z ] ) . the corrections were typically of the order of @xmath65 mag , as documented by the columns with ` _ cor _ ' subscripts in table [ tab : results_vz ] . here we also list the estimated uncertainties resulting from our extensive simulations . the distribution of corrected magnitudes is displayed in the lower panels of figure [ fig : errorsimpeak ] . the stacked image of individually unresolved host galaxies . top : grey scale original image and psf subtracted host galaxy image ( both show @xmath66 ) in the f606w - band . on top of the linear grey scale plot logarithmic contours with 0.5 dex spacing are overplotted . bottom : radial profiles of f606w - band . the curves show the upper data points with error bars from the original image , the psf ( dashed red line ) and peak subtracted host profile ( lower points with error bars from bootstrapping ) . the host galaxy is shown without correction for any oversubtraction of the nucleus . , width=321 ] for 14 agns we find that the individual psf - subtracted residuals are consistent with non - detections , i.e. the magnitudes of individual host galaxies lie below the 5% limit . these objects are marked by arrow symbols in figures [ fig : errorsimpeak ] and [ fig : nuchost ] . it is interesting to note that these flux limits on the hosts are by no means all outstandingly faint . for many , the reason why they were not detected is the high contrast with the agn which for these objects has above average brightness . only three objects show non - detections that indicate exceptionally faint host galaxies . we will further discuss the implications of the detection limits in section [ sec : colours ] . in an attempt to assess at least the mean host galaxy properties of the unresolved agn , we simply coadded the images of all 14 objects , one of which was observed in two frames . this yielded a very deep image with effectively 15 orbits of integration time . we also combined their psfs , weighted by the relative flux of the agns , and created combined variance and psf rms frames . the higher s / n in combination with the lower psf noise ( psf and agn position are sampled at 14 different subpixel points ) yields a nominal increase in surface brightness sensitivity of @xmath81.5 mag . the resulting ` object ' has a mean redshift of @xmath67 ( weighted by agn flux ) , and total magnitudes of @xmath68 and @xmath69 , respectively . this image is shown in figure [ fig : unresplot ] . with the higher sensitivity we now indeed find a host galaxy component in the f606w - band image after psf subtraction of 4.4% of the total flux . the radial surface brightness profile also shows a small excess over an unresolved point source . in both cases this flux is highly significant as we confirmed using a bootstrap simulation for the composition of the coadded frame from the 15 frames . in the bootstrap simulation we constructed 100 new sets of 15 frames each , drawn with repetition from the original 15 frames , coadded the images in each set and did the flux analysis as above . the uncertainty in the total flux estimated from these 100 realisations is @xmath70% of the total flux , or 25% in host galaxy flux . all realisations yielded substantial positive fluxes . the error is resulting from a combination of psf uncertainty and the noise inside the scaling aperture of 4 pixel diameter . we show the uncertainties in the radial surface brightness determined from bootstrapping as error bars for the derived host galaxy in figure [ fig : unresplot ] . the so extracted magnitudes for host galaxy and nucleus in the f606w - band are listed in the last row of table [ tab : results_vz ] ( the ` stack ' object ) . the f850lp - band stack , however , with its lower sensitivity showed a much weaker signal than the f606w - band , too faint to reliably be classified as resolved . cccccccccccc id&tile & @xmath71 & @xmath72 & @xmath73 & @xmath74 & n / h@xmath75 & @xmath76 & @xmath63 & @xmath77 & @xmath78 & n / h@xmath79 + 19965&23 & 20.59 & 23.9 & 23.3@xmath80 & 20.7 & 11.3 & 20.29 & 23.8 & 23.5 @xmath81 & 20.4&17.5 + 30792&82 & 21.60 & 24.6 & 24.4@xmath81 & 21.7 & 12.0 & 22.06 & 24.2 & 23.9 @xmath81 & 22.3 & 4.2 + 18324&19 & 22.00 & 23.6 & 23.0@xmath80 & 22.5 & 1.6 & 21.33 & 22.7 & 22.4 @xmath81 & 21.9 & 1.6 + 05498&01 & 22.91 & 25.5 & 25.2@xmath81 & 23.1 & 7.1 & 22.28 & 25.0 ? & 24.4 @xmath82 & 22.5 & 6.0 + 51835&55 & 23.01 & 26.0 & 25.7@xmath83 & 23.1 & 11.2 & 22.41 & 23.8 & 23.4 @xmath81 & 23.0 & 1.6 + 00784&05 & 23.28 & 25.0 & 24.7@xmath81 & 23.6 & 2.8 & 22.80 & 24.0 & 23.7 @xmath81 & 23.4 & 1.2 + 05696&02 & 22.73 & 24.7 & 24.4@xmath81 & 23.0 & 3.6 & 22.32 & 23.8 & 23.5 @xmath81 & 22.8 & 1.9 + 05696&03 & 23.14 & 24.2 & 23.9@xmath84 & 23.9 & 0.9 & 22.68 & 24.0 & 23.7 @xmath81 & 23.2 & 1.6 + 07671&07 & 22.35 & 25.6 & 25.2@xmath81 & 22.4 & 13.2 & 22.24 & 25.0 ? & 24.4 @xmath82 & 22.4 & 6.3 + 07671&15 & 22.36 & 25.5 & 25.1@xmath81 & 22.5 & 11.9 & 22.22 & 24.6 & 24.3 @xmath83 & 22.4 & 5.5 + 11922&11 & 22.65 & 24.9 & 24.6@xmath81 & 22.9 & 4.8 & 21.93 & 25.2 ? & 24.5 @xmath85 & 22.0 & 9.3 + stack & & 21.47 & 24.9 & 24.3@xmath80 & 21.6 & 12.3 & 21.60 & & & + ccccc id&tile&@xmath18&@xmath86&sfr f606w + & & & & [ @xmath87 + 19965 & 23 & 1.90&0.2@xmath81 & 11 + 30792 & 82 & 1.929 & 0.5@xmath83 & 4 + 18324 & 19 & 1.990 & 0.6@xmath81 & 15 + 05498 & 01 & 2.075 & 0.8@xmath82 & 2 + 51835 & 55 & 2.179 & 2.3@xmath88 & 1.5 + 00784 & 05 & 2.282 & 1.0@xmath83 & 4 + 05696 & 02 & 2.386 & 0.9@xmath83 & 6 + 05696 & 03 & 2.386 & 0.1@xmath83 & 9 + 07671 & 07 & 2.436 & 0.8@xmath82 & 3 + 07671 & 15 & 2.436 & 0.9@xmath88 & 3 + 11922 & 11 & 2.539 & 0.1@xmath85 & 5 + stack & & 2.3 & @xmath89 & 6 + observed colours ( @xmath90 ) of the sample from psf peak subtraction ( circles ) , the open symbol marks the upper limit for the ` stacked ' agn . overplotted are two single burst models from ( solar metallicity * ? ? ? * ) ( solid lines ) . the upper curve is for a passively evolving burst at @xmath91 , the lower for burst of 100 myr age , relative to each redshift . the dot - dashed lines are mixtures between the two , with a ( from top ) 0.1% , 1% and 10% fraction of mass of the 100 myr population on top of 99.9% , 99% and 90% of the @xmath91 population . , width=321 ] nuclear vs. host galaxy colours for the objects in the sample with available colours . the two objects imaged twice are connected with a dotted line ( in one case not visible to near identical colours ) . the dashed line is the 1:1 relation to guide the eye . , width=321 ] from the corrected magnitudes we have derived @xmath92 colours for the detected host galaxies . these are listed in table [ tab : colours ] , now including also correction for galactic dust extinction . however , with @xmath93 @xcite these values of @xmath94 and @xmath95 negligibly affect the colours . at these redshifts , the observed photometric bands correspond to the rest frame ultraviolet , ranging from 2160 at @xmath96 to 1616 at @xmath97 in f606w and 3150 to 2350 in f850lp , respectively , so for this redshift range pure rest - frame uv colours are observed . figure [ fig : ssp_colours_ab ] shows the measured values plotted against @xmath18 . there is no discernible colour trend with redshift . all points fall within a relatively narrow range of colours ; apart from combo-17 51835 , the objects occupy a band of @xmath98 . the open symbol represents our stacked ` average ' agn constructed from the 14 unresolved objects . although it was not resolved in the f850lp band , the upper limit on its colour is actually consistent with the values derived for several of the detected objects . one critical issue in measuring uv luminosities of barely resolved agn hosts is the lingering possibility that flux may have spilled over from the nuclei . this could have happened as a purely observational artefact due to imperfect psf removal , or physically by scattering of uv photons off dust in the the host galaxy . the latter phenomenon is known to be relevant in high redshift radio galaxies @xcite . independently of the underlying mechanism , any such cross - contamination should be visible in a correlation of host galaxy with nuclear colours . figure [ fig : nuc_host_col ] shows these colours plotted against each other . no correlation is visible and we conclude that a substantial contamination of the host galaxy light from the agn is very unlikely . notice also that when considering physical scattering in the hosts , powerful radio galaxies are huge massive entities of generally vastly different appearance compared to the relatively modest agn hosts featuring in our sample . uv colours can be converted to uv spectral slope @xmath99 independent of redshift , when assuming that the sed can be described in the form @xmath100 . with @xmath99 known , the absolute magnitude at 200 nm , @xmath101 , can be computed directly from the f606w- or f850lp - band apparent magnitudes . we do this for both the host and the nucleus and these values are shown for the full sample including upper limits in figure [ fig : nuchost ] , which illustrates the two main constraints for resolving a host galaxy , apart from compactness ( see sect . [ sec : unresolved ] ) . the reliability of host galaxy photometry is constrained by s / n , thus dependent on the filter which have different depths . this is marked by the horizontal dashed lines which show the @xmath101 magnitude of host galaxies at the @xmath61 mag and @xmath62 mag edges of the adopted regions of reliability ( see section [ sec : errors ] ) , at mean redshift and mean @xmath99 . as the second effect the maximum nucleus host contrast appears as the scatter of the unresolved objects around a line shifted by 3.2 mag from unity , corresponding to 5% of the total flux . here the scatter is only induced by the assumption that all unresolved host galaxies have a the @xmath99 value the upper limit determined for the stacked image . the diagonal dotted lines mark lines of 10% , 20% , 33% and 50% of the total flux associated with the host galaxy . thus in total the region right of the solid diagonal line is inaccessible to host galaxy detection with the current data and method of analysis . we would like to emphasise that the similarity of nuclear properties for resolved and unresolved host galaxies suggests that also the host galaxy properties are similar thus supporting that the data point for the coadded stacked object is not far off the individually resolved objects in all plots . nuclear vs. host galaxy absolute magnitude at 200 nm . shows are objects with resolved host galaxies ( solid symbols ) , the coadded stacked object ( open circle with upper limit arrow ) and upper limits for objects with unresolved host ( arrows ) . the diagonal lines mark positions of constant fraction of host galaxy light of 5% , 10% , 20% , 33% and 50% of the total light , assuming a constant spectral slope @xmath99 . the horizontal dashed lines show the magnitude of host galaxies at the @xmath61 mag and @xmath62 mag detection limit , at mean redshift and mean @xmath99 . , width=321 ] in figure [ fig : beta ] spectral slope is plotted against @xmath101 for the host galaxies , compared to the mean value for a sample of 794 lyman break galaxies at redshift @xmath102 @xcite , all values uncorrected for the influence of dust in the galaxies . the slight anticorrelation that seems to be visible between @xmath99 and uv luminosity suggest that the more luminous host galaxies are having the steeper spectral slopes or bluer colours . this would mean that more luminous host galaxies had bluer colours and thus more uv light from young stars . to test this we computed the spearman rank - order coefficient for this data set . the test gave a probability for the zero hypothesis uncorrelated data of 9% . thus the zero hypothesis can not even be rejected on a 2@xmath103 level and thus the anticorrelation is not significant . a few of the most uv - luminous host galaxies fall into the same magnitude colour regime as the lbgs , while a number of objects are substantially redder . thus we do not find a positive correlation of the amount of host uv light and luminosity . spectral slope @xmath99 in the uv ( @xmath100 ) vs. 200 nm absolute magnitude , uncorrected for dust . the open symbol marks the ` stacked ' object , the crossed square is the mean value obtained by @xcite for 794 lyman break galaxies at @xmath102 . , width=321 ] the top solid line in figure [ fig : ssp_colours_ab ] shows the theoretical colour of a galaxy that formed all its stars at @xmath91 and evolved passively afterwards . this colour was computed from single stellar population ( ssp ) models taken from @xcite . the assignment of solar metallicity in this context is arbitrary and subject to discussion ( see below ) . however , qualitatively there is no strong dependence on metallicity , the exact formation redshift or the particular choice of imf for a given model family ( @xcite or @xcite ) in this wavelength range . this comparison shows that the measured uv colours of our detected agn hosts as well as of the stacked ` mean ' host galaxy are markedly bluer than expected for an ` old ' population at that epoch , i.e.@xmath104 . any correction for dust in the host galaxies will strengthen this result . thus the blue light detected in these galaxies must come from relatively young stars . these stars could be forming continuously , in which case the uv luminosities can be interpreted as indicators of the star formation rate in these galaxies . we follow up on this option in the next subsection . alternatively , the uv flux could be the afterglow of a past starburst . this is the option we consider first . given the unknown dust absorption and metallicities in the host galaxies , a single uv colour is insufficient for performing a detailed age dating of starburst or a decomposition of stellar populations . however , we want to compare to the available theoretical colour range spanned by galaxy formation at @xmath91 and a very recent ( 100 myr ) starburst to illustrate how mixing of an old underlying population with most of the mass and a recent starburst influences the colour . in figure [ fig : ssp_colours_ab ] the solid lines mark these extremes . in between these the dot - dashed lines show how contribution of a 100 myr component would influence the colour of an otherwise old population . from top to bottom 0.1% , 1% and 10% in mass are added to the old population . there is a strong degeneracy between the choice of burst age and the mixing ratio . for 10 myr less than one tenth in mass is required to produce the same uv colour compared to 100 myr . if we choose 100 myr as a timescale similar to the dynamical timescale in galaxies , the masses involved in that starburst would be of the order of a few percent of the total stellar mass . for the assumption of only one single - aged population we can rule out a very high formation redshift @xmath91 corresponds to ages of 3.52.5 gyr at @xmath96@xmath105 in the chosen cosmology . for the adopted set of models , the resulting age estimates would range mostly between @xmath80.1 and @xmath80.7 gyr . we note that the colour tracks in figure [ fig : ssp_colours_ab ] are largely flat over the redshift range of interest , and that the two pure population and the different mixing ratios correspond to different colours almost independently of @xmath18 . since our measured colours are all quite similar , we conclude that the luminosity weighted ages of the the uv - dominating stellar population must be rather similar , unless younger ages and more reddening in some objects conspire . clearly , a single uv colour is insufficient to perform a reliable age dating , with all broad band colours being affected by various degeneracies with respect to dust and metallicities . however , we have reason to believe that at least the central lines of sight towards the agn are reasonably free of dust extinction ( because the agn sample is selected by optical / uv flux ) , and we therefore do not expect dust to play a major role . at any rate , significant dust extinction would make the host galaxies intrinsically bluer than what we observe . on the other hand , assuming a metallicity lower than solar would shift all curves in figure [ fig : ssp_colours_ab ] downward , resulting in older age estimates . reducing the metallicity to @xmath106 ( 1/5 solar ) gives an single burst age increase by a factor of two ( for bruzual & charlot models ) . we conclude that if the uv light in our host galaxies is emitted by a passively evolving population of young stars , this population is typically much younger than a gyr . the diagram in figure [ fig : nuchost ] shows that no correlation exists between the amount of stellar uv light from the host galaxies and the amounts of uv radiation produced by the nuclei . if the latter is taken as a measure of the amount of matter accreted by the nuclei then in the context of a recent starburst the size of the starburst and the amount of accreted matter must be governed by different mechanisms . if the accretion rate is primarily defined by the nuclear mass , and a correlation between galaxy and black hole mass is assumed , then the size of the starburst is independent of the bulk stellar mass of the host galaxy . other factors must be dominating the amount of ( gas ) mass involved in the starburst . this can be either the total amount of gas available in the galaxy , the size of the region involved in the starburst , the strength of an interaction responsible for the starburst , etc . in any case the amounts of gas involved are variable for a given nuclear luminosity . including the upper limits in figure [ fig : nuchost ] the host luminosities span e.g. @xmath84 mag at @xmath107 , i.e. the amount of gas involved can vary by a factor of @xmath108 or more . we now interpret the detected uv emission in the alternative framework of being due to young stars forming continuously in the agn host galaxies . under this assumption it is possible to estimate the star formation rate ( sfr ) of the host galaxies from the measured rest frame uv luminosities . following @xcite and using the conversion of ab magnitudes into monochromatic fluxes we obtain @xmath109 where @xmath110 is the luminosity distance to the agn in cm , and @xmath111 is the observed uv magnitude at an arbitrary wavelength between 15002800 . with this formula we can now convert our f606w band luminosities ( which are much deeper than the f850lpband data ) into star formation rates . as long as intrinsic dust attenuation is neglected , these values are of course mere lower limits . the resulting sfr values are listed in table [ tab : colours ] and generally amount to a few solar masses per year , with remarkably little variation . the maximum value found is @xmath815 @xmath112 , and the mean is @xmath86 @xmath112 , including the results from the stacked image of individually unresolved host galaxies . this value is thus representative for the full sample of 23 agn . we show the distribution of star formation rates vs. redshifts in figure [ fig : sfr ] . no trend emerges , consistent with the previously established observations that neither colours nor luminosities show any significant trend with @xmath18 . this is again compared to the star formation rates for lbgs at @xmath113 from @xcite , as determined from the uv flux and uncorrected for dust . there are a few host galaxies with higher uv flux while for the majority it is smaller by a factor of 23 . however , there is a principal caveat in our guiding assumption of this subsection : the uv light from galaxies with strong ongoing star formation is expected to be totally dominated by the youngest stars . @xcite pointed out that the spectral shape of galaxies with a constant sfr over at least @xmath114 myr is basically flat ( in @xmath115 ) between 15002800 , assuming a @xcite imf . this would lead to an expected uv colour for our objects of @xmath116 or spectral index @xmath117 , more or less independently of redshift ( within the @xmath18 range of our sample ) , inconsistent with our observations for most objects ( see previous subsection , figure [ fig : ssp_colours_ab ] and table [ tab : colours ] ) . in other words , most detected host galaxies of our sample have uv colours that are too ` red ' for a simple continuous star formation scenario , while the stacked host galaxy is roughly consistent . if for the resolved hosts the f850lp band fluxes were used instead of the f606w band , sfr values would be higher by roughly a factor of 2 . this apparent inconsistency could be resolved in several ways . the initial assumption could be wrong , and the uv flux originates not from freshly formed stars , but from a passively evolved starburst as outlined in sect . [ sec : starburst ] . the star formation rate might not have been constant over the past , so that varying amounts of stars of different masses and ages would have been formed ; such a configuration is always possible , and our only argument against it would invoke occam s razor . finally , as discussed in the previous section , there could be an underlying older stellar population contributing more to f850lp and less to f606w ( but note that by sample design , also f850lp is completely below the balmer jump at all relevant redshifts ) . this is the scenario that is favoured by our intermediate redshift data from gems @xcite . clearly , in this study our current data set of just two uv bands is insufficient to settle this ambiguity . however , in all three cases there would be no continuous star formation as inferred for lbgs . the blue light would be a result not of a continuous process but of one or more events in the past of the host galaxy that triggered star formation . whether galaxy interaction or merger incidents were responsible or the formation of bars or spiral arms is involved can not be investigated with the present set of data . even at low redshifts , colour data of agn host galaxies are relatively scant , except for rather low - luminosity agn where the host galaxy can be separated with relative ease . in those cases , colours were generally found to be consistent with morphological types , in particular for the prevailing disk - type host galaxies @xcite . however , when higher nuclear luminosities are observed ( which generally correlate with a larger bulge component ) , then a tendency towards abnormally blue colours and younger stellar populations emerges @xcite . this tendency appears to hold also at intermediate redshifts , as demonstrated in our companion _ paper @xcite , where we find more than half of the investigated host galaxies to have bluer colours than what would be expected from their morphological types . however , we can not confirm increasing amounts of blue light from young stars with increasing host luminosity as found by @xcite . our data are consistent with constant uv flux for all hosts . at high redshifts ( @xmath11 ) , colour information is available only for a handful of objects . in a ground - based study of six bright radio - loud quasars , @xcite found indications that the hosts were very blue , actively star - forming galaxies . for three high luminosity quasars , one radio - loud and two radio - quiet , @xcite claimed very luminous envelopes and star formation rates of several hundreds solar masses per year . more recently , @xcite presented optical _ _ observations of three radio - loud and two radio - quiet quasars and found less extreme , but qualitatively similar results . because of their high radio power and optical luminosities , many of the quasars observed in previous studies are probably quite different from the moderate - luminosity hosts of moderate - luminosity radio - quiet agn that we focus on . it is nevertheless interesting to see that the presence of a significantly enhanced uv continuum and young stars seems to be at least qualitatively similar between qsos of intermediate and high luminosities . we performed the hitherto largest study of host galaxies properties of a complete sample of high - redshift agn . we detected the hosts and extract colour information in 9 of the 23 agn , and we also achieved a statistical detection of the host in the remaining 14 from a stack analysis . the uv luminosities can be interpreted in three ways : either as contribution from a passively evolving population of relatively young stars , forming typically 0.5 gyrs ago , as a mix between a population of old ( e.g. @xmath104 ) stars and a small contribution of a recently formed young population ( e.g.0.1%10% in mass at an age of 100 myrs or 1/10th of this for age 10 myrs ) , or as an indicator of ongoing star formation at a level of @xmath8215@xmath118 ( uncorrected for internal dust attenuation ) . while the first possibility is very simplistic and appears unphysical , the uv colours actually favour the two burst interpretations ; but the possibility of on - going star formation can not be completely ruled out from our data . in the framework of combined old and young populations , it is remarkable how similar the host galaxy colours are within the sample , and , unless different mass age combinations conspire , hence the estimated stellar mass fractions and ages . the derived young population mass fractions and ages are also very similar to the values estimated in our companion gems study of agn at @xmath119 @xcite , where we find abnormally blue rest - frame @xmath120 colours for a substantial fraction of host galaxies , particularly the most luminous agn in the sample . even more , these colours and ages are in turn very close to the mean values obtained from our ground - based low-@xmath18 multicolour sample @xcite . while the stellar population diagnostics of @xcite are not immediately convertible into our simple colour indicators , their impressive and highly significant results point in exactly the same direction . while the results from all these redshift regimes are similar and point to a connection of nuclear activity and the presence of young stars , and that mass fractions of young stars are similar , we always find that a larger range of absolute masses is involved , showing as a range in uv luminosity . here we find a variation of a factor of @xmath12140 for a given nuclear luminosity . our host galaxy colours span a range that reaches the colours of lyman break galaxies for a few very luminous hosts , while , as mentioned , the colours of the majority are somewhat redder than these . a comparison of optical / uv properties to the general population of high redshift galaxies would be very illuminating , but large statistical samples only become available in the near future , e.g. from the goodsproject . this persistent trend to find agn to be associated with blue stellar colours is intriguing and suggests a close connection between enhanced star formation and nuclear activity . additional support for such a connection comes from the detection of submm co emission in a number of extremely luminous high - redshift qsos and radio galaxies @xcite , although the current sensitivity of submm telescopes is insufficient to perform this test for less luminous agn at high @xmath18 . while the fact that there is a relation can hardly be denied , its physical origin remains obscure . is the enhancement of star formation a prerequisite for nuclear activity ? is it a simultaneously occurring phenomenon , caused by the same trigger ? or is it a consequence of the agn ? galaxy merging and interaction are clearly two possible candidates to connect these two phenomena , but neither the only ones nor are the involved physics understood . much additional data will be required , in particular those helping reliably to reconstruct the star formation history in high - redshift galaxies , before any firm conclusions can be drawn . based on observations taken with the nasa / esa _ hubble space telescope _ , which is operated by the association of universities for research in astronomy , inc . ( aura ) under nasa contract nas5 - 26555 . support for the gems project was provided by nasa through grant number go-9500 from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa under contract nas5 - 26555 . efb and sfs ackowledge financial support provided through the european community s human potential program under contract hprn - ct-2002 - 00316 , sisco ( efb ) and hprn - ct-2002 - 00305 , euro3d rtn ( sfs ) . cw was supported by a pparc advanced fellowship . sj acknowledges support from the national aeronautics and space administration ( nasa ) under ltsa grant nag5 - 13063 issued through the office of space science . dhm acknowledges support from the national aeronautics and space administration ( nasa ) under ltsa grant nag5 - 13102 issued through the office of space science . 38 natexlab#1#1 aretxaga , i. , terlevich , r. j. , & boyle , b. j. 1998 , mnras , 296 , 643 , g. , & charlot , s. 2003 , mnras , 344 , 1000 , r. , s ' anchez , s. f. , gonz ' alez - serrano , j. i. , benn , c. r. , & vigotti , m. 1998 , aj , 115 , 1234 , g. 2003 , pasp , 115 , 763 de vaucouleurs , g. 1948 , ann . astrophys . , 11 , 247 , d. k. , shapley , a. e. , steidel , c. c. , pettini , m. , adelberger , k. l. , hunt , m. p. , moorwood , a. f. m. , & cuby , j. 2003 , apj , 591 , 101 , r. , kotilainen , j. k. , pagani , c. , scarpa , r. , & treves , a. 2004 , apj , 604 , 495 freeman , k. c. 1970 , apj , 160 , 812 gebhardt , k. , bender , r. , bower , g. , dressler , a. , faber , s. m. , filippenko , a. v. , green , r. , grillmair , c. , ho , l. c. , kormendy , j. , lauer , t. r. , magorrian , j. , pinkney , j. , richstone , d. , & tremaine , s. 2000 , apj , 539 , l13 giavalisco , m. , et al . 2004 , apj , 600 , l93 , n. a. , koekemoer , a. m. , schreier , e. j. , bergeron , j. , giacconi , r. , hasinger , g. , kewley , l. , norman , c. , rosati , p. , tozzi , p. , & zirm , a. 2003 , apj , 595 , 685 , j. b. 2003 , aj , 125 , 1053 hutchings , j. b. , crampton , d. , morris , s. l. , durand , d. , & steinbring , e. 1999 , aj , 117 , 1109 hutchings , j. b. , frenette , d. , hanisch , r. , mo , j. , dumont , p. j. , redding , d. c. , & neff , s. g. 2002 , aj , 123 , 2936 jahnke , k. , kuhlbrodt , b. , & wisotzki , l. 2004 , mnras ( in press ) , astro - ph/0311123 kauffmann , g. , heckman , t. m. , tremonti , c. , brinchmann , j. , charlot , s. , white , s. d. m. , ridgway , s. , brinkmann , j. , fukugita , m. , hall , p. , ivezic , z. , richards , g. , & schneider , d. 2003 , mnras , 346 , 1055 kennicutt , c. k. 1998 , araa , 36 , 189 , a. m. , grogin , n. a. , schreier , e. j. , giacconi , r. , gilli , r. , kewley , l. , norman , c. , zirm , a. , bergeron , j. , rosati , p. , hasinger , g. , tozzi , p. , & marconi , a. 2002 , apj , 567 , 657 kotilainen , j. k. , & ward , m. j. 1994 , mnras , 266 , 953 krist , j. , & hook , r. 2003 , stsci intrument science report acs 2003 - 06 kukula , m. j. , dunlop , j. s. , mclure , r. j. , miller , l. , percival , w. , baum , s. a. , & odea , c. p. 2001 , mnras , 326 , 1533 lehnert , m. d. , heckman , t. m. , chambers , k. c. , & miley , g. k. 1992 , apj , 393 , 68 , a. , beelen , a. , bertoldi , f. , cox , p. , carilli , c. l. , priddey , r. s. , mcmahon , r. g. , & isaak , k. g. 2003 , a&a , 398 , 857 peng , c. y. , ho , l. c. , impey , c. d. , & rix , h .- w . 2002 , aj , 124 , 266 ridgway , s. e. , heckman , t. m. , calzetti , d. , & lehnert , m. 2001 , apj , 550 , 122 rix , h .- w . , barden , m. , beckwith , s. v. w. , bell , e. f. , borch , a. , caldwell , j. a. r. , huler , b. , jahnke , k. , jogee , s. , mcintosh , d. h. , meisenheimer , k. , peng , c. y. , snchez , s. f. , somerville , r. s. , wisotzki , l. , & wolf , c. 2004 , apjs , 152 , 163 , e. e. 1955 , apj , 121 , 161 snchez , s. f. , & gonzlez - serrano , j. i. 2003 , a&a , 406 , 435 snchez , s. f. , jahnke , k. , wisotzki , l. , barden , m. , beckwith , s. v. w. , bell , e. f. , borch , a. , caldwell , j. a. r. , huler , b. , jogee , s. , mcintosh , d. h. , meisenheimer , k. , peng , c. y. , rix , h .- w . , somerville , r. s. , & wolf , c. 2004 , submitted to apj , astro - ph/0403645 schade , d. , boyle , b. j. , & letawsky , m. 2000 , mnras , 315 , 498 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , apj , 500 , 525 , e. j. , koekemoer , a. m. , grogin , n. a. , giacconi , r. , gilli , r. , kewley , l. , norman , c. , hasinger , g. , rosati , p. , marconi , a. , salvati , m. , & tozzi , p. 2001 , apj , 560 , 127 srsic , j. 1968 , atlas de galaxes australes , observatorio astronomico de cordoba shapley , a. e. , steidel , c. c. , pettini , m. , & adelberger , k. l. 2003 , apj , 588 , 65 , j. , fosbury , r. a. e. , villar - martn , m. , cohen , m. h. , cimatti , a. , di serego alighieri , s. , & goodrich , r. w. 2001 , a&a , 366 , 7 wolf , c. , meisenheimer , k. , kleinheinrich , m. , borch , a. , dye , s. , gray , m. , wisotzki , l. , bell , e. f. , rix , h .- w . , cimatti , a. , hasinger , g. , & szokoly , g. 2004 , submitted to a&a , astro - ph/0403666 wolf , c. , meisenheimer , k. , rix , h .- w . , borch , a. , dye , s. , & kleinheinrich , m. 2003 , a&a , 401 , 73 , c. , wisotzki , l. , borch , a. , dye , s. , kleinheinrich , m. , & meisenheimer , k. 2003 , a&a , 408 , 499 figure [ fig : allimages ] shows plots for each of the nine resolved object plus the composite ` stacked ' object . two objects appear twice as they appear in overlapping areas of gemstiles . figures [ fig : stars_v ] and [ fig : stars_z ] show a random selection of isolated stars used to show the zero case of a point sources without any host galaxy contribution , for comparison purposes .

we study the _ degrees of freedom ( dof ) _ of cellular networks in which a _ full duplex ( fd ) base station ( bs ) _ equipped with multiple transmit and receive antennas communicates with multiple mobile users . we consider two different scenarios . in the first scenario , we study the case when _ half duplex ( hd ) users _ , partitioned to either the uplink ( ul ) set or the downlink ( dl ) set , simultaneously communicate with the fd bs . in the second scenario , we study the case when _ fd users _ simultaneously communicate ul and dl data with the fd bs . unlike conventional hd only systems , inter - user interference ( within the cell ) may severely limit the dof , and must be carefully taken into account . with the goal of providing theoretical guidelines for designing such fd systems , we completely characterize the sum dof of each of the two different fd cellular networks by developing an achievable scheme and obtaining a matching upper bound . the key idea of the proposed scheme is to carefully allocate ul and dl information streams using interference alignment and beamforming techniques . by comparing the dofs of the considered fd systems with those of the conventional hd systems , we establish the dof gain by enabling fd operation in various configurations . as a consequence of the result , we show that the dof can approach the two - fold gain over the hd systems when the number of users becomes large enough as compared to the number of antennas at the bs . cellular network , degrees of freedom , full duplex , interference alignment , multiantenna technique .

1501.02889

current cellular communication systems operate in half - duplex ( hd ) mode by transmitting and receiving either at different time slots or over different frequency bands . the system is designed in such a way that the downlink ( dl ) and uplink ( ul ) traffics are structurally separated by time division duplexing ( tdd ) or frequency division duplexing ( fdd ) . the advantage of such design principle is that it avoids the high - powered self - interference that is generated during simultaneous transmission and reception . recent results @xcite , however , have demonstrated the feasibility of _ full - duplex _ ( fd ) wireless communication by suppressing or cancelling self - interference in the rf and baseband level . various practical designs to realize self - interference cancellation have been proposed in the literature , including adding additional antennas @xcite , adding auxiliary transmit rf chains @xcite or auxiliary receive rf chains @xcite , using polarization @xcite , employing balun circuits @xcite , and many more . for more details , see @xcite and the references therein . by enabling simultaneous transmission and reception , fd radio is expected to double the spectral efficiency of current hd systems @xcite , and is considered as one of the key technologies for next generation communication systems . evidently , in situations where the base station ( bs ) and the user simultaneously transmit bidirectionally as shown in figure [ fig : first ] , enabling fd doubles the overall spectral efficiency . this point - to - point bidirectional communication example , however , is just one instance of how a fd cellular system will function . in some practical cases , the system may have to support hd users which do not have fd radio due to extra hardware burden on mobile devices . in such case , the fd bs can simultaneously communicate with two sets of users , one receiving dl data from the bs and the other transmitting ul data to the bs ( figure [ fig : second ] ) . in another configuration shown in figure [ fig : third ] , for instance , when the bs has many more antennas compared to each user , the fd bs may wish to simultaneously communicate with multiple fd users using multi - user multiple - input and multiple - output ( mimo ) techniques . since the bs is simultaneously transmitting and receiving , there is potential to double the overall spectral efficiency compared to the conventional hd only systems . however , the configurations shown in figures [ fig : second ] and [ fig : third ] induce a new source of interference that does not arise in hd only networks . in figure [ fig : second ] , since user 1 is transmitting to the bs while user 2 is receiving from the bs , the transmission from user 1 causes interference to user 2 . similarly , in figure [ fig : third ] , the ul transmission of the users causes interference to the dl reception to each other . in cases where this type of interference is strong and proper interference mitigation techniques are not applied , the gain of having fd radios can be severely limited even when self - interference is completely removed . to manage inter - user interference and fully utilize wireless spectrum with fd operation , in this paper we employ signal space _ interference alignment _ ( ia ) schemes optimized for fd networks including the cases in figure 1 . initially proposed by the seminar works in @xcite , ia is a coding technique that efficiently deals with interference and is known to achieve the optimal dof for various interference networks @xcite . especially , it is shown that ia can be successfully applied to mitigate interference in various cellular networks , such as two - cell cellular networks @xcite and multiantenna ul dl cellular networks @xcite . furthermore , the idea of ia can also be applied to the ( multi - user ) bidirectional cellular network with ergodic phase fading @xcite , in which the achievable scheme is based on the ergodic ia scheme proposed in @xcite . motivated by the aforementioned previous works related to ia , we propose the optimal transmission schemes that attain the optimal sum dofs for two configurations : 1 ) a cellular network with a multiantenna fd bs and hd users ( figure [ fig : second ] ) ; 2 ) a cellular network with a multiantenna fd bs and fd users ( figure [ fig : third ] ) . the key idea of the proposed schemes is to carefully allocate the ul and dl information streams using ia and beamforming techniques . the ul data is sent to the bs using ia such that the inter - user interference is confined within a tolerated number of signal dimensions , while the bs transmits in the remaining signal dimensions via zero - forcing beamforming for the dl transmission . with the proposed schemes , our primary goal is to answer whether if fd operation can still double the overall spectral efficiency even in the presence of inter - user interference . we answer this question by providing matching upper bounds with the proposed achievable schemes , completely characterising the sum dofs of the considered networks . as a consequence of the result , even in the presence of inter - user interference , we show that the overall dof can approach the two - fold gain over hd only networks when the number of users becomes large as compared to the number of antennas at the bs . we further provide the dof gain of the fd systems by considering various configurations ( see sections iii and vi . ) . in @xcite , cadambe and jafar proposed a novel interference management technique called _ interference alignment _ ( ia ) , which achieves the optimal sum dof of @xmath0 for the @xmath1-user interference channel ( ic ) with time - varying channel coefficients . in addition , for the case in which all channel coefficients are constant , motahari et al . @xcite proposed a different type of ia scheme based on number - theoretic properties of rational and irrational numbers and showed that the optimal dof of @xmath0 is also achievable . later , alternative methods of aligning interference in the finite signal - to - noise regime has been also proposed in @xcite . the concept of ia has been successfully adapted to various network environments , e.g. , see @xcite and the references therein . the dof of cellular networks has been first studied by suh and tse for both ul and dl environments , where inter - cell interference exists @xcite . it was shown that , for two - cell networks having @xmath1 users in each cell , the sum dof of @xmath2 is achievable for both ul and dl . thus , multiple users at each cell are beneficial for improving the dof of cellular networks . these models were further extended to more general cases in terms of the number of users and the number of antennas at each bs @xcite . in addition , recently , the dof of the multiantenna ul dl cellular network consisting of dl and ul cells has been studied in @xcite . for a cellular network with fd operation in the absence of self - interference , the dof of the ( multi - user ) bidirectional case has been studied in @xcite for ergodic phase fading setting . the rest of this paper is organized as follows . in section ii , we describe the network model and the sum dof metric considered in this paper . in section iii , we present the main results of the paper and intuitively explain how fd operation can increase the dof . in sections iv and v , we provide the achievability and converse proofs of the main theorems , respectively . in section vi , we discuss the impacts of self - interference and scheduling on the dof . finally , we conclude in section vii . * notations * : we will use boldface lowercase letters to denote vectors and boldface uppercase letters to denote matrices . throughout the paper , @xmath3 $ ] denotes @xmath4 , @xmath5 denotes the @xmath6 all - zero vector , and @xmath7 denotes the @xmath8 identity matrix . for a real value @xmath9 , @xmath10 denotes @xmath11 . for a set of vectors @xmath12 , @xmath13 denotes the vector space spanned by the vectors in @xmath12 . for a vector @xmath14 , @xmath15 means that @xmath14 is orthogonal with all vectors in @xmath13 . for a set of matrices @xmath16 , @xmath17 denotes the block diagonal matrix consisting of @xmath16 . for a comprehensive understanding of the dof improvement by incorporating fd operation , we consider two types of network models : the first network model consists of a single fd bs which simultaneously transmits to a set of dl users ( in hd mode ) and receives from a set of ul users ( in hd mode ) ; the second model consists of a single fd bs communicating with a set of fd users . unless otherwise specified , we simply denote bs for fd bs in the rest of this paper . in this subsection , we formally define the network models for the two cases mentioned above . fd - bs hd - user cellular network . ] this network model consists of a mixture of a fd bs and hd users . the hd users are partitioned into two sets , in which one set of users are transmitting to the bs , and the other set of users are receiving from the bs simultaneously . this cellular network is depicted in figure [ fig : model1 ] . we assume that the fd bs is equipped with @xmath18 transmit antennas and @xmath19 receive antennas . on the user side , we assume that there are @xmath20 dl users and @xmath21 ul users , each equipped with a single antenna . here , each user is assumed to operate in hd mode . the bs wishes to send a set of independent messages @xmath22}_1,\cdots , w^{[{\sf d}]}_{n_1})$ ] to the dl users and at the same time wishes to receive a set of independent messages @xmath23}_1,\cdots , w^{[{\sf u}]}_{n_2})$ ] from the ul users . for @xmath24 $ ] , the received signal of dl user @xmath25 at time @xmath26 , denoted by @xmath27}_i(t)\in\mathbb{r}$ ] , is given by @xmath28}_i(t)=\mathbf{g}_i(t)\mathbf{x}^{[\sf bs]}(t)+\sum_{j=1}^{n_2}h_{ij}(t)x^{[\sf u]}_j(t)+z_i^{[\sf d]}(t)\end{aligned}\ ] ] and the received signal vector of the bs at time @xmath26 , denoted by @xmath29}(t)\in\mathbb{r}^{m_2\times 1}$ ] , is given by @xmath30}(t)=\sum_{j=1}^{n_2}\mathbf{f}_j(t)x_j^{[\sf u]}(t)+\mathbf{z}^{[\sf bs]}(t),\end{aligned}\ ] ] where @xmath31}(t)\in\mathbb{r}^{m_1\times1}$ ] is the transmit signal vector of the bs at time @xmath26 , @xmath32}_j[t]\in\mathbb{r}$ ] is the transmit signal of ul user @xmath33 at time @xmath26 , @xmath34 is the channel vector from the bs to dl user @xmath25 at time @xmath26 , @xmath35 is the scalar channel from ul user @xmath33 to dl user @xmath25 at time @xmath26 , and @xmath36 is the channel vector from ul user @xmath33 to the bs . the additive noises @xmath37}(t)\in\mathbb{r}$ ] and @xmath38}(t)\in\mathbb{r}^{m_2\times 1}$ ] are assumed to be independent of each other and also independent over time , and is distributed as @xmath37}(t)\sim \mathcal{n}(0,1)$ ] and @xmath38}(t)\sim \mathcal{n}(\mathbf{0}_{m_2},\mathbf{i}_{m_2})$ ] . we assume that channel coefficients are drawn i.i.d . from a continuous distribution and vary independently over time . it is further assumed that global channel state information ( csi ) is available at the bs and each ul and dl user . the bs and each ul user is assumed to satisfy an average transmit power constraint , i.e. , @xmath39}(t)\|^2\right]\leq p$ ] and @xmath40}(t)|^2\big]\leq p$ ] for all @xmath41 $ ] . in the rest of the paper , we denote this network as a @xmath42 fd - bs hd - user cellular network . we assume perfect self - interference suppression within the bs during fd operation . hence there is no self - interference for the input output relations in and . we will discuss how imperfect self - interference suppression effects the dof in section vi - a . @xmath43 fd - bs fd - user cellular network . ] in this model , we consider the case where both the bs and users have fd capability ( depicted in figure [ fig : model2 ] ) . as before , we assume that the bs is equipped with @xmath18 transmit antennas and @xmath19 receive antennas . however , unlike the fd - bs hd - user cellular network , there is a single set of @xmath44 fd users , each equipped with a single transmit and a single receive antenna , that simultaneously transmits to and receives from the bs . the bs wishes to send a set of independent messages @xmath22}_1,\cdots , w^{[{\sf d}]}_{n})$ ] to the users and at the same time wishes to receive a set of independent messages @xmath23}_1,\cdots , w^{[{\sf u}]}_{n})$ ] from the same users . for @xmath45 $ ] , the received signal of user @xmath25 at time @xmath26 is given by @xmath46}(t)+\sum_{j=1,j\neq i}^{n}h_{ij}(t)x_j(t)+z_i(t)\end{aligned}\ ] ] and the received signal vector of the bs at time @xmath26 is given by @xmath47}(t)=\sum_{j=1}^{n}\mathbf{f}_j(t)x_j(t)+\mathbf{z}^{[\sf bs]}(t).\end{aligned}\ ] ] as before , we assume that self - interference at the bs and each user is completely suppressed , which is reflected in the input output relations in and . the rest of the assumptions are the same as those of the @xmath42 fd - bs hd - user cellular network . in the rest of the paper , we denote this network as a @xmath48 fd - bs fd - user cellular network . for each network model , we define a set of length @xmath49 block codes and its achievable dof . let @xmath50}_i$ ] and @xmath51}_j$ ] be chosen uniformly at random from @xmath52}_i}]$ ] and @xmath53}_j}]$ ] respectively , where @xmath24 $ ] and @xmath41 $ ] . then a @xmath54}_1},\cdots,2^{nr^{[{\sf d}]}_{n_1}},2^{nr^{[{\sf u}]}_1},\cdots,2^{nr^{[{\sf u}]}_{n_2}};n)$ ] code consists of the following set of encoding and decoding functions : * _ encoding : _ for @xmath55 $ ] , the encoding function of the bs at time @xmath26 is given by @xmath56}(t)=\phi_t(w_1^{[{\sf d}]},\cdots , w_{n_1}^{[{\sf d}]},\mathbf{y}^{[{\sf bs}]}(1),\cdots \mathbf{y}^{[{\sf bs}]}(t-1)).\end{aligned}\ ] ] for @xmath55 $ ] , the encoding function of ul user @xmath33 at time @xmath26 is given by @xmath57}),\end{aligned}\ ] ] where @xmath41 $ ] . * _ decoding : _ upon receiving @xmath58}(1)$ ] to @xmath58}(n)$ ] , the decoding function of the bs is given by @xmath59}_j=\chi_j(\mathbf{y}^{[{\sf bs}]}(1),\cdots,\mathbf{y}^{[{\sf bs}]}(n),w_1^{[{\sf d}]},\cdots , w_{n_1}^{[{\sf d } ] } ) \text { for } j\in[1:n_2 ] . \end{aligned}\ ] ] upon receiving @xmath60 to @xmath61 , the decoding function of dl user @xmath25 is given by @xmath62}_i=\psi_i(y_i(1),\cdots , y_i(n)),\end{aligned}\ ] ] where @xmath24 $ ] . a rate tuple @xmath63}_1,\cdots , r^{[{\sf d}]}_{n_1},r^{[{\sf u}]}_1,\cdots , r^{[{\sf u}]}_{n_2})$ ] is said to be _ achievable _ for the fd - bs hd - user cellular network if there exists a sequence of @xmath54}_1},\cdots,2^{nr^{[{\sf d}]}_{n_1}},2^{nr^{[{\sf u}]}_1},\cdots,2^{nr^{[{\sf u}]}_{n_2}};n)$ ] codes such that @xmath64}_i\neq w^{[{\sf d}]}_i)\to 0 $ ] and @xmath65}_j\neq w^{[{\sf u}]}_j)\to 0 $ ] as @xmath49 increases for all @xmath24 $ ] and @xmath41 $ ] . then the achievable dof tuple is given by @xmath66}_1,\cdots , d_{n_1}^{[{\sf d}]},d^{[{\sf u}]}_1,\cdots , d_{n_2}^{[{\sf u}]})=\lim_{p\to\infty}\left(\frac{r^{[{\sf d}]}_1}{\frac{1}{2}\log p},\cdots,\frac{r^{[{\sf d}]}_{n_1}}{\frac{1}{2}\log p},\frac{r^{[{\sf u}]}_1}{\frac{1}{2}\log p},\cdots,\frac{r^{[{\sf u}]}_{n_2}}{\frac{1}{2}\log p}\right).\end{aligned}\ ] ] we further denote the maximum achievable sum dof of the fd - bs hd - user cellular network by @xmath67 , i.e. , @xmath68}_1,\cdots , d_{n_1}^{[{\sf d}]},d^{[{\sf u}]}_1,\cdots , d_{n_2}^{[{\sf u}]})\in\mathcal{d}}\left\{\sum_{i=1}^{n_1}d_i^{[{\sf d}]}+\sum_{j=1}^{n_2}d_j^{[{\sf u}]}\right\},\end{aligned}\ ] ] where @xmath69 denotes the dof region of the fd - bs hd - user cellular network . similar to the fd - bs hd - user cellular network , we can define an achievable dof tuple of the fd - bs fd - user cellular network . the key difference is that each user also operates in fd mode for this second model . specifically , the encoding function of user @xmath25 at time @xmath55 $ ] is given by @xmath70},y_i(1),\cdots , y_i(t-1))$ ] and the decoding function of user @xmath25 is given by @xmath71}_i=\psi_i(y_i(1),\cdots , y_i(n),w^{[{\sf u}]}_i)$ ] , where @xmath72 $ ] . then the definition of an achievable dof tuple @xmath73}_1,\cdots , d_{n}^{[{\sf d}]},d^{[{\sf u}]}_1,\cdots , d_{n}^{[{\sf u}]})$ ] is the same as that of the fd - bs hd - user cellular network . similarly , we denote the maximum achievable sum dof of the fd - bs fd - user cellular network by @xmath74 . in this section , we state the main results of this paper . we completely characterize the sum dofs of both the @xmath42 fd - bs hd - user cellular network and the @xmath48 fd - bs fd - user cellular network . [ thm : dof_1 ] for the @xmath42 fd - bs hd - user cellular network , @xmath75 the achievability proof is given in section [ sec : achievability ] and the converse proof is given in section [ sec : converse ] . we demonstrate the utility of theorem [ thm : dof_1 ] by the following example . [ ex : symmetric1 ] consider the @xmath76 fd - bs hd - user cellular network , i.e. , @xmath77 and @xmath78 . for this symmetric case , @xmath79 from theorem [ thm : dof_1 ] . on the other hand , if the bs operates in hd mode , we can easily see that the sum dof is limited by @xmath80 . by comparing the sum dofs , we can see that there is a two - fold gain by operating the bs in fd mode when we have enough number of users in the network , i.e. , @xmath81 . figure [ ex1 ] plots @xmath67 with respect to @xmath44 when @xmath82 . as shown in the figure , fd operation at the bs improves the sum dof as @xmath44 increases and eventually the sum dof is doubled compared to hd bs for large enough @xmath44 . @xmath43 and @xmath78 . ] for the fd - bs fd - user cellular network , we have the following theorem . [ thm : dof_2 ] for the @xmath48 fd - bs fd - user cellular network , @xmath83 from the network model and the dof definition in section [ sec : problem ] , any achievable sum dof in the @xmath84 fd - bs hd - user cellular network is also achievable for the @xmath48 fd - bs fd - user cellular network . in particular , the encoding functions at the bs are the same for both network models , and the bs also receives the same signal as shown in and . comparing the user encoders , we can see that the user encoding function for the fd - bs fd - user cellular network is more general than the encoding function for the fd - bs hd - user cellular network . furthermore , we can easily see that the received signal is `` better '' than the received signal for the fd - bs hd - user cellular network , in that it has less interference ( self - interference is suppressed for the fd user case ) . hence , from theorem [ thm : dof_1 ] , the sum dof of @xmath85 is achievable for the @xmath48 fd - bs fd - user cellular network , which coincides with @xmath74 in . the converse proof is given in section [ sec : converse ] . we demonstrate the utility of theorem [ thm : dof_2 ] by the following example . [ ex : symmetric2 ] consider the @xmath86 fd - bs fd - user cellular network , i.e. , @xmath77 . for this symmetric case , @xmath87 from theorem [ thm : dof_2 ] , which coincides with the sum dof of the symmetric fd - bs hd - user cellular network in example [ ex : symmetric1 ] . again , if both the bs and the users are limited to operate in hd mode , then the sum dof is limited by @xmath80 . @xmath43 to be fair , the @xmath86 fd - bs fd - user cellular network in example [ ex : symmetric2 ] has been considered in @xcite under the ergodic fading setting assuming that the phase of each channel coefficient in @xmath88,i\neq j}$ ] is drawn independently from a uniform phase distribution . for this case , it has been shown in ( * ? ? ? * theorem 1 ) that the achievable dof tuple satisfies : @xmath89}_i&\leq \min(m , n)\nonumber\\ \sum_{j=1}^{n}d^{[{\sf u}]}_j&\leq \min(m , n)\nonumber\\ \sum_{i=1}^{n}d^{[{\sf d}]}_i+\sum_{j=1}^{n}d^{[{\sf u}]}_j&\leq\min(2m , n),\end{aligned}\ ] ] where characterises the sum dof . this result in @xcite is general in that it provides a general achievable _ dof region _ , while our result in theorem [ thm : dof_2 ] generalizes the _ sum dof _ result in @xcite by considering arbitrary number of transmit and receive antennas at the bs , and also extends to any i.i.d . generic channel setting including the ergodic fading setting . in section [ sec : discussion ] , we discuss in detail regarding the dof improvement by enabling fd operation , and also the effect of imperfect self - interference suppression . in this section , we prove that the sum dof @xmath67 in theorem [ thm : dof_1 ] is achievable . to better illustrate the main insight of the coding scheme , we first consider the achievablity of theorem [ thm : dof_1 ] for the case @xmath90 in section [ subsec : case1 ] . the main component of the scheme utilizes ia via transmit beamforming with a finite symbol extension . for general @xmath20 , interference from multiple ul users should be simultaneously aligned at multiple dl users , which requires asymptotic ia , i.e. , an arbitrarily large symbol extension . in section [ subsec : case2 ] , we introduce transmit beamforming adopting such asymptotic ia for the general network configuration . fd - bs hd - user cellular network when @xmath91 . ] for the @xmath92 fd - bs hd - user cellular network , @xmath93 from theorem [ thm : dof_1 ] . for the proof on how can be evaluated from for the case @xmath90 , we refer to the proof in lemma [ lemma : sum_dof_conversion ] . in the following , we show that @xmath67 in is achievable by considering two cases , @xmath94 and @xmath95 . for the first case @xmath96 , we can easily achieve @xmath97 by simply utilizing only the ul transmission , i.e. , the bs receives from the @xmath21 ul users with @xmath19 receive antennas . now consider the second case where @xmath91 , which we explain with the help of figure [ fig : scheme1_model1 ] . for this case , communication takes place via transmit beamforming over a block of @xmath21 time slots , i.e. , @xmath21 symbol extension . denote @xmath98 where @xmath41 $ ] . the bs sends @xmath99 information symbols to the dl user via the @xmath100 beamforming vectors @xmath101}_k\}_{k\in[1:n_2-m_2]}$ ] . on the other hand , ul user @xmath41 $ ] sends @xmath19 information symbols to the bs via the @xmath102 beamforming vectors @xmath103}_{jk}\}_{k\in[1:m_2]}$ ] . we first construct @xmath101}_k\}_{k\in[1:n_2-m_2]}$ ] as a set of @xmath99 linearly independent random vectors . next , we construct linearly independent @xmath103}_{jk}\}_{k\in[1:m_2],j\in[1:n_2]}$ ] such that for each @xmath104 $ ] , all the @xmath21 information symbols that are indexed with @xmath104 $ ] are aligned at the dl user , i.e. , satisfying the ia condition @xmath105}_{1k}=\cdots=\bar{\mathbf{h}}_{1n_2}\bar{\mathbf{v}}^{[{\sf u}]}_{n_2k}$ ] for all @xmath104 $ ] . specifically , we first construct @xmath103}_{1k}\}_{k\in[1:m_2]}$ ] as a set of @xmath19 linearly independent random vectors . then , for a given @xmath103}_{1k}\}_{k\in[1:m_2]}$ ] , we construct @xmath106}_{jk}=(\bar{\mathbf{h}}_{1j})^{-1}\bar{\mathbf{h}}_{11}\bar{\mathbf{v}}^{[{\sf u}]}_{1k}$ ] for all @xmath104,j\in[2:n_2]$ ] . by such construction , the resulting @xmath103}_{jk}\}_{k\in[1:m_2],j\in[1:n_2]}$ ] are linearly independent almost surely . we now move on to the decoding step at the dl user . due to the previous ia procedure of the ul users , the number of dimensions occupied by the inter - user interference signals is given by @xmath19 . furthermore , the dl signals sent by the bs occupy @xmath99 dimensions and are linearly independent of the inter - user interference signals almost surely . hence , the dl user is able to decode its intended information symbols achieving one dof each . next , consider decoding at the bs . since @xmath103}_{jk}\}_{k\in[1:m_2],j\in[1:n_2]}$ ] are linearly independent , @xmath107}_{jk}\}_{k\in[1:m_2],j\in[1:n_2]}$ ] are also linearly independent almost surely . hence , the bs is able to decode the @xmath108 information symbols . finally , from the fact that a total of @xmath109 information symbols are communicated over @xmath21 time slots , @xmath110 is achievable for the case @xmath91 . fd - bs hd - user cellular network , where for convenience we assume @xmath111 in the figure . ] following the intuition in the previous subsection , with ia , we would like to confine the interference signals transmitted from multiple ul users into a preserved signal subspace at each dl user , leaving the rest of subspace for the intended signals sent from the bs . for general @xmath20 , this requires arbitrarily large number of symbol extensions @xcite . for this purpose , a recently developed ia technique in @xcite for the multiantenna ul dl cellular network can be applied for the @xmath42 fd - bs hd - user cellular network . to show how the scheme in @xcite fits into our problem , we begin with a brief overview of their network model . in @xcite , the authors consider a ul dl cellular network ( figure [ fig : uldl_model ] ) , where two cells co - exist ( each cell consists of one bs and a set of users ) . in one cell , a bs with @xmath18 antennas transmits to a set of @xmath20 dl users , while in the other cell a set of ul users transmit to a bs with @xmath19 antennas . thus , the network models the case when it can schedule each cell in dl or ul phase separately . the structural similarity with our fd - bs hd - user cellular network is apparent , and the key difference between them is that there is no inter - cell interference between the dl bs and ul bs ( since in the fd - bs hd - user cellular network , ul and dl is performed with a single fd bs ) . accordingly , the transmit signal vector of the dl bs in the ul dl model ( figure [ fig : uldl_model ] ) can also be used as the transmit signal vector of the fd bs in the fd - bs hd - user cellular network ( figure [ fig : model1 ] ) , and the transmit signal of each ul user in the ul dl model ( figure [ fig : uldl_model ] ) can also be used by each ul user in the fd - bs hd - user cellular network ( figure [ fig : model1 ] ) . therefore , the ia scheme stated in ( * ? ? ? * section iv - e ) is applicable to the @xmath42 fd - bs hd - user cellular network . however , due to the self - interference suppression capability in the fd bs case , the _ performance _ resulting from this scheme will be different for the two networks , and our contribution for achievability lies in the analysis of the sum dof of the scheme for the fd - bs hd - user cellular network . for completeness and better understanding , we briefly summarize how the ia scheme in ( * ? ? ? * section iv - e ) can be adapted to the @xmath42 fd - bs hd - user cellular network . we then give the analysis of its achievable sum dof . communication takes place over a block of @xmath112 time slots , i.e. , @xmath112 symbol extension . denote @xmath113 for @xmath24 $ ] and @xmath41 $ ] . each information symbol is transmitted through a length-@xmath112 time - extended beamforming vector . figure [ fig : scheme1_model1 ] is a conceptual illustration for this transmit beamforming . we refer to ( * ? ? ? * section iv - e ) for the detailed construction of beamforming vectors . suppose that @xmath114 $ ] and @xmath115 as @xmath112 increases . for @xmath24 $ ] , the bs sends @xmath116 information symbols to dl user @xmath25 using the set of @xmath112 time - extended beamforming vectors @xmath117}\}_{k\in[1:\lambda_1t(1-\epsilon_t)]}$ ] . similarly , ul user @xmath33 sends @xmath118 information symbols to the bs using the set of @xmath112 time - extended beamforming vectors @xmath119}\}_{k\in[1:\lambda_2t(1-\epsilon_t)]}$ ] , where @xmath41 $ ] . as seen in figure [ fig : scheme1_model1 ] , the set of beamforming vectors transmitted from each ul user is set to align its interference at each dl user . more specifically , by applying asymptotic ia for @xmath120}_{jk}\big\}_{j\in[1:n_2],k\in[1:\lambda_2 t(1-\epsilon_t)]}$ ] , we can guarantee that @xmath121}_{jk}\big\}_{j\in[1:n_2],k\in[1:\lambda_2 t(1-\epsilon_t)]}\right)$ ] occupies at most @xmath122 dimensional subspace in @xmath112 dimensional signal space for all @xmath24 $ ] almost surely in the limit of large @xmath112 , where @xmath115 as @xmath112 increases , see also ( * ? ? ? * lemma 2 ) . then the set of beamforming vectors transmitted from the bs is set to null out its interference at each dl user . more specifically , @xmath101}_{ik}\}_{i\in[1:n_1],k\in[1:\lambda_1t(1-\epsilon_t)]}$ ] is set to satisfy @xmath123}_{jk}\perp\operatorname{span}\left(\big\{\bar{\mathbf{g}}_{i}\bar{\mathbf{v}}^{[{\sf t(1-\epsilon_t)]}\right)$ ] for all @xmath124 $ ] satisfying @xmath125 and @xmath126 $ ] , i.e. , zero - forcing is performed using @xmath18 transmit antennas . in order to apply such dl interference nulling , @xmath127 should be satisfied . again , as seen in figure [ fig : scheme1_model1 ] , for reliable decoding at each dl user achieving one dof for each information symbol , @xmath128 should be satisfied . similarly , for reliable decoding at the bs achieving one dof for each information symbol , @xmath129 should be satisfied . therefore , the proposed scheme is able to deliver @xmath130 information symbols over @xmath112 time slots under the constraints to . finally , from the fact that @xmath115 as @xmath112 increases , its achievable sum dof is represented by the following optimization problem : @xmath131 region and the extreme points attaining the maximum sum dof . ] in the following , we prove that the sum dof attained by solving is given as @xmath67 stated in theorem [ thm : dof_1 ] . the linear program in is divided into five cases depending on the feasible region of @xmath132 as depicted in figure [ fig : feasible ] . obviously , one of the corner points , which are marked as points in figure [ fig : feasible ] , provides the maximum sum dof . hence , the maximum sum dof attained from is given by @xmath133 the following lemma then shows that is represented as @xmath67 in theorem [ thm : dof_1 ] , which completes the achievability proof of theorem [ thm : dof_1 ] . [ lemma : sum_dof_conversion ] the sum dof in is represented as @xmath134 for notational simplicity , denote @xmath135 then denote @xmath136 . in the following , we show that for each of the five cases in , @xmath137 is represented as in the corresponding dof expression in . * case i ( @xmath138 ) : obviously , @xmath139 . for @xmath140 , @xmath141 . for @xmath142 , @xmath143 . hence @xmath144 . in conclusion , @xmath145 for case i. * case ii ( @xmath146 ) : first consider the case where @xmath140 . then @xmath147 . also @xmath148 . since @xmath149 , @xmath150 . hence @xmath151 . next consider the case where @xmath142 . then @xmath152 and @xmath153 . also @xmath143 . hence @xmath154 . finally , from the relation that @xmath155 for @xmath140 and @xmath156 for @xmath142 , @xmath157 for case ii . * case iii ( @xmath158 ) : from the symmetric relation with case ii , @xmath159 for case iii . * case iv ( @xmath160 ) : the condition @xmath161 means that @xmath162 and @xmath163 . hence @xmath164 and @xmath165 , which show @xmath166 . for @xmath140 , @xmath167 and @xmath168 . similarly , @xmath169 and @xmath170 for @xmath142 hence @xmath171 . in conclusion , @xmath172 for case iv . * case v ( @xmath173 ) : for @xmath140 , @xmath174 and then @xmath175 . similarly , @xmath176 for @xmath142 . hence @xmath177 . the condition @xmath178 means that @xmath179 and @xmath180 . then @xmath181 and @xmath182 . hence @xmath183 . in conclusion , @xmath184 for case v. in conclusion , @xmath137 is represented as the corresponding sum dof in for all five cases , which completes the proof . in this section , we prove the converse of theorems [ thm : dof_1 ] and [ thm : dof_2 ] . recall the encoding and decoding functions of the fd bs and each fd user in section [ subsec : dof_definition ] . the key observation is that the received signals available for encoding the dl messages at the fd bs and the dl messages available for decoding the ul messages at the fd bs can not increase the sum dof . similarly , the received signals available for encoding its ul message at each fd user and its ul message available for decoding its dl message at each fd user can not increase the sum dof . to prove the converse of theorem [ thm : dof_1 ] , we introduce the two - user mimo z - ic with output feedback for encoding and message side information for decoding depicted in figure [ fig : mimo_zic ] . the received signal vectors of receivers 1 and 2 at time @xmath26 are respectively given by @xmath185 where @xmath186 , @xmath187 , and @xmath188 denote the channel matrices from transmitter 1 to receiver 1 , from transmitter 2 to receiver 1 , and from transmitter 2 to receiver 2 , respectively . the rest of the assumptions are the same as those of the fd - bs hd - user cellular network in section [ sebsec : fd_bs_hd_user ] . obviously , the capacity of the two - user mimo z - ic is an outer bound on the capacity of the @xmath42 fd - bs hd - user cellular network , since it corresponds to the fd - bs hd - user cellular network with full cooperation among the dl users and among the ul users . [ lemma : mimo_zic ] consider the two - user mimo z - ic with output feedback for encoding and message side information for decoding in figure [ fig : mimo_zic ] . then the dof region is given by the set of all dof pairs ( @xmath189 ) satisfying @xmath190 the achievability immediately follows from that in ( * ? ? ? * theorem 1 ) , which corresponds to the two - user mimo z - ic without output feedback for encoding and message side information for decoding . next , we show the converse . obviously @xmath191 and also @xmath192 since side information of @xmath193 at receiver 2 can not increase the dof more than @xmath194 , which gives . now substitute @xmath20 antennas with @xmath195 antennas at receiver 1 . assume that both receivers are able to recover @xmath193 and @xmath196 respectively with arbitrarily small probabilities of error . then , after subtracting @xmath197 from @xmath198 ( @xmath197 is obtained from re - encoding @xmath193 ) , receiver 1 constructs @xmath199 , where @xmath200 . since receiver 2 recovers @xmath196 from @xmath201 , where @xmath202 , receiver 1 can also recover @xmath196 from @xmath203 from the fact that @xmath204 . as a result , receiver 1 is able to decode both @xmath193 and @xmath196 with @xmath195 antennas . because output feedback can not increase the sum dof of the mimo multiple - access channel ( mac ) , @xmath205 , which provides . in conclusion , lemma [ lemma : mimo_zic ] holds . since the sum dof of the @xmath42 fd - bs hd - user cellular network is upper bounded by the sum dof of the two - user mimo z - ic , @xmath206 from lemma [ lemma : mimo_zic ] , which is yet not enough to show the converse . in a more refined way of applying lemma [ lemma : mimo_zic ] , we prove the converse of theorem [ thm : dof_1 ] in the following . denote @xmath207}_{\sigma}=\sum_{i=1}^{n_1}d^{[\sf d]}_i$ ] and @xmath208}_{\sigma}=\sum_{j=1}^{n_2}d^{[\sf u]}_j$ ] . first consider the case where @xmath140 . for this case , choose a subset of dl users in @xmath209}\in[1:n_1]$ ] satisfying @xmath210})=n_2 $ ] . then , by applying lemma [ lemma : mimo_zic ] only for the dl users in @xmath209}$ ] ( and for the entire ul users ) , we have @xmath211}}d^{[\sf d]}_i+d^{[\sf u]}_{\sigma}\leq n_2.\end{aligned}\ ] ] by summing over all possible @xmath209}$ ] satisfying @xmath210})=n_2 $ ] , we have @xmath212}_{\sigma}+n_1d^{[\sf u]}_{\sigma}\leq n_1n_2.\end{aligned}\ ] ] therefore , @xmath213}_{\sigma}\leq \min(m_1,n_1)\\ d^{[\sf u]}_{\sigma}\leq \min(m_2,n_2)\\ n_2d^{[\sf d]}_{\sigma}+n_1d^{[\sf u]}_{\sigma}\leq n_1n_2}}\{d^{[\sf d]}_{\sigma}+d^{[\sf u]}_{\sigma}\}.\end{aligned}\ ] ] now consider the case where @xmath142 . for this case , choose a subset of ul users in @xmath214}\in[1:n_2]$ ] satisfying @xmath215})=n_1 $ ] . then applying lemma [ lemma : mimo_zic ] for all possible @xmath214}$ ] satisfying @xmath215})=n_1 $ ] and summing them provides the same upper bound in . as a result , also holds for @xmath142 . by solving the linear program in a similar manner as in section [ sec : achievability ] , we have @xmath216 note that the upper bound in is exactly the same as in . therefore , from lemma [ lemma : sum_dof_conversion ] , @xmath217 which completes the converse proof of theorem [ thm : dof_1 ] . in this subsection , we prove the converse of theorem [ thm : dof_2 ] . we first show that @xmath218 in section [ subsec : two_way_bound ] and then show that @xmath219 in section [ subsec : x_network_bound ] . combining the above two bounds , we have the desired bound @xmath220 , which completes the converse proof . by allowing full cooperation among the @xmath44 users in the @xmath48 fd - bs fd - user cellular network , we obtain a mimo two - way network depicted in figure [ upperbound_twoway ] . clearly , the considered mimo two - way network provides an upper bound on @xmath74 . therefore , from the result in @xcite , we have @xmath221 users . ] we now prove @xmath219 by using the result of four - node x networks in @xcite . in order to apply the result in @xcite , we convert the original @xmath48 fd - bs fd - user cellular network into the corresponding four - node x network as follows : * * step 1 * : we first transform the @xmath222 fd - bs fd - user cellular network into the equivalent two - cell cellular network consisting of one dl cell and one ul cell depicted in figure [ upperbound_xnetwork_step1 ] . specifically , the fd bs is decomposed into the bs@xmath223 and the bs@xmath224 and fd user @xmath25 is decomposed into user@xmath225 and user@xmath226 , where @xmath72 $ ] . there exists output feedback from the bs@xmath224 to the bs@xmath223 and from user@xmath226 to user@xmath225 for all @xmath72 $ ] , which can be used as side information for encoding . in addition , @xmath22}_1\cdots , w^{[{\sf d}]}_n)$ ] is available at the bs@xmath224 and @xmath51}_i$ ] is available at user@xmath226 for all @xmath72 $ ] , which can be used as side information for decoding . we refer to the encoding and decoding functions in section [ sebsec : dof_fd_bs_fd_user ] . the channel coefficients from the bs@xmath223 to the bs@xmath224 and from user@xmath225 to user@xmath226 are set to zeros due to perfect self - interference suppression in the original network . the validity of this transformation is also proved by ( * ? ? ? * lemma 1 ) . + * * step 2 * : as shown in figure [ upperbound_xnetwork_step2 ] , we allow full cooperation among user@xmath227 and user@xmath228 , among bs@xmath223 , user@xmath229 to user@xmath230 , among bs@xmath224 , user@xmath231 to user@xmath232 , and among user@xmath233 and user@xmath234 , each of which is called nodes 1,2,3 , and 4 respectively . because of such cooperation , the set of @xmath22}_2,\cdots , w^{[{\sf d}]}_{n-1},w^{[{\sf u}]}_2,\cdots , w^{[{\sf u}]}_{n-1})$ ] is priorly known at node 3 as side information , so that node 3 is able to attain those messages without communication . hence , we delete those messages in the figure without loss of generality . in the end , node 1 wishes to send @xmath235}$ ] and estimate @xmath236}$ ] , node 2 wishes to send @xmath237},w_n^{[{\sf d}]})$ ] with the help of output feedback from node 3 , i.e. , the set of all output signals received by the components consisting of node 3 , node 3 wishes to estimate @xmath238},w_n^{[{\sf u}]})$ ] with the help of message side information @xmath237},w_n^{[{\sf d}]})$ ] , and node 4 wishes to send @xmath239}$ ] and estimate @xmath240}$ ] . since the network in figure [ upperbound_xnetwork_step2 ] assumes cooperation between some nodes and allow more information for encoding and decoding , it provides an outer bound on the dof region of the network in figure [ upperbound_xnetwork_step1 ] . + } $ ] and @xmath241}$ ] and create more links and output feedback . ] * * step 3 * : we now focus on an upper bound on @xmath208}_1+d^{[\sf d]}_n$ ] . we first eliminate all the messages except @xmath242}$ ] and @xmath241}$ ] , which does not decrease @xmath208}_1+d^{[\sf d]}_n$ ] @xcite . then we provide @xmath243 receive antennas at node 2 and @xmath244 transmit antennas at node 3 and allow fd operation at all nodes , which creates more links illustrated as dashed links in figure 13 . we further assume that output feedback from nodes 3 and 4 is available at nodes 1 and 2 . obviously , adding more antennas at some nodes , allowing fd operation , and providing more output feedback for encoding do not decrease @xmath208}_1+d^{[\sf d]}_n$ ] . as a result , the converted network in figure [ upperbound_xnetwork_step3 ] provides an upper bound on @xmath208}_1+d^{[\sf d]}_n$ ] achievable by the original ( @xmath18 , @xmath19 , @xmath44 ) fd - bs fd - user cellular network . note that the converted network in figure [ upperbound_xnetwork_step3 ] corresponds to the four - node x network studied in @xcite except the fact that @xmath241}$ ] is provided to node 3 through a genie . as stated in ( * ? ? ? * section iv ) , providing this side information does not increase the sum dof and , therefore , we have @xmath208}_1+d^{[\sf d]}_n\leq 1 $ ] from the result in @xcite . in the same manner , we can establish @xmath245}_i+d^{[\sf d]}_n\leq j\end{aligned}\ ] ] for @xmath246 $ ] with @xmath125 . by summing for all @xmath246 $ ] with @xmath125 , we finally have @xmath247}_{i}+\sum_{i=1}^{n}d^{[\sf d]}_{i}\leq n.\end{aligned}\ ] ] in this section , we briefly discuss about the impacts of self - interference and ul and dl scheduling on dof . , @xmath248 , and @xmath249 . ] throughout the paper , we assumed that there is no self - interference within the bs during fd operation . however , in a practical fd bs , the amount of residual self - interference may not be negligible due to insufficient self - interference suppression or imperfect self - interference cancellation from the priorly known message information at the receiver side @xcite . in this subsection , we will discuss the impacts of such self - interference on the sum dof . note that when there exists self - interference within the bs of the @xmath42 fd - bs hd - user cellular network , the sum dof is given by @xmath250 from the result of @xcite , by interpreting inter - bs interference in @xcite as self - interference within the bs . obviously , if we restrict for the bs to operate either ul or dl only , then the sum dof is given by @xmath251 to see the effect of self - interference on the sum dof , let us consider the case where @xmath252 , @xmath248 , and @xmath249 as an example . we plot the sum dofs as a function of the number of total users @xmath253 in figure [ dof_fullduplex ] . for comparison , we also plot the sum dof of the fd - bs fd - user cellular network when the number of fd users is given by @xmath44 . as shown in the figure , the fd - bs hd - user cellular network is able to achieve the same sum dof attained by the fd - bs fd - user cellular network when @xmath44 is large enough . however , fd capability at the user side is beneficial to improve the sum dof for small @xmath44 . interestingly , even when there exists self - interference , fd operation at the bs alone can increase the sum dof in a certain regime . however , the sum dof collapses to that of the hd - bs hd - user cellular network when @xmath44 is large enough . note that similar tendencies can be observed for general @xmath42 . therefore , from these observations , self - interference suppression or cancellation is of crucial importance for fully utilising the potential of fd networks . in this subsection , we discuss the effects of hd user scheduling on the sum dof . suppose that there exist total @xmath44 hd users and we are able to coordinate the operational mode of each of these users , i.e. , dividing them into @xmath20 dl users and @xmath21 ul users , where @xmath254 . obviously , the sum dof varies with the values of @xmath20 and @xmath21 from theorem [ thm : dof_1 ] . , @xmath248 , and @xmath255 . ] and @xmath248 . ] as an example , consider again the case where @xmath252 and @xmath248 . first , we fix the total number of users @xmath44 ( @xmath256 ) as 50 and plot the sum dof of the fd - bs hd - user cellular network with and without self - interference suppression as a function of @xmath20 in figure [ dof_uplink ] . for comparison , we also plot the sum dofs of the fd - bs fd - user cellular network and the hd - bs hd - user cellular network . as depicted in figure [ dof_uplink ] , except the fd - bs fd - user cellular network , the achievable sum dofs vary with @xmath20 , and we can maximize the sum dof of each network by optimally choosing @xmath20 and @xmath21 . now , we plot the sum dofs as a function of the number of total users @xmath44 in figure [ dof_uplink2 ] . here , for each @xmath44 , we choose @xmath20 and @xmath21 to achieve the optimal sum dofs . as seen in figure [ dof_uplink2 ] , when there is no self - interference , the optimal sum dof of the fd - bs hd - user cellular network approaches to that of the fd - bs fd - user cellular network and reaches the same sum dof when @xmath44 is large enough . however , when there exists self - interference , the optimal sum dofs of the fd - bs hd - user cellular network and the hd - bs hd - user cellular network are the same for any @xmath44 . this statement is also true for general @xmath18 and @xmath19 since the optimal scheduling for the fd - bs hd - user cellular network with self - interference is to operate all hd users as either ul or dl , which can be easily verified from . therefore , for the case in which the optimal scheduling is allowed , fd operation at the bs is not required in terms of dof if there exists self - interference . in this paper , we have studied the sum dofs of cellular networks with a multiantenna fd bs and hd mobile users and with a multiantenna fd bs and fd mobile users . for our main contribution , we have completely characterized the sum dofs of these networks . to be specific , for achievability , the key idea was to fully utilize the intended signal dimensions by minimizing the inter - user interference dimensions via ia for the ul transmission and by minimizing the intra - cell interference dimensions via multiantenna nulling for the dl transmission . for converse , we have provided a matching upper bound that shows the optimality of the proposed scheme . as a consequence of the result , we have shown that even when inter - user interference exists , fd operation at the bs can double the sum dof over the hd only networks when the number of users becomes large enough as compared to the number of antennas at the bs , for both the fd - bs hd - user cellular network and the fd - bs fd - user cellular network . our work can be extended to several interesting directions : ( 1 ) extending to multi - cell scenarios in which inter - cell interference exists ; ( 2 ) extending to the case in which mobile users have multiple antennas ; ( 3 ) extending to the cases in which channel state information at transmitters ( csit ) is not available or delayed . j. i. choi , m. jain , k. srinivasan , p. levis , and s. katti , `` achieving single channel , full duplex wireless communication , '' in _ proc . 16th annual international conference on mobile computing , networking , and communications ( mobicom ) _ , new york , ny , aug . 2010 . e. aryafar , m. a. khojastepour , k. sundaresan , s. rangarajan , and m. chiang , `` midu : enabling mimo full duplex , '' in _ proc . 18th annual international conference on mobile computing , networking , and communications ( mobicom ) _ , istanbul , turkey , aug . 2012 . m. duarte and a. sabharwal , `` full - duplex wireless communications using off - the - shelf radios : feasibility and first results , '' in _ proc . asilomar conference on signals , systems and computers _ , pacific grove , ca , nov . 2010 . m. jainy , j. i. choi , t. m. kim , d. bharadia , s. seth , k. srinivasan , p. levis , s. katti , and p. sinha , `` practical , real - time , full duplex wireless , '' in _ proc . 17th annual international conference on mobile computing , networking , and communications ( mobicom ) _ , las vegas , nv , sep . 2011 . s. hong , j. brand , j. choi , m. jain , j. mehlman , s. katti , and p. levis , `` applications of self - interference cancellation in 5 g and beyond , '' _ ieee communications magazine _ , vol . 52 , pp . 114121 , feb . 2014 . m. a. maddah - ali , a. s. motahari , and a. k. khandani , `` communication over mimo x channels : interference alignment , decomposition , and performance analysis , '' _ ieee trans . inf . theory _ 34573470 , aug . 2008 . v. s. annapureddy , a. el gamal , and v. v. veeravalli , `` degrees of freedom of interference channels with comp transmission and reception , '' _ ieee trans . inf . theory _ , 57405760 , sep . 2012 . t. gou , s. a. jafar , c. wang , s .- w . jeon , and s .- y . chung , `` aligned interference neutralization and the degrees of freedom of the @xmath259 interference channel , '' _ ieee trans . inf . theory _ 43814395 , jul . 2012 . w. shin , n. lee , j .- b . kim , c. shin , and k. jang , `` on the design of interference alignment scheme for two - cell mimo interfering broadcast channels , '' _ ieee trans . wireless commun . _ , vol . 437442 , feb .

we investigate the radiative transition in the framework of qcd sum rules . in particular , we calculate the transition form factors responsible for this decay in both weak annihilation and electromagnetic penguin channels using the quark condensate , mixed and two - gluon condensate diagrams as well as propagation of the soft quark in the electromagnetic field as non - perturbative corrections . these form factors are then used to estimate the branching ratios of the channels under consideration . the total branching ratio of the transition is obtained to be in order of , and the dominant contribution comes from the weak annihilation channel . pacs numbers : 11.55.hx , 13.20.-v , 13.20.he

1207.1676

the @xmath2 is the only heavy meson consisting of two heavy quarks with different flavors , hence the decay properties of this meson are of special interest . the difference in heavy quark flavors forbids annihilation of this meson into gluons , so the excited @xmath2 states undergo pionic or radiative transition to the pseudoscalar ( ps ) ground state when these states lie below the threshold of the decay into the pair of heavy @xmath3 and @xmath4 mesons . the resulting ps ground state is more stable compared to the corresponding quarkonia and decays mostly weakly . because of this phenomenon , it is expected that the experimental study of the @xmath2 meson and its decay properties will constitute an important part of the physics program at lhcb . the study of the heavy mesons will not only provide a window in extracting the most accurate values of the cabbibo- kobayashi- maskawa ( ckm ) matrix elements as the sources of the @xmath5-violation in the standard model ( sm ) but also will help us better understand the perturbative and non - perturbative aspects of qcd . in the present study , we work out the rare radiative @xmath6 transition in the framework of the qcd sum rules @xcite . here , the @xmath7 is the axial vector charmed - strange meson with quantum numbers @xmath8 and the interpolating current @xmath9 . this transition proceeds via both weak annihilation ( wa ) and electromagnetic penguin ( ep ) of flavor changing neutral current ( fcnc ) transition , based on the @xmath10 at quark level . we calculate the transition form factors responsible for this decay in both wa and ep modes using the quark condensate , mixed and two - gluon condensate diagrams , as well as propagation of the soft quark in the electromagnetic field as non - perturbative corrections . we then use these form factors to estimate the branching ratios in both modes as well as the total branching fraction of the @xmath6 transition . as expected , the dominant contribution comes from the weak annihilation channel . note that similar decays like the @xmath11 transition have been studied in the same framework @xcite . some other radiative channels of the @xmath12 meson like , @xmath13 and @xmath14 have also been previously studied using the qcd sum rules technique @xcite . for analysis of other decay channels of the @xmath12 meson see , for instance , @xcite . the outline of the paper is as follows . in section ii , we consider the radiation of the photon from both @xmath12 and @xmath15 mesons , to construct the transition amplitude for the wa channel in terms of four relevant form factors . two of the form factors @xmath16 and @xmath17 , responsible for the emission of the photon from the initial state , are calculated in @xcite , and the remaining two form factors @xmath18 and @xmath19 , representing the emission of the photon from @xmath15 meson , are calculated in section iii . in section iv , we consider the two gluon condensate contributions , to calculate the transition form factors responsible for the ep mode . finally , section v is devoted to the numerical analysis of the form factors and , calculation of the decay rates and branching ratios for the modes under consideration . we also present results for the total decay rate and branching ratio of the @xmath6 transition . this section also contains our concluding remarks . in this section , we construct the wa amplitude for the radiative @xmath0 transition . considering the quark contents of the initial and final mesonic states , the possible diagrams are shown in figure 1 . taking into account these diagrams , the transition amplitude for the radiative decay under consideration is written as @xmath20 where @xmath21 is the fermi weak coupling constant , @xmath22 are elements of the ckm matrix , @xmath23 ; and @xmath24 , @xmath25 and @xmath26 are the momenta of the @xmath15 meson , photon and @xmath12 meson , respectively . to proceed further , we use the factorization hypothesis and write the transition matrix element in eq.([weakamp ] ) as @xmath27 where we have divided the matrix element into two separate parts : the emission of the photon from the @xmath12 meson ( diagrams ( i ) and ( ii ) in figure 1 ) , represented by the covariant tensor @xmath28 and the emission of the photon from the @xmath15 meson denoted by the tensor @xmath29 ( see diagrams ( iii ) and ( iv ) in figure 1 ) . in eq.([2 ] ) , @xmath30 ( @xmath31 ) is the decay constant of the @xmath12 ( @xmath15 ) meson and @xmath32 ( @xmath33 ) is the polarization vector of the photon ( @xmath15 meson ) . the covariant tensors @xmath28 and @xmath29 are defined as @xmath34 @xmath35 where @xmath36 is the electromagnetic current and @xmath37 is the time ordering operator . applying the ward identity for the electromagnetic current , using @xmath38 for the real photon , @xmath39 and @xmath40 , similar to what is done in @xcite , we get the following results corresponding to the emission of the photon from the initial and final mesonic states in terms of form factors : @xmath41if_{a}^{(b_{c})}\nonumber \\&&+if_{b_{c}}\big(\varepsilon.\varepsilon^{(d_{s1})}\big ) + \varepsilon_{\nu\mu\lambda\sigma}\varepsilon^{(d_{s1})_{\nu}}\varepsilon^{\mu}p^{\lambda}q^{\sigma}f_{v}^{(b_{c})}\bigg\},\end{aligned}\ ] ] @xmath42if_{a}^{(d_{s1})}\nonumber \\&+&f_{d_{s1}}m_{d_{s1}}\big(\varepsilon.\varepsilon^{(d_{s1})}\big ) + \varepsilon_{\nu\mu\lambda\sigma}\varepsilon^{(d_{s1})_{\nu}}\varepsilon^{\mu}p^{\lambda}q^{\sigma}f_{v}^{(d_{s1})}\bigg\},~~\end{aligned}\ ] ] where @xmath43 and @xmath44 are the transition form factors . using eqs ( [ first term ] ) , ( [ second term final ] ) and ( [ 2 ] ) we find the wa transition amplitude to be @xmath45if_{a}^{(b_{c } ) } + if_{b_{c}}\big(\varepsilon.\varepsilon^{(d_{s1})}\big ) \notag \\ & & \ + \varepsilon_{\nu\mu\lambda\sigma}\varepsilon^{(d_{s1})\nu}\varepsilon^{\mu}p^{\lambda}q^{\sigma}f_{v}^{(b_{c})}\bigg\ } -if_{b_{c}}\bigg\{\bigg[\big(\varepsilon.\varepsilon^{(d_{s1})}\big)(p.q)-(\varepsilon.p)\big(\varepsilon^{(d_{s1})}.q\big)\bigg]if_{a}^{(d_{s1})}\notag \\ & & \ + f_{d_{s1}}m_{d_{s1 } } \big(\varepsilon.\varepsilon^{(d_{s1})}\big ) + \varepsilon_{\nu\mu\lambda\sigma}\varepsilon^{(d_{s1})\nu}\varepsilon^{\mu}p^{\lambda}q^{\sigma}f_{v}^{(d_{s1})}\bigg\}\bigg).\end{aligned}\ ] ] as mentioned in section i , the form factors @xmath16 and @xmath17 are calculated in @xcite , so what remains to be calculated are the form factors @xmath18 and @xmath19 , which we discuss in the next section . to calculate the transition form factors @xmath18 and @xmath19 via qcd sum rules formalism , we start considering the following correlation function : @xmath46 where @xmath47 . the basic idea in this method is to calculate this correlation function first in hadronic language , called phenomenological or physical side and second in terms of the qcd degrees of freedom using the operator product expansion in deep euclidean space , called the theoretical or qcd side . the two representations are then matched in order to get the qcd sum rules for the form factors . to suppress the contributions coming from the higher energy states and continuum we apply a borel transformation as well as continuum subtraction which bring two auxiliary parameters : namely the borel mass parameter and the continuum threshold . we shall find their working regions requiring that the physical observables be independent of these parameters . first , we focus on calculation of the phenomenological side . for this aim , we insert a full set of hadronic @xmath15 state into eq.([cf1 ] ) and perform the four - integral over @xmath48 to get @xmath49 the matrix element , @xmath50 is defined in terms of the decay constant and the polarization vector of the @xmath15 meson as @xmath51 while the transition matrix element is parametrized in terms of form factors , @xmath52\frac{f_{a}^{(d_{s1})}(q^2)}{m_{(d_{s1})}^2}\bigg\}.\end{aligned}\ ] ] substituting eqs . ( [ mtrx1 ] ) and ( [ mtrx2 ] ) into eq . ( [ cf2 ] ) and summing over the polarization vector of the @xmath15 meson , we find the following result for the phenomenological part of the correlation function : @xmath53\frac{f_{a}^{(d_{s1})}(q^2)}{m_{d_{s1}}^2}\bigg\}.\end{aligned}\ ] ] we now compute the qcd side of the correlation function within the deep euclidean region in terms of the qcd parameters . we start by writing the correlation function in terms of the two selected structures as @xmath54\pi_{2},\ ] ] where each function @xmath55 or @xmath56 has perturbative and non - perturbative parts , i.e. , @xmath57 to calculate the perturbative parts , we consider diagrams ( a ) and ( b ) in figure 2 where the photon can be radiated from both the charm and strange quarks . for the non - perturbative parts , we take into account the quark condensate and mixed diagrams [ diagrams 2 ( c ) , 2 ( d ) and 2 ( e ) ] as well as diagram 2 ( f ) for the interaction of the photon with the soft quark . the perturbative part in each case can be written via the dispersion relation as @xmath58 where @xmath59 are the spectral densities . our main task is now to calculate these spectral densities using the diagrams ( a ) and ( b ) in figure 2 . here , we use a method based on both feynman and schwinger parameterizations with several borel transformations ( see also @xcite ) . the feynman amplitude for the diagram ( a ) can be written as @xmath60\bigg\},\ ] ] where @xmath61 is the number of colors and @xmath62 is the charge of the strange quark . using the feynman parameterization , we perform the four - integral over @xmath63 and then we use the schwinger parameterization @xmath64 to write the denominators in exponential forms . as a result , we get @xmath65\int_{0}^{\infty } d\alpha e^{-\alpha \delta}\bigg\},\ ] ] @xmath66\int_{0}^{\infty } d\alpha e^{-\alpha \delta}\bigg\},\ ] ] where @xmath67 , and @xmath68 . applying a double borel transformation @xmath69 on @xmath70 , that transforms @xmath71 and @xmath72 , we obtain @xmath73 @xmath74 where @xmath75 and we have used @xmath76 now , we perform a second double borel transformation on @xmath77 in order to transform @xmath78 and @xmath79 to the new variables @xmath80 and @xmath81 using @xmath82 in the calculations , we also use the relations @xmath83 and @xmath84 the final expressions for the spectral densities are then calculated via the following formula : @xmath85 after lengthy calculations , we get the following spectral densities corresponding to the diagram ( a ) : @xmath86- m_s^2 x^2\overline{x } \big[(x-10 ) q^2 \notag \\ & & \ + ( x-2 ) ( s - q^2)\big ] -4 x^2 \overline{x}^2 q^2(s - q^2 ) -4 x^2 \overline{x}^2 q^4\bigg\},\nonumber\\\end{aligned}\ ] ] @xmath87 + m_s^2 x^2\overline{x } \big[(x-10 ) q^2 \notag \\ & & \ + ( x-2)(s - q^2)\big ] + 4 x^2 \overline{x}^2 q^2(s - q^2 ) + 4 x^2 \overline{x}^2 q^4 \bigg\},\end{aligned}\ ] ] where the integral boundaries @xmath88 and @xmath89 satisfy the following inequality : @xmath90 which comes from the definition of the heaviside - theta function arising in these calculations . similarly , we calculate the contribution of the diagram 2(b ) . the final expressions for the spectral densities corresponding to the two selected structures are @xmath91\bigg ) \notag \\ & & \ + 2 \bigg(4 m_c m_s ( s - q^2 ) + 8\alpha s q^2 + \alpha ( 1 - 4\alpha + 9\beta)s^2\bigg ) \ln\big(\frac { 1 + \alpha - \beta - \lambda } { 1 + \alpha - \beta + \lambda}\big)\bigg ] \notag \\ & & \ + q_c \bigg [ \lambda \bigg(4(5 \beta - 5 \alpha-1)sq^2 + s^2 \big[3 \beta ( \beta-3)- \alpha(6 \beta+1 ) + 3 \alpha^2\big]\bigg ) \notag \\ & & \ + 2 \bigg(4 m_c m_s ( s - q^2 ) + 8\beta s q^2 + \beta ( 1 - 4\beta + 9\alpha)s^2\bigg ) \ln\big(\frac { 1 + \beta - \alpha - \lambda } { 1 + \beta - \alpha + \lambda}\big)\bigg]\bigg\},\end{aligned}\ ] ] @xmath92\bigg ) \notag \\ & & \ + 2 \bigg(4 m_c m_s ( s - q^2 ) - 8 s \alpha q^2 - \alpha ( 1 - 4 \alpha + 9 \beta)s^2\bigg ) \ln\big(\frac { 1 + \alpha - \beta - \lambda } { 1 + \alpha - \beta + \lambda}\big)\bigg ] \notag \\ & & \ + q_c \bigg [ \lambda \bigg(4(- 5 \beta + 5 \alpha+1)s q^2 + s^2 \big [ -3\beta(\beta-3)+ \alpha(6 \beta+1)- 3\alpha^2\big]\bigg ) \notag \\ & & \ + 2 \bigg(4 m_c m_s ( s - q^2 ) - 8 s \beta q^2 - \beta ( 1 - 4 \beta + 9 \alpha)s^2\bigg ) \ln\big(\frac { 1 + \beta - \alpha - \lambda } { 1 + \beta - \alpha + \lambda}\big)\bigg ] \bigg\},\end{aligned}\ ] ] where @xmath93 , @xmath94 and @xmath95 . for the non - perturbative parts , we begin by calculatting contributions of the quark condensate and mixed diagrams ( diagrams ( c ) , ( d ) and ( e ) in figure 2 ) and obtain @xmath96\notag \\ & & \ + \frac{m_{s}^2 } { 2}\langle\overline{s}s\rangle\big[\frac { 2 m_{c}^3}{r^2 r^6 } - \frac{8 m_{c}^5}{r^6 r^4 } + \frac{2 m_{c}^3}{r^4 r^4 } - \frac { 3 m_{c}}{r^2 r^4 } + \frac{2 m_{c}^3}{r^6 r^2 } - \frac{m_{c}}{r^4 r^2}\big]\notag \\ & & \ + \frac { m_{0}^2 } { 12}\langle\overline{s}s\rangle\big[\frac{- 6 m_{c}^3}{r^2 r^6 } + \frac{24 m_{c}^5}{r^6 r^4 } - \frac { 6 m_{c}^3}{r^4 r^4 } + \frac{8 m_{c}}{r^2 r^4 } - \frac{6 m_{c}^3 } { r^6 r^2 } + \frac{3 m_{c}}{r^4 r^2}\big ] - \frac{4 m_{c}^3}{r^2 r^4}\langle\overline{s}s\rangle,\nonumber\\\end{aligned}\ ] ] @xmath97\notag \\ & & \ + \frac{m_{s}^2 } { 2}\langle\overline{s}s\rangle\big[-\frac{2 m_{c}^3}{r^2 r^6 } + \frac{8 m_{c}^5}{r^6 r^4 } - \frac{2 m_{c}^3 } { r^4 r^4 } + \frac{3 m_{c } } { r^2 r^4 } - \frac{2 m_{c}^3}{r^6 r^2 } + \frac{m_{c } } { r^4 r^2}\big]\notag \\ & & \ + \frac{m_{0}^2 } { 4}\langle\overline{s}s\rangle\big[\frac{2 m_{c}^3 } { r^2 r^6 } - \frac{8 m_{c}^5}{r^6 r^4 } + \frac{2 m_{c}^3 } { r^4 r^4 } - \frac{4 m_{c}}{r^2 r^4 } + \frac{2 m_{c}^3}{r^6 r^2 } - \frac{m_{c } } { r^4 r^2}\big ] + \frac{4 m_{c}^3}{r^2 r^4}\langle\overline{s}s\rangle,\nonumber\\\end{aligned}\ ] ] where @xmath98 and @xmath99 . the final contribution to the wa mode is that of diagram ( f ) . this diagram corresponds to the propagation of the soft quark in the external electromagnetic field . here we need to make use of the light - cone version of the qcd sum rules and photon distribution amplitudes ( das ) . the relevant correlation function is of the form : @xmath100 contracting the @xmath101-quark lines in eq.([cflc1bir ] ) and using the propagator of the heavy quark in momentum space , we obtain @xmath102 to relate the matrix element in the above equation to the photon das , we use the identities @xmath103 the relevant photon das of twist 2 , 3 , and 4 @xcite are @xmath104 where @xmath105 is the field strength tensor of the electromagnetic field and is defined by @xmath106 and @xmath107 the wave function @xmath108 is defined in terms of the magnetic susceptibility @xmath109 at a renormalization scale ( @xmath110 ) in the following manner : @xmath111 the remaining functions @xmath112 , @xmath113 , @xmath114 , and @xmath115 are also defined as @xcite @xmath116,\notag \\ \textbf{b}(u ) & = & 40\int_{0}^{u}d\alpha(4-\alpha)(1 + 3k^{+})[-\frac{1}{2}+\frac{3}{2}(2\alpha-1)^2],\end{aligned}\ ] ] where @xmath63 , @xmath117 , @xmath118 , @xmath119 and @xmath120 are constants ( see @xcite ) . putting the above equations all together and after performing the four - integrals over @xmath48 and @xmath63 , the coefficients of the corresponding structures , @xmath121 and @xmath122 $ ] are obtained as follows : @xmath123\big\},\notag \\\end{aligned}\ ] ] @xmath124\big\}.\end{aligned}\ ] ] now , to find the qcd sum rules for the form factors we match the coefficients of the selected structures from both phenomenological and qcd sides and perform the borel transformation with respect to the momentum of @xmath15 meson @xmath125 . to further suppress the contributions of the higher energy states and continuum we also perform the continuum subtraction and use the quark - hadron duality assumption and find @xmath126 where @xmath127 is the continuum threshold and the @xmath128 ( @xmath129 ) on the left - hand side corresponds to the 1 ( 2 ) on the right - hand side . to obtain the expressions for the above sum rules in the borel scheme , we perform the borel transformation using the standard rule @xmath130 at the quark level , the fcnc based ep transition of the @xmath131 proceeds via @xmath132 whose effective hamiltonian is written as @xmath133bf^{\mu\nu}.\ ] ] the amplitude of this mode is obtained from @xmath134 hence to proceed further , we need to calculate the following matrix elements : @xmath135 which can be parametrized in terms of two gauge invariant form factors @xmath136 and @xmath137 in the case of real photon , i.e. @xmath138t_{2}(0),\nonumber\\\end{aligned}\ ] ] where these two from factors are not independent from each other . using the relation , @xmath139 , we see @xmath140 . therefor , we need to calculated just one of them , and we choose to calculate the form factor @xmath141 . the corresponding correlation function is chosen as @xmath142 where @xmath143 and @xmath144 are the interpolating currents of the initial and final mesonic states , respectively , and @xmath145 is the transition current . using the general philosophy of the qcd sum rules we calculate this correlation function in two different languages : namely the hadronic language and the quark - gluon language . for the hadronic , or phenomenological side , we get @xmath146t_{2}(0)\bigg\}+ .... ,\end{aligned}\ ] ] where @xmath147 denotes contributions of the higher energy states and continuum which will be suppressed by applying the borel transformation as well as the continuum subtraction , and we have used the following definition of the decay constant of the @xmath2 meson : @xmath148 to calculate the form factor @xmath149 , we choose the structure @xmath150 . in the qcd side , the correlation function is written in terms of the selected structure as @xmath151 where @xmath152 here , the perturbative part is related to the spectral density , @xmath153 by a double dispersion integral , @xmath154 and for the non - perturbative contributions we will calculate the two - gluon condensate diagrams . now , we focus our attention on calculating the spectral density . using the cutkosky method @xcite , we get @xmath155 - \delta ' [ ( m_{b } - m_{c } ) ( m_{c } + m_{s } ) + s ' ] \notag \\ & & \ + 2 m_{c } [ ( m_{c } + m_{s } ) s ' + m_{b } t - m_{c } t ] - m_{c } ( m_{b } + m_{s } ) u\big ] + 2 a1 ( -2 s ' + u)\big\},\end{aligned}\ ] ] where @xmath156,\notag \\ \delta & = & s ' + m_{c}^2 - m_{b}^2,\notag \\ \delta ' & = & t + m_{c}^2 - m_{s}^2,\notag \\ u & = & t + s ' - q^2.\end{aligned}\ ] ] note that , to obtain the above spectral density , we have performed the integrals over the delta - functions which restricts the boundaries of the integrals on the @xmath157 and @xmath158 as : @xmath159 where @xmath160 and @xmath161 are the continuum thresholds in the initial and final channels in the case of ep mode . there are several sources for non - perturbative contributions , such as quark , quark - gluon , and gluon condensates , however , the quark - quark and quark - gluon condensates give zero contributions after applying the double borel transformation with respect to @xmath162 ( @xmath71 ) and @xmath163 ( @xmath72 ) . therefore , the remaining source of the non - perturbative contributions would be the gluon condensates [ see fig(3 ) ] . the calculation of such contributions is lengthy but standard . for the non - perturbative part in the borel scheme , we get @xmath164 where @xmath165 is the wilson coefficient of the gluon condensates that is defined as @xmath166 the explicit expressions of the @xmath167 are given in the appendix . using a similar procedure to that presented in the previous section , we get the sum rule for the form factor @xmath168 to be @xmath169.\end{aligned}\ ] ] this section is devoted to the numerical analysis of the form factors , estimating the branching ratio in the wa and ep channels and the total branching fraction of the @xmath6 transition . for this aim , we use the quark and mesons masses as @xmath170 , @xmath171 @xcite , @xmath172 @xcite , @xmath173 , @xmath174 @xcite . for the values of the decay constants , we use @xmath175 and @xmath176 @xcite . the values of the condensates are @xcite : @xmath177 , @xmath178 , @xmath179 and @xmath180 . the parameters entered the photon das are also taken as @xmath181 , @xmath182 , @xmath183 , @xmath184 , @xmath185 , @xmath186 , @xmath187 , @xmath188 , @xmath189 and @xmath190 @xcite . the remaining parameters are chosen as @xmath191 , @xmath192 , @xmath193 , @xmath194 @xcite , @xmath195 @xcite , and @xmath196 . the sum rules for the form factors contain also the continuum thresholds and the borel mass parameters as auxiliary objects . we find working regions for these parameters such that the physical observables are practically independent of them . the continuum thresholds are not completely arbitrary but are correlated with the energy of the first excited states in the initial and final mesonic channels . our numerical results show that the results depend weakly on the thresholds in the intervals @xmath197 and @xmath198 . the working regions for the borel parameters are obtained by demanding that not only the contributions of the higher states and continuum are effectively suppressed , but also the contributions of the higher order operators and higher twist das are small , i.e. , the series of the sum rules converge . these conditions lead to the intervals , @xmath199 , @xmath200 and @xmath201 for the borel mass parameters . now , we proceed to find the fit functions of the form factors using the aforesaid working regions for the auxiliary parameters . here we would like to mention that , for the decay rates , we need only the values of the form factors @xmath18 and @xmath19 at @xmath202 , @xmath16 and @xmath17 at @xmath203 and @xmath204 at @xmath38 . however , we determine their fit functions in general and give their values at these fixed points . the fit functions for the form factors @xmath18 and @xmath19 are @xmath205 where @xmath206 , @xmath207 and @xmath208 are the fit parameters whose values are @xmath209 the values of these form factors at @xmath202 are @xmath210 where the errors on the values are due to the uncertainties in determination of the working regions for the auxiliary parameters as well as those coming from the das and other input parameters . also the fit functions for the form factors @xmath211 are @xcite @xmath212 where the fit parameters are @xmath213 the values of the form factors @xmath214 calculated at @xmath203 are @xmath215 for the form factor induced by the ep at @xmath38 , we obtain @xmath216 at the end of this section we would like to calculate the decay widths and branching ratios . using the amplitudes of each decay mode , we find the following expressions for the decay rates at fixed points in wa and ep channels as well as for the total decay rate of the transition under consideration : @xmath217 + 2 f_{b_{c } } f_{d_{s1 } } f_{v}^{(b_{c } ) } f_{v}^{(d_{s1})}\frac{m_{d_{s1}}}{m_{b_{c}}^{2 } } \notag \\ & & \ + f_{d_{s1}}^{2 } m_{d_{s1}}^{2}\big[\frac{(f_{a}^{(b_{c})})^{2}}{m_{b_{c}}^{4 } } + \frac{(f_{v}^{(b_{c})})^{2}}{m_{b_{c}}^{4}}\big]\bigg\ } , \end{aligned}\ ] ] @xmath218^{2 } , \end{aligned}\ ] ] @xmath219 \notag \\ & & \ + |c_{7}|^2 |v_{tb}v^*_{ts}|^2 \big(16(m_b - m_s)^2+(m_b+m_s)^{2}\big ) [ t_{2}(0)]^2 \notag \\ & & \ + 16\pi^2 t_{2}(0 ) |v_{cb}v^*_{cs}| |v_{tb}v^*_{ts}| \bigg [ f_{d_{s1 } } m_{d_{s1 } } x \big\{4\frac{f_{a}^{(b_{c})}}{m_{b_{c}}^2 } ( m_{b } - m_{s } ) \notag \\ & & \ + \frac{f_{v}^{(bc)}}{m_{b_{c}}^2 } ( m_{b } + m_{s})\big\ } + f_{b_{c } } \big\{f_{v}^{(d_{s1 } ) } ( m_{b } + m_{s } ) x \notag \\ & & \ -4 f_{a}^{(d_{s1 } ) } ( m_{b } - m_{s } ) y\big\}\bigg]\bigg\ } , \end{aligned}\ ] ] where @xmath220 and @xmath221 are the real and imaginary parts of the wilson coefficient @xmath222 , respectively . in these formulas , the fixed point values of the form factors are used . finally the numerical values of the corresponding branching ratios for the radiative decay under consideration are obtained as follows : @xmath223 where the dominant contribution to each channel comes from the perturbative part . from these values , we also see that the @xmath6 transition proceeds mostly via the wa mode . the order of the total branching ratio indicates that this decay channel can be detected at lhcb in near future . any measurement on this decay and the comparison of the obtained data with our predictions in the present work can give valuable information about the nature and internal structure of the participating particles , especially the @xmath15 meson . one of the authors ( a. r. o. ) would like to thank r. khosravi and s. zarepour for useful discussions . also partial support of shiraz university research council is appreciated . the explicit expressions for @xmath224 are given as follows : @xmath225 - 3m_{b}^2i_{0}[1,4,1]+2m_{b}^2 ( m_{b}-m_{s})(m_{b}+m_{s})i_{0}[1,5,1 ] ] \notag \\ & & \ + 2m_{c}(i_{0}[1,2,1]-4m_{b}^2 i_{0}[1,3,1]-m_{b}m_{s}i_{0}[1,3,1]+5m_{b}^4 i_{0}[1,4,1]\notag \\ & & \ + 3m_{b}^3 m_{s}i_{0}[1,4,1]-3m_{b}^2 m_{s}^2 i_{0}[1,4,1]+2m_{b}^3 ( -m_{b}+m_{s})(m_{b } + m_{s})^2 i_{0}[1,5,1])\notag \\ & & \ + 2m_{s}[i_{0}[1,2,1]-2m_{b}^6 i_{0}[1,5,1]+m_{b}^4 ( 5i_{0}[1,4,1]+2i_{0}^{[0,1]}[1,5,1 ] ) \notag \\ & & \ + m_{b}^2 ( -4i_{0}[1,3,1]-3i_{0}^{[0,1]}[1,4,1]+3i_{0}^{[1,0]}[1,3,1 ] ) ] \notag \\ & & \ + m_{b}(3i_{0}^{[0,1]}[1,3,1]+i_{0}^{[0,2]}[1,4,1]+i_{0}^{[1,0]}[1,3,1]+2m_{b}^4 ( 3i_{0}^{[0,1]}[1,5,1]+i_{0}^{[1,0]}[1,5,1 ] ) \notag \\ & & \ -m_{b}^2 ( 9i_{0}^{[0,1]}[1,4,1]+2i_{0}^{[0,2]}[1,5,1]+3i_{0}^{[1,0]}[1,4,1]-2i_{0}^{[2,0]}[1,4,1 ] ) \notag \\ & & \ -4i_{3}^{[0,1]}[1,5,1]+4i_{3}^{[1,0]}[1,3,1])\bigg\},\end{aligned}\ ] ] @xmath226+m_{b}^3 i_{0}[1,1,4]+m_{c}i_{0}[1,1,4]+m_{b}m_{c}i_{0}[1,1,4 ] \notag \\ & & \ -m_{b}^2 m_{c}i_{0}[1,1,4]-m_{b}^3 m_{s}i_{0}[1,1,4]-m_{c}m_{s}i_{0}[1,1,4]+2m_{b}m_{c}m_{s}i_{0}[1,1,4 ] \notag \\ & & \ + m_{b}^2 m_{c}m_{s}i_{0}[1,1,4 ] -m_{c}i_{0}[1,1,5]-m_{b}m_{c}i_{0}[1,1,5]+m_{b}^3 m_{c}i_{0}[1,1,5]-m_{b}^2 m_{c}^2 i_{0}[1,1,5 ] \notag \\ & & \ + m_{c}m_{s}i_{0}[1,1,5]-2m_{b}m_{c}m_{s}i_{0}[1,1,5]+2m_{b}^3 m_{c}m_{s}i_{0}[1,1,5]-2m_{b}^2 m_{c}^2 m_{s}i_{0}[1,1,5 ] \notag \\ & & \ -m_{b}i_{0}^{[0,1]}[1,1,4]+m_{s}i_{0}^{[0,1]}[1,1,4]-3/2 i_{0}^{[0,1]}[1,1,5]+3/2m_{b}^2 i_{0}^{[0,1]}[1,1,5 ] \notag \\ & & \ + m_{b}m_{s}i_{0}^{[0,1]}[1,1,5]-m_{s}i_{0}^{[0,2]}[1,1,2]+3/2 i_{0}^{[1,0]}[1,1,3 ] + m_{b}i_{0}^{[1,0]}[1,1,3 ] \notag \\ & & \ + 3m_{s}i_{0}^{[1,0]}[1,1,3]+1/2 m_{b}^2 i_{0}^{[1,0]}[1,1,4]-m_{b}m_{s}i_{0}^{[1,0]}[1,1,4]+1/2 i_{0}^{[2,0]}[1,1,4 ] \notag \\ & & \ + m_{s}i_{0}^{[2,0]}[1,1,4],\end{aligned}\ ] ] @xmath227+m_{c}^2 i_{0}[3,1,2]+i_{0}^{[0,1]}[3,1,1 ] ) \notag \\ & & \ + m_{c}(m_{c}^2 i_{0}[3,2,2]+i_{0}^{[0,1]}[3,1,2])]-3i_{0}^{[0,1]}[3,2,1]+i_{0}^{[0,1]}[3,2,2]+3i_{0}^{[0,2]}[3,1,2 ] \notag \\ & & \ -2i_{0}^{[0,2]}[3,2,1 ] + i_{0}^{[0,3]}[3,2,2]+3i_{0}^{[1,0]}[3,1,1]-i_{0}^{[1,0]}[3,1,2 ] \notag \\ & & \ + 2m_{c}^3 m_{s}\{i_{0}[3,2,1]-i_{0}[3,2,2]-4i_{0}^{[0,1]}[3,1,2]+3i_{0}^{[0,1]}[3,2,1]+3i_{0}^{[1,0]}[3,1,1 ] \notag \\ & & \-2i_{0}^{[1,0]}[3,2,1]\ } -i_{0}^{[1,1]}[3,1,1]+i_{0}^{[1,1]}[3,1,2]-2m_{b}^3 [ -m_{c}(-2i_{0}^{[0,1]}[3,2,2]+i_{0}^{[1,0]}[3,1,2 ] \notag \\ & & \ + m_{c}^2 ( i_{0}[3,1,2]-2i_{0}[3,2,2]-3(i_{0}^{[0,1]}[3,2,2]+i_{0}^{[1,0]}[3,2,2]))+i_{0}^{[1,1]}[3,2,1 ] ) \notag \\ & & \ + m_{s}((1 + 2m_{c}^2)i_{0}[3,2,2]+2i_{0}^{[0,1]}[3,1,1]-2i_{0}^{[0,1]}[3,2,1]+4m_{c}^2 i_{0}^{[0,1]}[3,2,2 ] \notag \\ & & \ + i_{0}^{[0,2]}[3,1,2]+i_{0}^{[2,1]}[3,1,2]))-m_{b}^2 ( 2m_{c}^6 i_{0}[3,2,2]+i_{0}^{[0,1]}[3,1,2]-6i_{0}^{[0,1]}[3,2,1 ] \notag \\ & & \ -2i_{0}^{[0,2]}[3,1,2 ] + 6i_{0}^{[0,2]}[3,2,2]+i_{0}^{[0,3]}[3,2,2]+2m_{c}^3 m_{s}(i_{0}[3,1,2]-5i_{0}^{[0,1]}[3,2,2 ] \notag \\ & & \ -i_{0}^{[1,0]}[3,2,1])+3i_{0}^{[1,0]}[3,2,1]-2i_{0}^{[1,0]}[3,2,2 ] -7m_{c}^4 i_{0}^{[1,0]}[3,2,2 ] \notag \\ & & \ + 2m_{c}m_{s}(i_{0}[3,1,2]-i_{0}[3,2,1]+2i_{0}^{[0,2]}[3,1,2]-i_{0}^{[1,1]}[3,2,1 ] ) + 2i_{0}^{[1,1]}[3,2,1 ] \notag \\ & & \ -3i_{0}^{[1,2]}[3,2,2]+2i_{0}^{[2,0]}[3,2,1]-2i_{0}^{[2,1]}[3,2,2])+i_{0}^{[2,1]}[3,2,2 ] \notag \\ & & \ + m_{c}^2 ( -2i_{0}^{[0,1]}[3,2,2 ] + 10i_{0}^{[0,2]}[3,1,2]-2i_{0}^{[0,2]}[3,2,1]+i_{0}^{[1,0]}[3,2,1]+i_{0}^{[1,0]}[3,2,2 ] \notag \\ & & \ -14i_{0}^{[1,1]}[3,1,2]+14i_{0}^{[1,1]}[3,2,2]+3i_{0}^{[1,2]}[3,2,2]+2i_{0}^{[2,0]}[3,1,2 ] -10i_{0}^{[2,0]}[3,2,1 ] \notag \\ & & \ -3i_{0}^{[2,1]}[3,2,2]-3i_{0}^{[3,0]}[3,2,2])-i_{0}^{[3,0]}[3,2,2]\bigg\},\end{aligned}\ ] ] @xmath228 + 2m_{c}^3 m_{s}i_{0}[3,1,1]+24m_{b}^7 ( m_{c}+m_{s})i_{0}[3,2,2]+i_{0}^{[0,1]}[3,2,2 ] \notag \\ & & \ + 2m_{b}^6 ( 8m_{c}m_{s}i_{0}[3,1,1]+8m_{c}^2 i_{0}[3,2,2 ] + 3i_{0}^{[0,1]}[3,1,2]+i_{0}^{[1,0]}[3,1,1 ] ) \notag \\ & & \ + 6m_{b}^5 ( 2m_{c}^3 i_{0}[3,1,1]+2m_{c}^2 m_{s}i_{0}[3,2,2]-2m_{s}(-2i_{0}[3,1,2 ] + 2m_{s}^2i_{0}[3,2,1 ] \notag \\ & & \ + 2i_{0}^{[0,1]}[3,2,2])+m_{c}-9i_{0}^{[0,1]}[3,2,1 ] + i_{0}^{[1,0]}[3,1,2]))+3i_{0}^{[1,0]}[3,2,1]-i_{0}^{[1,0]}[3,2,2 ] \notag \\ & & \ -2m_{c}m_{s}(2i_{0}^{[0,1]}[3,1,2]-i_{0}^{[1,0]}[3,1,1 ] -i_{0}^{[1,0]}[3,1,2])+2i_{0}^{[1,1]}[3,2,2]+i_{0}^{[1,2]}[3,1,2 ] \notag \\ & & \ -m_{c}^2 ( -2i_{0}[3,1,2]+2i_{0}[3,2,2]+2i_{0}^{[0,1]}[3,2,2]+i_{0}^{[0,2]}[3,2,2]-5i_{0}^{[1,0]}[3,2,1 ] \notag \\ & & \ + 3i_{0}^{[1,0]}[3,2,2]-i_{0}^{[2,0]}[3,2,2])-i_{0}^{[2,0]}[3,2,2]+m_{b}^4 ( 4m_{c}^4 i_{0}[3,2,1]+4m_{c}^3 m_{s}i_{0}[3,2,2 ] \notag \\ & & \ + 2i_{0}^{[0,1]}[3,2,1]-15i_{0}^{[0,1]}[3,2,2]-2i_{0}^{[0,2]}[3,1,1 ] + 5i_{0}^{[0,1]}[3,2,2]-4i_{0}^{[1,0]}[3,1,2 ] ) \notag \\ & & \ + 6i_{0}^{[1,0]}[3,2,2 ] + 3(i_{0}^{[0,1]}[3,2,2]+i_{0}^{[1,0]}[3,2,2])+6i_{0}^{[1,1]}[3,1,2]+4i_{0}^{[2,0]}[3,2,2 ] ) \notag \\ & & \ + 3m_{b}(2m_{c}^2 m_{s}(-i_{0}[3,2,1]-2i_{0}^{[0,1]}[3,2,2]+2i_{0}^{[1,0]}[3,2,2])-4i_{0}[3,2,2 ] \notag \\ & & \ + 3i_{0}^{[0,1]}[3,2,1 ] -9i_{0}^{[0,1]}[3,2,2]-3i_{0}^{[0,2]}[3,1,2]+5i_{0}^{[1,0]}[3,1,2]+i_{0}^{[1,0]}[3,2,1 ] \notag \\ & & \ + 3i_{0}^{[2,0]}[3,1,2 ] ) -4m_{s}(i_{0}[3,1,2]+i_{0}^{[0,1]}[3,2,2]+i_{0}^{[1,0]}[3,1,2]-2i_{0}^{[1,0]}[3,2,1 ] \notag \\ & & \ -i_{0}^{[1,1]}[3,2,1]+i_{0}^{[2,0]}[3,2,2 ] ) -3m_{b}^3 ( -4m_{c}^3i_{0}[3,2,2]+2m_{c}^2 m_{s}-4i_{0}^{[0,1]}[3,2,2 ] \notag \\ & & \ + 4i_{0}^{[1,0]}[3,1,2 ] ) -4m_{s}(-3i_{0}[3,2,1]+4i_{0}[3,2,2]+3i_{0}^{[0,1]}[3,2,2]-4i_{0}^{[1,0]}[3,2,2 ] \notag \\ & & \ -2i_{0}^{[1,1]}[3,2,2 ] + 2i_{0}^{[2,0]}[3,2,2])+m_{c}(-16i_{0}[3,1,2 ] + 12i_{0}[3,2,2]-27i_{0}^{[0,1]}[3,1,2]\notag \\ & & \ + 6i_{0}^{[0,1]}[3,2,1]-6i_{0}^{[0,2]}[3,2,1]+3i_{0}^{[1,0]}[3,1,1 ] + 10i_{0}^{[1,0]}[3,2,2]+6i_{0}^{[2,0]}[3,2,2 ] ) \notag \\ & & \ + m_{b}^2 ( m_{c}^4 ( -6i_{0}[3,1,2]+4i_{0}[3,2,2])+m_{c}^3 ( 4m_{s}i_{0}[3,1,2]-6m_{s}i_{0}[3,2,1 ] ) \notag \\ & & \ -3i_{0}^{[0,1]}[3,1,1]+12i_{0}^{[0,1]}[3,1,2]-9i_{0}^{[1,0]}[3,1,2]+4i_{0}^{[1,0]}[3,2,1 ] \notag \\ & & \ + 2m_{c}m_{s}(7i_{0}[3,1,1]+9i_{0}^{[0,1]}[3,2,2]-2i_{0}^{[1,0]}[3,2,1]-6i_{0}^{[1,0]}[3,2,2])-9i_{0}^{[1,1]}[3,1,1 ] \notag \\ & & \ -2i_{0}^{[1,2]}[3,2,1]+m_{c}^2 ( 14i_{0}[3,1,2]-12i_{0}[3,2,2]+9i_{0}^{[0,1]}[3,1,2 ] \notag \\ & & \ -2i_{0}^{[0,1]}[3,2,2]+2i_{0}^{[0,2]}[3,1,2 ] -10i_{0}^{[1,0]}[3,1,2]+9i_{0}^{[1,0]}[3,2,1]-2i_{0}^{[2,0]}[3,1,2 ] ) \notag \\ & & \ -6i_{0}^{[2,0]}[3,1,2]+6i_{0}^{[2,0]}[3,2,2 ] + 2i_{0}^{[3,0]}[3,1,0])-i_{0}^{[3,0]}[3,2,2]\bigg\},\end{aligned}\ ] ] @xmath229+i_{0}[1,3,3])+4m_{b}^5 ( -m_{c}i_{0}[1,2,3]-2m_{s}i_{0}[1,3,3 ] ) \notag \\ & & \ -2m_{c}^2 [ i_{0}[1,2,2]-i_{0}[1,3,3]]-3i_{0}^{[0,1]}[1,1,3]+i_{0}^{[0,1]}[1,2,2]-i_{0}^{[0,2]}[1,2,3 ] \notag \\ & & \ -i_{0}^{[1,0]}[1,2,2 ] + 3i_{0}^{[1,0]}[1,3,1]-2m_{b}^4 ( -2m_{c}m_{s}i_{0}[1,2,1]-2m_{c}^2i_{0}[1,3,3 ] ) \notag \\ & & \ + 3i_{0}^{[0,1]}[1,3,3]+i_{0}^{[1,0]}[1,2,2])-2m_{b}^3 [ m_{c}(-3i_{0}[1,2,3]+2i_{0}[1,3,3 ] ) \notag \\ & & \ + 2m_{s}(-i_{0}[1,3,3]-2i_{0}^{[0,1]}[1,2,2 ] + 2i_{0}^{[1,0]}[1,3,2 ] ) ] \notag \\ & & \ + 2m_{b}\big\{m_{c}[-i_{0}[1,2,2]+i_{0}[1,3,2]]+2m_{s}[-i_{0}[1,2,3]+i_{0}[1,3,2 ] \notag \\ % \nonumber to remove numbering ( before each equation ) & & -i_{0}^{[0,1]}[1,1,3 ] + i_{0}^{[1,0]}[1,2,3 ] -3i_{0}[1,3,1]+2i_{0}^{[0,1]}[1,2,3]-2i_{0}^{[1,0]}[1,3,3])]\big\}+i_{0}^{[2,0]}[1,3,2 ] \notag \\ & & \ + m_{b}^2 ( 2m_{c}^2 ( 2i_{0}[1,2,3]-3i_{0}[1,3,2])+2m_{c}m_{s}(2i_{0}[1,1,2]-3i_{0}[1,3,3 ] ) + 9i_{0}^{[0,1]}[1,2,3 ] \notag \\ & & \ -2i_{0}^{[0,1]}[1,3,2]+2i_{0}^{[0,2]}[1,3,3]-6i_{0}^{[1,0]}[1,1,3]+3i_{0}^{[1,0]}[1,3,2 ] -2i_{0}^{[2,0]}[1,3,3])\bigg\},\end{aligned}\ ] ] @xmath230)+m_{b}^2 m_{c}(m_{c}^2i_{0}[2,1,4]-2i_{0}^{[0,1]}[2,1,4 ] ) \notag \\ & & \ + m_{b}(i_{0}^{[0,1]}[2,1,3]-2i_{0}^{[0,1]}[2,1,4 ] + i_{0}^{[0,2]}[2,1,2]+i_{0}^{[1,0]}[2,1,3]+m_{c}^2 i_{0}^{[1,0]}[2,1,3 ] \notag \\ & & \ -i_{0}^{[1,1]}[2,1,4 ] ) + m_{c}(2i_{0}^{[0,1]}[2,1,3]+i_{0}^{[0,2]}[2,1,3]-3i_{0}^{[1,0]}[2,1,4])\bigg\},\end{aligned}\ ] ] where we have ignored terms with higher powers of the strange quark mass . the functions , @xmath231 $ ] and @xmath232}[a , b , c]$ ] are defined as : @xmath233 & = & \frac{(-1)^{a+b+c}}{16\pi^2 \gamma(a ) \gamma(b ) \gamma(c)}(m_{1}^2)^{2-a - b}(m_{2}^2)^{2-a - c}\emph{\textbf{u}}_{0}(a+b+c-4,1-c - b ) , \notag \\ i_{1}[a , b , c ] & = & \frac{(-1)^{a+b+c+1}}{16\pi^2 \gamma(a ) \gamma(b ) \gamma(c)}(m_{1}^2)^{2-a - b}(m_{2}^2)^{3-a - c}\emph{\textbf{u}}_{0}(a+b+c-5,1-c - b ) , \notag \\ i_{2}[a , b , c ] & = & \frac{(-1)^{a+b+c+1}}{16\pi^2 \gamma(a ) \gamma(b ) \gamma(c)}(m_{1}^2)^{3-a - b}(m_{2}^2)^{2-a - c}\emph{\textbf{u}}_{0}(a+b+c-5,1-c - b ) , \notag \\ i_{n}^{[i , j]}[a , b , c ] & = & [ m_{1}^2]^{i}[m_{2}^2]^{j}\frac{d^{i}}{d(m_{1}^2)^i}\frac{d^{j}}{d(m_{2}^2)^j}[m_{1}^2]^{i}[m_{2}^2]^{j}i_{n}[a , b , c].\end{aligned}\ ] ] where @xmath234 is given by @xmath235,\ ] ] and @xmath236 , \notag \\ b_{0 } & = & \frac{1}{m_{1}^2m_{2}^2}[m_{1}^2m_{c}^2+m_{2}^2(m_{c}^2+m_{b}^2 ) ] , \notag \\ b_{1 } & = & \frac{m_{c}^2}{m_{1}^2m_{2}^2}.\end{aligned}\ ] ] 99 m. a. shifman , a. i. vainshtein and v. i. zakharov , nucl phys . b * 147 * , 385 ( 1979 ) . p. colangelo and a. khodjamirian , in _ at the frontier of particle physics / handbook of qcd _ , edited by m. shifman ( world scientifinc , singapore , 2001 ) , vol . 3 , p. 1495 . k. azizi , v. bashiry , phys . d. * 76 * , 114007 ( 2007 ) . t. m. aliev and m. savci , phys . b * 434 * , 358 ( 1998 ) . t. m. aliev and m. savci , phys . b * 480 * , 97 ( 2000 ) . k. azizi , r. khosravi , v. bashiry eur . j. c * 56 * , 357 ( 2008 ) . r. khosravi , k. azizi , m. ghanaatian , f. falahati , j. phys . g * 36 * , 095003 ( 2009 ) . v. v. kiselev , arxiv : hep - ph/0211021 ; arxiv : hep - ph/0308214 . v. v. kiselev , a. e. kovalsky , a. k.likhoded , nucl . b * 585 * , 353 ( 2000 ) . a. khodjamirian and d. wyler , to be published in sergei matinian festschrift from integrable models to guage theories edited by v. gurzadian and a. sedrakyan ( world scientific , singapore , 2002 ) ; arxiv : hep - ph/0111249 .

this work studies the joint design of precoding and backhaul compression strategies for the downlink of cloud radio access networks . in these systems , a central encoder is connected to multiple multi - antenna base stations ( bss ) via finite - capacity backhaul links . at the central encoder , precoding is followed by compression in order to produce the rate - limited bit streams delivered to each bs over the corresponding backhaul link . in current state - of - the - art approaches , the signals intended for different bss are compressed independently . in contrast , this work proposes to leverage joint compression , also referred to as multivariate compression , of the signals of different bss in order to better control the effect of the additive quantization noises at the mobile stations ( mss ) . the problem of maximizing the weighted sum - rate with respect to both the precoding matrix and the joint correlation matrix of the quantization noises is formulated subject to power and backhaul capacity constraints . an iterative algorithm is proposed that achieves a stationary point of the problem . moreover , in order to enable the practical implementation of multivariate compression across bss , a novel architecture is proposed based on successive steps of minimum mean - squared error ( mmse ) estimation and per - bs compression . robust design with respect to imperfect channel state information is also discussed . from numerical results , it is confirmed that the proposed joint precoding and compression strategy outperforms conventional approaches based on the separate design of precoding and compression or independent compression across the bss . cloud radio access network , constrained backhaul , precoding , multivariate compression , network mimo .

1304.3179

cellular systems are evolving into heterogeneous networks consisting of distributed base stations ( bss ) covering overlapping areas of different sizes , and thus the problems of interference management and cell association are becoming complicated and challenging @xcite . one of the most promising solutions to these problems is given by so called _ cloud radio access networks _ , in which the encoding / decoding functionalities of the bss are migrated to a central unit . this is done by operating the bss as `` soft '' relays that interface with the central unit via backhaul links used to carry only baseband signals ( and not `` hard '' data information ) @xcite-@xcite . cloud radio access networks are expected not only to effectively handle the inter - cell interference but also to lower system cost related to the deployment and management of the bss . however , one of the main impairments to the implementation of cloud radio access networks is given by the capacity limitations of the digital backhaul links connecting the bss and the central unit . these limitations are especially pronounced for pico / femto - bss , whose connectivity is often afforded by last - mile cables @xcite@xcite , and for bss using wireless backhaul links @xcite . in the _ uplink _ of cloud radio access networks , each bs compresses its received signal to the central unit via its finite - capacity backhaul link . the central unit then performs joint decoding of all the mobile stations ( mss ) based on all received compressed signals . recent theoretical results have shown that _ distributed compression _ schemes @xcite can provide significant advantages over the conventional approach based on independent compression at the bss . this is because the signals received by different bss are statistically correlated @xcite-@xcite , and hence distributed source coding enables the quality of the compressed signal received from one bs to be improved by leveraging the signals received from the other bss as side information . note that the correlation among the signals received by the bss is particularly pronounced for systems with many small cells concentrated in given areas . while current implementations @xcite@xcite employ conventional independent compression across the bss , the advantages of distributed source coding were first demonstrated in @xcite , and then studied in more general settings in @xcite-@xcite . related works based on the idea of computing a function of the transmitted codewords at the bss , also known as compute - and - forward , can be found in @xcite@xcite . in the _ downlink _ of cloud radio access networks , the central encoder performs joint encoding of the messages intended for the mss . then , it independently compresses the produced baseband signals to be transmitted by each bs . these baseband signals are delivered via the backhaul links to the corresponding bss , which simply upconvert and transmit them through their antennas . this system was studied in @xcite@xcite . in particular , in @xcite , the central encoder performs dirty - paper coding ( dpc ) @xcite of all mss signals before compression . a similar approach was studied in @xcite by accounting for the effect of imperfect channel state information ( csi ) . reference @xcite instead proposes strategies based on compute - and - forward , showing advantages in the low - backhaul capacity regime and high sensitivity of the performance to the channel parameters . for a review of more conventional strategies in which the backhaul links are used to convey message information , rather than the compressed baseband signals , we refer to @xcite-@xcite . in this work , we propose a novel approach for the compression on the backhaul links of cloud radio access networks in the downlink that can be seen as the counterpart of the distributed source coding strategy studied in @xcite-@xcite for the uplink . moreover , we propose the joint design of precoding and compression . a key idea is that of allowing the quantization noise signals corresponding to different bss to be correlated with each other . the motivation behind this choice is the fact that a proper design of the correlation of the quantization noises across the bss can be beneficial in limiting the effect of the resulting quantization noise seen at the mss . in order to create such correlation , we propose to jointly compress the baseband signals to be delivered over the backhaul links using so called _ multivariate compression _ * ch . 9 ) . we also show that , in practice , multivariate compression can be implemented without resorting to joint compression across all bss , but using instead a successive compression strategy based on a sequence of minimum mean squared error ( mmse ) estimation and per - bs compression steps . after reviewing some preliminaries on multivariate compression in sec . [ sec : preliminaries ] , we formulate the problem of jointly optimizing the precoding matrix and the correlation matrix of the quantization noises with the aim of maximizing the weighted sum - rate subject to power and the backhaul constraints resulting from multivariate compression in sec . [ sec : problem formulation ] . there , we also introduce the proposed architecture based on successive per - bs steps . we then provide an iterative algorithm that achieves a stationary point of the problem in sec . [ sec : joint ] . moreover , we compare the proposed joint design with the more conventional approaches based on independent backhaul compression @xcite-@xcite or on the separate design of precoding and ( multivariate ) quantization in sec . [ sec : separate ] . the robust design with respect to imperfect csi is also discussed in detail . in sec . [ sec : numerical - results ] , extensive numerical results are provided to illustrate the advantages offered by the proposed approach . the paper is terminated with the conclusion in sec . [ sec : conclusions ] . _ notation _ : we adopt standard information - theoretic definitions for the mutual information @xmath0 between the random variables @xmath1 and @xmath2 , conditional mutual information @xmath3 between @xmath1 and @xmath2 conditioned on random variable @xmath4 , differential entropy @xmath5 of @xmath1 and conditional differential entropy @xmath6 of @xmath1 conditioned on @xmath2 @xcite . the distribution of a random variable @xmath1 is denoted by @xmath7 and the conditional distribution of @xmath1 conditioned on @xmath2 is represented by @xmath8 . all logarithms are in base two unless specified . the circularly symmetric complex gaussian distribution with mean @xmath9 and covariance matrix @xmath10 is denoted by @xmath11 . the set of all @xmath12 complex matrices is denoted by @xmath13 , and @xmath14 represents the expectation operator . we use the notations @xmath15 and @xmath16 to indicate that the matrix @xmath17 is positive semidefinite and positive definite , respectively . given a sequence @xmath18 , we define a set @xmath19 for a subset @xmath20 . the operation @xmath21 denotes hermitian transpose of a matrix or vector , and notation @xmath22 is used for the correlation matrix of random vector @xmath23 , i.e. , @xmath24 $ ] ; @xmath25 represents the cross - correlation matrix @xmath26 $ ] ; @xmath27 is used for the conditional correlation matrix , i.e. , @xmath28 $ ] . downlink communication in a cloud radio access network in which there are @xmath29 multi - antenna bss and @xmath30 multi - antenna mss . the @xmath29 bss include both macro - bss and pico / femto - bss . the @xmath30 mss are distributed across all the cells.,width=566,height=434 ] we consider the downlink of a cloud radio access network as illustrated in fig . [ fig : block diagram ] . in the system , a central encoder communicates to @xmath30 mss through @xmath29 distributed bss . the message @xmath31 for each @xmath32th ms is uniformly distributed in the set @xmath33 , where @xmath34 is blocklength and @xmath35 is the information rate of message @xmath31 in bits per channel use ( c.u . ) . each ms @xmath32 has @xmath36 receive antennas for @xmath37 , and each bs @xmath38 is equipped with @xmath39 antennas for @xmath40 . note that the bss can be either macro - bss or pico / femto - bss and that the mss are arbitrarily distributed across the cells . each @xmath38th bs is connected to the central encoder via digital backhaul link with finite - capacity @xmath41 bits per c.u . for notational convenience , we define @xmath42 as the total number of transmitting antennas , @xmath43 as the total number of receive antennas , and the sets @xmath44 and @xmath45 . as shown in fig . [ fig : central encoder ] , each message @xmath31 is first encoded by a separate channel encoder , which produces a coded signal @xmath46 . the signal @xmath47 corresponds to the @xmath48 vector of encoded symbols intended for the @xmath32th ms for a given c.u . , and we have @xmath49 . we assume that each coded symbol @xmath46 is taken from a conventional gaussian codebook so that we have @xmath50 . the signals @xmath51 are further processed by the central encoder in two stages , namely _ precoding _ and _ compression_. as is standard practice , precoding is used in order to control the interference between the data streams intended for the same ms and for different mss . instead , compression is needed in order to produce the @xmath29 rate - limited bit streams delivered to each bs over the corresponding backhaul link . specifically , recall that each bs @xmath38 receives up to @xmath41 bits per c.u . on the backhaul link from the central encoder . further discussion on precoding and compression can be found in sec . [ sec : problem formulation ] . on the basis of the bits received on the backhaul links , each bs @xmath38 produces a vector @xmath52 for each c.u . , which is the baseband signal to be transmitted from its @xmath39 antennas . we have the per - bs power constraints\leq\delta_{l}$ ] for @xmath53 , where the matrix @xmath54 is a non - negative definite matrix ( see , e.g. , ( * ? ? ? ii - c ) ) . ] @xmath55\leq p_{i},\,\,\mathrm{for}\,\ , i\in\mathcal{n_{b}}.\label{eq : perbs power constraint}\ ] ] assuming flat - fading channels , the signal @xmath56 received by ms @xmath32 is written as @xmath57 where we have defined the aggregate transmit signal vector @xmath58^{\dagger}$ ] , the additive noise @xmath59 to obtain ( [ eq : received signal ms ] ) . ] , and the channel matrix @xmath60 toward ms @xmath32 as @xmath61,\label{eq : channel to ms k}\ ] ] with @xmath62 denoting the channel matrix from bs @xmath38 to ms @xmath32 . the channel matrices remain constant for the entire coding block duration . we assume that the central encoder has information about the global channel matrices @xmath63 for all @xmath64 and that each ms @xmath32 is only aware of the channel matrix @xmath63 . the bss must also be informed about the compression codebooks used by the central encoder , as further detailed later . the case of imperfect csi at the central encoder will be discussed in sec . [ sub : robust - design - with ] . based on the definition given above and assuming single - user detection at each ms , the rates @xmath65 can be achieved for each ms @xmath64 . this section reviews some basic information - theoretical results concerning multivariate compression , which will be leveraged in the analysis of the proposed backhaul compression strategy in sec . [ sub : multivariate - backhaul - compression ] . illustration of ( _ a _ ) conventional compression ; ( _ b _ ) multivariate compression.,width=457,height=289 ] to fix the ideas and the notation , we first consider the conventional compression problem illustrated in fig . [ fig : compression examples]-(a ) . the compression unit compresses a random sequence @xmath66 of @xmath34 independent and identically distributed ( i.i.d . ) samples with distribution @xmath7 at a rate of @xmath67 bits per symbol . specifically , the compressor selects a codeword @xmath68 within a codebook @xmath69 of size @xmath70 and sends the corresponding index , of @xmath71 bits , to the decompression unit . at the decompression unit , the sequence @xmath72 indicated by the received index is recovered . using the standard information - theoretic formulation , the compression strategy is specified by a conditional distribution @xmath73 , which is referred to as the _ test channel _ ( see , e.g. , ( * ? ? ? 2 ) ) . for a given test channel , compression consists in selecting a sequence @xmath72 that is jointly typical and @xmath74 are called jointly typical with respect to a distribution @xmath75 if their joint empirical distribution ( i.e. , normalized histogram with step size @xmath76 ) does not deviate much from @xmath75 ( see , e.g. , ( * ? ? ? 2 ) for a formal definition ) . ] with the sequence @xmath66 with respect to the given joint distribution @xmath77 . compression is hence successful if the encoder is able to find a jointly typical sequence @xmath68 in the codebook @xmath69 . a classical result in information theory is that this happens with arbitrarily large probability as the block length @xmath34 grows large if the inequality @xmath78 is satisfied ( * ? ? ? * ch . 3)@xcite . we now review the more general multivariate compression illustrated in fig . [ fig : compression examples]-(b ) . here , the sequence @xmath66 is compressed into @xmath79 indices with the goal of producing correlated compressed versions @xmath80 . each @xmath38th index indicates a codeword within a codebook @xmath81 of size @xmath82 , and is sent to the @xmath38th decompression unit for @xmath83 . each @xmath38th decompression unit then recovers a sequence @xmath84 corresponding to the received index . we emphasize that the choice of the codewords @xmath80 is done jointly at the compression unit . in particular , the specification of the compression strategy is given by a test channel @xmath85 . this implies that the compression unit wishes to find codewords @xmath80 that are jointly typical with the sequence @xmath66 with respect to the given joint distribution @xmath86 . the following lemma provides a sufficient condition for multivariate compression to be successful ( we refer to ( * ? ? ? * lemma 8.2 ) for a more precise statement ) . [ lem : multivariate covering ] _ ( * ? ? ? 9 ) _ consider an i.i.d . sequence @xmath66 and @xmath34 large enough . then , there exist codebooks @xmath87 with rates @xmath88 , that have at least one tuple of codewords @xmath89 jointly typical with @xmath66 with respect to the given joint distribution @xmath86 with probability arbitrarily close to one , if the inequalities @xmath90 are satisfied . see ( * ? ? ? 9 ) for a proof . we observe that , for a given test channel @xmath85 , the inequalities ( [ eq : multivariate covering condition ] ) impose joint conditions on the rate of all codebooks . this is due to the requirement of finding codewords @xmath91 that are jointly correlated according to the given test channel @xmath85 . also , note that the vector @xmath66 may be such that each @xmath92 is itself a vector and that the distortion requirements at each decompression unit prescribe that the decompression unit be interested in only a subset of entries in this vector . the connection between the multivariate set - up in fig . [ fig : compression examples]-(b ) and the system model under study in fig . [ fig : block diagram ] will be detailed in the next section . in this section , we first propose a novel precoding - compression strategy based on multivariate compression for the downlink of a cloud radio access network . we then establish the problem definition . finally , a novel architecture that implements multivariate compression via a sequence of mmse estimation and per - bs compression steps is proposed . illustration of the operation at the central encoder.,width=642,height=283 ] as mentioned in the previous section , the operation at the central encoder can be represented by the block diagram in fig . [ fig : central encoder ] . specifically , after channel encoding , the encoded signals @xmath93^{\dagger}$ ] undergo precoding and compression , as detailed next . * 1 . precoding : * in order to allow for interference management both across the mss and among the data streams for the same ms , the signals in vector @xmath94 are linearly precoded via multiplication of a complex matrix @xmath95 . the precoded data can be written as @xmath96 where the matrix @xmath97 can be factorized as @xmath98,\label{eq : whole beamformer}\ ] ] with @xmath99 denoting the precoding matrix corresponding to ms @xmath32 . the precoded data @xmath100 can be written as @xmath101^{\dagger}$ ] , where the signal @xmath102 is the @xmath103 precoded vector corresponding to the @xmath38th bs and given as @xmath104 with the matrix @xmath105 having all zero elements except for the rows from @xmath106 to @xmath107 which contain an @xmath108 identity matrix . note that non - linear precoding using dpc techniques can also be considered , as discussed in remark [ rem : dpc ] below . compression : * each precoded data stream @xmath102 for @xmath109 must be compressed in order to allow the central encoder to deliver it to the @xmath38th bs through the backhaul link of capacity @xmath41 bits per c.u . each @xmath38th bs then simply forwards the compressed signal @xmath110 obtained from the central encoder . note that this implies that the bss need not be aware of the channel codebooks and of the precoding matrix @xmath97 used by the central encoder . instead , they must be informed about the quantization codebooks selected by the central encoder . using standard rate - distortion considerations , we adopt a gaussian test channel to model the effect of compression on the backhaul link . in particular , we write the compressed signals @xmath110 to be transmitted by bs @xmath38 as is seemingly more general than ( [ eq : gaussian test channel each bs ] ) , but this can be captured by adjusting the matrix @xmath97 in ( [ eq : precoding only ] ) . ] @xmath111 where the compression noise @xmath112 is modeled as a complex gaussian vector distributed as @xmath113 . overall , the vector @xmath58^{\dagger}$ ] of compressed signals for all the bss is given by @xmath114 where the compression noise @xmath115^{\dagger}$ ] is modeled as a complex gaussian vector distributed as @xmath116 . the compression covariance @xmath117 is given as @xmath118,\label{eq : compression covariance}\ ] ] where the matrix @xmath119 is defined as @xmath120 $ ] and defines the correlation between the quantization noises of bs @xmath38 and bs @xmath121 . rate - distortion theory guarantees that compression codebooks can be found for any given covariance matrix @xmath122 under appropriate constraints imposed on the backhaul links capacities . this aspect will be further discussed in sec . [ sub : multivariate - compression theory ] . with the described precoding and compression operations , the achievable rate ( [ eq : rate ms k ] ) for ms @xmath32 is computed as @xmath123 [ rem : compression]in the conventional approach studied in @xcite-@xcite , the signals @xmath102 corresponding to each bs @xmath38 are compressed independently . this corresponds to setting @xmath124 for all @xmath125 in ( [ eq : compression covariance ] ) . a key contribution of this work is the proposal to leverage correlated compression for the signals of different bss in order to better control the effect of the additive quantization noises at the mss . the design of the precoding matrix @xmath97 and of the quantization covariance @xmath117 can be either performed separately , e.g. , by using a conventional precoder @xmath97 such as zero - forcing ( zf ) or mmse precoding ( see , e.g. , @xcite-@xcite ) , or jointly . both approaches will be investigated in the following . [ rem : dpc]if non - linear precoding via dpc @xcite is deployed at the central encoder with a specific encoding permutation @xmath126 of the ms indices @xmath127 , the achievable rate @xmath128 for ms @xmath129 is given as @xmath130 in lieu of ( [ eq : rate ms k ] ) and can be calculated as @xmath131 with the function @xmath132 given as @xmath133 note that the dpc is designed based on the knowledge of the noise levels ( including the quantization noise ) in order to properly select the mmse scaling factor @xcite . as explained above , due to the fact that the bss are connected to the central encoder via finite - capacity backhaul links , the precoded signals @xmath102 in ( [ eq : precoding bs - wise ] ) for @xmath109 are compressed before they are communicated to the bss using the gaussian test channels ( [ eq : gaussian test channel each bs ] ) . in the conventional case in which the compression noise signals related to the different bss are uncorrelated , i.e. , @xmath124 for all @xmath134 as in @xcite-@xcite , the signal @xmath110 to be emitted from bs @xmath38 can be reliably communicated from the central encoder to bs @xmath38 if the condition @xmath135 is satisfied for @xmath109 . this follows from standard rate - distortion theoretic arguments ( see , e.g. , @xcite and sec . [ sub : compression ] ) . we emphasize that ( [ eq : independent comp condition ] ) is valid under the assumption that each bs @xmath38 is informed about the quantization codebook used by the central encoder , as defined by the covariance matrix @xmath136 . in this paper , we instead propose to introduce correlation among the compression noise signals , i.e. , to set @xmath137 for @xmath125 , in order to control the effect of the quantization noise at the mss . as discussed in sec . [ sec : preliminaries ] , introducing correlated quantization noises calls for joint , and not independent , compression of the precoded signals of different bss . as seen , the family of compression strategies that produce descriptions with correlated compression noises is often referred to as _ multivariate compression . _ by choosing the test channel according to ( [ eq : whole encoding operation ] ) ( see sec . [ sub : multivariate - compression theory ] ) , we can leverage lemma [ lem : multivariate covering ] to obtain sufficient conditions for the signal @xmath110 to be reliably delivered to bs @xmath38 for all @xmath109 . in lemma [ lem : multivariate ] , we use @xmath138 to denote the matrix obtained by stacking the matrices @xmath139 for @xmath140 horizontally . [ lem : multivariate ] the signals @xmath141 obtained via the test channel ( [ eq : whole encoding operation ] ) can be reliably transferred to the bss on the backhaul links if the condition @xmath142 is satisfied for all subsets @xmath143 . the proof follows by applying lemma [ lem : multivariate covering ] by substituting @xmath144 for the signal @xmath1 to be compressed , and @xmath141 for the compressed versions @xmath91 . comparing ( [ eq : independent comp condition ] ) with ( [ eq : multivariate computed ] ) shows that the introduction of correlation among the quantization noises for different bss leads to additional constraints on the backhaul link capacities . assuming the operation at the central encoder , bss and mss detailed above , we are interested in maximizing the weighted sum - rate @xmath145 subject to the backhaul constraints ( [ eq : multivariate computed ] ) over the precoding matrix @xmath97 and the compression noise covariance @xmath117 for given weights @xmath146 , @xmath64 . this problem is formulated as [ eq : original problem ] @xmath147 the condition ( [ eq : original problem backhaul ] ) corresponds to the backhaul constraints due to multivariate compression as introduced in lemma [ lem : multivariate ] , and the condition ( [ eq : original problem power ] ) imposes the transmit power constraints ( [ eq : perbs power constraint ] ) . it is noted that the problem ( [ eq : original problem ] ) is not easy to solve due to the non - convexity of the objective function @xmath148 in ( [ eq : original problem objective ] ) and the functions @xmath149 in ( [ eq : original problem backhaul ] ) with respect to @xmath150 . in sec . [ sec : joint ] , we will propose an algorithm to tackle the solution of problem ( [ eq : original problem ] ) . proposed architecture for multivariate compression based on successive steps of mmse estimation and per - bs compression.,width=623,height=347 ] in order to obtain correlated quantization noises across bss using multivariate compression , it is in principle necessary to perform joint compression of all the precoded signals @xmath102 corresponding to all bss @xmath38 for @xmath109 ( see sec . [ sub : multivariate - compression theory ] ) . if the number of bss is large , this may easily prove to be impractical . here , we argue that , in practice , joint compression is not necessary and that the successive strategy based on mmse estimation and per - bs compression illustrated in fig . [ fig : successive comp ] is sufficient . the proposed approach works with a fixed permutation @xmath151 of the bss indices @xmath152 . the central encoder first compresses the signal @xmath153 using the test channel ( [ eq : gaussian test channel each bs ] ) , namely @xmath154 , with @xmath155 , and sends the bit stream describing the compressed signal @xmath156 over the backhaul link to bs @xmath157 . then , for any other @xmath109 with @xmath158 , the central encoder obtains the compressed signal @xmath159 for bs @xmath160 in a successive manner in the given order @xmath161 by performing the following steps : * ( _ a _ ) * * _ estimation _ : * the central encoder obtains the mmse estimate @xmath162 of @xmath159 given the signal @xmath163 and the previously obtained compressed signals @xmath164 . this estimate is given by @xmath165\label{eq : mmse estimate x_hat_i}\\ & = \mathbf{\sigma}_{\mathbf{x}_{\pi(i)},\mathbf{u}_{\pi(i)}}\mathbf{\sigma}_{\mathbf{u}_{\pi(i)}}^{-1}\mathbf{u}_{\pi(i)},\nonumber\end{aligned}\ ] ] where we defined the vector @xmath166^{\dagger}$ ] , and the correlation matrices @xmath167 and @xmath168 are given as @xmath169\end{aligned}\ ] ] and @xmath170,\end{aligned}\ ] ] with @xmath171 for subsets @xmath172 and the set @xmath173 defined as @xmath174 . * ( _ b _ ) * * _ compression _ : * the central encoder compresses the mmse estimate @xmath175 to obtain @xmath159 using the test channel @xmath176 where the quantization noise @xmath177 is independent of the estimate @xmath175 and distributed as @xmath178 with @xmath179 note that the first equality in ( [ eq : conditional covariance ] ) follows from the fact that the mmse estimate @xmath175 is a sufficient statistic for the estimation of @xmath159 from @xmath180 ( see , e.g. , @xcite ) . moreover , the compression rate @xmath181 required by the test channel ( [ eq : compression after mmse ] ) is given by @xmath182 to see why the structure in fig . [ fig : successive comp ] described above realizes multivariate compression , we need the following lemma . [ lem : contra polymatroid ] the region of the backhaul capacity tuples @xmath183 satisfying the constraints ( [ eq : original problem backhaul ] ) is a _ _ contrapolymatroid _ _ and @xmath184 with @xmath185 being the set of all subsets of @xmath186 . then , the polyhedron @xmath187 is a contrapolymatroid if the function @xmath188 satisfies the conditions : ( _ a _ ) @xmath189 ; ( _ b _ ) @xmath190 if @xmath191 ; ( _ c _ ) @xmath192 ( * ? ? ? therefore , it has a corner point for each permutation @xmath161 of the bs indices @xmath152 , and each such corner point is given by the tuple @xmath193 with @xmath194 for @xmath40 . moreover , the corner point @xmath193 in ( [ eq : corner points ] ) is such that the constraints ( [ eq : original problem backhaul ] ) are satisfied with equality for the subsets @xmath195 @xmath196 . this lemma follows immediately by the definition and properties of contrapolymatroids as summarized in ( * ? ? ? moreover , the second equality of ( [ eq : corner points ] ) holds due to the fact that the mmse estimate @xmath175 is a sufficient statistic for the estimation of @xmath159 from @xmath180 ( see , e.g. , @xcite ) , or equivalently from the markov chain @xmath197 . illustrative example of the contrapolymatroidal region of backhaul capacities @xmath198 satisfying the constraint ( [ eq : original problem backhaul ] ) imposed by multivariate compression for @xmath199 . the two corner points are given by ( [ eq : corner points]).,width=453,height=347 ] lemma [ lem : contra polymatroid ] shows that the region of backhaul capacities that guarantees correct delivery of the compressed signals ( [ eq : gaussian test channel each bs ] ) to the bss , as identified in lemma [ lem : multivariate ] , is a type of polyhedron known as contrapolymatroid , as exemplified in fig . [ fig : corner points ] . a specific feature of contrapolymatroid is that the corner points can be easily characterized as in ( [ eq : corner points ] ) . from standard rate - distortion theory arguments , the equality between ( [ eq : compression rate mmse compression ] ) and ( [ eq : corner points ] ) implies that the corner point @xmath193 can be obtained for any permutation @xmath161 by the successive estimation - compression strategy outlined above and illustrated in fig . [ fig : successive comp ] . overall , the discussion above shows that , for any correlation matrix @xmath117 in ( [ eq : compression covariance ] ) , multivariate compression is feasible by using the successive estimation - compression architecture in fig . [ fig : successive comp ] if the backhaul capacities satisfy the corner point condition ( [ eq : corner points ] ) for the given bs order @xmath161 . note that , in general , conditions ( [ eq : corner points ] ) are more restrictive than ( [ eq : original problem backhaul ] ) , which allows for any backhaul capacities in the contrapolymatroid . this is because the solution to the optimization problem ( [ eq : original problem ] ) can be seen by contradiction to lie necessarily on the boundary of the region ( [ eq : original problem backhaul ] ) but possibly not on the corner points . further discussion on this point can be found in sec . [ sub : practical - implementation ] , where we observe that , in practice , this limitation is not critical . in this section , we aim at jointly optimizing the precoding matrix @xmath97 and the compression covariance @xmath117 by solving problem ( [ eq : original problem ] ) . in sec . [ sec : separate ] , we will then consider the generally suboptimal strategy in which the precoding matrix is fixed according to standard techniques , such as zf @xcite , mmse @xcite or weighted sum - rate maximizing precoding @xcite@xcite by neglecting the compression noise , and only the compression noise matrix @xmath117 is optimized . as mentioned , the optimization ( [ eq : original problem ] ) is a non - convex problem . to tackle this issue , we will now show how to obtain an efficient scheme that is able to achieve a stationary point of problem ( [ eq : original problem ] ) . to this end , we first make a change of variable by defining the variables @xmath200 for @xmath64 . then , we define the functions @xmath201 and @xmath202 with respect to the variables @xmath203 which are obtained by substituting @xmath204 into the functions @xmath205 and @xmath149 in problem ( [ eq : original problem ] ) , respectively , and the transmit power constraint as @xmath206 for @xmath109 . the so - obtained problem over the variables @xmath203 and @xmath117 is still non - convex due to the second term in @xmath201 and the first term in @xmath202 , which are concave in the variables @xmath203 and @xmath117 . however , we observe that it falls into the class of difference - of - convex ( dc ) problems @xcite . among various algorithms having desirable properties for the solution of dc problems @xcite , we adopt the majorization minimization ( mm ) scheme @xcite , which solves a sequence of convex problems obtained by linearizing non - convex parts in the objective function @xmath201 and the constraint function @xmath202 . it is shown that the mm algorithm converges to a stationary point of the original non - convex problems ( see , e.g. , ( * ? ? ? * theorem 1 ) , ( * ? ? ? 1.3.3 ) and ( * ? ? * theorem 3 ) ) . the proposed algorithm is summarized in table algorithm 1 where we define the functions @xmath207 and @xmath208 as @xmath209 and @xmath210 with the function @xmath211 given as @xmath212 \1 . initialize the matrices @xmath213 and @xmath214 to arbitrary feasible positive semidefinite matrices for problem ( [ eq : original problem ] ) and set @xmath215 . update the matrices @xmath216 and @xmath217 as a solution of the following ( convex ) problem . @xmath218 \3 . go to step 4 if a convergence criterion is satisfied . otherwise , set @xmath219 and go back to step 2 . calculate the precoding matrices @xmath220 for @xmath64 , where @xmath221 is a diagonal matrix whose diagonal elements are the nonzero eigenvalues of @xmath222 and the columns of @xmath223 are the corresponding eigenvectors . as we have discussed in sec . [ sec : problem formulation ] , given the solution @xmath150 obtained from the proposed algorithm , the central processor should generally perform joint compression in order to obtain the signals @xmath110 to be transmitted by the bss . however , as seen in sec . [ sub : successive - compression ] , if the solution is such that the corner point conditions ( [ eq : corner points ] ) are satisfied for a given permutation @xmath161 of the bss indices , then the simpler successive estimation - compression structure of fig . [ fig : successive comp ] can be leveraged instead . we recall , from lemma [ lem : contra polymatroid ] , that in order to check whether the conditions ( [ eq : corner points ] ) are satisfied for some order @xmath161 , it is sufficient to observe which inequalities ( [ eq : original problem backhaul ] ) are satisfied with equality : if these inequalities correspond to the subsets @xmath224 for a given permutation @xmath161 , then the given solution corresponds to the corner point ( [ eq : corner points ] ) with the given @xmath161 . in our extensive numerical results , we have consistently found this condition to be verified . as a result , in practice , one can implement the compression strategy characterized by the calculated covariance @xmath117 by employing the implementation of fig . [ fig : successive comp ] with the obtained ordering @xmath161 . for reference , it is useful to consider the weighted sum - rate maximization problem with independent quantization noises as in @xcite-@xcite . this is formulated as ( [ eq : original problem ] ) with the additional constraints @xmath225 since the constraints ( [ eq : independent quantization ] ) are affine , the mm algorithm in table algorithm 1 is still applicable by simply setting to zero matrices @xmath124 for @xmath125 as per ( [ eq : independent quantization ] ) . so far , we have assumed that the central encoder has information about the global channel matrices @xmath63 for @xmath64 . in this subsection , we discuss the robust design of the precoding matrix @xmath97 and the compression covariance @xmath117 in the presence of uncertainty at the central encoder regarding the channel matrices @xmath63 for @xmath64 . specifically , we focus on deterministic worst - case optimization under two different uncertainty models , namely singular value uncertainty @xcite and ellipsoidal uncertainty models ( see @xcite ( * ? ? ? 4.1 ) and references therein ) . while the singular value uncertainty model can be related via appropriate bounds to any normed uncertainty on the channel matrices , as discussed in ( v ) , the ellipsoidal uncertainty model is more accurate when knowledge of the covariance matrix of the csi error , due , e.g. , to estimation , is available ( * ? ? ? 4.1 ) . in the following , we briefly discuss both models . considering multiplicative uncertainty model of ( ii - a ) , the actual channel matrix @xmath63 toward each ms @xmath32 is modeled as @xmath226 where the matrix @xmath227 is the csi known at the central encoder and the matrix @xmath228 accounts for the multiplicative uncertainty matrix . the latter is bounded as @xmath229 where @xmath230 is the largest singular value of matrix @xmath17 . then , the problem of interest is to maximizing the worst - case weighted sum - rate over all possible uncertainty matrices @xmath231 for @xmath64 subject to the backhaul capacity ( [ eq : original problem backhaul ] ) and power constraints ( [ eq : original problem power ] ) , namely [ eq : worst - case problem ] @xmath232 the following lemma offers an equivalent formulation for problem ( [ eq : worst - case problem ] ) . [ lem : problem singular uncertainty ] the problem ( [ eq : worst - case problem ] ) is equivalent to the problem ( [ eq : original problem ] ) with the channel matrix @xmath63 replaced with @xmath233 for @xmath64 . we first observe that the uncertainty matrix @xmath231 affects only the corresponding rate function @xmath234 in ( [ eq : rate ms k computed ] ) . therefore , the minimization versus matrices @xmath231 for @xmath64 in ( [ eq : worst - case problem objective ] ) can be performed separately for each @xmath32 by solving the problem @xmath235 . it can be now easily seen , following ( * ? ? ? * theorem 1 ) , that the result of this minimization is obtained when @xmath231 is such that @xmath236 . this concludes the proof . based on lemma [ lem : problem singular uncertainty ] , one can hence solve problem ( [ eq : worst - case problem ] ) by using the mm algorithm in table algorithm 1 with only change of the channel matrices from @xmath237 to @xmath238 . we now consider the ellipsoidal uncertainty model . to this end , for simplicity and following related literature ( see , e.g. , ( * ? ? ? 4.1 ) ) , we focus on multiple - input single - output ( miso ) case where each ms is equipped with a single antenna , i.e. , @xmath239 for @xmath64 . thus , we denote the channel vector corresponding to each ms @xmath32 by @xmath240 . the actual channel @xmath241 is then modeled as @xmath242 with @xmath243 and @xmath244 being the presumed csi available at the central encoder and the csi error , respectively . the error vector @xmath244 is assumed to be bounded within the ellipsoidal region described as @xmath245 for @xmath64 with the matrix @xmath246 specifying the size and shape of the ellipsoid @xcite . following the standard formulation , we consider here the `` dual '' problem of power minimization under signal - to - interference - plus - noise ratio ( sinr ) constraints for all mss ( see ( * ? ? ? 4.1 ) and references therein ) . this problem is stated as [ eq : problem pmin ] @xmath247 where the coefficients @xmath248 are arbitrary weights , @xmath249 is the sinr constraint for ms @xmath32 , and we recall that we have @xmath200 for @xmath64 . the problem ( [ eq : problem pmin ] ) is challenging since it contains infinite number of constraints in ( [ eq : problem pmin sinr constraint ] ) . but , following the conventional _ s - procedure _ * appendix b.2 ) , we can translate the constraints ( [ eq : problem pmin sinr constraint ] ) into a finite number of linear constraints by introducing auxiliary variables @xmath250 for @xmath64 , as discussed in the following lemma . [ lem : s lemma ] the constraints ( [ eq : problem pmin sinr constraint ] ) hold if and only if there exist constants @xmath251 such that the condition @xmath252-\beta_{k}\left[\begin{array}{cc } \mathbf{c}_{k } & \mathbf{0}\\ \mathbf{0 } & -1 \end{array}\right]\succeq\mathbf{0}\label{eq : s lemma constraint}\ ] ] is satisfied for all @xmath64 where we have defined @xmath253 for @xmath64 . it directly follows by applying the s - procedure ( * ? ? ? * appendix b.2 ) . by transforming the constraint ( [ eq : problem pmin sinr constraint ] ) into ( [ eq : s lemma constraint ] ) , we obtain a problem that falls again in the class of dc problems @xcite . therefore , one can easily derive the mm algorithm , similar to table algorithm 1 , by linearizing the non - convex terms in the constraint ( [ eq : problem pmin backhaul constraints ] ) . the algorithm is guaranteed to converge to a stationary point of problem ( [ eq : problem pmin ] ) ( see , e.g. , ( * ? ? ? * theorem 1 ) , ( * ? ? ? 1.3.3 ) and ( * ? ? * theorem 3 ) ) . in this section , we discuss a simpler approach in which the precoding matrix @xmath97 is fixed a priori to some standard scheme , such as zf , mmse or weighted sum - rate maximizing precoding , by neglecting the compression noise . the compression covariance @xmath117 is then designed separately so as to maximize the weighted sum - rate . the precoding matrix @xmath97 is first selected according to some standard criterion @xcite-@xcite by neglecting the compression noise . a subtle issue arises when selecting the precoding matrix @xmath97 that requires some discussion . specifically , the design of @xmath97 should be done by assuming a reduced power constraint , say @xmath254 for some @xmath255 for @xmath109 . the power offset factor @xmath255 is necessary since the final signal @xmath110 transmitted by each bs @xmath38 is given by ( [ eq : gaussian test channel each bs ] ) and is thus the sum of the precoded signal @xmath256 and the compression noise @xmath112 . therefore , if the power of the precoded part @xmath256 is selected to be equal to the power constraint @xmath257 , the compression noise power would be forced to be zero . but this is possible only when the backhaul capacity grows to infinity due to ( [ eq : original problem backhaul ] ) . as a result , in order to make the compression feasible , one needs to properly select the parameters @xmath258 depending on the backhaul constraints . having fixed the precoding matrix @xmath97 , the problem then reduces to solving problem ( [ eq : original problem ] ) only with respect to the compression covariance @xmath117 . the obtained problem is thus a dc problem which can be tackled via the mm algorithm described in table algorithm 1 by limiting the optimization at step 2 only to matrix @xmath117 . it is observed that , as discussed above , this problem may not be feasible if the parameters @xmath259 , @xmath109 , are too large . in practice , one can set these parameters using various search strategies such as bisection . in this section , we present numerical results in order to investigate the advantage of the proposed approach based on multivariate compression and on the joint design of precoding and compression as compared to the conventional approaches based on independent compression across the bss and separate design . we will focus on the sum - rate performance @xmath260 ( i.e. , we set @xmath261 in ( [ eq : original problem objective ] ) ) . we also assume that there is one ms active in each cell and we consider three cells , so that we have @xmath262 . every bs is subject to the same power constraint @xmath263 and has the same backhaul capacity @xmath264 , i.e. , @xmath265 and @xmath266 for @xmath109 . per - cell sum - rate versus the backhaul capacity @xmath264 for the circular wyner model @xcite with @xmath267 db and @xmath268.,width=453,height=340 ] we start by considering as a benchmark the performance in a simple circulant wyner model . in this model , all mss and bss have a single antenna and the channel matrices @xmath269 reduce to deterministic scalars given as @xmath270 for @xmath271 and @xmath272 $ ] for @xmath273 @xcite . in fig . [ fig : graph wyner ] , we compare the proposed scheme with joint design of precoding and compression with state - of - the - art techniques , namely the compressed dpc of @xcite , which corresponds to using dpc precoding with independent quantization , and reverse compute - and - forward ( rcof ) @xcite . we also show the performance with linear precoding for reference . it is observed that multivariate compression significantly outperforms the conventional independent compression strategy for both linear and dpc precoding . moreover , rcof in @xcite remains the most effective approach in the regime of moderate backhaul @xmath264 , although multivariate compression allows to compensate for most of the rate loss of standard dpc precoding in the low - backhaul regime is due to the integer constraints imposed on the function of the messages to be computed by the mss @xcite . ] . in this subsection , we evaluate the average sum - rate performance as obtained by averaging the sum - rate @xmath274 over the the realization of the fading channel matrices . the elements of the channel matrix @xmath275 between the ms in the @xmath32th cell and the bs in the @xmath38th cell are assumed to be i.i.d . complex gaussian random variables with @xmath276 in which we call @xmath277 the inter - cell channel gain . moreover , each bs is assumed to use two transmit antennas while each ms is equipped with a single receive antenna . in the separate design , we assume that the precoding matrix @xmath97 is obtained via the sum - rate maximization scheme in @xcite under the power constraint @xmath278 for each bs with @xmath279 selected as discussed in sec . [ sub : separate precoding ] . note that the algorithm of @xcite finds a stationary point for the sum - rate maximization problem using the mm approach , similar to table algorithm 1 without consideration of the quantization noises . average sum - rate versus the power offset factor @xmath280 for the separate design of linear precoding and compression in sec . [ sec : separate ] with @xmath281 db and @xmath282 db.,width=453,height=340 ] fig . [ fig : graph gamma ] demonstrates the impact of the power offset factor @xmath280 on the separate design of linear precoding and compression described in sec . [ sec : separate ] with @xmath281 db and @xmath282 db . increasing @xmath280 means that more power is available at each bs , which generally results in a better sum - rate performance . however , if @xmath280 exceeds some threshold value , the sum - rate is significantly degraded since the problem of optimizing the compression covariance @xmath117 given the precoder @xmath97 is more likely to be infeasible as argued in sec . [ sub : separate precoding ] . this threshold value grows with the backhaul capacity , since a larger backhaul capacity allows for a smaller power of the quantization noises . throughout the rest of this section , the power offset factor @xmath280 is optimized via numerical search . average sum - rate versus the transmit power @xmath263 for linear precoding with @xmath283 bit / c.u . and @xmath282 db.,width=453,height=340 ] in fig . [ fig : graph snr ] , the average sum - rate performance of the linear precoding and compression schemes is plotted versus the transmit power @xmath263 with @xmath283 bit / c.u . and @xmath282 db . it is seen that the gain of multivariate compression is more pronounced when each bs uses a larger transmit power . this implies that , as the received snr increases , more efficient compression strategies are called for . in a similar vein , the importance of the joint design of precoding and compression is more significant when the transmit power is larger . moreover , it is seen that multivariate compression is effective in partly compensating for the suboptimality of the separate design . for reference , we also plot the cutset bound which is obtained as @xmath284 where @xmath285 is the sum - capacity achievable when the bss can fully cooperate under per - bs power constraints . we have obtained the rate @xmath285 by using the inner - outer iteration algorithm proposed in ( * ? ? ? it is worth noting that only the proposed joint design with multivariate compression approaches the cutset bound as the transmit power increases . average sum - rate versus the transmit power @xmath263 for the joint design in sec . [ sec : joint ] with @xmath283 bit / c.u . and @xmath282 db.,width=453,height=340 ] in fig . [ fig : graph dpc ] , we compare two precoding methods , dpc and linear precoding , by plotting the average sum - rate versus the transmit power @xmath263 for the joint design in sec . [ sec : joint ] with @xmath283 bit / c.u . and @xmath282 db . for dpc , we have applied algorithm 1 with a proper modification for all permutations @xmath286 of mss indices @xmath127 and took the largest sum - rate . unlike the conventional broadcast channels with perfect backhaul links where there exists constant sum - rate gap between dpc and linear precoding at high snr ( see , e.g. , @xcite ) , fig . [ fig : graph dpc ] shows that dpc is advantageous only in the regime of intermediate @xmath263 due to the limited - capacity backhaul links . this implies that the overall performance is determined by the compression strategy rather than precoding method when the backhaul capacity is limited at high snr . average sum - rate versus the backhaul capacity @xmath264 for linear precoding with @xmath281 db and @xmath282 db.,width=453,height=340 ] fig . [ fig : graph ci ] plots the average sum - rate versus the backhaul capacity @xmath264 for linear precoding with @xmath281 db and @xmath282 db . it is observed that when the backhaul links have enough capacity , the benefits of multivariate compression or joint design of precoding and compression become negligible since the overall performance becomes limited by the sum - capacity achievable when the bss are able to fully cooperate with each other . it is also notable that the separate design with multivariate compression outperforms the joint design with independent quantization for backhaul capacities larger than @xmath287 bit / c.u . average sum - rate versus the inter - cell channel gain @xmath277 for linear precoding with @xmath283 bit / c.u . and @xmath281 db.,width=453,height=340 ] finally , we plot the sum - rate versus the inter - cell channel gain @xmath277 for linear precoding with @xmath283 bit / c.u . and @xmath281 db in fig . [ fig : graph alpha ] . we note that the multi - cell system under consideration approaches the system consisting of @xmath29 parallel single - cell networks as the inter - cell channel gain @xmath277 decreases . thus , the advantage of multivariate compression is not significant for small values of @xmath277 , since introducing correlation of the quantization noises across bss is helpful only when each ms suffers from a superposition of quantization noises emitted from multiple bss . in this work , we have studied the design of joint precoding and compression strategies for the downlink of cloud radio access networks where the bss are connected to the central encoder via finite - capacity backhaul links . unlike the conventional approaches where the signals corresponding to different bss are compressed independently , we have proposed to exploit multivariate compression of the signals of different bss in order to control the effect of the additive quantization noises at the mss . the problem of maximizing the weighted sum - rate subject to power and backhaul constraints was formulated , and an iterative mm algorithm was proposed that achieves a stationary point of the problem . moreover , we have proposed a novel way of implementing multivariate compression that does not require joint compression of all the bss signals but is based on successive per - bs estimation - compression steps . robust design with imperfect csi was also discussed . via numerical results , it was confirmed that the proposed approach based on multivariate compression and on joint precoding and compression strategy outperforms the conventional approaches based on independent compression and separate design of precoding and compression strategies . this is especially true when the transmit power or the inter - cell channel gain are large , and when the limitation imposed by the finite - capacity backhaul links is significant . 15 j. andrews , `` the seven ways hetnets are a paradigm shift , '' to appear in _ ieee comm . mag . _ , mar . 2013 . intel cor . , `` intel heterogeneous network solution brief , '' solution brief , intel core processor , telecommunications industry . j. segel and m. weldon , `` lightradio portfolio - technical overview , '' technology white paper 1 , alcatel - lucent . s. liu , j. wu , c. h. koh and v. k. n. lau , `` a 25 gb / s(/km@xmath288 ) urban wireless network beyond imt - advanced , '' _ ieee comm . mag . 122 - 129 , feb . 2011 . china mobile , `` c - ran : the road towards green ran , '' white paper , ver . 2.5 , china mobile research institute , oct . 2011 . t. flanagan , `` creating cloud base stations with ti s keystone multicore architecture , '' white paper , texas inst . , oct . 2011 . ericsson , `` heterogeneous networks , '' ericsson white paper , feb . 2012 . p. marsch , b. raaf , a. szufarska , p. mogensen , h. guan , m. farber , s. redana , k. pedersen and t. kolding , `` future mobile communication networks : challenges in the design and operation , '' _ ieee veh . tech . mag . _ , vol . 7 , no 16 - 23 , mar . v. chandrasekhar , j. g. andrews and a. gatherer , `` femtocell networks : a survey , '' _ ieee comm . mag . 59 - 67 , sep . 2008 . i. maric , b. bostjancic and a. goldsmith , `` resource allocation for constrained backhaul in picocell networks , '' in _ proc . ita 11 _ , ucsd , feb . s. h. lim , y .- h . kim , a. e. gamal and s .- y . chung , `` noisy network coding , '' _ ieee trans . inf . theory _ 3132 - 3152 , may 2011 . a. e. gamal and y .- h . network information theory _ , cambridge university press , 2011 . a. sanderovich , o. somekh , h. v. poor and s. shamai ( shitz ) , `` uplink macro diversity of limited backhaul cellular network , '' _ ieee trans . inf . theory _ , 55 , no . 8 , pp . 3457 - 3478 , aug . a. sanderovich , s. shamai ( shitz ) and y. steinberg , `` distributed mimo receiver - achievable rates and upper bounds , '' _ ieee trans . inf . theory _ , 55 , no . 10 , pp . 4419 - 4438 , oct . a. del coso and s. simoens , `` distributed compression for mimo coordinated networks with a backhaul constraint , '' _ ieee trans . wireless comm . _ , vol . 8 , no . 9 , pp . 4698 - 4709 , sep . l. zhou and w. yu , `` uplink multicell processing with limited backhaul via successive interference cancellation , '' in _ proc . ieee glob . ( globecom 2012 ) _ , anaheim , ca , dec . 2012 . s .- h . park , o. simeone , o. sahin and s. shamai ( shitz ) , `` robust and efficient distributed compression for cloud radio access networks , '' _ ieee trans 692 - 703 , feb . 2013 . m. grieger , s. boob and g. fettweis , `` large scale field trial results on frequency domain compression for uplink joint detection , '' in _ proc . ieee glob . ( globecom 2012 ) _ , anaheim , ca , dec . b. nazer , a. sanderovich , m. gastpar and s. shamai ( shitz ) , `` structured superposition for backhaul constrained cellular uplink , '' in _ proc . ieee intern . inf . theory ( isit 2009 ) _ , seoul , korea , jun . hong and g. caire , `` quantized compute and forward : a low - complexity architecture for distributed antenna systems , '' in _ proc . theory workshop ( itw 2011 ) _ , paraty , brazil , oct . o. simeone , o. somekh , h. v. poor and s. shamai ( shitz ) , `` downlink multicell processing with limited - backhaul capacity , '' _ eurasip j. adv . proc . _ , 2009 . p. marsch and g. fettweis , `` on downlink network mimo under a constrained backhaul and imperfect channel knowledge , '' in _ proc . ieee glob . ( globecom 2009 ) _ , honolulu , hawaii , nov . 2009 . m. h. m. costa , `` writing on dirty paper , '' _ ieee trans . inf . theory _ , 439 - 441 , may 1983 . hong and g. caire , `` reverse compute and forward : a low - complexity architecture for downlink distributed antenna systems , '' in _ proc . ieee intern . inf . theory ( isit 2012 ) _ , cambridge , ma , jul . b. l. ng , j. s. evans , s. v. hanly and d. aktas , `` distributed downlink beamforming with cooperative base stations , '' _ ieee trans . inf . theory _ , vol . 5491 - 5499 , dec . 2008 . i. sohn , s. h. lee and j. g. andrews , `` belief propagation for distributed downlink beamforming in cooperative mimo cellular networks , '' _ ieee trans . wireless comm . 4140 - 4149 , dec . 2011 . r. zakhour and d. gesbert , `` optimized data sharing in multicell mimo with finite backhaul capacity , '' _ ieee trans proc . _ , vol . 6102 - 6111 , dec . o. simeone , n. levy , a. sanderovich , o. somekh , b. m. zaidel , h. v. poor and s. shamai ( shitz ) , `` cooperative wireless cellular systems : an information - theoretic view , '' _ foundations and trends in communications and information theory _ , vol . 8 , nos . 1 - 2 , pp . 1 - 177 , 2012 . l. zhang , r. zhang , y .- c . liang , y. xin and h. v. poor , `` on gaussian mimo bc - mac duality with multiple transmit covariance constraints , '' _ ieee trans . inf . theory _ , 2064 - 2078 , apr . 2012 . x. zhang , j. chen , s. b. wicker and t. berger , `` successive coding in multiuser information theory , '' _ ieee trans . inf . theory _ , 53 , no . 6 , pp . 2246 - 2254 , jun . r. zhang , `` cooperative multi - cell block diagonalization with per - base - station power constraints , '' _ ieee j. sel . areas comm . 1435 - 1445 , dec . 2010 . m. hong , r .- y . sun , h. baligh and z .- q . luo , `` joint base station clustering and beamformer design for partial coordinated transmission in heterogeneous networks , '' _ ieee j. sel . areas comm . 226 - 240 , feb . 2013 . c. t. k. ng and h. huang , `` linear precoding in cooperative mimo cellular networks with limited coordination clusters , '' _ ieee j. sel . areas comm . 9 , pp . 1446 - 1454 , dec . m. hong , q. li , y .- f . liu and z .- q . luo , `` decomposition by successive convex approximation : a unifying approach for linear transceiver design in interfering heterogeneous networks , '' arxiv:1210.1507 . u. erez and s. t. brink , `` a close - to - capacity dirty paper coding scheme , '' _ ieee trans . inf . theory _ , 3417 - 3432 , oct . g. d. forney , jr . , `` shannon meets wiener ii : on mmse estimation in successive decoding schemes , '' arxiv:0409011v2 . d. n. c. tse and s. v. hanly , `` multiaccess fading channels - part i : polymatroid structure , optimal resource allocation and throughput capacities , '' _ ieee trans . inf . theory _ , 44 , no . 7 , pp . 2796 - 2815 , nov . a. beck and m. teboulle , `` gradient - based algorithms with applications to signal recovery problems , '' in _ convex optimization in signal processing and communications _ , y. eldar and d. palomar , eds . 42 - 88 , cambridge university press . 2010 . g. scutari , f. facchinei , p. song , d. p. palomar and j .- s . pang , `` decomposition by partial linearization : parallel optimization of multi - agent systems , '' arxiv:1302.0756v1 . s. loyka and c. d. charalambous , `` on the compound capacity of a class of mimo channels subject to normed uncertainty , '' _ ieee trans . inf . theory _ , vol . 2048 - 2063 , apr . 2012 . c. shen , t .- h . chang , k .- y . wang , z. qiu and c .- y . chi , `` distributed robust multicell coordinated beamforming with imperfect csi : an admm approach , '' _ ieee trans . sig . proc . 60 , no . 6 , pp . 2988 - 3003 , jun . 2012 . e. bjornson and e. jorswieck , `` optimal resource allocation in coordinated multi - cell systems , '' _ foundations and trends in communications and information theory _ , vol . 9 , nos . 2 - 3 , pp . 113 - 381 , 2013 . s. boyd and l. vandenberghe , _ convex optimization _ , cambridge university press , 2004 . d. gesbert , s. hanly , h. huang , s. shamai ( shitz ) , o. simeone and w. yu , `` multi - cell mimo cooperative networks : a new look at interference , '' _ ieee j. sel . areas comm . 1380 - 1408 , dec . 2010 h. huh , h. c. papadopoulos and g. caire , `` multiuser miso transmitter optimization for intercell interference mitigation , '' _ ieee trans . 4272 - 4285 , aug . j. lee and n. jindal , `` high snr analysis for mimo broadcast channels : dirty paper coding versus linear precoding , '' _ ieee trans . inf . theory _ , 4787 - 4792 , dec . 2007

this note contributes to the point calculus of persistent homology by extending alexander duality to real - valued functions . given a perfect morse function $ ] and a decomposition such that is an-manifold , we prove elementary relationships between the persistence diagrams of restricted to , to , and to .

1109.5052

persistent homology is a recent extension of the classical theory of homology ; see e.g. @xcite . given a real - valued function on a topological space , it measures the importance of a homology class by monitoring when the class appears and when it disappears in the increasing sequence of sublevel sets . a technical requirement is that the function be _ tame _ , which means it has only finitely many homological critical values , and each sublevel set has finite rank homology groups . pairing up the births and deaths , and drawing each pair of values as a _ dot _ ( a point in the plane ) , we get a multiset which we refer to as the _ persistence diagram _ of the function . it is a combinatorial summary of the homological information contained in the sequence of sublevel sets . if we substitute reduced for standard homology groups , we get a slightly modified _ reduced persistence diagram _ of the function . as between homology groups , we can observe relationships between persistence diagrams . a prime example is poincar duality , which says that the @xmath8-th and the @xmath9-th homology groups of an @xmath3-manifold are isomorphic . more precisely , this is true if homology is defined for field coefficients , which is what we assume throughout this paper . the extension to functions says that the diagram is symmetric with respect to reflection across the vertical axis ; see @xcite . here , we change the homological dimension of a dot from @xmath8 to @xmath10 whenever we reflect it across the axis . this paper contributes new relationships by extending alexander duality from spaces to functions . to state our results , we assume a perfect morse function , @xmath0 $ ] , which for the sphere has no critical points other than a minimum and a maximum , and a decomposition of the @xmath11-dimensional sphere into two subsets , @xmath1 , whose intersection is an @xmath3-manifold , @xmath2 . our first result says that the reduced persistence diagrams of @xmath4 restricted to @xmath5 and to @xmath6 are reflections of each other . we call this the land and water theorem . our second result relates land with shore . ignoring some modifications , it says that the persistence diagram of @xmath4 restricted to @xmath7 is the disjoint union of the diagram of @xmath4 restricted to @xmath5 and of its reflection . the modifications become unnecessary if we assume that the minimum and maximum of @xmath4 both belong to a common component of @xmath6 . we call this the euclidean shore theorem . in the example that justifies the title of this paper , and the names of our theorems , we let @xmath5 be the planet earth , not including the water and the air . to a coarse approximation , @xmath5 is homeomorphic to a @xmath12-ball , sitting inside the universe , which we model as a @xmath12-sphere , @xmath13 . the function we consider is the negative gravitational potential of the earth , which is defined on the entire universe . the sea is then a sublevel set of this function restricted to @xmath6 , which is the closure of @xmath14 . with these definitions , our results relate the persistence diagram of the gravitational potential restricted to the earth with the shape of the sea as its water level rises . also , the euclidean shore theorem applies , unmodified , expanding the relationship to include the sea floor , which is swept out by the shoreline as the water level rises . besides developing the mathematical theory of persistent homology , there are pragmatic reasons for our interest in the extension of alexander duality to functions . persistence has fast algorithms , so that the bulk of the work is often in the construction of the space and the function for which we compute persistence . a point in case is the analysis of the biological process of cell segregation started in @xcite . modeling the process as a subset of space - time , the function of interest is time which , after compactifying space - time to @xmath15 , has no critical points other than a minimum and a maximum . the subset @xmath5 of @xmath15 is a union of cells times time , whose boundary is a @xmath12-manifold . we can represent @xmath5 by a @xmath16-parameter family of alpha complexes , whose disjoint union has the same homotopy type ; see e.g. ( * ? ? ? * chapter iii ) . however , the boundary of that disjoint union is not necessarily a @xmath12-manifold . using our euclidean shore theorem , we can compute the persistence diagram of the function on the @xmath12-manifold without ever constructing the @xmath12-manifold . [ [ outline . ] ] outline . + + + + + + + + sections [ sec2 ] and [ sec3 ] introduce the necessary background on homology and persistent homology . sections [ sec4 ] and [ sec5 ] present our two results . section [ sec6 ] concludes the paper . starting with a brief introduction of classical homology groups , we present the relevant background on alexander duality and mayer - vietoris sequences . more comprehensive discussions of these topics can be found in textbooks of algebraic topology , such as @xcite . [ [ background . ] ] background . + + + + + + + + + + + the @xmath8-dimensional homology of a topological space , @xmath17 , is a mathematical language to define , count , and reason about the @xmath8-dimensional connectivity of @xmath17 . there are different but essentially equivalent theories depending on the choices one makes in the representation of the space , the selection of cycles , and the meaning of addition . for our purpose , the most elementary of these theories will suffice : a simplicial complex , @xmath18 , that triangulates @xmath17 , formal sums of @xmath8-simplices with zero boundary as @xmath8-cycles , and adding with coefficients in a field , @xmath19 . most algorithms on homology assume this model , in particular the ones developed within persistent homology , which is defined only for field coefficients . in this model , we call a formal sum of @xmath8-simplices a _ @xmath8-chain _ , a _ @xmath8-cycle _ if its boundary is empty , and a _ @xmath8-boundary _ if it is the boundary of a @xmath20-chain . the @xmath8-boundaries form a subgroup of the @xmath8-cycles , which form a subgroup of the @xmath8-chains : @xmath21 . @xmath8-th homology group _ is the quotient of the @xmath8-cycles over the @xmath8-boundaries : @xmath22 . its elements are sets of @xmath8-cycles that differ from each other by @xmath8-boundaries . more fully , we denote the @xmath8-th homology group by @xmath23 , or by @xmath24 to emphasize that the group is independent of the simplicial complex we choose to triangulate the space . however , we will fix an arbitrary field @xmath19 and drop it from the notation . for field coefficients , @xmath25 is necessarily a vector space , which is fully described by its rank , @xmath26 such that @xmath27 . this rank is called the _ @xmath8-th betti number _ of @xmath17 . finally , we will often drop the dimension from the notation by introducing the direct sum , @xmath28 . besides standard homology , we will frequently use _ reduced homology groups _ , @xmath29 , which are isomorphic to the non - reduced groups except possibly for dimensions @xmath30 . to explain the difference , we note that @xmath31 counts the components of @xmath17 . in contrast , @xmath32 counts the gaps between components or , equivalently , the edges that are needed to merge the components into one . hence , @xmath33 , except when @xmath17 is empty , in which case @xmath34 . to distinguish this case from a single component , we have @xmath35 equal to @xmath16 if @xmath36 , and equal to @xmath37 , otherwise . furthermore , we use relative homology , which is defined for pairs of spaces , @xmath38 . taking a pair relaxes the requirement of a chain to be called a cycle , namely whenever its boundary is contained in @xmath39 , which includes the case when the boundary is empty . we write @xmath40 for the _ @xmath8-th relative homology group _ of the pair , and @xmath41 for the _ @xmath8-th relative betti number_. as before , we will suppress the dimension from the notation by introducing @xmath42 . as examples , consider the @xmath11-dimensional sphere , @xmath43 , and a closed hemisphere , @xmath44 , which is a ball of dimension @xmath45 . in standard homology , we have @xmath46 while all other betti numbers are zero . in reduced homology , we have @xmath47 while all other reduced betti numbers are zero . in particular , @xmath48 for all @xmath8 . in relative homology , we have @xmath49 while all other relative betti numbers are zero . in particular , @xmath50 , which may be confusing at first but makes sense because every point on the sphere can be connected by a path to a point in the hemisphere and is thus a @xmath37-boundary . [ [ alexander - duality . ] ] alexander duality . + + + + + + + + + + + + + + + + + + recall that a _ perfect morse function _ is one whose number of critical points equals the sum of betti numbers of the space . for a sphere , this number is @xmath51 : a minimum and a maximum . throughout the remainder of this section , we assume a perfect morse function @xmath52 $ ] , whose values at the minimum and the maximum are @xmath37 and @xmath16 . we also assume _ complementary subsets _ whose union is the sphere , @xmath53 , and whose intersection is an @xmath3-manifold , @xmath54 . we assume that the restriction of @xmath4 to @xmath7 is tame , which implies that its restrictions to @xmath5 and @xmath6 are also tame . for each @xmath55 , we write @xmath56 $ ] for the _ sublevel set _ of @xmath4 , and @xmath57 , @xmath58 , @xmath59 for the sublevel sets of the restrictions of @xmath4 to @xmath5 , @xmath6 , @xmath7 . similarly , we write @xmath60 $ ] for the _ superlevel set _ , and @xmath61 , @xmath62 , @xmath63 for the superlevel sets of the restrictions . the basic version of alexander duality is a statement about two complementary subsets of the sphere ; see @xcite . more specifically , it states that @xmath64 and @xmath65 are isomorphic , where @xmath66 , except for @xmath67 and @xmath68 when the @xmath37-dimensional group has an extra generator . this implies @xmath69 for @xmath70 . we will also need the version that deals with two complementary subsets of the @xmath11-dimensional ball , @xmath71 ; see @xcite . let @xmath72 denote a regular value of @xmath73 . note first that @xmath74 is homeomorphic to @xmath71 . by excision , the homology groups of @xmath75 relative to @xmath76 are isomorphic to those of @xmath5 relative to @xmath77 . alexander duality states that @xmath78 and @xmath79 are isomorphic , where @xmath80 , as before , except for @xmath81 , when @xmath82 has an extra component . this implies @xmath83 for @xmath84 . [ [ mayer - vietoris . ] ] mayer - vietoris . + + + + + + + + + + + + + + + we can connect the homology groups of @xmath5 and @xmath6 with those of @xmath7 and @xmath13 using the mayer - vietoris sequence of the decomposition . this sequence is _ exact _ , meaning the image of every map equals the kernel of the next map . counting the trivial homology groups , the sequence is infinitely long , with three terms per dimension : @xmath85 where the maps between homology groups of the same dimension are induced by the inclusions . the only non - trivial homology groups of @xmath13 are in dimensions @xmath37 and @xmath45 , with ranks @xmath86 . it follows that for @xmath70 , the groups defined by @xmath7 , @xmath5 , @xmath6 are surrounded by trivial groups . this implies that the groups of @xmath7 and the direct sums of the groups of @xmath5 and @xmath6 are isomorphic . for @xmath81 and @xmath87 , the non - trivial homology groups of @xmath13 prevent this isomorphism , and we get @xmath88 for @xmath70 . we also have a mayer - vietoris sequence for the spaces @xmath74 , @xmath75 , @xmath82 and @xmath89 : @xmath90 for @xmath91 , @xmath74 is a point or a closed @xmath11-dimensional ball , and its only non - trivial homology group is of dimension @xmath37 , with rank @xmath92 . it follows that for @xmath84 , the groups defined by @xmath89 , @xmath75 , @xmath82 are surrounded by trivial groups , which implies that the groups of @xmath89 and the direct sums of the groups of @xmath75 and @xmath82 are isomorphic . we thus get @xmath93 for @xmath84 . we may also consider the relative homology groups , again connected by a mayer - vietoris sequence : @xmath94 for @xmath95 , the only non - trivial homology group of the pair @xmath96 is in dimension @xmath45 , which implies @xmath97 for @xmath98 . [ [ separating - manifold . ] ] separating manifold . + + + + + + + + + + + + + + + + + + + + combining alexander duality with mayer - vietoris , we get relations between the homology of the sublevel sets of @xmath7 and @xmath5 . more specifically , we get @xmath99 for all @xmath100 . here , we combine , , with , , to get . similarly , we combine , with , to get . finally , we exploit the symmetry between @xmath5 and @xmath6 and combine , with , to get . note that and imply @xmath101 , which is a consequence of lefschetz duality ; see @xcite . [ [ an - example . ] ] an example . + + + + + + + + + + + we illustrate the above relationships with an example . let @xmath7 be a @xmath51-dimensional torus in @xmath102 . accordingly , @xmath5 and @xmath6 are the two solid tori that decompose the @xmath12-sphere and intersect in @xmath7 . this is sketched in figure [ fig : torus ] , where we assume that @xmath5 is the part of space surrounded by @xmath7 , while @xmath6 is the space surrounding @xmath7 . the only non - zero betti numbers of the solid torus are @xmath103 . we thus get @xmath104 from . these are the correct betti numbers of the @xmath51-dimensional torus . next , choose @xmath105 so that the sublevel set of @xmath5 is half the solid torus . its only non - zero betti number is @xmath106 , and the only non - zero relative betti number of the pair is @xmath107 . we thus get @xmath108 from . these are the correct betti numbers of the sublevel set of @xmath7 . finally , we get @xmath109 from . these are the correct relative betti numbers of the pair @xmath110 . starting with a brief introduction of persistent homology , we review its recent combinatorial expression as a point calculus ; see @xcite and @xcite for more comprehensive discussions . [ [ background.-1 ] ] background . + + + + + + + + + + + we take the step from homology to persistent homology by replacing a space with the sequence of ( closed and open ) sublevel sets of a function on the space . to explain this , let @xmath17 be compact and @xmath111 $ ] tame . as before , we write @xmath112 $ ] for the sublevel set and @xmath113 $ ] for the superlevel set defined by @xmath105 . since @xmath114 is tame , it has finitely many homological critical values , @xmath115 to @xmath116 . we assume w.l.o.g . that @xmath117 and @xmath118 , else , we add them as additional critical values . we interleave the @xmath119 with homological regular values @xmath120 , such that @xmath121 . taking the homology of closed and open sublevel sets at the regular values , we get the filtration @xmath122 where @xmath123 is @xmath124 , if @xmath125 , @xmath126 , if @xmath127 , and @xmath128 , if @xmath129 . for notational convenience , we add the trivial groups , @xmath130 and @xmath131 , at the beginning and end of the filtration . the maps connecting the homology groups are induced by the inclusions @xmath132 and @xmath133 , whenever @xmath134 . the maps compose and we write @xmath135 . reading the filtration from left to right , we see homology classes appear and disappear . to understand these events , we say a class @xmath136 is _ born _ at @xmath123 if it does not belong to the image of @xmath137 . if furthermore @xmath138 belongs to the image of @xmath139 but @xmath140 does not belong to the image of @xmath141 , then we say @xmath142 _ dies entering _ @xmath143 . since the filtration starts and ends with trivial groups , every homology class has well - defined birth and death values . every event is associated with the immediately preceding homological critical value , namely with @xmath144 if the event is at @xmath123 , and with @xmath145 if the event is at @xmath146 , for @xmath125 . events at @xmath147 are associated with @xmath118 . we represent a class by a dot in the plane whose two coordinates mark its birth and its death . more precisely , the coordinates signal the increase and decrease of betti numbers , and the dot represents an entire coset of classes that are born and die with @xmath142 . all dots have coordinates in @xmath148 $ ] , by construction . note that we use each @xmath119 twice , once during the first pass in which the sublevel set grows from @xmath149 to @xmath17 , and then again during the second pass in which @xmath17 minus the superlevel set shrinks back from @xmath17 to @xmath149 . when we collect the dots , we separate the passes by drawing each coordinate axis twice . the result is a multiset of dots , which we refer to as the _ persistence diagram _ of the function , and denote as @xmath150 , or @xmath151 if we restrict ourselves to the dots representing @xmath8-dimensional classes ; see figure [ fig : rdiagram ] , where we label each dot with the dimension of the classes it represents . we further distinguish four regions within the diagram : the _ ordinary _ , the _ horizontal _ , the _ vertical _ , and the _ relative subdiagrams _ , denoted as @xmath152 , @xmath153 , @xmath154 , and @xmath155 . for example , a dot belongs to the ordinary subdiagram if its birth and death both happen during the first pass ; see again figure [ fig : rdiagram ] . [ [ reduced - persistence - diagrams . ] ] reduced persistence diagrams . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + it is sometimes convenient to use reduced instead of standard homology groups in the filtration . there are small differences caused by the @xmath156-dimensional class , which exists if the space is empty . as a first step , we introduce reduced versions of the relative homology groups , which are isomorphic to the standard relative groups , for all @xmath8 . to define them , let @xmath157 be a new ( dummy ) vertex , write @xmath158 for the cone of @xmath157 over @xmath159 , and let @xmath160 be the result of gluing the cone along @xmath159 to @xmath17 . then the _ reduced relative homology group _ of the pair is @xmath161 . for example , if @xmath162 , then @xmath157 forms a separate component , so that the reduced betti number is equal to the number of components of @xmath17 , just as @xmath163 . the _ reduced persistence diagram _ is now defined as before , but for the filtration of reduced homology groups . to describe this , we suppress the dimension and write @xmath164 for @xmath165 , if @xmath166 , for @xmath167 , if @xmath127 , and for @xmath168 , if @xmath129 . the resulting sequence of reduced homology groups , from @xmath169 to @xmath170 , is connected from left to right by maps induced by inclusion . finally , we define @xmath171 by matching up the births and the deaths and by drawing each pair as a dot in the plane . similar to before , we write @xmath172 when we restrict ourselves to homology classes of dimension @xmath8 . for dimension @xmath173 , the reduced diagrams are the same as the standard ones , simply because the groups are the same . more formally , the persistence diagrams are the same because the two filtrations form a commutative diagram with vertical isomorphisms : @xmath174 see the persistence equivalence theorem @xcite . this is not true for @xmath175 , where the standard diagram is empty , while the reduced diagram contains a single dot marking the transition from an empty to a non - empty space . to describe the difference for @xmath81 , we call a dot @xmath176 in the @xmath37-th standard persistence diagram _ extreme _ if no other dot has a first coordinate smaller than @xmath177 and a second coordinate larger than @xmath178 . here , we assume for simplicity that no two dots in the @xmath37-th diagram share the same first coordinate or the same second coordinate . we note that only dots in the horizontal subdiagram can be extreme . the non - extreme dots also belong to the reduced diagram of @xmath114 , while the extreme dots exchange their coordinates to form new dots in the reduced diagram ; see figure [ fig : rdiagram ] . the way the coordinates are exchanged will be important later , so we describe the details of this process now . [ [ cascades . ] ] cascades . + + + + + + + + + we let @xmath179 be the number of extreme dots and , for a reason that will become clear shortly , denote their coordinates with different indices as @xmath180 , for @xmath181 , ordering them such that @xmath182 . by definition of extreme , the corresponding second coordinates satisfy @xmath183 . observe that @xmath184 is the globally minimum value and @xmath185 is the globally maximum value . let us now construct the reduced diagram , focusing on dimension @xmath37 . we start the upward pass with the birth of the @xmath156-dimensional homology class at @xmath37 , which dies at @xmath184 , so we have @xmath186 in @xmath187 . the minima of all other components mark the births of @xmath37-dimensional classes . they can be interpreted as gaps between components . we start the downward pass with the birth of a @xmath37-dimensional class at @xmath16 , which dies at @xmath185 , so we have @xmath188 in @xmath189 . the maxima of all other components mark the deaths of @xmath37-dimensional classes . if @xmath178 is the second coordinate of a non - extreme component , then it marks the death of the gap that opened up when we passed the corresponding first coordinate , @xmath177 , during the upward pass . hence , @xmath190 belongs to @xmath189 , and we note that this dot also belongs to the standard diagram . however , if @xmath191 is the second coordinate of a extreme dot , then it marks the death of the gap that has opened up when we passed the minimum of the next component with extreme dot during the upward pass . this minimum value is @xmath192 , so @xmath193 belongs to @xmath189 , for @xmath194 . in summary , we see that contiguous extreme dots exchange one pair of coordinates , using @xmath37 on the left and @xmath16 on the right to complete the process . more precisely , the reduced diagram is the same as the standard one , except for substituting the dots for the dots : @xmath195 for later use , we call this substitution a _ cascade _ and write @xmath196 ; see figure [ fig : cascade ] . note that all dots belong to the horizontal and vertical subdiagrams of dimension @xmath37 , except for @xmath188 , which belongs to the relative subdiagram of dimension @xmath37 , and @xmath186 , which belongs to the ordinary subdiagram of dimension @xmath197 . we will often have @xmath198 and @xmath199 , in which case the first and the last dots in lie on the diagonal and can be ignored . [ [ point - calculus . ] ] point calculus . + + + + + + + + + + + + + + + as described in @xcite , the persistence diagram can be harvested for a wealth of homological information . this includes the ranks of the homology groups of the sets @xmath200 and of the pairs @xmath201 . to explain this , we write @xmath202 for the multiset of dots in the rectangle with lower corner @xmath203 on the base edge of @xmath204 . similarly , we write @xmath205 for the multiset of dots if the lower corner , @xmath203 , lies on the base edge of @xmath206 ; see figure [ fig : diagram ] . often , we drop the dimension from the notation and write @xmath207 for the disjoint union of the multisets @xmath202 , over all @xmath8 , and similar for @xmath208 . we can read the betti numbers of @xmath200 by collecting the dots in @xmath207 , and we can read the relative betti numbers of @xmath201 by collecting the dots in @xmath208 . writing @xmath209 and @xmath210 for the corresponding multisets in the reduced diagrams , we note that the same relations hold between the reduced betti numbers and the reduced persistence diagrams . [ betti - numbers - and - persistence - diagrams . ] ] betti numbers and persistence diagrams . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to motivate the first result of this paper , we rewrite to for reduced betti numbers : @xmath211 for all @xmath100 and all regular values @xmath105 . observe also that @xmath212 iff @xmath213 , which implies that also holds for @xmath175 . in words , for every dimension @xmath8 and every regular value @xmath105 , the rectangles @xmath214 and @xmath215 have equally many dots . this @xmath16-parameter family of relations is satisfied if the reduced diagrams of @xmath216 and @xmath217 are reflections of each other . this is our first result , stated as the land and water theorem in section [ sec4 ] . we see an illustration in figure [ fig : rdiagram ] , which shows the reduced and non - reduced diagrams of the function @xmath4 in figure [ fig : torus ] restricted to the solid torus , @xmath5 , and to the complementary solid torus , @xmath6 . removing the black dots , we are left with the two reduced diagrams , which are indeed reflections of each other . to motivate our second result , we consider the relations to . they say that for every regular value @xmath105 , the rectangle @xmath218 has the same number of dots as @xmath219 and @xmath220 together . similarly , @xmath221 has the same number of dots as @xmath222 and @xmath223 together . this @xmath16-parameter family of relations is satisfied if the persistence diagram of @xmath224 is the disjoint union of the diagram of @xmath217 and of its reflection . this is our second result , stated as the euclidean shore theorem in section [ sec5 ] . it is illustrated in figure [ fig : diagram ] , which shows the persistence diagram of the function @xmath224 in figure [ fig : torus ] . comparing figures [ fig : rdiagram ] and [ fig : diagram ] , we see that the two circle dots in figure [ fig : diagram ] also belong to the diagram of @xmath217 , whereas the two square dots are reflections of the circles . while to and to motivate our two results , these relations are not sufficient to prove them . indeed , to hold in general , but the euclidean shore theorem requires that the minimum and maximum of the perfect morse function belong to a common component of @xmath6 . if this condition is violated , then we can have coordinate exchanges among the dots that contradict the euclidean shore theorem without affecting the relations to . an example of this phenomenon can be seen in figure [ fig : counterexample ] , which will be discussed in section [ sec5 ] . in this section , we present our first result , which extends alexander duality from spaces to functions . its proof needs a general algebraic result about contravariant filtrations , which we present first . [ [ compatible - pairings . ] ] compatible pairings . + + + + + + + + + + + + + + + + + + + + a _ pairing _ between two finite - dimensional vector spaces @xmath225 and @xmath226 over the field @xmath19 is a bilinear map @xmath227 . the pairing is _ non - degenerate _ if every @xmath228 has at least one @xmath229 with @xmath230 , and , symmetrically , every element @xmath229 has at least one @xmath228 with @xmath230 . the term is justified by the fact that a non - degenerate pairing implies an isomorphism between the vector spaces . to see this , we choose arbitrary bases of @xmath225 and @xmath226 , writing @xmath231 and @xmath232 for their sizes , and represent the pairing by its values between the basis elements . this gives an invertible @xmath37-@xmath16 matrix with @xmath231 rows and @xmath232 columns . reducing the matrix to diagonal form gives the claimed isomorphism . now consider two filtrations , each consisting of @xmath233 finite - dimensional vector spaces over @xmath19 , and non - degenerate pairings connecting the filtrations contravariantly : @xmath234 it will be convenient to assume that the filtrations begin and end with the trivial vector space : @xmath235 . write @xmath236 and @xmath237 for the maps upstairs and downstairs . we call the pairings _ compatible _ with these maps if @xmath238 for every @xmath239 and @xmath240 , where @xmath241 and @xmath242 . by what we said above , we have isomorphisms @xmath243 , for all @xmath244 , but we need more , namely isomorphisms that satisfy a compatibility condition with the horizontal maps . before defining what we mean , we prove a few properties of the compatible pairings . to begin , we split the vector spaces connected by @xmath245 : @xmath246 and similarly for @xmath247 and @xmath248 ; see figure [ fig : split ] . here , we have of course @xmath249 and @xmath250 . the restriction of the pairing to the kernel of @xmath245 times the image of @xmath251 is well - defined because elements of the kernel and the image are also elements of the vector spaces . in contrast , an element of a quotient is an entire coset , whose members may have inconsistent values under the pairing . to define the restriction , we assume a basis that splits along the same lines , so that the coset in @xmath252 has a unique representation of the form @xmath253 , in which the projection of @xmath254 to the kernel is zero . finally , we define the _ restricted pairing _ using @xmath254 as a proxy for the coset it defines . similarly , we define the restriction of the pairing to the cokernel of @xmath245 . we have the following properties : 1 . the pairing restricted to @xmath255 is trivial , that is : all pairs map to @xmath37 . 2 . we can find a basis of the cokernel such that the pairing restricted to @xmath256 is trivial . to prove ( a ) , let @xmath257 and @xmath258 both be non - zero . the image of @xmath259 is @xmath37 , and a preimage , @xmath254 , of @xmath260 exists . this implies @xmath261 , as claimed . to prove ( b ) , let @xmath262 and @xmath263 both be non - zero , with zero projection of @xmath264 to @xmath265 and of @xmath266 to @xmath267 . recall that @xmath264 and @xmath266 act as proxies for the cosets @xmath268 in @xmath269 and @xmath270 in @xmath271 . if @xmath272 , then there is nothing to do , else we take steps to set the value of the pairing for the cosets to zero . assume without loss of generality that @xmath264 is an element of the chosen basis of @xmath273 . by the non - degeneracy of the pairing @xmath274 , there exists @xmath275 with @xmath276 , where @xmath242 . by ( a ) , we can assume that the projection of @xmath254 to the @xmath277 is zero . by compatibility , we have @xmath230 , where @xmath241 . now we have @xmath278 , and @xmath279 because one belongs to the cokernel and the other to the image of @xmath245 . replacing @xmath264 by @xmath280 in the basis of @xmath273 , we get @xmath281 by bilinearity . we note that we still have a basis that splits as required . the operation achieves its goal by changing the restricted pairing without affecting the pairing . repeating the operation for other basis elements of the cokernel , we eventually get a basis such that the pairing restricted to @xmath256 is trivial , as claimed . [ [ compatible - isomorphisms . ] ] compatible isomorphisms . + + + + + + + + + + + + + + + + + + + + + + + + we are now ready to specify the compatibility condition for the vertical isomorphisms . first , we require that they split along the same lines as the spaces . specifically , we consider indices @xmath282 and the maps @xmath245 and @xmath283 , and we define @xmath284 see figure [ fig : four - way ] . for the case @xmath285 , these spaces represent the classes that end at @xmath143 , that extend before and after @xmath143 , that exist only at @xmath143 , and that begin at @xmath143 , in this sequence . similarly , we consider the anti - parallel maps , @xmath286 and @xmath251 , and we define the subspaces @xmath287 , @xmath288 , @xmath289 , @xmath290 downstairs . we call the vertical isomorphisms _ compatible _ with the horizontal maps if @xmath291 splits into isomorphisms @xmath292 and the maps between the subspaces compose to the identity , for every @xmath244 ; see figure [ fig : split ] . more formally , this means that the maps satisfy @xmath293 for all @xmath294 and @xmath295 . we now prove that the compatible pairings imply the existence of compatible isomorphisms . the proof is constructive . to start , we choose bases of @xmath143 and @xmath247 that split four ways , as do the spaces . representing the pairing between @xmath143 and @xmath247 by the thus defined @xmath37-@xmath16 matrix , we get a decomposition into @xmath296 blocks ; see figure [ fig : blocks ] . we use ( a ) and ( b ) to show that the bases can be chosen so that all non - diagonal blocks are zero . indeed , the upper - right four blocks are zero because @xmath297 and @xmath298 . the lower - left four blocks are zero because @xmath299 and @xmath300 . repeating the argument with reversed roles ( swapping @xmath273 with @xmath301 ) , we conclude that the remaining four non - diagonal blocks are also zero . since the entire matrix is invertible , this implies that all four diagonal blocks are invertible . we finally get the isomorphisms by turning the diagonal blocks into unit matrices . all we need here are standard row and column operations , which preserve the pairing while changing the basis for its description . the compatibility condition is therefore but a reformulation of the compatibility condition for the pairings . this completes the argument for the existence of compatible isomorphisms . the reason for our interest is the following straightforward result . let @xmath302 to @xmath147 and @xmath303 to @xmath304 be two filtrations contravariantly connected by compatible isomorphisms . then there is a bijection between the classes upstairs and downstairs such that a class is born at @xmath123 and dies entering @xmath143 iff the corresponding class is born at @xmath305 and dies entering @xmath306 . proof . recall that @xmath291 splits into four isomorphisms , for each choice of triple indices @xmath307 . the case @xmath285 implies that the births and deaths upstairs correspond to the deaths and births downstairs . condition implies that the births and deaths are matched upstairs the same way as downstairs , only backward . we see that the compatibility condition is for contravariantly connected filtrations what commutativity is for covariantly connected filtrations . while the ( covariant ) persistence equivalence theorem @xcite implies equal persistence diagrams , the contravariant p(ersistence ) e(quivalence ) theorem implies reflected persistence diagrams . [ [ alexander - duality - for - functions . ] ] alexander duality for functions . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + after getting the abstract prerequisites settled , we are ready to state and prove our first result . let @xmath52 $ ] be a perfect morse function on the @xmath11-dimensional sphere , and @xmath308 a decomposition into complementary subsets whose intersection , @xmath2 , is an @xmath3-manifold . assuming @xmath224 is tame , we have only finitely many homological critical values , including @xmath37 and @xmath16 . letting these critical values be @xmath115 to @xmath116 , we interleave them with homological regular values to get @xmath309 . we are interested in the filtrations defined by the restrictions of @xmath4 to @xmath5 and to @xmath6 . to describe them , we write @xmath310 for the @xmath8-th reduced homology group of @xmath311 , if @xmath312 , of @xmath5 , if @xmath127 , and of @xmath313 , if @xmath129 . in the same way , we define @xmath314 for @xmath315 with respect to @xmath6 . using alexander duality to form isomorphisms , we get two contravariantly connected filtrations : @xmath316 where @xmath317 . we will prove shortly that the isomorphisms implied by alexander duality are compatible with the horizontal maps . assuming this much , we get our first result , which we formulate using the superscript ` @xmath318 ' for the operation that reflects a dot across the vertical axis of a persistence diagram and , at the same time , changes its dimension from @xmath8 to @xmath319 . let @xmath5 and @xmath6 be complementary subsets of @xmath43 , and let @xmath52 $ ] be a perfect morse function whose restriction to the @xmath3-manifold @xmath2 is tame . then @xmath320 . proof . by the contravariant pe theorem , we have a bijection between the classes upstairs and downstairs that respects the matching of births and deaths . specifically , if a class @xmath142 is born at @xmath310 and dies entering @xmath321 then its corresponding class is born at @xmath322 and dies entering @xmath323 . if @xmath324 , then the class @xmath142 is represented by @xmath325 in the @xmath8-th ordinary subdiagram of @xmath217 . the corresponding class is represented by @xmath326 in the @xmath327-th relative subdiagram of @xmath216 . the reflection maps the first and second coordinate - axes of the ordinary subdiagram to the second and first coordinate - axes of the relative subdiagram . it follows that it maps @xmath325 in the former to @xmath326 in the latter subdiagram . other cases are similar , and we conclude that @xmath328 and @xmath329 are reflections of each other . writing this more succinctly gives the claimed relationship between the reduced persistence diagrams of @xmath4 restricted to the two complementary subsets of the sphere . [ [ alexander - pairing . ] ] alexander pairing . + + + + + + + + + + + + + + + + + + we fill the gap in the proof of the land and water theorem by establishing compatible pairings between the groups upstairs and downstairs . for the sake of simplicity , we restrict ourselves to the case @xmath330 . while the argument for general fields is similar , it requires oriented simplices , which is a technical formalism we prefer to avoid . we begin with the pairing implicit in lefschetz duality for manifolds with possibly non - empty boundary : @xmath331 which is defined by mapping two classes to the parity of the number of intersections between representing cycles . we get such a pairing for every regular value @xmath105 , and these pairings are compatible with the horizontal maps induced by inclusion of the sublevel sets ; see @xcite . next , we reduce the dimension of the second factor using a mapping , @xmath332 , which we compose from four simpler mappings : @xmath333 the first mapping , @xmath334 , is an isomorphism defined by excision . next , @xmath335 , is the connecting homomorphism of the exact sequence of the pair @xmath336 . since the @xmath327-th reduced homology of @xmath13 is trivial , for all @xmath337 , @xmath335 is an isomorphism for all @xmath338 . it is a surjection for @xmath339 . the third mapping , @xmath340 , is induced by inclusion . it occurs in the exact sequence of the pair @xmath341 . for @xmath95 , the reduced homology groups of @xmath342 are all trivial , implying that @xmath340 is an isomorphism . for @xmath343 , @xmath340 is the trivial isomorphism . finally , @xmath344 is again an isomorphism defined by excision . in summary , we get @xmath345 , which is an isomorphism for @xmath338 and a surjection for @xmath339 . more specifically , each class @xmath346 has two preimages . indeed , an @xmath3-cycle representing the corresponding class in @xmath347 partitions the components of @xmath75 into two subsets , and each subset generates an @xmath11-dimensional relative homology class that maps to @xmath348 . note that the roles of @xmath5 and @xmath6 can be interchanged to get a mapping from @xmath349 to @xmath350 . with this , we are ready to construct the _ alexander pairing _ : @xmath351 defined by @xmath352 , where @xmath353 is a preimage of @xmath348 under @xmath332 . this fixes the pairing of @xmath310 and @xmath354 for @xmath312 . similarly , we define @xmath355 for @xmath356 which fixes the pairing for @xmath357 . the pairing is clearly well - defined for @xmath358 , where @xmath332 is an isomorphism . the remaining two cases are symmetric , and we consider @xmath81 and @xmath346 . as noted before , @xmath348 has two preimages , @xmath353 and @xmath359 , which are generated by complementary subsets of @xmath75 . every @xmath360 is represented by an even number of points , which the complementary subsets partition into two even or two odd subsets . in either case , we have @xmath361 , so the pairing is well - defined in all cases . moreover , the compatibility of the alexander pairings within the left and right halves follows from the compatibility of the lefschetz pairings . it remains to define the paring in the middle of the sequence , for @xmath362 and @xmath363 . to this end , let @xmath142 be a reduced @xmath8-cycle in @xmath5 , @xmath348 a reduced @xmath327-cycle in @xmath6 , and @xmath353 a @xmath364-chain whose boundary is @xmath348 . we define @xmath365 by counting the intersections between @xmath142 and @xmath353 . note that @xmath353 is the preimage of @xmath348 under the previously defined mapping , @xmath345 , for any regular value @xmath105 . therefore , the pairing in the middle is compatible with the pairings in left half , again exploiting the compatibility of the lefschetz pairing . alternatively , we can define the pairing by taking a @xmath20-chain @xmath366 with boundary @xmath142 and counting its intersections with @xmath348 . in this case , @xmath366 is the preimage of @xmath142 under the map @xmath367 , and compatibility with the pairings in the right half follows . indeed , both definitions are equivalent , as already observed by lefschetz @xcite . we give a simple proof for the case @xmath330 : let @xmath142 and @xmath348 be non - intersecting reduced cycles on the @xmath11-sphere whose dimensions add up to @xmath3 . let @xmath366 and @xmath353 be chains whose boundaries are @xmath142 and @xmath348 , respectively . then @xmath368 . we can assume that @xmath142 and @xmath348 do not intersect . the intersection of @xmath366 and @xmath353 is a @xmath16-chain . its boundary consists of an even number of points , and is the disjoint union of intersections of @xmath366 with @xmath348 and of @xmath142 with @xmath353 . this implies that both types of intersections occur with the same parity . this section presents our second result , which extends the combination of alexander duality and mayer - vietoris sequences from spaces to functions . [ [ mayer - vietoris - sequence - of - filtrations . ] ] mayer - vietoris sequence of filtrations . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + assuming that @xmath224 is tame , we write @xmath309 for the interleaved sequence of homological critical and regular values , as before . the main tool in this section is the diagram obtained by connecting the filtrations of @xmath4 and its restrictions with mayer - vietoris sequences . we describe this using shorthand notation for the groups . consistent with earlier notation , we write @xmath369 for the @xmath8-th homology group of @xmath370 , for @xmath312 , of @xmath13 , for @xmath127 , and of @xmath371 , for @xmath129 . similarly , we write @xmath372 for the @xmath8-th homology group of @xmath373 , of @xmath7 , and of @xmath374 . finally , we write @xmath375 for the direct sum of the @xmath8-th homology groups of @xmath311 and @xmath376 , of @xmath5 and @xmath6 , and of @xmath313 and @xmath377 . drawing the filtrations from left to right and the mayer - vietoris sequences from top to bottom , we get @xmath378 all squares commute , which is particularly easy to see for the squares that connect groups of the same dimension , whose maps are all induced by inclusion . for @xmath84 , all groups @xmath369 are trivial . by exactness of the mayer - vietoris sequences , this implies that the maps @xmath379 are isomorphisms , for @xmath70 . the persistence diagram of @xmath224 is therefore the disjoint union of the persistence diagrams of @xmath217 and @xmath216 . more generally , we claim : let @xmath3 be a positive integer , let @xmath5 and @xmath6 be complementary subsets of @xmath43 , and let @xmath52 $ ] be a perfect morse function whose restriction to the @xmath3-manifold @xmath2 is tame . then @xmath380^c , \label{eqn : dgm-0 } \\ { { \ifmmode{\sf dgm}\else{\mbox{\(\sf dgm\)}}\fi}}_p ( f|_{{\ifmmode{{\mathbb m}}\else{\mbox{\({\mathbb m}\)}}\fi } } ) & = & { { \ifmmode{\sf dgm}\else{\mbox{\(\sf dgm\)}}\fi}}_p ( f|_{{\ifmmode{{\mathbb u}}\else{\mbox{\({\mathbb u}\)}}\fi } } ) \sqcup { { \ifmmode{\sf dgm}\else{\mbox{\(\sf dgm\)}}\fi}}_p ( f|_{{\ifmmode{{\mathbb v}}\else{\mbox{\({\mathbb v}\)}}\fi } } ) , \label{eqn : dgm } \\ { { \ifmmode{\sf dgm}\else{\mbox{\(\sf dgm\)}}\fi}}_n ( f|_{{\ifmmode{{\mathbb m}}\else{\mbox{\({\mathbb m}\)}}\fi } } ) & = & [ { { \ifmmode{\sf dgm}\else{\mbox{\(\sf dgm\)}}\fi}}_0 ( f|_{{\ifmmode{{\mathbb u}}\else{\mbox{\({\mathbb u}\)}}\fi } } ) \sqcup { { \ifmmode{\sf dgm}\else{\mbox{\(\sf dgm\)}}\fi}}_0 ( f|_{{\ifmmode{{\mathbb v}}\else{\mbox{\({\mathbb v}\)}}\fi } } ) ] ^{ct } , \label{eqn : dgm - n}\end{aligned}\ ] ] for @xmath70 , where @xmath381 stands for applying the cascade and @xmath318 stands for reflecting the diagram . we note that the assumption of @xmath3 being positive is necessary since the formulas do not hold for @xmath382 . concerning the proof of the theorem , note that is clear , and that follows from using the duality of persistence diagrams from @xcite . we will need to study the impact of the non - trivial groups @xmath383 to prove . [ [ latitudinal - manifolds . ] ] latitudinal manifolds . + + + + + + + + + + + + + + + + + + + + + + call the minimum of @xmath4 the _ south - pole _ and the maximum the _ north - pole _ of the sphere . let @xmath384 be a component of @xmath7 , and note that it is an @xmath3-manifold that decomposes @xmath13 into two complementary subsets . if it separates the two poles , we refer to @xmath384 as a _ latitudinal @xmath3-manifold_. assuming neither pole lies on @xmath7 , we order the latitudinal @xmath3-manifolds from south to north as @xmath385 . letting @xmath192 and @xmath386 be the minimum and maximum values of @xmath4 restricted to @xmath387 , we get @xmath388 as well as @xmath389 . for each component @xmath384 of @xmath7 , there are _ neighboring components _ @xmath390 of @xmath5 and @xmath391 of @xmath6 defined such that @xmath392 . a single component of @xmath5 or @xmath6 can be neighbor to an arbitrary number of @xmath3-manifolds , but not to more than two latitudinal @xmath3-manifolds . specifically , there are components @xmath393 of @xmath5 and @xmath6 such that @xmath394 for each @xmath194 . we refer to the @xmath395 as _ latitudinal components_. for example , in figure [ fig : counterexample ] , we have @xmath396 latitudinal @xmath16-manifolds and @xmath397 latitudinal components . setting @xmath198 and @xmath199 , the minimum and maximum values of @xmath4 restricted to @xmath395 are @xmath192 and @xmath398 , for @xmath181 . assuming @xmath395 belongs to @xmath5 , it gives rise to zero or more dots in the ordinary subdiagram of @xmath399 and to exactly one dot , @xmath180 , in the horizontal subdiagram . we say the dot in the horizontal subdiagram _ represents _ the @xmath37-dimensional homology class defined by @xmath395 . note that its coordinates are indexed consistently with the notation used in the introduction of the cascade . there is indeed a connection , namely the dots in the @xmath37-th horizontal subdiagram representing latitudinal components are the extreme ones in the multiset : the latitudinal components of @xmath5 and @xmath6 correspond bijectively to the extreme dots in the disjoint union of @xmath399 and @xmath400 . we omit the proof , which is not difficult . the statement includes the case in which there is no latitudinal @xmath3-manifold so that @xmath401 is the only latitudinal component . it contains both poles and is therefore represented by @xmath402 , which is the only extreme dot in the disjoint union of @xmath37-th diagrams . [ [ proof - of- . ] ] proof of . + + + + + + + + + + we compare @xmath403 with the disjoint union of @xmath404 and @xmath400 , noting that their dots all belong to the ordinary and horizontal subdiagrams . consider first a dot @xmath405 in the ordinary subdiagram of @xmath406 . it represents a component in the sublevel set that is born at a minimum @xmath407 , with @xmath408 , and that dies at a saddle point @xmath409 , with @xmath410 . let @xmath411 be the connected @xmath3-manifold that contains @xmath412 and @xmath413 , and recall that @xmath414 . assume the neighboring component @xmath390 of @xmath5 lies above @xmath412 . in the sequence of sublevel sets of @xmath217 , we see the birth of a component at @xmath408 and its death at @xmath410 . it follows that @xmath405 is also a dot in the ordinary subdiagram of @xmath399 and therefore of the disjoint union of the diagrams of @xmath217 and @xmath216 . the argument can be reversed , which implies that the @xmath37-th ordinary subdiagrams are the same . consider second a dot @xmath405 in the horizontal subdiagram of @xmath406 . it represents a connected @xmath3-manifold @xmath411 , which splits @xmath13 into two subsets . if @xmath384 is non - latitudinal , then one subset contains both poles and the other contains neither . the latter subset contains a ( non - latitudinal ) neighboring component , which is represented by @xmath405 in the horizontal subdiagram of @xmath399 or of @xmath400 . again , the argument can be reversed . if on the other hand , @xmath384 is latitudinal , then @xmath415 , for some @xmath416 , where we write @xmath417 for the minimum values and @xmath418 for the maximum values of the latitudinal @xmath3-manifolds , as before . more generally , we get the dots @xmath419 in the horizontal subdiagram of @xmath406 , for @xmath420 . adding @xmath198 and @xmath199 , we get the dots @xmath421 in the horizontal subdiagram of @xmath422 , for @xmath420 . after adding @xmath423 and @xmath424 , which are both irrelevant , we get precisely the dots specified in and in . this implies that the two diagrams are related to each other by a cascade , which completes the proof of . as an alternative to the above geometric proof of , we could give an algebraic argument that considers the filtrations of meyer - vietoris sequences ; see figure [ fig : pattern ] , which draws the filtrations as rows in a matrix , shading the area of non - trivial groups . the obstacle to @xmath425 and @xmath426 being isomorphic , and thus the reason for the cascade , are the non - trivial groups @xmath383 in the left half of the filtration . because each non - trivial @xmath383 has exactly one generator , @xmath426 must have one more generator than @xmath425 . indeed , by comparing the groups , before and after the cascade , we see that in the left half , each @xmath426 has one more generator than @xmath425 . [ [ euclidean - case . ] ] euclidean case . + + + + + + + + + + + + + + + considering that to give elegant relations between the betti numbers of the shore and the land , it is perhaps surprising that we need cascades to formulate a similar result for persistence diagrams . indeed , to suggest that the persistence diagrams of @xmath224 be the disjoint union of the persistence diagram of @xmath217 and of its reflection . the example in figure [ fig : counterexample ] shows that this simple relation does not hold in general . indeed , we have the dots @xmath427 , @xmath428 , @xmath429 , @xmath430 in the standard diagram of @xmath224 , while the standard diagram of @xmath217 contains @xmath431 and @xmath432 . this clearly violates the suggested relation . however , there is a natural setting in which the relation does hold , which we now describe . let @xmath3 be a positive integer , let @xmath433 be a compact set whose boundary is an @xmath3-manifold in @xmath434 , and suppose that @xmath435 has no homological critical values and its restriction to @xmath436 is tame . then @xmath437 . we can extend @xmath438 to a perfect morse function @xmath439 and @xmath433 to a subset @xmath5 of @xmath43 such that the persistence diagrams of @xmath438 restricted to @xmath433 and to @xmath436 are the same as those of @xmath4 restricted to @xmath5 and @xmath440 . it thus suffices to show that the persistence diagram of @xmath224 is the disjoint union of the diagram of @xmath217 and of its reflection . for @xmath441 , this follows from , from @xmath442 as stated in the land and water theorem , and the fact that the reduced diagrams are equal to the non - reduced ones . for @xmath81 , we start with @xmath443 and note that there is only one extreme component in @xmath422 , namely the one of @xmath6 that contains both the minimum and the maximum of @xmath4 . it follows that the cascade leaves @xmath399 unchanged while it turns @xmath400 into @xmath444 . using the land and water theorem again , we obtain : @xmath445^{c } \\ & = & { { \ifmmode{\sf dgm}\else{\mbox{\(\sf dgm\)}}\fi}}_0 ( f|_{{\ifmmode{{\mathbb u}}\else{\mbox{\({\mathbb u}\)}}\fi } } ) \sqcup { { \ifmmode{\sf \tilde{d}gm}\else{\mbox{\(\sf \tilde{d}gm\)}}\fi}}_0 ( f|_{{\ifmmode{{\mathbb v}}\else{\mbox{\({\mathbb v}\)}}\fi } } ) \\ & = & { { \ifmmode{\sf dgm}\else{\mbox{\(\sf dgm\)}}\fi}}_0 ( f|_{{\ifmmode{{\mathbb u}}\else{\mbox{\({\mathbb u}\)}}\fi } } ) \sqcup { { \ifmmode{\sf \tilde{d}gm}\else{\mbox{\(\sf \tilde{d}gm\)}}\fi}}_n ( f|_{{\ifmmode{{\mathbb u}}\else{\mbox{\({\mathbb u}\)}}\fi}})^t \\ & = & { { \ifmmode{\sf dgm}\else{\mbox{\(\sf dgm\)}}\fi}}_0 ( f|_{{\ifmmode{{\mathbb u}}\else{\mbox{\({\mathbb u}\)}}\fi } } ) \sqcup { { \ifmmode{\sf dgm}\else{\mbox{\(\sf dgm\)}}\fi}}_n ( f|_{{\ifmmode{{\mathbb u}}\else{\mbox{\({\mathbb u}\)}}\fi}})^t . \end{aligned}\ ] ] finally , for @xmath87 , we exploit poincar duality for manifolds , which implies @xmath446 ; see @xcite . the @xmath447-dimensional version of the theorem , for @xmath448 , brings us back full circle to the motivation for this work , namely the computation of the persistence diagram of the space - time shape formed by a moving collection of biological cells @xcite . modeling space - time as @xmath449 , the shape is compact , and we consider the time function restricted to that shape . this data satisfies the assumptions of the euclidean shore theorem , so we can infer the persistence diagram of the function on the boundary from the diagram of the function on the solid @xmath447-dimensional shape . the main contributions of this paper are two extensions of alexander duality from spaces to functions . the first extension is direct and relates the persistence diagrams of a perfect morse function restricted to two complementary subsets of the @xmath11-dimensional sphere with each other . the second extension relates the persistence diagram of the function restricted to the intersection of the complementary subsets with the diagram of the function restricted to one subset . a key tool in its proof is the filtration of mayer - vietoris sequences ( or the mayer - vietoris sequence of filtrations ) . this suggests the study of more general filtrations of exact sequences . besides the hope to develop a general purpose device that pervades persistent homology in the same way exact sequences pervade homology , we motivate the study with a few concrete questions . * is it possible to relate the extent to which a function @xmath4 on @xmath450 is not perfect morse with the severity with which this function violates our two extensions of alexander duality ? * does every morse function @xmath4 on @xmath43 have a decomposition into complementary subsets , @xmath308 , such that up to small modifications , the disjoint union of the diagrams of @xmath4 and @xmath224 is equal to the disjoint union of the diagrams of @xmath217 and @xmath216 ? * can these or similar relations be developed into a divide - and - conquer algorithm for computing the persistence diagram of @xmath4 on a sphere ?

for a hopf algebra , we endow the heisenberg double with the structure of a module algebra over the drinfeld double . based on this property , we propose that is to be the counterpart of the algebra of fields on the quantum - group side of the kazhdan lusztig duality between logarithmic conformal field theories and quantum groups . as an example , we work out the case where is the taft hopf algebra related to the quantum group that is kazhdan lusztig - dual to logarithmic conformal models . the corresponding pair is `` truncated '' to , where is a module algebra that turns out to have the form{\otimes}{\mathbb{c}}[\lambda]/(\lambda^{2p}-1)$ ] , where $ ] is the-module algebra with the relations , , and .

0905.2215

the `` logarithmic '' kazhdan lusztig duality a remarkable correspondence between logarithmic conformal field theories and quantum groups is based on a drinfeld double construction on the quantum group side @xcite . the starting point is the hopf algebra @xmath0 generated by the screening(s ) in a logarithmic model and diagonal , `` zero - mode - like '' element(s ) ( see @xcite for the two - screening case , which is relatively complicated by modern standards ) . the strategy is then to construct the drinfeld double of this quantum group and to `` slightly truncate '' it , to produce the kazhdan lusztig - dual quantum group . various aspects of the `` logarithmic '' kazhdan lusztig duality were developed in @xcite . the resulting correspondence ( ranging up to the coincidence ) in the properties of the symmetry algebra of the logarithmic model and the dual quantum group is `` circumstantial '' in that it is seen to work nicely in particular cases , although no general argument for its existence has been developed or attempted . that the drinfeld double of @xmath0 plays a crucial role in this correspondence was a serendipitous finding in @xcite . modulo the `` slight truncation '' mentioned above , the drinfeld double is a counterpart of the symmetry algebra ( `` the '' triplet @xcite or a higher one @xcite ) of a given logarithmic conformal field model . in this paper , we propose another algebraic object that may play a role in the logarithmic kazhdan lusztig duality , being a counterpart of the algebra of fields in logarithmic models . we here mean the fields describing logarithmic models in manifestly quantum - group - invariant terms ( i.e. , `` carrying quantum - group indices '' ) , as a generalization of the symplectic fermions @xcite . the necessary algebraic requirement is that the quantum group act `` covariantly '' on products of fields , which is expressed as the module algebra axiom @xmath13 , where we use the sweedler notation @xmath14 for the coproduct . we now describe a @xmath2-module algebra that is to play the role of fields on the algebraic side . for a hopf algebra @xmath0 , the drinfeld double @xmath2 is @xmath15 as a vector space . the same vector space admits another characteristic algebraic structure , a ( semisimple ) associative algebra given by the smash product with respect to the ( left ) regular action of @xmath0 on @xmath16 , or , in the established terminology traced back to @xcite , a heisenberg double ( see , e.g. , @xcite ) , specifically , the heisenberg double @xmath17 of @xmath16 . the main observation in this paper is that _ for any hopf algebra @xmath0 with bijective antipode , @xmath1 is a @xmath2-module algebra_. this requires introducing a new @xmath2 action ( which may be termed `` heterotic '' because it is obtained by combining , in a sense , a left and a right @xmath2 actions ) . as is the case with the drinfeld double @xmath2 , the heisenberg double @xmath1 turns out to be `` slightly too big '' for the correspondence with logarithmic models , but for the @xmath18-dimensional quantum group @xmath3 at the @xmath19th root of unity dual to the @xmath4 logarithmic conformal models , the corresponding @xmath1 nicely allows a `` truncation '' to a @xmath18-dimensional @xmath3-module algebra . we prove the general statement in sec . [ sec : thm ] and detail the @xmath3 example in sec . [ sec : sl2 ] . the definition of the drinfeld double is recalled in appendix [ app : d - double ] . in appendix [ app : lcftrem ] , we collect some motivation coming from logarithmic conformal field theories . let @xmath0 be a hopf algebra . in this section , we make @xmath1 into a @xmath2-module algebra . for this , we combine two well - known @xmath2 actions , which can be taken from different sources , among which we prefer the beautiful paper @xcite . [ [ section ] ] we use the `` tickling '' notation for the left and right regular actions : for a hopf algebra @xmath20 , its left and right regular actions on @xmath21 are respectively given by @xmath22 and @xmath23 , where @xmath24 and @xmath25 . it follows that @xmath21 is an @xmath20-bimodule under these actions ( and @xmath26 is the evaluation ) . we also have the left and right actions of @xmath21 on @xmath20 , @xmath27 and @xmath28 . [ [ section-1 ] ] we recall that the heisenberg double @xmath1 is the smash product @xmath29 with respect to the left regular action of @xmath0 on @xmath16 , which means that the composition in @xmath1 is given by @xmath30 we now describe the @xmath2 action on @xmath1 making it into a @xmath2-module algebra . first , the @xmath2 action on @xmath16the first factor in @xmath31is given by the restriction of the left regular action of @xmath2 on @xmath32 , which is @xcite @xmath33 restricting this to @xmath34 gives @xmath35 under which @xmath16 is an @xmath36-commutative @xmath2-module algebra @xcite ( also see @xcite ) . carrying an action of a quasitriangular hopf algebra @xmath20 is called @xmath36-commutative , or quantum commutative @xcite if @xmath37 for all @xmath38 , where the dot denotes the action and @xmath39 is the universal @xmath36-matrix . ] second , the @xmath2 action on @xmath0 is obtained by restricting the right regular action of @xmath2 on @xmath32 to @xmath40 and using the antipode to convert it into a left action @xcite . with the right regular action of @xmath2 on @xmath41 given by @xcite @xmath42 its restriction to @xmath0 is @xmath43 . replacing @xmath44 here with @xmath45 we readily calculate @xmath46 , which defines the left action @xcite @xmath47 under which @xmath0 is an @xmath36-commutative @xmath2-module algebra ( also see @xcite ) . we now define a @xmath2 action on @xmath1 , also denoted by @xmath48 , simply by setting , and hence , in accordance with the drinfeld double construction , @xmath49 , with the coproducts of @xmath16 and @xmath0 in the right - hand side . ] @xmath50 that is , @xmath51 \notag \mu{\otimes}m\in{\mathscr{d}}(b),\quad \alpha{\mathbin{\hash}}a\in{\mathscr{h}}(b^*),\end{aligned}\ ] ] and prove that @xmath1 is then a @xmath2-module algebra . because each factor in @xmath52 is already a @xmath2-module algebra , it suffices to show that @xmath53 we evaluate the left - hand side : @xmath54 & \phantom{{}= { } } \text{\mbox{}\hfill(because $ \delta((\mu{\otimes}m){\mathbin{\mbox{\small${\vartriangleright}$}}}a ) = \bigl(m ' a ' s(m''''){\kern1pt{\leftharpoonup}\kern1pt}{{s^*}^{-1}}(\mu)\bigr ) { \otimes}m '' a '' s(m''')$ ) } \\[2pt ] & = ( \mu'{\otimes}m^{(5)}{\kern1pt{\rightharpoonup}\kern1pt}\beta)'{\mathbin{\hash}}m^{(2 ) } a '' s(m^{(3 ) } ) { \langle{{s^*}^{-1}}(\mu'')(\mu'{\otimes}m^{(5)}{\kern1pt{\rightharpoonup}\kern1pt}\beta)'',\ , m^{(1 ) } a ' s(m^{(4)})\rangle\,}\\*[-2pt ] & \phantom{{}= { } } \text{(simply because $ ( a{\kern1pt{\leftharpoonup}\kern1pt}\alpha){\kern1pt{\rightharpoonup}\kern1pt}\beta = \beta'{\langle\alpha\beta'',\,a\rangle\,}$ ) } \\[2pt ] & = \mu^{(3)}\beta'{{s^*}^{-1}}(\mu^{(2)}){\mathbin{\hash}}m^{(2 ) } a '' s(m^{(3 ) } ) \\ * & \qquad\qquad\qquad{}\times { \langle{{s^*}^{-1}}(\mu^{(5)})\mu^{(4)}(m^{(5)}{\kern1pt{\rightharpoonup}\kern1pt}\beta '' ) { { s^*}^{-1}}(\mu^{(1)}),\,m^{(1 ) } a ' s(m^{(4)})\rangle\ , } \\*[-2pt ] & \phantom{{}= { } } \text{(because $ \delta((\mu{\otimes}m){\kern1pt{\rightharpoonup}\kern1pt}\beta ) = \mu'''\beta'{{s^*}^{-1}}(\mu''){\otimes}\mu''''(m{\kern1pt{\rightharpoonup}\kern1pt}\beta '' ) { { s^*}^{-1}}(\mu')$ ) } \\[2pt ] & = \mu^{(3)}\beta'{{s^*}^{-1}}(\mu^{(2)}){\mathbin{\hash}}m^{(3)}a ' '' s(m^{(4 ) } ) \\ * & \qquad\qquad\qquad{}\times { \langlem^{(7)}{\kern1pt{\rightharpoonup}\kern1pt}\beta'',\,m^{(1 ) } a ' s(m^{(6)})\rangle\ , } { \langle{{s^*}^{-1}}(\mu^{(1)}),\,m^{(2 ) } a '' s(m^{(5)})\rangle\ , } \\ & = \mu'''(m'a'{\kern1pt{\rightharpoonup}\kern1pt}\beta){{s^*}^{-1}}(\mu '' ) { \mathbin{\hash}}\bigl((m '' a '' s(m''')){\kern1pt{\leftharpoonup}\kern1pt}{{s^*}^{-1}}(\mu ' ) \bigr ) \end{aligned } \\=\bigl((m'{\otimes}\mu''){\kern1pt{\rightharpoonup}\kern1pt}(a'{\kern1pt{\rightharpoonup}\kern1pt}\beta)\bigr ) { \mathbin{\hash}}\bigl((m''{\otimes}\mu'){\mathbin{\mbox{\small${\vartriangleright}$}}}a''\bigr),\end{gathered}\ ] ] which is the desired result . as already noted , each of the subalgebras @xmath55 and @xmath56 in @xmath1 is known to be @xmath36-commutative with respect to the corresponding action or of @xmath2 . but @xmath1 is _ not _ @xmath36-commutative with respect to the action in in general : the @xmath36-commutativity axiom is satisfied for only `` half '' the cross - relations , @xmath57 but not for the other half : @xmath58 in general . in this section , we consider the pair @xmath5 for the taft hopf algebra @xmath0 that underlies the kazhdan lusztig correspondence with the @xmath4 logarithmic conformal field theory models . by `` truncation , '' @xmath2 yields the @xmath3 quantum group that is kazhdan lusztig - dual to the @xmath4 logarithmic models . this quantum group first appeared in @xcite and was rediscovered , together with its role in the kazhdan lusztig correspondence , in @xcite ; its further properties were considered in @xcite and , notably , recently in @xcite ( also see @xcite for a somewhat larger quantum group ) . we recall this in * [ the - b]*. we evaluate @xmath1 in * [ hd - relations ] * , and in * [ to - sl2 ] * `` truncate '' @xmath5 to a pair @xmath59 , where @xmath7 is a @xmath3-module algebra . its structure is detailed in * [ sec : structure]*. for an integer @xmath61 , we set @xmath62 and recall some of the results in @xcite . let @xmath63 be the @xmath60-dimensional hopf algebra generated by @xmath64 and @xmath65 with the relations @xmath66 and with the comultiplication , counit , and antipode given by @xmath67 we next introduce elements @xmath68 as @xmath69 then @xcite @xmath70 moreover , straightforward calculation shows @xcite that the drinfeld double @xmath2 ( see appendix [ app : d - double ] ) is the hopf algebra generated by @xmath64 , @xmath71 , @xmath65 , and @xmath72 with the relations given by * relations in @xmath0 , * the relations @xmath73 in @xmath16 , and * the cross - relations @xmath74={\raisebox{.8pt}{\mbox{\small$\displaystyle\frac { k^2-\varkappa^2}{{\mathfrak{q}}-{\mathfrak{q}}^{-1}}$}}}. \end{gathered}\ ] ] here , in accordance with writing @xmath75 , @xmath64 and @xmath65 are of course understood as @xmath76 and @xmath77 , and @xmath71 and @xmath72 as @xmath78 and @xmath79 . then , for example , the last relation in is to be rewritten as@xmath80 dropping the @xmath81 inside @xmath2 again , we have the hopf - algebra structure @xmath82 given by and @xmath83 ( we reiterate that the coalgebra structure on @xmath2 is the direct product of those on @xmath84 and @xmath0 ) . for the above @xmath0 , @xmath1 is spanned by @xmath85 where @xmath86 , @xmath87 , @xmath88 , and @xmath89 . to evaluate the product in @xmath1 , defined in , we first write the left regular action of @xmath0 on @xmath16 , @xmath90 : @xmath91}{0pt}{}{a}{m}$}}}{{\genfrac{[}{]}{0pt}{}{a}{m } } } { { \genfrac{[}{]}{0pt}{}{a}{m}}}{{\genfrac{[}{]}{0pt}{}{a}{m}}}}{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{{[m]!\,}}{({\mathfrak{q}}- { \mathfrak{q}}^{-1})^m}$}}}\ , { \mathfrak{q}}^{-(b+2a)\frac{n}{2 } - m(a + b ) + { \mathchoice{{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{1}{2}$}}}}{\frac{1}{2}}{\frac{1}{2}}{\frac{1}{2}}}m(m+1 ) } f^{a - m}\varkappa^b.\end{gathered}\ ] ] it then follows that @xmath92}{0pt}{}{m}{s}$}}}{{\genfrac{[}{]}{0pt}{}{m}{s } } } { { \genfrac{[}{]}{0pt}{}{m}{s}}}{{\genfrac{[}{]}{0pt}{}{m}{s } } } } { \mathchoice { { \mbox{\footnotesize$\displaystyle \genfrac{[}{]}{0pt}{}{a}{s}$}}}{{\genfrac{[}{]}{0pt}{}{a}{s } } } { { \genfrac{[}{]}{0pt}{}{a}{s}}}{{\genfrac{[}{]}{0pt}{}{a}{s}}}}{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{{[s]!\,}}{({\mathfrak{q}}- { \mathfrak{q}}^{-1})^s}$}}}\ , { \mathfrak{q}}^{-(b+2a)\frac{n}{2 } + s(m - a - b ) } f^{a - s}\varkappa^b { \mathbin{\hash}}e^{m - s}k^{2s+n}\end{gathered}\ ] ] ( the sum is limited above by @xmath93 due to the binomial coefficient vanishing ) . in particular , @xmath94 and also @xmath95 , @xmath96 , and @xmath97@xmath98 . for the future reference , we write the general case , obtained from immediately : @xmath99}{0pt}{}{m}{u}$}}}{{\genfrac{[}{]}{0pt}{}{m}{u } } } { { \genfrac{[}{]}{0pt}{}{m}{u}}}{{\genfrac{[}{]}{0pt}{}{m}{u } } } } { \mathchoice { { \mbox{\footnotesize$\displaystyle \genfrac{[}{]}{0pt}{}{a}{u}$}}}{{\genfrac{[}{]}{0pt}{}{a}{u } } } { { \genfrac{[}{]}{0pt}{}{a}{u}}}{{\genfrac{[}{]}{0pt}{}{a}{u}}}}{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{{[u]!\,}}{({\mathfrak{q}}- { \mathfrak{q}}^{-1})^u}$}}}\ , { \mathfrak{q}}^{- { \mathchoice{{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{1}{2}$}}}}{\frac{1}{2}}{\frac{1}{2}}{\frac{1}{2}}}b n + c n + a ( s - n ) + u(2 c - a - b + m - s ) } \\*[-6pt ] { } \times f^{a+r - u}\varkappa^{b+s } { \mathbin{\hash}}e^{m + c - u}k^{n+d+2u}.\end{gathered}\ ] ] ( this is an associative product for generic @xmath100 as well . ) we next evaluate the @xmath2 action on @xmath1 . the @xmath2 action on @xmath16 in , rewritten in terms of the comultiplication and antipode of the double , @xmath101 factors into the action of @xmath102 in times the action of @xmath103 given by @xmath104}\ ; f^{i+a}\varkappa^{b}.\ ] ] the @xmath2 action on @xmath0 in , @xmath105 , with @xmath106 and @xmath107 , factors through the adjoint action of @xmath108 , @xmath109}\bigr ) e^{a+m } k^{b-2m},\ ] ] and the action of @xmath110 , given by @xmath111 : @xmath112}{0pt}{}{a}{i}$}}}{{\genfrac{[}{]}{0pt}{}{a}{i } } } { { \genfrac{[}{]}{0pt}{}{a}{i}}}{{\genfrac{[}{]}{0pt}{}{a}{i}}}}{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{{[i]!\,}}{({\mathfrak{q}}-{\mathfrak{q}}^{-1})^i}$}}}\ , { \mathfrak{q}}^{\frac{b j}{2 } - { \mathchoice{{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{1}{2}$}}}}{\frac{1}{2}}{\frac{1}{2}}{\frac{1}{2}}}i(i+1 ) + i(j+a ) } e^{a - i } k^{2i+b}.\ ] ] the action in is therefore given by @xmath113}{0pt}{}{m}{s}$}}}{{\genfrac{[}{]}{0pt}{}{m}{s } } } { { \genfrac{[}{]}{0pt}{}{m}{s}}}{{\genfrac{[}{]}{0pt}{}{m}{s}}}}{\mathchoice { { \mbox{\footnotesize$\displaystyle \genfrac{[}{]}{0pt}{}{a}{s}$}}}{{\genfrac{[}{]}{0pt}{}{a}{s } } } { { \genfrac{[}{]}{0pt}{}{a}{s}}}{{\genfrac{[}{]}{0pt}{}{a}{s}}}}{[s]!\,}\\[-4pt ] & \quad{}\times \bigl(\prod_{\ell=1}^{m - s}{[\ell-1-{\raisebox{.9pt}{\mbox{\scriptsize$\displaystyle\frac{d}{2}$}}}]}\bigr ) ( { \mathfrak{q}}-{\mathfrak{q}}^{-1})^{m-2s } f^{a - s}\varkappa^b{\mathbin{\hash}}e^{c+m - s } k^{d - 2 m + 2 s } , \\ k{\mathbin{\mbox{\small${\vartriangleright}$}}}(f^a\varkappa^b{\mathbin{\hash}}e^c k^d ) & = { \mathfrak{q}}^{-a + c - \frac{b}{2}}(f^a\varkappa^b{\mathbin{\hash}}e^c k^d ) , \\ \varkappa{\mathbin{\mbox{\small${\vartriangleright}$}}}(f^a\varkappa^b{\mathbin{\hash}}e^c k^d ) & = { \mathfrak{q}}^{a + \frac{d}{2}}(f^a\varkappa^b{\mathbin{\hash}}e^c k^d ) , \\ f^i{\mathbin{\mbox{\small${\vartriangleright}$}}}(f^a\varkappa^b{\mathbin{\hash}}e^c k^d ) & = { \mathfrak{q}}^ { { \mathchoice{{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{1}{2}$}}}}{\frac{1}{2}}{\frac{1}{2}}{\frac{1}{2}}}i(i-1)}\sum_{s{\geqslant}0}(-1)^i{\mathfrak{q}}^{-s^2}{\mathchoice { { \mbox{\footnotesize$\displaystyle \genfrac{[}{]}{0pt}{}{i}{s}$}}}{{\genfrac{[}{]}{0pt}{}{i}{s } } } { { \genfrac{[}{]}{0pt}{}{i}{s}}}{{\genfrac{[}{]}{0pt}{}{i}{s}}}}{\mathchoice { { \mbox{\footnotesize$\displaystyle \genfrac{[}{]}{0pt}{}{c}{s}$}}}{{\genfrac{[}{]}{0pt}{}{c}{s } } } { { \genfrac{[}{]}{0pt}{}{c}{s}}}{{\genfrac{[}{]}{0pt}{}{c}{s}}}}{[s]!\ , } { \mathfrak{q}}^ { { \mathchoice{{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{1}{2}$}}}}{\frac{1}{2}}{\frac{1}{2}}{\frac{1}{2}}}b(i - s ) + a i + a s + s c}\\[-2pt ] & \quad{}\times \bigl(\prod_{\ell=1}^{i - s}{[\ell + a - 1 + { \raisebox{.9pt}{\mbox{\scriptsize$\displaystyle\frac{b}{2}$}}}]}\bigr ) ( { \mathfrak{q}}- { \mathfrak{q}}^{-1})^{i - 2 s } f^{a + i - s}\varkappa^b{\mathbin{\hash}}e^{c - s } k^{d + 2 s}.\end{aligned}\ ] ] the `` truncation '' whereby @xmath2 yields @xmath3 @xcite consists of two steps : first , taking the quotient @xmath114 by the hopf ideal generated by the central element @xmath115 and , second , identifying @xmath3 as the subalgebra in @xmath116 spanned by @xmath117 ( tensor product omitted ) with @xmath118 and @xmath119 . it follows from the above formulas for @xmath120 and from formulas for the antipode that @xmath3 is a hopf algebra . in @xmath1 , dually , we take a subalgebra and then a quotient , as follows . first , dually to taking the quotient in , we identify the subspace @xmath121 on which @xmath122 acts by unity . it follows from the above formulas for the @xmath2 action that @xmath123 two nice properties immediately follow : from , @xmath124 is a subalgebra , and from * [ dd - action ] * , the @xmath2 action restricts to @xmath124 . second , dually to the restriction @xmath125 , we take a quotient of @xmath124 . it follows from @xmath126 that the eigenvalues of @xmath127 are not all different for @xmath128 ; we can impose the additional relation @xmath129 in @xmath124 , is central in @xmath124 , which suffices for our purposes , although it is not central in @xmath1 , where @xmath130 and @xmath131 . ] i.e. , pass to the quotient by the relations @xmath132 this defines the @xmath18-dimensional algebra @xmath7 , which is a @xmath3 module algebra . [ [ change - basis ] ] being a semisimple associative algebra , a heisenberg double decomposes into matrix algebras . for our @xmath1 , we choose the generators as @xmath133 , where @xmath72 is understood as @xmath134 and we set @xmath135 the relations in @xmath1 are then equivalent to @xmath136 ( where the unity in is of course @xmath137 in the detailed nomenclature used above ) . clearly , @xmath138 , @xmath139 , and @xmath140 generate a subalgebra , which is in fact @xmath124 . its quotient by @xmath141 gives @xmath7 . it follows that as an associative algebra,@xmath142}{\otimes}({\mathbb{c}}[\lambda]/(\lambda^{2p}-1)),\ ] ] where @xmath143}$ ] is the @xmath144-dimensional algebra defined by relations and . it is indeed isomorphic to the full matrix algebra @xmath145 @xcite ( also see @xcite ) ; hence , @xmath146/(\lambda^{2p}-1)\bigr)$ ] . the @xmath3 action on the new generators of @xmath1 is readily seen to be given by @xmath147}\,\lambda^n\,z , & k^2{\mathbin{\mbox{\small${\vartriangleright}$}}}\lambda^n&={\mathfrak{q}}^{-n}\lambda , & f{\mathbin{\mbox{\small${\vartriangleright}$}}}\lambda^n&=-{\mathfrak{q}}^{\frac{n}{2}}{[{\raisebox{.9pt}{\mbox{\scriptsize$\displaystyle\frac{n}{2}$}}}]}\,\lambda^n\,{\partial } , \\ e{\mathbin{\mbox{\small${\vartriangleright}$}}}z^m & = -{\mathfrak{q}}^m { [ m ] } z^{m+1 } , & k^2{\mathbin{\mbox{\small${\vartriangleright}$}}}z^m & = { \mathfrak{q}}^{2m}\,z^m , & f{\mathbin{\mbox{\small${\vartriangleright}$}}}z^m & = { [ m ] } { \mathfrak{q}}^{1-m}\,z^{m-1 } , \\ e{\mathbin{\mbox{\small${\vartriangleright}$}}}{\partial}^n & = { \mathfrak{q}}^{1-n}{[n]}{\partial}^{n-1},\quad & k^2{\mathbin{\mbox{\small${\vartriangleright}$}}}{\partial}^n & = { \mathfrak{q}}^{-2 n } { \partial}^n,\quad & f{\mathbin{\mbox{\small${\vartriangleright}$}}}{\partial}^n & = -{\mathfrak{q}}^n { [ n ] } { \partial}^{n+1}\end{aligned}\ ] ] as we have already noted ( and as is very clearly seen now ) , the action restricts to @xmath124 and then pushes forward to @xmath7 . there , it restricts to the subalgebra @xmath143}$ ] , and the isomorphism @xmath148}\cong{\mathrm{mat}}_p({\mathbb{c}})\ ] ] is actually that of @xmath3-module algebras @xcite . [ [ sec : czd - decomp ] ] under the above action , @xmath143}$ ] decomposes into indecomposable @xmath3 representations as @xcite @xmath149}={{\mathscr}{p}}^+_1\oplus{{\mathscr}{p}}^+_3\oplus\dots\oplus{{\mathscr}{p}}^+_\nu,\ ] ] where @xmath150 if @xmath151 is even and @xmath152 if @xmath151 is odd , and where @xmath153 is the projective cover of the @xmath3 irreducible representation with weight @xmath154 ( in particular , @xmath155 is the cover of the trivial representation ; see @xcite for a detailed description ) . the @xmath19-dimensional projective module @xmath155 in has the remarkable structure @xmath156}}$}}}\,z^i\,{\partial}^i \ar^(.7){f}[dr ] \ar_(.7){e}[dl ] \\ z^{p-1 } & \kern-10pt\rightleftarrows\kern-10pt & z^{p-2 } & \kern-10pt\rightleftarrows \ldots \rightleftarrows\kern-10pt & z \ar_{f}[dr ] & { } & { \partial}\ar^{e}[dl ] & \kern-10pt\rightleftarrows \ldots \rightleftarrows\kern-10pt & { \partial}^{p-2 } & \kern-10pt\rightleftarrows\kern-10pt & { \partial}^{p-1 } \\ & & & & & 1 } \ ] ] where the horizontal left right arrows denote the action of @xmath64 ( to the left ) and @xmath71 ( to the right ) up to nonzero factors ( and there are no maps inverse to the tilted arrows ) . as regards all of @xmath7 , its decomposition into indecomposable @xmath3 representations involves not just the `` odd '' projective modules as in but actually _ all projective @xmath3 modules with the multiplicity of each equal to the dimension of its irreducible quotient _ : @xmath157 where @xmath158 is the projective cover of the irreducible representation with weight @xmath159 . the multiplicities in are identical to those in the regular representation decomposition . the sum in is nothing but the @xmath138-independent subalgebra in @xmath7 . decomposition follows by first noting the evident fact that the @xmath3 action on @xmath7 does not change the degree in @xmath138 , and then proceeding much as in @xcite . for example , one of the two copies of @xmath160 involved in is given by @xmath161&\kern-10pt\rightleftarrows\kern-10pt & t_-\ar^(.7){f}[dr ] \\ l_{p-2 } & \kern-10pt\rightleftarrows \ldots \rightleftarrows\kern-10pt & l_1 \ar_{f}[dr ] & & & & r_1 \ar^{e}[dl ] & \kern-10pt\rightleftarrows \ldots \rightleftarrows\kern-10pt & r_{p-2 } \\ & & & b_+&\kern-10pt\rightleftarrows\kern-10pt&b_- } \ ] ] where @xmath162 \intertext{with } \alpha_{i } = \smash[t]{\sum_{j=1}^{i}{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{{\mathfrak{q}}^{j + { \mathchoice{{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{1}{2}$}}}}{\frac{1}{2}}{\frac{1}{2}}{\frac{1}{2}}}}}{{[j - { \mathchoice{{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{1}{2}$}}}}{\frac{1}{2}}{\frac{1}{2}}{\frac{1}{2}}}]}}$ } } } } , \quad c_{i } = { \mathfrak{q}}^{\frac{i}{2}}\smash[t]{\prod_{n=1}^{i } { \raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{{[n - { \mathchoice{{\raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{1}{2}$}}}}{\frac{1}{2}}{\frac{1}{2}}{\frac{1}{2}}}]}}{{[n]}}$ } } } } , \\[-6pt ] \intertext{and } l_1 = { \raisebox{.5pt}{\mbox{\footnotesize$\displaystyle\frac{{\mathfrak{q}}^2}{{\mathfrak{q}}^2 + 1}$ } } } \smash[t]{\sum_{i=0}^{p-3}}c_{i}\,\lambda\,z^{i+2}\,{\partial}^{i } , \quad b_+ = \smash[t]{\sum_{i=0}^{p - 2 } } c_{i}\,\lambda z^{i + 1}\,{\partial}^{i}.\end{gathered}\ ] ] this construction , being a linear - in-@xmath138 analogue of , does not fully share its utmost simplicity , except possibly at one point : @xmath163 in is proportional to @xmath164 ; in the other copy of @xmath160 in , linear in @xmath165 , @xmath166 is proportional to @xmath167 . we also note that the subspace of degree @xmath151 in @xmath138 decomposes into the sum @xmath168 of @xmath169 modules with multiplicities @xmath170 ; because @xmath141 , there is the _ subalgebra _ { { \mathscr}{p}}^+_1\oplus{{\mathscr}{p}}^-_1\oplus{{\mathscr}{p}}^+_3\oplus{{\mathscr}{p}}^-_3\oplus \dots\oplus{{\mathscr}{p}}^+_\nu\oplus{{\mathscr}{p}}^-_\nu\ ] ] of _ all _ `` odd '' projective modules in @xmath7 . [ [ czd - diff ] ] we also recall from @xcite that @xmath143}$ ] extends to a _ differential _ @xmath3-module algebra @xmath171}$ ] ( a quantum de rham complex of @xmath143}$ ] ) , which is the unital algebra with the generators @xmath139 , @xmath140 , @xmath172 , @xmath173 and the relations ( in addition to and ) @xmath174 the differential acting as @xmath175 ( and @xmath176 ) commutes with the @xmath3 action if this is defined on @xmath172 and @xmath173 as @xmath177}z\,{\underline{dz}},&\qquad k^2{\mathbin{\mbox{\small${\vartriangleright}$}}}{\underline{dz}}&= { \mathfrak{q}}^2 { \underline{dz}},&\qquad f{\mathbin{\mbox{\small${\vartriangleright}$}}}{\underline{dz}}&= 0,\\ e{\mathbin{\mbox{\small${\vartriangleright}$}}}{\underline{d{\partial}}}&= 0 , & k^2{\mathbin{\mbox{\small${\vartriangleright}$}}}{\underline{d{\partial}}}&= { \mathfrak{q}}^{-2 } { \underline{d{\partial } } } , & f{\mathbin{\mbox{\small${\vartriangleright}$}}}{\underline{d{\partial}}}&= -{\mathfrak{q}}^{2}{[2]}{\partial}\,{\underline{d{\partial}}}\end{alignedat}\ ] ] and is then extended to all of @xmath171}$ ] in accordance with the module algebra property . in fact , _ the entire @xmath7 extends to a differential @xmath3-module algebra_. let @xmath178 be the algebra on @xmath139 , @xmath140 , @xmath138 , @xmath172 , @xmath173 , and @xmath179 with the relations given by those in @xmath171}$ ] and @xmath7 and the following ones : * @xmath180 , @xmath181 , * @xmath179 commutes with @xmath139 and @xmath140 and anticommutes with @xmath172 and @xmath173 , * @xmath182 ( whence , in particular , @xmath183 ) , * @xmath138 commutes with @xmath172 and @xmath173 . then @xmath178 endowed with the @xmath3 action @xmath184 is a differential @xmath3-module algebra . we expect not only the drinfeld double @xmath2 but also the pair @xmath5 , with @xmath1 being a @xmath2-module algebra , to play a fundamental role on the quantum group side of the logarithmic kazhdan lusztig duality . based on the general recipe in sec . [ sec : thm ] , the contents of sec . [ sec : sl2 ] must have a counterpart for the quantum group @xmath185 that is kazhdan lusztig - dual to the @xmath186 logarithmic conformal field models @xcite ; hopefully , a `` truncation '' of the appropriate drinfeld double would also allow its dual version for the corresponding heisenberg double , yielding the pair @xmath187 , where @xmath188 is a @xmath185-module algebra . this paper was finished in the very inspiring atmosphere of the lcft meeting at eth , rh.21 zurich ( may 2009 ) , and it is a pleasure to thank m. gaberdiel for the kind hospitality . i am grateful to j. fuchs , a. gainutdinov , v. gurarie , p. mathieu , j. rasmussen , p. ruelle , i. runkel , and c. schweigert for stimulating discussions . special thanks , also for stimulation , go to g. mutafyan . i thank the referee for the useful comments . this work was supported in part by the rfbr grant 07 - 01 - 00523 , the rfbr cnrs grant 09 - 01 - 93105 , and the grant lss-1615.2008.2 . we recall that for a hopf algebra @xmath0 with bijective antipode , its drinfeld double @xmath2 is @xmath15 as a vector space , endowed with the structure of a quasitriangular hopf algebra as follows . the coalgebra structure is that of @xmath189 , the algebra structure is @xmath190 for all @xmath191 and @xmath192 , the antipode is given by @xmath193 and the universal @xmath36-matrix is @xmath194 where @xmath195 is a basis of @xmath0 and @xmath196 its dual basis in @xmath16 . referring to the @xmath4 logarithmic conformal models , we here emphasize several features that may be captured on the algebraic side by @xmath143}$]the `` noncommutative part '' of @xmath7and its de rham complex @xmath171}$ ] ( secs . * [ sec : czd - decomp ] * and * [ czd - diff ] * ) . the basic observation is that the fields of the simplest , @xmath197 logarithmic model arrange under the action of @xmath64 and @xmath71 into the same diagrams as certain elements of @xmath171}$ ] at @xmath198 ( i.e. , @xmath199 ) . guided by the quantum group symmetry , we then expect that for @xmath200 , the corresponding diagrams for @xmath171}$ ] capture some structures to be observed in the @xmath4 models when these are described in terms of fields manifestly covariant under @xmath3 . the quantum group analysis may thus help find this ( so far hypothetical ) description of logarithmic conformal models . we proceed from the free - fermion description of the @xmath197 logarithmic conformal field model . the starting point is the usual system of two free fermion fields @xmath201 and @xmath202 with the respective conformal weights @xmath203 and @xmath170 , whose ope is @xmath204 virasoro generators with central charge @xmath205 are the modes of the energy momentum tensor @xmath206 where @xmath207 and the normal - ordered product is understood in the right - hand side . it follows that two screenings operators commuting with this virasoro algebra are given by @xmath208 ( the `` long '' screening ) . the relevant complex of ( feigin fuchs ) virasoro modules is @xmath209{0pt}{7pt}\bullet}\ar@{}|(.3){1}[0,0];[0,0]+<0pt,12pt>;}="1s"&{*{\rule[-1pt]{0pt}{7pt}\circ}\ar@{}|(.3){\xi}[0,0];[0,0]+<0pt,12pt>;}="m1ms"\\ & { * { \rule[-1pt]{0pt}{7pt}\bullet}\ar@{}|(.3){\eta}[0,0];[0,0]+<0pt,12pt>;}="1ms"&{*{\rule[-1pt]{0pt}{7pt}\circ}\ar@{}|(.3){\quad\eta\xi}[0,0];[0,0]+<0pt,12pt>;}="1s2"&{*{\rule[-1pt]{0pt}{7pt}\bullet}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="m1ms2 " & { * { \rule[-1pt]{0pt}{7pt}\circ}\ar@{}|(.3){{d}\xi\,\xi}[0,0];[0,0]+<0pt,12pt>;}="m1s"\\ { * { \rule[-1pt]{0pt}{7pt}\bullet}\ar@{}|(.3){{d}\eta\eta}[0,0];[0,0]+<0pt,12pt>;}="3s"&{*{\rule[-1pt]{0pt}{7pt}\circ}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="1ms2"&{*{\rule[-1pt]{0pt}{7pt}\bullet}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="1s3"&{*{\rule[-1pt]{0pt}{7pt}\circ}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="m1ms3"&{*{\rule[-1pt]{0pt}{7pt}\bullet}\ar@{}|(.3){\qquad\quad{d}^2\xi{d}\,\xi}[0,0];[0,0]+<0pt,12pt>;}="m1s2"&{*{\rule[-1pt]{0pt}{7pt}\circ}\ar@{}|(.3){\quad{d}^2\xi{d}\,\xi\,\xi}[0,0];[0,0]+<0pt,12pt>;}="m3ms"\\ { * { \rule[-1pt]{0pt}{7pt}\circ}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="3s2"&{*{\rule[-1pt]{0pt}{7pt}\bullet}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="1ms3"&{*{\rule[-1pt]{0pt}{7pt}\circ}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="1s4"&{*{\rule[-1pt]{0pt}{7pt}\bullet}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="m1ms4"&{*{\rule[-1pt]{0pt}{7pt}\circ}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="m1s3"&{*{\rule[-1pt]{0pt}{7pt}\bullet}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="m3ms2"&{*{\rule[-1pt]{0pt}{7pt}\circ}\ar@{}|(.3){}[0,0];[0,0]+<0pt,12pt>;}="m3s " \ar"1ms2";"3s"_e \ar"1s2";"1ms"_e \ar"m1ms";"1s"_e \ar"m1s";"m1ms2"_e \ar"m3ms";"m1s2"^e \ar"m3s";"m3ms2"_e \ar"1s4";"1ms3"_e \ar"m1ms3";"1s3"_e \ar"m1s3";"m1ms4"_e \ar"1ms2";"1ms " \ar"1ms2";"1ms3 " \ar"3s2";"3s " \ar"1s2";"1s " \ar"1s2";"1s3 " \ar"1s4";"1s3 " \ar"m1ms";"m1ms2 " \ar"m1ms3";"m1ms2 " \ar"m1ms3";"m1ms4 " \ar"m1s";"m1s2 " \ar"m1s3";"m1s2 " \ar"m3ms";"m3ms2 " \ar"3s2";"3s2"+<-30pt,0pt > } \ ] ] where vertical arrows ( directed towards _ _ sub__modules ) indicate embedding of subquotients in feigin fuchs modules . the picture continues to the left and to the right ( and downward ) indefinitely . the weight-@xmath210 fields @xmath211 and @xmath212 are the triplet algebra generators . the algebra of fields is then extended by a field @xmath213 such that @xmath214\\ \eta(u ) } \ ] ] it is @xmath215 and @xmath201 that are in fact the symplectic fermions @xcite . weight-@xmath203 _ fields generate two standard first - order systems : our starting @xmath216 and @xmath217@xmath218 ( cf . @xcite ) . the fields @xmath219 allow constructing a logarithmic partner @xmath220 of the identity operator ; diagram then extends such that the top level ( split vertically for visual clarity ) becomes @xmath221^f\ar[1,-4]_e&&&&&\\ \delta(u)\ar[1,4]_f&&&&&&&&\xi(u)\ar[1,-4]^e&\\ & & & & 1 & & & & & } \ ] ] there are also two characteristic diagrams of _ weight-@xmath170 _ fields . we recall that if the fermions are bosonized through a free bosonic field , @xmath222 then the long - screening _ current _ ( the `` integrand '' in ) is @xmath223 ( which is a weight-@xmath170 field ) , and we have @xmath224^f&&e^{2\varphi(u)}\ar[1,-1]_e\\ & { d}\xi(u ) } \ ] ] similarly , there is an alternative bosonization through the scalar field introduced as @xmath225 . this gives the diagram @xmath226^f&&\eta(u)\xi(u)\ar[1,-1]_e\\ & \eta(u ) } \ ] ] ( once again , @xmath227 , which makes the two diagrams symmetric to each other ) . the @xmath197 logarithmic model corresponds to @xmath199 in . relations and in @xmath143}$ ] are then indeed those mimicking free fermions : @xmath228 based on the @xmath3 symmetry , we conjecture that for general @xmath151 , @xmath143}$ ] similarly allows expressing the relations among the @xmath4-model fields `` with an explicit quantum - group index . '' following @xcite , we call such fields _ parafermions _ ( the term is somewhat overloaded by different meanings ; its usage for fields transforming under a quantum group action goes back to @xcite ) . on the quantum - group side , clearly , is the general-@xmath151 counterpart of under the correspondence @xmath229 for the @xmath230-component `` parafermion '' fields @xmath231 and @xmath232 generalizing the symplectic fermions @xmath219 . the constituents of satisfy commutation relations generalizing the fermionic ones that occur for @xmath198 : for general @xmath151 , we have @xmath233}{0pt}{}{j}{\ell}$}}}{{\genfrac{[}{]}{0pt}{}{j}{\ell } } } { { \genfrac{[}{]}{0pt}{}{j}{\ell}}}{{\genfrac{[}{]}{0pt}{}{j}{\ell } } } } { \mathchoice { { \mbox{\footnotesize$\displaystyle \genfrac{[}{]}{0pt}{}{i}{\ell}$}}}{{\genfrac{[}{]}{0pt}{}{i}{\ell } } } { { \genfrac{[}{]}{0pt}{}{i}{\ell}}}{{\genfrac{[}{]}{0pt}{}{i}{\ell } } } } { [ \ell]!\ , } \left({\mathfrak{q}}- { \mathfrak{q}}^{-1}\right)^\ell z^{i-\ell}{\partial}^{j-\ell}.\end{gathered}\ ] ] moreover , the counterparts of and for general @xmath151 are the diagrams that are easily established using , essentially by applying the differential to , with the resulting modules extended by the `` cohomology corners '' @xmath234 and @xmath235 : @xmath236{\sum_{i=1}^{p - 1 } } { \raisebox{.9pt}{\mbox{\scriptsize$\displaystyle\frac{1}{{[i]}}$}}}z^{i}\,d({\partial}^{i } ) \ar^(.6)f[dr ] & & & & & & { \partial}^{p-1}\,{\underline{d{\partial}}}\ar_e[dl ] \\ & { \underline{d{\partial}}}&\kern-10pt\rightleftarrows\kern-10pt & { \partial}\,{\underline{d{\partial}}}&\kern-6pt\rightleftarrows \ldots \rightleftarrows\kern-6pt & { \partial}^{p-2}\,{\underline{d{\partial } } } } \\ \intertext{and } \xymatrix@=12pt@c=5pt { z^{p-1}\,{\underline{dz}}\ar^f[dr ] & & & & & & \displaystyle\smash[t]{\sum_{i=1}^{p - 1 } } { \raisebox{.9pt}{\mbox{\scriptsize$\displaystyle\frac{1}{{[i]}}$}}}d(z^{i})\,{\partial}^{i } \ar_(.75)e[dl ] \\ & z^{p-2}\,{\underline{dz}}&\kern-6pt\rightleftarrows \ldots \rightleftarrows\kern-6pt & z\,{\underline{dz}}&\kern-10pt\rightleftarrows\kern-10pt & { \underline{dz } } & } \end{gathered}\ ] ] ( as before , horizontal left right arrows represent the action of @xmath64 and @xmath71 up to nonzero factors and tilted arrows have no inverse maps ) . the cohomology corners are hopf - algebra counterparts of the screening currents in the two bosonizations , and the bottom elements , of the differentials @xmath237 , , @xmath238 and @xmath239 , , @xmath240 . m.r.gaberdiel and h.g.kausch , _ indecomposable fusion products _ , nucl . b477 ( 1996 ) 293318 [ hep - th@xmath2419604026 ] ; _ a rational logarithmic conformal field theory _ , b 386 ( 1996 ) 131137 [ hep - th@xmath2419606050 ] . b.l.feigin , a.m.gainutdinov , a.m.semikhatov , and i.yu.tipunin , _ modular group representations and fusion in logarithmic conformal field theories and in the quantum group center _ , commun . 265 ( 2006 ) 4793 [ hep - th@xmath2410504093 ] . b.l.feigin , a.m.gainutdinov , a.m.semikhatov , and i.yu.tipunin , _ logarithmic extensions of minimal models : characters and modular transformations _ , nucl . phys . b 757 ( 2006 ) 303343 [ hep - th@xmath2410606196 ] . d.adamovi and a.milas , _ on the triplet vertex algebra @xmath245 _ , adv . ( 2008 ) 26642699 [ arxiv:0707.1857v2 [ math.qa ] ] ; _ the @xmath246 triplet vertex operator superalgebras _ , commun . math . ( 2009 ) 225270 [ arxiv:0712.0379 [ math.qa ] ] . b.l.feigin , a.m.gainutdinov , a.m.semikhatov , and i.yu.tipunin , _ kazhdan lusztig - dual quantum group for logarithmic extensions of virasoro minimal models _ , j. math . 48 ( 2007 ) 032303 [ math.qa@xmath2410606506 ] . b.l.feigin , a.m.gainutdinov , a.m.semikhatov , and i.yu.tipunin , _ kazhdan lusztig correspondence for the representation category of the triplet @xmath244-algebra in logarithmic cft _ , theor . math . 148 ( 2006 ) 12101235 [ math.qa@xmath2410512621 ] . p.v.bushlanov , b.l.feigin , a.m.gainutdinov , and i.yu.tipunin , _ lusztig limit of quantum @xmath252 at root of unity and fusion of @xmath253 virasoro logarithmic minimal models _ , arxiv:0901.1602 [ hep - th ] . f.panaite , _ doubles of ( quasi ) hopf algebras and some examples of quantum groupoids and vertex groups related to them _ , in : _ contemporary mathematics 441 : hopf algebras and generalizations _ , eds : l.h.kauffman , d.e.radford , and f.j.o.souza . ams 2007 [ math.qa@xmath2410101039 ] . k.erdmann , e.l.green , n.snashall , and r.taillefer , _ representation theory of the drinfeld doubles of a family of hopf algebras _ , j. pure and applied algebra 204 ( 2006 ) 413454 [ math.rt@xmath2410410017 ] .

the glimpse survey has revealed a number of extended green objects " ( egos ) which display extended emission at 4.5 . these egos are potential candidates for high mass protostellar outflows . we have used high resolution ( ) h 1 - 0 s(1 ) line , , and-band images from the united kingdom infrared telescope to study 34 egos to investigate their nature . we found that 12 egos exhibit h outflows ( two with chains of h knotty structures ; five with extended h bipolar structures ; three with extended h lobes ; two with pairs of h knots ) . in the 12 egos with h outflows , three of them exhibit similar morphologies between the 4.5 and h emission . however , the remaining 9 egos show that the h features are more extended than the continuum features , and the h emission is seldom associated with continuum emission . furthermore , the morphologies of the near - infrared continuum and 4.5 emission are similar to each other for those egos with-band emission , implying that at least a part of the irac - band continuum emission of egos comes from scattered light from the embedded ysos .

1203.2416

massive stars play an important role in the evolution of galaxies . once they are born , their outflows , intense uv radiation , and stellar winds influence the nearby interstellar medium , and their feedback could induce the next generation of star formation @xcite . however , the formation mechanism of massive stars is still unclear . one of reasons for this is that massive protostars are very embedded , and hence can only be seen in infrared and radio . since the @xmath0 space telescope was launched , it has provided a good chance to study this early phase of star formation . @xmath0 provides better resolution than previous infrared space telescopes ( @xmath6 at irac bands ) , and is able to survey large regions . in the galactic legacy infrared mid - plane survey extraordinaire ( glimpse ) i survey at irac wavelengths ( @xmath7 ; @xmath8 , * ? ? ? * ) , @xcite identified more than 300 massive outflow candidates in the galactic plane . these candidates show excess emission at 4.5 @xmath1 , and have been dubbed `` extended green objects ( egos ) '' due to the common practice of assigning blue , green and red to 3.6 , 4.5 , and 8 @xmath1 bands in images . the most common explanation is that they are outflows from massive young stars , due to their association with irdcs , thought to be the birthplace of massive stars , and with class ii 6.7 ghz masers which are radiatively pumped by massive ysos @xcite . the 4.5 @xmath1 emission may come from h@xmath3 ( @xmath9 = 00 , s(9 , 10 , 11 ) ) and/or co ( @xmath9 = 10 ) band heads @xcite which can be excited by shocks from outflows @xcite . since their identification , there have been several attempts to investigate the nature of these egos . approximately two - third of egos are associated with class i 44 ghz methanol masers @xcite which are likely to be an outflow tracer . later , high resolution vla observations showed that class ii methanol masers are tightly concentrated around 24 @xmath1 sources , and class i masers tracing molecular outflows are widely distributed around the diffuse 4.5 @xmath1 emission @xcite . @xcite and @xcite used outflow tracers ( hco@xmath10 ) to survey egos , and their line profiles also indicate the presence of outflows . egos ( or green fuzzies , * ? ? ? * ) have been seen in known massive star forming regions ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , irdc cores @xcite , and jets associated with low - mass protostars ( e.g. , * ? ? ? * ; * ? ? ? molecular line interferometric observations can also help to identify outflows from massive ysos , and provide higher resolution and kinematic information for outflows ( see review by * ? ? ? ego g11.92@xmath110.61 and g19.01@xmath110.03 were observed using sma and carma @xcite . both of them have bipolar molecular outflows , coincident with the 4.5 @xmath1 lobes . the above mentioned studies give evidence that some egos are related to outflows ; however it is unclear if the 4.5 @xmath1 emission in egos comes from line emission or not . the best way to answer this question is to take spectra of egos in the irac 4.5 @xmath1 band . the infrared spectrograph ( irs ) on - board @xmath0 can observe only wavelengths longer than 5.3 @xmath1 , so irs itself can not directly determine the source of diffuse 4.5 @xmath1 emission . to determine the origin of what they denote as `` green fuzzies '' , @xcite took spectra of two egos ( ego g19.88@xmath110.53 and ego g49.27@xmath110.34 ) in @xmath12- ( 2.95 4.12 @xmath1 ) and @xmath13- ( 4.47 5.75 @xmath1 ) bands . one of them ( ego g19.88 - 0.53 ) does exhibit pure molecular hydrogen line emission at 4.695 @xmath1 ( @xmath14 @xmath15 ) , which dominates in the wavelength range 45 @xmath1 ( no co band heads are detected ) . this result supports the idea that the green fuzzy emission is excited by outflow shocks . however , the other source ( ego g49.27 - 0.34 ) only shows continuum emission in the above spectral ranges without line emission . they suggested that the green fuzzy emission is the result of extinction and a magnification in the color stretch in the 4.5 @xmath1 band . in this paper , we use h@xmath3 , @xmath5- , and @xmath4-band observations from the united kingdom infrared telescope ( ukirt ) to study the near - infrared properties of egos . near - infrared h@xmath3 1 - 0 s(1 ) observations have been used to study shock - excited outflows from ysos @xcite , and provide a good resolution ( @xmath161@xmath17 ) which allows us to study the morphologies of h@xmath3 outflows in detail . compared with continuum images of spitzer , our continuum - subtracted h@xmath3 images , which focus on shock - excited h@xmath3 line emission , are helpful for determining if the egos are associated with h@xmath3 outflows or not . this paper is organized as follows . we describe the observations and results in section 2 and 3 , respectively . we discuss the near - infrared properties of egos in section 4 , and summarize our conclusions in section 5 . narrow - band images of h@xmath3 @xmath9 = 1 - 0 s(1 ) emission ( 2.122 @xmath1 ) were obtained through the ukirt widefield infrared survey for h@xmath3 ( uwish2 , * ? ? ? * ) . uwish2 began in july 2009 and was completed in august 2011 ; it surveyed the inner galactic plane ( @xmath18 ) with the wide field camera ( wfcam ) at ukirt . the observations were taken with a narrow band filter ( @xmath19 ) at 2.122 @xmath1 with 720 seconds integration time per pixel . the median seeing in h@xmath3 images is @xmath160.7@xmath17 ; only a few images were taken with seeing @xmath20 . the 5@xmath21 detection limit of point sources is @xmath1618 mag in @xmath4 , and the surface brightness limit is @xmath22 w m@xmath23 arcsec@xmath23 at the typical seeing @xcite . the @xmath4- and @xmath5-band images were obtained through the wsa archive for the ukirt infrared deep sky survey galactic plane survey ( ukidss gps , * ? ? ? * ) . gps is one of the five surveys within ukidss , and will survey more than 1800 square degrees of the northern equatorial galactic plane in @xmath24 , @xmath5 and @xmath4-bands with wfcam . the 5 @xmath21 detection limit of point sources are @xmath5=19 and @xmath4=18.5 mag , and a seeing @xmath25 1@xmath17 in @xmath4 band is requested by the survey . @xmath4-band ( 2.2 @xmath1 ) images were used to construct our continuum - subtracted h@xmath3 images of egos , and we compared their morphologies with the h@xmath3 and continuum ( @xmath5-band and 4.5 @xmath1 ) emission . the uwish2 data for 12 egos were obtained in the earliest phase of the survey as high priority targets . these targets were selected as follows . we investigated gps @xmath4-band images of egos in table 1 of @xcite , before the h@xmath3 observations . if there was @xmath4-band continuum emission within a 20@xmath17 radius of the ego s positions , we included them as high priority targets for uwish2 observations in 2009 . the criterion is not only restricted to the ego s positions , but also extended to a 20@xmath17 radius , as the morphologies of h@xmath3 outflows may be more extended than those of @xmath4-band continuum . besides these preselected 12 egos , another 22 egos were covered by the uwish2 survey . in this paper , we studied a total of 34 egos which were observed by uwish2 in 2009 . table [ tab : h2 ] shows the list of these targets . the gps covers all the targets in @xmath4-band and @xmath5-band except for ego g35.20 - 0.74 ( @xmath4-band only ) . data reduction was done by the cambridge astronomical survey unit ( casu ) . at casu , the wfcam pipeline is responsible for flat fielding and sky correction of the image data , and stacking frames as described by @xcite . reduced @xmath4-band and h@xmath3 images were accessed via the wide field astronomy unit . the continuum - subtracted h@xmath3 images were obtained as follows . using idl , we shifted and rotated each @xmath4-band image to match the coordinates of the corresponding h@xmath3 image . then we compared the full width at half maximum ( fwhm ) of the @xmath4-band and h@xmath3 images for each ego , and smoothed the smaller fwhm to the larger one . before subtraction , we scaled the @xmath4-band fluxes to match their stellar fluxes to those in the h@xmath3 image . then , a continuum - subtracted h@xmath3 image of each ego was generated by subtracting the @xmath4-band from the h@xmath3 image . in addition to the h@xmath3 and @xmath4-band images described above , @xmath0 irac and mips 24 @xmath1 images from the glimpse @xcite and mipsgal @xcite legacy surveys , respectively , were obtained from the archive data for comparison . the data had been reduced with post basic calibrated data ( post - bcd ) piplines developed by the infrared processing and analysis center ( ipac ) . the irac images are used to compare the spatial distribution of the egos with continuum - subtracted h@xmath3 and @xmath4-band images . the mips 24 @xmath1 images will be used to identify ysos from thermal dust emission . of our 34 object sample , 23 sources show clear detections of positive ( i.e. , h@xmath3 emission ) or / and negative ( presumably due to continuum , as described below in detail ) valued features , in the continuum - subtracted h@xmath3 images . figures [ fig : g10.34 - 0.14 ] - [ fig : g58.09 - 0.34 ] show their images in the irac bands ( 3.6 , 4.5 , and 8.0 @xmath1 ) , mips 24 @xmath1 , continuum - subtracted h@xmath3 , and @xmath4-band . table [ tab : h2 ] summarizes the detections of h@xmath3 , @xmath4 , and @xmath5-band emission . based on the continuum - subtraced h@xmath3 images , 34 egos are categorized as follows : those associated with h@xmath3 outflows ( y , 12 objects ) ; those associated with possible h@xmath3 outflows , ( y ? , 4 objects ) ; and finally , those with no h@xmath3 detection ( n , 18 objects ) . in addition , we include distances of egos @xcite in table [ tab : h2 ] . molecular hydrogen emission - line objects ( mhos , * ? ? ? * ) numbers are assigned for newly identified h@xmath3 outflows in table [ tab : mho ] . in sections 3.1 and 3.2 , we describe the details of the results for the h@xmath3 and continuum emission ( @xmath4 and @xmath5-band ) , respectively . the details for the 23 selected individual objects are presented in the appendix . in the continuum - subtracted h@xmath3 images , 16 objects show positive valued features due to h@xmath3 emission ( seen in black in figures [ fig : g10.34 - 0.14 ] - [ fig : g58.09 - 0.34 ] ) . seven egos are associated with an extended bipolar structure or a set of aligned knots in these images ( ego g19.88 - 0.53 , g35.04 - 0.47 , g35.13 - 0.74 , g35.15 + 0.80 , g35.20 - 0.74 , g35.79 - 0.17 , and g35.83 - 0.20 ) . two egos exhibit two h@xmath3 knots ( ego 19.01 - 0.03 and g54.45 + 1.01 ) . three egos only show one lobe in h@xmath3 emission ( ego g11.92 - 0.61 , g16.61 - 0.24 , and g35.68 - 0.18 ) . four egos show possible h@xmath3 outflows ( ego g12.02 - 0.21 , g12.42 + 0.50 , g12.91 - 0.03 , and g16.59 - 0.05 ) , exhibiting either extended h@xmath3 emission or an isolated h@xmath3 knot . the extended emission in the continuum - subtracted h@xmath3 image of ego g12.42 + 0.50 could be due to residuals from the subtraction of bright continuum emission of the ysos . in most of the above objects , the h@xmath3 emission appears to be associated with a monopolar or bipolar h@xmath3 outflow driven by a single protostar . the protostellar outflows have a bipolar geometry in most cases , and the absence of the counterpart in some of the outflows is attributed to obscuration of the counterflow by extinction ( see * ? ? ? * ; * ? ? ? * for review ) however , ego g19.88 - 0.53 and g35.13 - 0.74 seem to be associated with more than one h@xmath3 outflow . generally , the h@xmath3 outflows of the egos usually show small opening angles ( highly collimated outflows ) . in contrast , the h@xmath3 outflow of ego g35.20 - 0.74 shows an hourglass shape with a large opening angle , @xmath26 . in most of the above objects , the h@xmath3 emission is more extended than the 4.5 @xmath1 emission , and faint or absent at the regions where the 4.5 @xmath1 emission is bright . the 4.5 @xmath1 emission shows a counterpart to the h@xmath3 features in the following three egos . in g19.88 - 0.53 , diffuse 4.5 @xmath1 emission is observed at the base of the western lobe of the h@xmath3 outflow . @xcite conducted ground - based spectroscopy of this emission component at 2.9 - 4.1 @xmath1 and 4.5 - 5.8 @xmath1 , and their spectra indicate that the emission in the irac 4.5 @xmath1 should be dominated by h@xmath3 emission . in g35.68 - 0.18 , the h@xmath3 feature `` b '' indicated in figure [ fig : g35.68 - 0.18 ] shows a faint counterpart in the 4.5 @xmath1 emission . in g35.83 - 0.20 , three bright h@xmath3 knots in figure [ fig : g35.83 - 0.20 ] are also seen in the 4.5 @xmath1 image . it is notable , however , the glimpse image also contains a dozen brighter features associated with the 4.5 @xmath1 emission . figure [ fig : hist ] shows histograms of the distances for egos with h@xmath3 ( dashed line histogram ) and without ( dotted line histogram ) detections . we only include those egos with distances in table [ tab : h2 ] , but exclude possible h@xmath3 outflows in the histogram . in general , the distances of the egos without h@xmath3 outflow detections seem to be larger than those with h@xmath3 outflow detections . beyond four kpc , the numbers of h@xmath3 outflow detections drop , but the numbers of non - detections increase . therefore , our h@xmath3 outflow detection has a bias toward distance , and we probably miss h@xmath3 outflow detections for some distant egos . in the continuum - subtracted h@xmath3 images ( figures [ fig : g10.34 - 0.14 ] - [ fig : g58.09 - 0.34 ] ) , the positions of extended @xmath4-band emission ( see section 3.2 and appendix for details ) usually show negative valued features ( seen in white in figures [ fig : g10.34 - 0.14 ] - [ fig : g58.09 - 0.34 ] ) that are likely to represent continuum emission with a large infrared excess and high extinction , and may be light scattered from the ysos . this type of emission can also be seen around other h@xmath3 outflow bases @xcite . there are 12 egos associated with negative valued features around the positions of the egos in the continuum - subtracted h@xmath3 images . six of them are associated with h@xmath3 outflows ( ego g11.92 - 0.61 , g16.61 - 0.24 , g19.88 - 0.53 , g35.13 - 0.74 , g35.20 - 0.74 , and g54.45 + 1.01 ) , and the other six are not ( ego g10.34 - 0.14 , g12.20 - 0.03 , g19.61 - 0.12 , g20.24 + 0.07 , g28.83 - 0.25 , and g58.09 - 0.34 ) . figures [ fig : g10.34 - 0.14 ] - [ fig : g58.09 - 0.34 ] show that at least 9 of the 23 egos show similar morphologies in the @xmath4 and 4.5 @xmath1 emission ( ego g12.42 + 0.50 , g19.01 - 0.03 , g28.83 - 0.25 , g35.13 - 0.74 , g35.15 + 0.80 , g35.20 - 0.74 , g35.68 - 0.18 , g35.83 - 0.20 , and g54.45 + 1.01 ) , although the signal to noise ratios of the @xmath4-band emission are lower than those of the 4.5 @xmath1 emission . in the remaining 13 egos , 10 egos exhibit extended @xmath4-band emission at the position of the egos ( ego g10.34 - 0.14 , g11.92 - 0.61 , g12.20 - 0.03 , g12.91 - 0.03 , g16.61 - 0.24 , g19.61 - 0.12 , g19.88 - 0.53 , g20.24 + 0.07 , g35.79 - 0.17 , and g58.09 - 0.34 ) , and three egos show a @xmath4-band point source around the 4.5 @xmath1 peak of each ego ( ego g12.02 - 0.21 , g16.59 - 0.05 , and g35.04 - 0.47 ) . one or more 24 @xmath1 sources are usually located near the 4.5 @xmath1 peak of the majority of the egos . in the 23 ego sample , 7 are detected in @xmath5-band . figure [ fig : hk ] shows these objects in @xmath5- , @xmath4-band , 3.6 , and 4.5 @xmath1 . in these figures , the flux distribution gradually changes with wavelength . in ego g11.92 - 0.61 , g28.83 - 0.25 and g35.68 - 0.18 , the extended emission is marginal in the @xmath5-band , and more apparent at longer wavelengths . in ego g12.42 + 0.50 the @xmath5-band emission is observed only at the peak of the ego at 3.6 and 4.5 @xmath1 . the remaining extended components , in particular that to the southwest , are apparent only at 3.6 and 4.5 @xmath1 . in ego 19.88 - 0.53 , the bright point source is seen at the center only at 3.6 and 4.5 @xmath1 . the simplest explanation of the above characteristics is the effect of extinction at different wavelengths . high - mass protostellar outflows show a variety of morphologies , and their driving mechanism is less clear than for low - mass protostellar outflows . well known energetic outflows like the orion bn / kl and dr 21 outflows show a wide opening angle and complicated shock structures ( see e.g. , * ? ? ? * ; * ? ? ? in contrast , some high - mass protostar drive a relatively collimated outflow analogous to that associated with low - mass protostars @xcite . these tend to be associated with a protostar with spectral types of late o or b , not with the earlier spectral types ( see * ? ? ? * ; * ? ? ? * for review ) . @xcite proposed an evolution scenario of outflows from massive stars . in their scenario , an outflow from a massive star is collimated in their early evolutionary phases . as it grows by accretion , it reaches the main sequence , associated with an ultra compact region . the opening angle of the outflow could then be increased by the ionized wind from the massive star . as mentioned in section 3.1 , the h@xmath3 emission in most of the egos is associated with a collimated outflow . this suggests that they are relatively young , or their masses are not very high , according to the scenario proposed by @xcite . in contrast , ego g35.20 - 0.74 exhibits a large opening angle h@xmath3 outflow . the hourglass shape outflow is unlikely to be a composite of two collimated outflows at different directions , because the shapes , lengths , and intensities of mho 2431c and d are similar ( figure [ fig : g35.20 - 0.74 ] ) . in addition , co interferometric observations also support that it is a single outflow with large opening angle @xcite . the h@xmath3 outflow is likely to be driven by a yso which is associated with an ultra compact region @xcite at the outflow base , and it is the only h@xmath3 outflow with a known region in our sample . our results support the evolution scenario of @xcite . in our sample , two egos are likely to be associated with h@xmath3 outflows from multiple protostars ( ego g19.88 - 0.53 and g35.13 - 0.74 ) . three radio sources are located near the outflow base of ego g19.88 - 0.53 @xcite , and ego g35.13 - 0.74 is associated with a young star cluster mercer 14 @xcite . in this section , we discuss the emission mechanisms of those egos with near - infrared counterparts , based on the 23 egos mentioned in 3 . originally , @xcite suggested that the extended 4.5 @xmath1 emission of egos traces shocked molecular gas in outflows . however , some studies suggest that the 4.5 @xmath1 emission come from scattered light . for example , the appearance of ngc 6334 v is similar to that of egos in the @xmath0 three - color image . in the 2 @xmath1 polarimetric image , ngc 6334 v shows two highly polarized reflection nebulae ( scattered light ) coinciding with the 4.5 @xmath1 emission and the bipolar outflow @xcite . another source , ego g35.20 - 0.74 known as g35.2n , also shows a highly polarized extended reflection nebulae in @xmath4-band @xcite . these two cases suggest that @xmath4-band emission represents scattered emission from embedded ysos , which is consistent with results from @xcite and @xcite . iras 17527 - 2439 has an s - shaped h@xmath3 outflow which may be caused by the precession of a jet @xcite , and its appearance in the @xmath0 image is similar to that of egos . interestingly , the morphologies of the continuum ( @xmath4 and @xmath0 bands ) and h@xmath3 emission of iras 17527 - 2439 are different . the continuum emission in @xmath4 and @xmath0 bands in the direction of the outflow is rotated counterclockwise with respect to the h@xmath3 outflow , suggesting that the continuum emission emerges from the outflow cavity of the yso @xcite . recently , @xcite took spectra of egos at 5 10 @xmath1 wavelength , and found that the 4.5 @xmath1 emission is not due to h@xmath3 lines which would be too faint to contribute the irac 4.5 @xmath1 emission . in our sample , only three of the 12 egos with h@xmath3 outflows show a distribution of the h@xmath3 emission similar to that of the 4.5 @xmath1 emission ( ego g35.68 - 0.18 , g35.83 - 0.20 , and west lobe of ego g19.88 - 0.53 ) . the 4.5 @xmath1 emission of these three egos likely comes from h@xmath3 emission . the @xmath12- and @xmath13-band spectrum of ego g19.88 - 0.53 show h@xmath3 emission @xcite , supporting this idea . in contrast , for the remaining 9 egos with h@xmath3 outflows in our sample , we found that the morphologies of the 4.5 @xmath1 and h@xmath3 emission are different . the peaks of the 4.5 @xmath1 emission are rarely associated with h@xmath3 emission , and the 4.5 @xmath1 emission generally appears close to the bases of the outflows ( embedded ysos ) . the h@xmath3 emission is more extended than the 4.5 @xmath1 emission which is usually weak or not detected at the peak of the h@xmath3 emission . this suggests that the origins of the h@xmath3 and 4.5 @xmath1 emission are different . for those egos with @xmath4-band emission , the morphologies of the extended @xmath4-band emission and the 4.5 @xmath1 emission resemble each other , suggesting the same origin . furthermore , seven of the sources with extended @xmath4-band emission also show extended emission in the @xmath5-band , with flux distributions similar to the @xmath4-band emission . this @xmath5-band emission is not likely to be due to shocked emission , since ( 1 ) the flux distribution is significantly different between h@xmath3 and @xmath4-band emission ; and ( 2 ) shocked emission in the @xmath5-band should be so faint that it is not likely to be observed ( e.g. , * ? ? ? * ; * ? ? ? these observed characteristics can be consistently explained if the extended emission at @xmath5 , @xmath4 , 3.6 , and 4.5 @xmath1 is due to scattered continuum @xcite . indeed , all of these sources are associated with 24 @xmath1 source(s ) , i.e. , a protostar or multiple protostars , which are the sources of the scattered light . in figure [ fig : hk ] , the gradual change in flux distribution with wavelength is consistent with extinction at different wavelengths . in ego g35.20 - 0.74 , the emission in @xmath4-band and 4.5 @xmath1 shows a bipolar distribution about the position of the protostar seen at 24 @xmath1 . this implies that the scattering occurs in an outflow cavity @xcite , as in low - mass protostars ( e.g. , * ? ? ? * ; * ? ? ? * ) . in our 23 ego sample , only three are consistent with the 4.5 @xmath1 emission coming from the h@xmath3 emission , and the rest are likely associated with scattered light from the embedded ysos or mixture of both . throughout , we suggest that the 4.5 @xmath1 emission could result from scattered light or / and h@xmath3 emission . however , the 4.5 @xmath1 emission itself can not tell which emission mechanism dominates , and we need other wavelength observations ( e.g. , narrow h@xmath3- and @xmath4-band ) to distinguish the contributions of these two emission mechanisms . we analyzed the data from h@xmath3 narrow band , @xmath4- , and @xmath5-band observations from ukirt for 34 egos ( 12 egos are preselected and 22 are covered by the uwish2 survey ) . our narrow band h@xmath3 observations provide high resolution images ( @xmath27 ) of egos , dedicated to shock - excited h@xmath3 emission , @xmath9 = 1 - 0 s(1 ) . the h@xmath3 images allow us to determine if these egos are associated with h@xmath3 outflows . we found that 12 egos show h@xmath3 outflows and four egos are h@xmath3 outflow candidates . within this group of 12 egos with h@xmath3 outflows two objects show a chain of h@xmath3 knotty structures , 5 display extended h@xmath3 bipolar structures , three have extended h@xmath3 lobes , and two have a pair of h@xmath3 knots . most of them appear to be associated with a collimated outflow driven by a single protostar . in contrast , ego g35.20 - 0.74 shows an hourglass shape bipolar outflow , and two egos ( ego g19.88 - 0.53 and g35.13 - 0.74 ) exhibit multiple outflows . of the 12 egos with h@xmath3 outflows , three egos show similar distributions between the 4.5 @xmath1 and h@xmath3 emission , implying similar origins . for the remaining of the 9 egos with h@xmath3 outflows , however , we found that the peaks of the 4.5 @xmath1 emission are rarely associated with the h@xmath3 emission . the h@xmath3 features are more extended than the continuum features . in our sample , we found similar morphologies between the @xmath4-band and 4.5 @xmath1 emission . seven egos also exhibit extended emission in @xmath5-band . the morphologies of the extended emission gradually changes with wavelength at @xmath5-band , @xmath4-band , 3.6 , and 4.5 @xmath1 . this can be explained if the emission is associated with scattered continuum , e.g. , in the outflow cavity . the different morphologies at different wavelengths may be attributed to extinction . we thank drs . christopher davis and bringfried stecklum for suggestions and comments . this research made use of the simbad data base operated at cds , strasbourg , france , and nasa s astrophysics data system abstract service . mt is supported from national science council of taiwan ( grant no . nsc-100 - 2112-m-001 - 007-my3 ) . * ego g10.34@xmath110.14 * ( figure [ fig : g10.34 - 0.14 ] ) . the 24 @xmath1 image is marginally extended in the ego region , presumably due to the presence of multiple sources . no h@xmath3 feature is detected in the continuum - subtracted h@xmath3 image . two negative valued features are seen in the continuum - subtracted h@xmath3 image . one is located at the 4.5 @xmath1 peak flux position of the ego , the other @xmath28 northeast of the ego . in the @xmath4-band image , both extended and point sources are seen at the position of the ego , and coincide with the 24 @xmath1 source . * ego g11.92@xmath110.61 * ( figure [ fig : g11.92 - 0.61 ] ) . a molecular outflow along the ne - sw direction was discovered by @xcite . the ego exhibits two peaks in the ne - sw direction with a separation of @xmath29 in the 4.5 @xmath1 contour map . a bright 24 @xmath1 source is associated with the ne 4.5 @xmath1 peak of the ego . h@xmath3 emission is identified near the sw peak of the ego , and extends to the sw direction . this flow component is presumably associated with the blue lobe of the molecular outflow discovered by @xcite . the distribution of h@xmath3 emission differs from that of the @xmath4-band and 4.5 @xmath1 continuum , while the distributions of the @xmath4-band and 4.5 @xmath1 continuum match more closely . in the @xmath4-band image , there are two extended sources at the two peaks of the 4.5 @xmath1 emission . * ego g12.02@xmath110.21 * ( figure [ fig : g12.02 - 0.21 ] ) . ego g12.02 - 0.21 is associated with a 24 @xmath1 source located at the 4.5 @xmath1 peak . there is a faint h@xmath3 knot and another possible h@xmath3 knot located 3@xmath17 and 6@xmath17 north of the 24 @xmath1 source , respectively . the @xmath4-band image shows a continuum source at the 4.5 @xmath1 peak of the ego . * ego g12.20@xmath110.03 * ( figure [ fig : g12.20 - 0.03 ] ) . there are two barely resolved 24 @xmath1 sources with a 5@xmath17 separation associated with 4.5 @xmath1 peaks in the nw - se direction . the se 4.5 @xmath1 peak coincides with an ultra - compact region , [ wbh2005 ] g012.199 - 0.034 @xcite . in the continuum - subtracted h@xmath3 image , no h@xmath3 feature is detected . we identified two continuum sources in @xmath4-band image at the two peaks of the 4.5 @xmath1 emission , seen as negative valued features in the continuum - subtracted h@xmath3 image . * ego g12.42@xmath300.50 * ( figure [ fig : g12.42 + 0.50 ] ) . ego g12.42 + 0.50 is associated with a very bright 24 @xmath1 source which saturates the detector . the 24 @xmath1 source coincides with an region , [ uhp2009 ] vla g012.4180 + 00.5038 @xcite , and is located at the peak of 4.5 @xmath1 emission . there is extended emission in the continuum - subtracted h@xmath3 image . this may be due to residuals from the continuum subtraction , rather than true h@xmath3 emission . in the @xmath4-band image , there is an extended @xmath4-band source at the position of the 24 @xmath1 source . the location of the brightest @xmath4-band flux coincide in the 4.5 @xmath1 and @xmath4-band images . * ego g12.91@xmath110.03 * ( figure [ fig : g12.91 - 0.03 ] ) . ego g12.91 - 0.03 is associated with a 24 @xmath1 source at the peak of the 4.5 @xmath1 emission . there is a possible knot in the h@xmath3 emission 1@xmath17 to the north of the peak in the 4.5 @xmath1 emission . the @xmath4-band image shows an extended continuum source very close to the 4.5 @xmath1 peak and the 24 @xmath1 source . * ego g16.59@xmath110.05 * ( figure [ fig : g16.59 - 0.05 ] ) . ego g16.59 - 0.05 is associated with iras 18182 - 1433 , a star - forming region with radio free - free emission @xcite . @xcite suggested that this emission is either due to an optically thin region or a thermal jet ( or stellar wind ) . ego g16.59 - 0.05 is associated with a 24 @xmath1 source coinciding with the peak of the 4.5 @xmath1 emission . the h@xmath3 knot in the continuum - subtracted h@xmath3 image , the @xmath4-band source , and the 4.5 @xmath1 source are close to each other , but not exactly coincident . * ego g16.61@xmath110.24 * ( figure [ fig : g16.61 - 0.24 ] ) . ego g16.61 - 0.24 is associated with a 24 @xmath1 source located near the peak of the 4.5 @xmath1 emission . in the continuum - subtracted h@xmath3 image , there is a faint h@xmath3 lobe to the east ; the tip of the h@xmath3 outflow may be associated with faint 4.5 @xmath1 emission but the latter detection is not convincing . the spatial distributions of the h@xmath3 outflow and bright 4.5 @xmath1 emission in the east side are different , with position angles of 80@xmath31 and 70@xmath31 , respectively . the continuum - subtracted h@xmath3 image shows negative valued features at the peak of the 4.5 @xmath1 emission . there is a @xmath4-band extended source at the position of the 24 @xmath1 source . * ego g19.01@xmath110.03 * ( figure [ fig : g19.01 - 0.03 ] ) . ego g19.01 - 0.03 is associated with a 24 @xmath1 source at the 4.5 @xmath1 peak . the ego is elongated in the 4.5 @xmath1 , and there is a group of 44 ghz methanol masers distributed along the ego @xcite . a molecular outflow discovered by @xcite has a morphology similar to that of the 4.5 @xmath1 emission . we detected two h@xmath3 knots near the two lobes of the molecular outflow , and the northern h@xmath3 knot coincides with the position of the 44 ghz methanol masers . the symmetric distribution of the two h@xmath3 knots with respect to the 24 @xmath1 source suggests a bipolar outflow . the @xmath4-band extended emission is distributed to the both sides of the outflow . the morphology of the 4.5 @xmath1 emission is similar to that of the @xmath4-band continuum , but is different from that of the h@xmath3 emission . * ego g19.61@xmath110.12 * ( figure [ fig : g19.61 - 0.12 ] ) . in the 4.5 @xmath1 contour map , there are two peaks , each associated with a 24 @xmath1 source . the southern peak is associated with the ego identified by @xcite . in the continuum - subtracted h@xmath3 image , ego g19.61 - 0.12 is not associated with any h@xmath3 feature . there are two @xmath4-band sources exhibiting negative valued features in the continuum - subtracted h@xmath3 image . one is associated with the ego ( southern 4.5 @xmath1 peak ) , and the other is located to the s - e direction with a @xmath32 separation . * ego g19.88@xmath110.53 * ( figure [ fig : g19.88 - 0.53 ] ) . ego g19.88 - 0.53 is related to the iras 18264 - 1152 star - forming region . the crosses in the glimpse image show positions where @xcite took spectra in the 3 - 6 @xmath1 range . these authors show that the 4.5 @xmath1 emission at the black cross ( i.e. , the position of a massive yso ) is due to continuum , while it is due to h@xmath3 emission at the white crosses . a 24 @xmath1 source is at the peak of the 4.5 @xmath1 emission . in our image , ego g19.88 - 0.53 shows a clear bipolar h@xmath3 outflow with an angular scale of @xmath33 in the e - w direction . h@xmath3 knots are also located further out to the northeast and north of the ego , probably due to multiple outflow sources . these are marked as mho 2203 , 2204 , 2205 , and 2245 in the figure . all except mho 2245 have been identified in previous narrow - band h@xmath3 observations by @xcite . the position of the h@xmath3 emission close to the driving source matches the diffuse components at 4.5 @xmath1 . the 4.5 @xmath1 and @xmath4-band emission are not as extended as the h@xmath3 emission . the @xmath4-band image shows a relatively bright extended component at the base of the eastern lobe . there are also some faint extended component in the @xmath4-band image in the west side of the protostar(s ) . * ego g20.24@xmath300.07 * ( figure [ fig : g20.24 + 0.07 ] ) . there are two sources seen in the glimpse image ; the central source is associated with the ego @xcite , while the other source is located at the north east . both sources show strong 24 @xmath1 emission . no h@xmath3 emission is seen in our continuum - subtracted h@xmath3 image , however negative valued features due to continuum emission are associated with both 4.5 @xmath1 sources . in the @xmath4-band image , there is an extended source around the 24 @xmath1 source associated with the ego . * ego g28.83@xmath110.25 * ( figure [ fig : g28.83 - 0.25 ] ) . ego g28.83 - 0.25 is associated with a 24 @xmath1 source at the peak of the 4.5 @xmath1 emission . no h@xmath3 emission is seen in the continuum - subtracted h@xmath3 image , however negative valued features due to continuum emission are seen , with a morphology similar to the 4.5 @xmath1 emission . diffuse @xmath4-band emission is observed at the peak of the ego and a faint feature @xmath28 east . * ego g35.04@xmath110.47 * ( figure [ fig : g35.04 - 0.47 ] ) . ego g35.04 - 0.47 has a 24 @xmath1 source at the peak of the 4.5 @xmath1 emission . its glimpse image shows that it is extended along an n - s axis . however , our continuum - subtracted h@xmath3 image shows an h@xmath3 outflow along ne - sw axis which is not detected in the 4.5 @xmath1 image . the sw lobe of the h@xmath3 outflow is much brighter than the ne one which can be explain if the ne lobe is obscured by the irdc . the morphology of the h@xmath3 emission is more extended than that of the 4.5 @xmath1 emission . a faint @xmath4-band source is observed at the 4.5 @xmath1 peak ( and the 24 @xmath1 source ) . * ego g35.13@xmath110.74 * ( figure [ fig : g35.13 - 0.74 ] ) . ego g35.13 - 0.74 is likely associated with the embedded cluster [ mcm2005b ] 14 ( mercer 14 , * ? ? ? there is a 24 @xmath1 source close to the ego , but it is not located at the peak of the 4.5 @xmath1 emission . the continuum - subtracted h@xmath3 image shows a well collimated h@xmath3 outflow around the ego in the nw - se direction , and also a number of h@xmath3 knots surrounding it . the 4.5 @xmath1 image shows a few possible faint counterparts for these h@xmath3 knots . in contrast , the 4.5 @xmath1 and @xmath4-band images only show extended emission , with similar morphologies . for further information regarding h@xmath3 knots ( mho 2423 - 2428 ) and their possible driving sources , see @xcite in detail . * ego g35.15@xmath300.80 * ( figure [ fig : g35.15 + 0.80 ] ) . ego g35.15 + 0.80 is associated with a 24 @xmath1 source at the 4.5 @xmath1 peak . the ego is at the center of the image and is extended in the ne - sw direction . in the h@xmath3 image , ego g35.15 + 0.80 shows a set of aligned h@xmath3 knots in the nw - se direction , different from the orientation of the 4.5 @xmath1 emission . the 4.5 @xmath1 image shows a few possible faint counterparts to these h@xmath3 knots . the @xmath4-band extended emission is located @xmath1610@xmath17 se of the 24 @xmath1 source . the morphology of 4.5 @xmath1 emission resembles that of @xmath4-band continuum emission , except there is no @xmath4-band point source at the position of the 24 @xmath1 source ( figure [ fig : g35.15 + 0.80b ] ) . * ego g35.20@xmath110.74 * ( figure [ fig : g35.20 - 0.74 ] ) . ego g35.20 - 0.74 is also known as g35.2n . @xcite reported that there are two compact regions , located near the saturated 24 @xmath1 source . in the figure , the lowest contour around the ego roughly represents its shape . our continuum - subtracted h@xmath3 image shows an hourglass outflow with a wide opening angle ( @xmath26 ) in the ne - sw direction . in contrast to the extended h@xmath3 emission , the 4.5 @xmath1 and @xmath4-band continuum emission are brighter at the base of the outflow and are similar in morphology . the tips of the h@xmath3 outflow are associated with 4.5 @xmath1 emission , but the 4.5 @xmath1 emission at the tips is much fainter than that near the base . the peak of the @xmath4-band extended emission coincides with the 24 @xmath1 source . * ego g35.68@xmath110.18 * ( figure [ fig : g35.68 - 0.18 ] ) . a 24 @xmath1 source is located near the 4.5 @xmath1 peak of ego g35.68 - 0.18 . the continuum - subtracted h@xmath3 image shows two emission components : one close to the 4.5 @xmath1 peak ( `` a '' in the figure ) , the other extending toward the east ( `` b '' ) . the position of `` a '' is offset from the 4.5 @xmath1 peak by @xmath162@xmath17 . the position of `` b '' approximately matches a diffuse component in the 4.5 @xmath1 image . the @xmath4-band image shows emission similar to the h@xmath3 , but the peak position of the component close to `` a '' matches the 4.5 @xmath1 peak . this component is much brighter than `` b '' at 4.5 @xmath1 and @xmath4-band , while `` a '' and `` b '' are of similar brightness in h@xmath3 emission . * ego g35.79@xmath110.17 * ( figure [ fig : g35.79 - 0.17 ] ) . in the 4.5 @xmath1 contour map of ego g35.79 - 0.17 , there are two peaks with a separation of 3@xmath17 . the ne peak is associated with a 24 @xmath1 source . the continuum - subtracted h@xmath3 image exhibits an h@xmath3 outflow in the ne - sw direction with the ne lobe brighter than the sw one . in contrast to the more extended h@xmath3 emission , the glimpse image only shows bright emission close to the 24 @xmath1 source . the @xmath4-band continuum image shows an extended source at the position of the sw 4.5 @xmath1 peak which does not coincide with the position of the 24 @xmath1 source . * ego g35.83@xmath110.20 * ( figure [ fig : g35.83 - 0.20 ] ) . there are several faint 24 @xmath1 sources around ego g35.83 - 0.20 , but it is not clear which one is physically related to the ego . the continuum - subtracted h@xmath3 image shows a chain of a few h@xmath3 knots ( mho 2434a - c ) in the ne - sw direction . one of the h@xmath3 knot ( mho 2434b ) coincides with the ego . these h@xmath3 knots are also seen in the @xmath0 image , however , the @xmath0 image contains a significantly larger number of emission features not seen in the continuum - subtracted h@xmath3 image . this makes the presence of the outflow ambiguous in the @xmath0 image . in the @xmath4-band image , there is diffuse emission associated with the ego . * ego g54.45@xmath301.01 and g54.45 + 1.02 * ( figure [ fig : g54.45 + 1.01 ] ) . a 24 @xmath1 source is associated with g54.45 + 1.01 . there are two h@xmath3 knots near the ego to the north west and south east located @xmath34 and @xmath28 away from the 24 @xmath1 source , respectively . the alignment of these knots , and the lack of other nearby 24 @xmath1 sources , implies that they are associated with ego g54.45 + 1.01 . a @xmath4-band continuum extended source is related to the ego . the orientation of the @xmath4-band extended source resembles that of the 4.5 @xmath1 emission . for g54.45 + 1.02 , an extended @xmath4-band source is associated with the 4.5 @xmath1 emission ; however no 24 @xmath1 source is seen . * ego g58.09@xmath110.34 * ( figure [ fig : g58.09 - 0.34 ] ) . ego g58.09 - 0.34 is associated with a 24 @xmath1 source which coincides with the peak of the 4.5 @xmath1 emission . the continuum - subtracted h@xmath3 image does not show any h@xmath3 features , but does show continuum emission with negative valued feature at the 4.5 @xmath1 peak . in the @xmath4-band image , there are two extended sources with a separation of 3@xmath17 to the north and south . the northern source is located on the 4.5 @xmath1 peak . in the continuum - subtracted h@xmath3 image , the northern source shows negative valued feature , but the southern one does not . arce , h. g. , shepherd , d. , gueth , f. , lee , c .- f . , bachiller , r. , rosen , a. , & beuther , h. 2007 , protostars and planets v , 245 bally , j. , reipurth , b. , & davis , c. j. 2007 , protostars and planets v , 215 benjamin , r. a. , churchwell , e. , babler , b. l. , et al . 2003 , , 115 , 953 beuther , h. , & shepherd , d. 2005 , cores to clusters : star formation with next generation telescopes , 105 birks , j. r. , fuller , g. a. , & gibb , a. g. 2006 , , 458 , 181 carey , s. j. , et al . 2009 , , 121 , 76 chambers , e. t. , jackson , j. m. , rathborne , j. m. , & simon , r. 2009 , , 181 , 360 chen , x. , shen , z .- q . , li , j .- j . , xu , y. , & he , j .- h . 2010 , , 710 , 150 chen , x. , ellingsen , s. p. , & shen , z .- q . 2009 , , 396 , 1603 churchwell , e. , et al . 2009 , , 121 , 213 churchwell , e. , povich , m. s. , allen , d. , et al . 2006 , , 649 , 759 connelley , m. s. , reipurth , b. , & tokunaga , a. t. 2007 , , 133 , 1528 cyganowski , c. j. , brogan , c. l. , hunter , t. r. , churchwell , e. , & zhang , q. 2011 , , 729 , 124 cyganowski , c. j. , brogan , c. l. , hunter , t. r. , & churchwell , e. 2009 , , 702 , 1615 cyganowski , c. j. , et al . 2008 , , 136 , 2391 cyganowski , c. j. , brogan , c. l. , & hunter , t. r. 2007 , , 134 , 346 davis , c. j. , gell , r. , khanzadyan , t. , smith , m. d. , & jenness , t. 2010 , , 511 , a24 davis , c. j. , et al . 2009 , , 496 , 153 davis , c. j. , kumar , m. s. n. , sandell , g. , froebrich , d. , smith , m. d. , & currie , m. j. 2007 , , 374 , 29 davis , c. j. , & smith , m. d. 1996 , , 310 , 961 de buizer , j. m. , & vacca , w. d. 2010 , , 140 , 196 dent , w. r. f. , little , l. t. , & white , g. j. 1984 , , 210 , 173 de zeeuw , p. t. , hoogerwerf , r. , de bruijne , j. h. j. , brown , a. g. a. , & blaauw , a. 1999 , , 117 , 354 dye , s. , et al . 2006 , , 372 , 1227 everett , m. e. , depoy , d. l. , & pogge , r. w. 1995 , , 110 , 1295 froebrich , d. , et al . 2011 , , 413 , 480 froebrich , d. , & ioannidis , g. 2011 , , 418 , 1375 gibb , e. l. , whittet , d. c. b. , boogert , a. c. a. , & tielens , a. g. g. m. 2004 , , 151 , 35 gredel , r. 1994 , , 292 , 580 hodapp , k .- w . 1994 , , 94 , 615 kaifu , n. , usuda , t. , hayashi , s. s. , et al . 2000 , , 52 , 1 karr , j. l. , manoj , p. , & ohashi , n. 2009 , , 697 , 133 lawrence , a. , et al . 2007 , , 379 , 1599 lee , h .- , & chen , w. p. 2007 , , 657 , 884 lee , h .- t . , chen , w. p. , zhang , z .- w . , & hu , j .- y . 2005 , , 624 , 808 lucas , p. w. , et al . 2008 , , 391 , 136 mercer , e. p. , et al . 2005 , , 635 , 560 minier , v. , ellingsen , s. p. , norris , r. p. , & booth , r. s. 2003 , , 403 , 1095 reach , w. t. , et al . 2006 , , 131 , 1479 robitaille , t. p. , whitney , b. a. , indebetouw , r. , & wood , k. 2007 , , 169 , 328 robitaille , t. p. , whitney , b. a. , indebetouw , r. , wood , k. , & denzmore , p. 2006 , , 167 , 256 rosolowsky , e. , et al . 2010 , , 188 , 123 seale , j. p. , & looney , l. w. 2008 , , 675 , 427 shepherd , d. s. , et al . 2007 , , 669 , 464 simpson , j. p. , burton , m. g. , colgan , s. w. j. , cotera , a. s. , erickson , e. f. , hines , d. c. , & whitney , b. a. 2009 , , 700 , 1488 simpson , j. p. , cotera , a. s. , burton , m. g. , et al . 2012 , , 419 , 211 smith , h. a. , hora , j. l. , marengo , m. , & pipher , j. l. 2006 , , 645 , 1264 stecklum , b. , caratti o garatti , a. , davis , c. , linz , h. , stanke , t. , & zinnecker , h. 2009 , protostellar jets in context , 619 takami , m. , karr , j. l. , koh , h. , chen , h .- h . , & 2010 , , 720 , 155 tobin , j. j. , hartmann , l. , calvet , n. , & dalessio , p. 2008 , , 679 , 1364 urquhart , j. s. , et al . 2009 , , 501 , 539 varricatt , w. p. 2011 , , 527 , a97 varricatt , w. p. , davis , c. j. , ramsay , s. , & todd , s. p. 2010 , , 404 , 661 voronkov , m. a. , caswell , j. l. , ellingsen , s. p. , & sobolev , a. m. 2010 , , 405 , 2471 voronkov , m. a. , brooks , k. j. , sobolev , a. m. , et al . 2006 , , 373 , 411 walther , d. m. , aspin , c. , & mclean , i. s. 1990 , , 356 , 544 watson , c. , povich , m. s. , churchwell , e. b. , et al . 2008 , , 681 , 1341 zapata , l. a. , rodrguez , l. f. , ho , p. t. p. , beuther , h. , & zhang , q. 2006 , , 131 , 939 zhang , m. , & wang , h. 2009 , , 138 , 1830 zinnecker , h. , & yorke , h. w. 2007 , , 45 , 481 irac image showing 3.6 @xmath1 ( blue ) , 4.5 @xmath1 ( green ) , and 5.8 @xmath1 ( red ) ; upper - right panel : @xmath0 mips 24 @xmath1 image ; lower - left panel : continuum - subtracted h@xmath3 image ; lower - right : gps @xmath4-band image . in addition , 4.5 @xmath1 contours are superposed . the contour levels are arbitrarily chosen for the best comparisons of the flux distribution between the 4.5 @xmath1 emission and the others . in the continuum - subtracted h@xmath3 image , compact features with a combination of positive and negative valued features are the residual of continuum subtraction of point sources . in addition to those features , values lower than the background are observed at the peak of the ego and @xmath28 northeast . this is due to continuum whose flux ratio between the h@xmath3 narrow band and @xmath4-band filters are different from that of foreground stars . the resolutions of irac , mips , and continuum - subtracted h@xmath3 images are @xmath162@xmath17 , 6@xmath17 , and less than 1@xmath17 , respectively . ] , but for ego g35.15 + 0.80 . the lowest contour around the ego roughly represents its shape . the dashed circles illustrate the h@xmath3 knots of the h@xmath3 outflow ( nho 2430a f ) . the white box is the region which will be shown in figure [ fig : g35.15 + 0.80b ] . ] , but for ego g35.20 - 0.74 . two diamonds label the positions of the ultra compact regions @xcite . the dashed ellipse marks the h@xmath3 outflow ( mho 2431a d ) . the white part in the 24 @xmath1 source is saturated . ] -band emission ; @xmath5 , @xmath4 , 3.6 @xmath1 , and 4.5 @xmath1 images in the first , second , third , and fourth columns , respectively.,title="fig : " ] -band emission ; @xmath5 , @xmath4 , 3.6 @xmath1 , and 4.5 @xmath1 images in the first , second , third , and fourth columns , respectively.,title="fig : " ] -band emission ; @xmath5 , @xmath4 , 3.6 @xmath1 , and 4.5 @xmath1 images in the first , second , third , and fourth columns , respectively.,title="fig : " ] -band emission ; @xmath5 , @xmath4 , 3.6 @xmath1 , and 4.5 @xmath1 images in the first , second , third , and fourth columns , respectively.,title="fig : " ] -band emission ; @xmath5 , @xmath4 , 3.6 @xmath1 , and 4.5 @xmath1 images in the first , second , third , and fourth columns , respectively.,title="fig : " ] -band emission ; @xmath5 , @xmath4 , 3.6 @xmath1 , and 4.5 @xmath1 images in the first , second , third , and fourth columns , respectively.,title="fig : " ] -band emission ; @xmath5 , @xmath4 , 3.6 @xmath1 , and 4.5 @xmath1 images in the first , second , third , and fourth columns , respectively.,title="fig : " ] g10.29 - 0.13 & 2.2 & n & y & n & + g10.34 - 0.14 & 2.0 & n & y & n & continuum emission near the ego + g11.92 - 0.61 & 3.9 & y & y & y & an h@xmath3 outflow lobe + g12.02 - 0.21 & 17.0 & y ? & n & n & an h@xmath3 knot located at the north of the ego + g12.20 - 0.03 & 4.6 & n & y & n & continuum emission near the ego + g12.42 + 0.50 & 2.4 & y ? & y & y & extended h@xmath3 emission , hii [ uhp2009 ] g012.4180 + 00.5038 + g12.68 - 0.18 & & n & n & n & + g12.91 - 0.03@xmath35 & 4.7 & y ? & y & n & a faint h@xmath3 knot + g14.63 - 0.58@xmath35 & 2.2 & n & n & n & + g16.58 - 0.08 & 3.6 & n & n & n & + g16.59 - 0.05 & 4.4 & y ? & n & n & an h@xmath3 knot + g16.61 - 0.24@xmath35 & & y & y & n & a faint h@xmath3 outflow lobe + g18.67 + 0.03 & 5.0 & n & y & n & + g18.89 - 0.47@xmath35 & 4.5 & n & n & n & + g19.01 - 0.03@xmath35 & 4.3 & y & y & n & an h@xmath3 outflow with two h@xmath3 knots + g19.61 - 0.12 & 4.1 & n & y & n & continuum emission near the ego + g19.61 - 0.14 & & n & y & n & + g19.88 - 0.53@xmath35 & 3.4 & y & y & y & a bipolar h@xmath3 outflow , iras 18264 - 1152 + g20.24 + 0.07 & 4.6 & n & y & n & continuum emission near the ego + g28.83 - 0.25 & 5.1 & n & y & y & continuum emission near the ego + g28.85 - 0.23 & & n & n & n & + g35.04 - 0.47 & 3.4 & y & n & n & a bipolar h@xmath3 outflow + g35.13 - 0.74@xmath35 & 2.4 & y & y & y & a bipolar h@xmath3 outflow + g35.15 + 0.80@xmath35 & 4.7 & y & y & n & a set of aligned h@xmath3knots + g35.20 - 0.74@xmath35 & 2.4 & y & y & & g35.2 - 0.74n , an hourglass shape bipolar h@xmath3 outflow + g35.68 - 0.18@xmath35 & 2.1 & y & y & y & an h@xmath3 outflow lobe + g35.79 - 0.17 & & y & y & n & a bipolar h@xmath3 outflow + g35.83 - 0.20 & 2.1 & y & y & n & a set of aligned h@xmath3 knots + g36.01 - 0.20 & 5.9 & n & n & n & + g54.11 - 0.04 & 4.9 & n & n & n & + g54.11 - 0.05 & & n & n & n & + g54.45 + 1.01 & 3.7 & y & y & n & an h@xmath3 outflow with two h@xmath3 knots + g54.45 + 1.02@xmath35 & & n & y & n & + g58.09 - 0.34@xmath35 & & n & y & y & continuum emission near the ego + [ tab : h2 ] 2303 & 18 13 57.5 & -18 54 29 & an h@xmath3 outflow lobe from ego g11.92 - 0.61 + 2243 & 18 21 53.6 & -14 35 50 & an h@xmath3 outflow lobe from ego g16.61 - 0.24 + 2244 & 18 25 44.8 & -12 22 46 & a pair of h@xmath3 knots near ego g19.01 - 0.03 + 2245 & 18 29 14.8 & -11 50 08 & a faint h@xmath3 knot near ego g19.88 - 0.53 + 2429 & 18 56 58.1 & + 01 39 37 & a bipolar outflow from ego g35.04 - 0.47 + 2430 & 18 52 36.6 & + 02 20 26 & a chain of h@xmath3 knots from ego g35.15 + 0.80 + 2431 & 18 58 12.9 & + 01 40 33 & an hourglass h@xmath3 outflow from ego g35.20 - 0.74 + 2432 & 18 57 05.0 & + 02 22 00 & an h@xmath3 outflow lobe from ego g35.68 - 0.18 + 2433 & 18 57 16.7 & + 02 27 56 & a bipolar outflow from ego g35.79 - 0.17 + 2434 & 18 57 26.9 & + 02 29 00 & a chain of h@xmath3 knots from ego g35.83 - 0.20 + 2622 & 19 28 26.4 & + 19 32 15 & two h@xmath3 knots near ego g54.45 + 1.01 + [ tab : mho ]

we present convincing evidence of unburned carbon at photospheric velocities in new observations of 5 type ia supernovae ( sne ia ) obtained by the nearby supernova factory . these sne are identified by examining 346 spectra from 124 sne obtained before d relative to maximum . detections are based on the presence of relatively strong absorption `` notches '' in multiple spectra of each sn , aided by automated fitting with the ` synapps ` code . four of the 5 sne in question are otherwise spectroscopically unremarkable , with ions and ejection velocities typical of sne ia , but spectra of the fifth exhibits high - velocity ( km s ) and features . on the other hand , the light curve properties are preferentially grouped , strongly suggesting a connection between carbon - positivity and broad band light curve / color behavior : three of the 5 have relatively narrow light curves but also blue colors , and a fourth may be a dust - reddened member of this family . accounting for signal - to - noise and phase , we estimate that% of sne iaexhibit spectroscopic signatures as late as d with respect to maximum . we place these new objects in the context of previously recognized carbon - positive sne ia , and consider reasonable scenarios seeking to explain a physical connection between light curve properties and the presence of photospheric carbon . we also examine the detailed evolution of the detected carbon signatures and the surrounding wavelength regions to shed light on the distribution of carbon in the ejecta . our ability to reconstruct the feature in detail under the assumption of purely spherical symmetry casts doubt on a `` carbon blobs '' hypothesis , but does not rule out all asymmetric models . a low volume filling factor for carbon , combined with line - of - sight effects , seems unlikely to explain the scarcity of detected carbon in sne ia by itself .

1109.1312

there is a well - established core population of type ia supernovae ( sne ia ) distinguished by striking spectroscopic homogeneity @xcite . the maximum light ultraviolet - optical - infrared spectra of these sne carry signatures of predominantly intermediate - mass and iron - peak elements ( often identified : oxygen , magnesium , silicon , sulfur , calcium , and iron ) . the consistent presence of these features in the spectra of the sn ia core population contrasts with the intermittent detection of carbon lines . such carbon is detected in the earliest data , but fades away by the time of peak brightness . sne ia probably arise from the thermonuclear incineration of carbon / oxygen white dwarfs in binary systems ( see * ? ? ? * for a review ) . since carbon is consumed during the explosion , constraints on its quantity , distribution , and incidence in sn ia ejecta help identify viable explosion mechanisms . for example , turbulent deflagration models predict that a large amount of unprocessed carbon may be left over ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) while models where the deflagration is followed by a detonation phase can more completely burn it ( e.g. , * ? ? ? * ; * ? ? ? interpreting oxygen signatures is more ambiguous , since oxygen is produced as well as consumed in the explosion . thus , constraints on carbon in sne ia are expected to be an important piece of the explosion mechanism puzzle . unfortunately , multiple factors hinder an unbiased accounting of carbon signatures in sn ia spectra . challenges inherent in both discovering and coordinating time - series spectroscopic follow - up of early , nearby sne ia result in relatively poor statistics at phases optimal for studying carbon ( i.e. , within 10 days after explosion or before about a week prior to maximum light ) : early observations probe the outermost layers of the sn , the region of the ejecta where carbon is most likely to be found . optical carbon features are seldom conspicuous even at the earliest phases , so noise may muddle detection in `` screening '' spectra of young sne ia . to achieve large numbers of sn candidates with a high `` purity '' of early sn ia discoveries , sn surveys must cover as much sky as possible while maintaining a few - day cadence ; a difficult endeavor . such surveys with associated on - demand access to spectroscopy resources , like the nearby supernova factory ( snfactory , * ? ? ? * ) or the palomar transient factory ( ptf , * ? ? ? * ) , provide a direct way to address the relative lack of spectroscopic time series sustained from the earliest phases . such data sets provide opportunities to obtain new constraints on sn ia physics . by comparison , spectra obtained at and after maximum light are already plentiful and well - studied . the need to understand the sn ia explosion mechanism and progenitor system is as critical as ever , especially because of the important role sne ia play in observational cosmology . observations of high - redshift sne ia resulted in the discovery of dark energy , a revelation that has changed our view of the universe radically over the past decade . the universe is not only expanding , but the rate of expansion itself is _ accelerating _ , in contrast to previous expectations that it would be found to be decelerating @xcite . new techniques show promise for improving sn ia distance measurement precision @xcite . however , the fact that sne ia couple dark energy results to stellar evolution astrophysics amplifies concerns that drift in emergent sn properties as a function of redshift could be confused with a cosmology signal . for example , the demographics of the underlying progenitor population may drift over time , leading to a shift in measured sn properties such as decline rate @xcite . overcoming such problems requires building a coherent picture of the progenitor system and explosion mechanism . in support of this effort , we consider the occurrence of unburned carbon in a large set of early phase spectra obtained by the snfactory . our analysis supplements that of @xcite who conducted the first comprehensive carbon signature survey using publicly available ( and some new ) sn ia spectroscopy . together , our studies find or verify several `` definite '' or `` probable '' @xmath16580 detections in sne ia that are neither super - chandrasekhar candidates nor otherwise obviously peculiar . six of 27 normal sne ia with snfactory observations before @xmath5 d and with sufficient signal - to - noise ( s / n ) have definite @xmath16580 detections . such normal sne are the foundation of the sn ia hubble diagram ; studying observables correlated with the incidence of carbon may provide new constraints on the explosion mechanism , leading to new systematics controls or ruling out the need for some . in particular , we are interested in addressing the question of whether carbon - positivity in sne ia is idiosyncratic ( distributed randomly within the population without regard to other observables like light curve shape or color ) or specifically correlated with other parameters . for example , if the frequency ( or strength , or persistence ) of carbon signatures is somehow correlated with luminosity or light - curve shape , explosion models need to reproduce this relationship . our results are the strongest indication so far of such a connection between carbon - positivity and broad - band photometric properties in normal sne ia . the remainder of this article is arranged as follows . in [ sec : selection ] , the initial carbon detection procedure is outlined and in [ sec : data ] , the selected sample is presented . [ sec : analysis ] covers our analysis of the pre - maximum evolution of the identified carbon features , overall characteristics of the selected spectra , and light - curve / color properties of these sne . details on individual objects are deferred to the appendix . [ sec : discussion ] brings together our results , provides the first measurement of the @xmath16580 detection rate , and discusses our findings in broader contexts . in [ sec : conclusion ] , we draw conclusions . the spectra used in this study were obtained by the snfactory between 2004 and 2009 with the supernova integral field spectrograph ( snifs , * ? ? ? snifs is a fully integrated instrument optimized for automated observation of point sources on a structured background over the full ground - based optical window at moderate spectral resolution . it consists of a high - throughput wide - band pure - lenslet integral field spectrograph ( ifs , la tiger ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , a multi - filter photometric channel to image the field in the vicinity of the ifs for atmospheric transmission monitoring simultaneous with spectroscopy , and an acquisition / guiding channel . the ifs possesses a fully - filled @xmath6 spectroscopic field of view subdivided into a grid of @xmath7 spatial elements , a dual - channel spectrograph covering 32005200 and 510010000 simultaneously , and an internal calibration unit ( continuum and arc lamps ) . snifs is mounted on the south bent cassegrain port of the university of hawaii 2.2 m telescope on mauna kea , and is operated remotely . spectra of all targets were reduced using the snfactory s dedicated data reduction pipeline , similar to that presented in 4 of @xcite . a key point is that all spectra analyzed and presented have been corrected for telluric absorption . a brief discussion of the software pipeline is presented in @xcite and is updated in @xcite . a detailed description of the instrument and reduction pipeline will be published in the future . the starting - point for the analysis is the sample of snifs spectra obtained prior to @xmath0 d with respect to rest - frame @xmath8-band maximum , as determined by salt2 light - curve fits @xcite to sne iaobserved on 5 or more nights . this selection yields 346 epochs of spectrophotometry from 124 individual sne ia . we have excluded the previously published carbon - positive sn 2006d @xcite , but will include it in the discussion in [ sec : discussion ] . the sample includes spectroscopically `` normal '' sne ia @xcite , overluminous sn 1991t - like and sn 1999aa - like @xcite sne ia , and candidate super - chandrasekhar - mass sne ia @xcite . a range of light curve stretch values @xmath9 @xcite is represented ( corresponding post - peak decline rate @xmath10 , * ? ? ? . however , there are no subluminous sn 1991bg - like events following the most strict definition ( i.e. , exhibiting a prominent absorption trough at maximum light ; * ? ? ? * ) . to help identify candidate carbon features , we systematically conduct fits to the relevant portion of each spectrum using the ` synapps ` code @xcite . ` synapps ` uses a highly parameterized but fast sn spectrum synthesis technique , and as such does not solve the full model sn atmospheres problem , in contrast to programs like ` phoenix ` @xcite , ` cmfgen ` @xcite , or ` sedona ` @xcite . the other codes mentioned are examples designed for full / partial non - local thermodynamic equilibrium radiative transfer , enabling detailed post - processing of stellar explosion models for comparison to observations . ` synapps ` is useful in the opposite sense , producing constraints on explosion models from observations , since it identifies ion species and estimates their characteristic ejection velocities . given the schematic radiative transfer model that ` synapps ` uses and its as yet unquantified systematic uncertainty , we employ the fit results as an adjunct in our examination of the spectra rather than as a direct quantitative discriminant . the radiative transfer assumptions underlying ` synapps ` match those of the well - known ` synow ` sn spectrum synthesis program @xcite : spherical symmetry , a sharply - defined blackbody - emitting pseudo - photosphere , pure - resonance line transfer under sobolev formalism , with line opacity parameterized radially using convenient functional forms and in wavelength assuming boltzmann excitation . ` synapps ` combines this parameterized spectrum synthesis calculation with a parallel non - linear optimization framework @xcite to reduce the need for tedious interactive adjustment of fit parameters and to assure more systematic sampling of the parameter space . we use @xmath16580 as our primary carbon indicator . we consider it the best option , for reasons we now describe . @xmath1@xmath14745 and 7234 may also be useful ( and have been previously identified , e.g. , * ? ? ? * ) but are much weaker at the temperatures of interest , and @xmath14745 is very likely to be blended with other lines . elsewhere , @xmath14267 has been considered as a carbon tracer ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) but it is problematic : it does not seem to produce a reliable , distinct feature in the spectrum , and is vastly weaker than @xmath16580 at the relevant temperatures . there is some evidence for the high - excitation @xmath14649 line in the spectra of sn 1999aa - like @xcite and sn 1991t - like @xcite sne ia , but it is the only strong optical feature and is not reliably detected in the spectra of less luminous objects . lines are mostly too red or too weak ( e.g. , @xmath1@xmath19087 , 9406 , 10693 ) for us to consider with snifs data @xcite . thus , @xmath16580 seems to be the best choice , especially considering the observational constraints . as @xcite demonstrated ( their figure 9 ) , @xmath16580 absorptions caused by carbon at higher velocities ( much faster than that at the photosphere at the phases of interest ) will disappear into the @xmath16355 absorption trough where they can not be detected reliably . hence , inferences about carbon from the @xmath16580 notch are limited to lower velocities . fortunately , the low velocity regime is what is of greatest interest for constraining the explosion mechanism . carbon at the highest velocities may not be surprising ; the presence of carbon is much more variable at lower velocities in existing models . a final , but related , issue worthy of mention is the systematic error that could be incurred when estimating the ejection velocity of carbon directly from the position of the absorption notch instead of with the aid of modeling , that of center - to - limb variation @xcite . without the aid of any kind of spectral modeling , the velocity of the carbon layer could be underestimated by up to a factor of 2 . however , ` synapps ` ( at least ) handles line blending explicitly , and the velocity at the photosphere will be estimated by a fit to the full spectrum , so we can be reasonably confident that the carbon is located near or just above the photosphere . the wavelength range selected for initial fitting extends from 5300 to 7300 . the pseudo - continuum level and line features are simultaneously fit by ` synapps ` , and all fits include lines from , , and . lines from these ions are the most important to consider in this wavelength region while the sn is on the rise . strong iron - peak lines only become important after the phases considered here . a second - order polynomial warping function multiplied against the synthetic spectrum is included to mitigate limitations of the parameterized model continuum , which does not affect our ability to detect a @xmath16580 absorption `` notch , '' a much smaller scale feature . since ` synapps ` models the sn spectrum but not the intervening gas , we mask out rest - frame wavelengths between 5875 and 5915 , a generous allowance to exclude host galaxy interstellar ( or circumstellar ) d contamination , based on the width of the strongest d feature observed in the sample ( snf20080720 - 001 , * ? ? ? all spectra presented and fit by ` synapps ` are corrected for observer - frame milky way ( @xmath11 ) extinction using the ccm @xcite dust law , with color excess estimates from @xcite . three fits are performed to each spectrum to find candidate features . the first set of fits including only , , and provides a carbon - free baseline . the second set of fits adds with a `` photospheric '' sobolev optical depth profile , meaning that the minimum velocity of the profile is constrained to equal the velocity at the photosphere . the third set of fits substitutes a `` detached '' optical depth profile , with the minimum velocity of the profile unconstrained with respect to the photospheric velocity . in all 3 fits , the sobolev opacity declines with velocity exponentially . a set of fits allowing opacity to be detached was also performed , but in practically all cases ` synapps ` found a better fit without detachment . some of the targets included in the analysis have previously been identified or suspected of being carbon - positive , in particular snf20080514 - 002 @xcite . to reduce ( somewhat ) the risk of confirmation bias , the process of organizing and executing the fits includes concealing the identity of the parent sn of each spectrum and its phase . this is accomplished by shuffling the spectra and assigning a unique integer identifier to each . the mapping from spectrum to sn and phase is kept in a hash file that is only consulted after the carbon detections have been made . this procedure , along with the ` synapps`fitting , forced us to consider more spectra more carefully . it also concealed the measured light curve fit parameters while identifying strong carbon features , ensuring that any correlation between them and the presence of carbon would also be uncovered without bias . for a detection to be considered definitive , we require that be detected in at least 2 spectra of an object s time - series , either from the same night of observation or across multiple consecutive nights . the final identifications are confirmed by visual inspection and comparison of the fits in the region of the @xmath16355 emission peak . cases where the fits with and without are basically identical are considered non - detections ( the addition of opacity does not improve the fit ) . fits with that differ from the baseline fit , and moreover improve upon it , are flagged for closer scrutiny . several example fits are shown in figure [ fig : feature ] . while the detailed agreement in all cases is not perfect , the ` synapps ` fits greatly improve our confidence in the identifications we otherwise would have made purely by visual inspection . ` synapps ` also highlighted cases we might have overlooked , but in practice few of these cases ultimately made it into the sample , usually because they fail the 2-spectrum cut or were less convincing . lllrrcl sn 2005di & mgc -04 - 52 - 46 & @xmath12 & @xmath13 & @xmath14 & 2005 - 08 - 26.2 & @xmath15 + sn 2005el & ngc 1819 & @xmath16 & @xmath17 & @xmath18 & 2005 - 10 - 03.6 & @xmath19 + sn 2005ki & ngc 3332 & @xmath20 & @xmath21 & @xmath22 & 2005 - 12 - 01.2 & @xmath23 + snf20080514 - 002 & ugc 8472 & @xmath24 & @xmath25 & @xmath26 & 2008 - 05 - 26.5 & @xmath27 + sn 2005cf & mcg -01 - 39 - 003 & @xmath28 & @xmath29 & @xmath30 & 2005 - 06 - 12.9 & @xmath31 [ tab : list ] the fit survey yielded 6 of 124 sne ia with definitive @xmath16580 absorption notch detections . in one of these cases , sn 2005cf , there is only a single - spectrum detection , but this sn is difficult to discard since the detection is corroborated by other external data sets . of the 124 sne ia in the sample , 4 are super - chandrasekhar - mass sn iacandidates ( including sn 2007if , the only clear @xmath16580 detection among the 4 , * ? ? ? removing these leaves 5 out of 120 carbon - positive sne ia : sne 2005di , 2005el , 2005ki , snf20080514 - 002 , and sn 2005cf ( a special case ) . for our dataset , spectra having the requisite s / n generally come from sne at the lowest redshifts . due to the resulting small volume sampled by untargeted surveys out to such redshifts , the sample here is dominated by sne iafrom targeted searches . these new detections are the focus of our detailed analysis in the rest of this article . basic target information is summarized in table [ tab : list ] and selected pre - maximum spectra of each object , up to the point where the carbon signatures disappear , are shown in figure [ fig : spectra ] . a few characteristics of the spectra presented in figure [ fig : spectra ] are worthy of specific mention . the spectra of sn 2005di are quite red , but there is evidence ( discussed below ) that this is due to extrinsic dust extinction . sn 2005el , sn 2005ki , and snf20080514 - 002 all possess what appear to be pronounced flux excess blueward of about 3600 . the spectra of these 3 objects are also dramatically similar in terms of detailed feature morphology . sn 2005cf clearly stands apart , with its more modest relative flux levels blueward of 3600 , but broader h&k and 6150 absorptions , and apparent high velocity ( hv ) infrared triplet . snifs observed sn 2005di on 10 occasions between aug . 14 and sep . 22 . the first 3 spectra of sn 2005di appear in figure [ fig : spectra ] . the @xmath32 d spectrum used here is a co - addition of 2 consecutive exposures on the same night . close examination of the time series reveals the presence of narrow rest - frame d absorption ( see figure [ fig : carbon]a ) . this and the relatively red appearance of the spectral energy distribution lead us to conclude that sn 2005di is heavily extinguished by line - of - sight dust in its host galaxy or circumstellar environment . since snifs exposure times were generally not adjusted to compensate for high extinction , the spectra of sn 2005di have lower s / n than typical sne ia observed by snifs at the same redshift . still , @xmath16580 is detected in the first spectrum and possibly in the second . initial spectroscopic classifications of sn 2005di were published by 2 groups . @xcite reported that a spectrum obtained aug . 16 resembles the spectrum of another sn ia a few days past peak , and @xcite state that their aug . 14 spectrum suggests that sn 2005di is `` young , '' but neither report discusses . the snifs quick - reduction pipeline classification indicated that sn 2005di was a sn ia about 1 week before maximum , and highly extinguished with an estimated @xmath33 . revisiting our quick - reduction spectrum we find that the @xmath16580 absorption notch could have been identified , but its significance was not obvious at the time . sn 2005di was used in the snfactory spectral flux ratio luminosity indicator study @xcite . that study found the ratio of flux measured at 642 nm to that at 443 nm at maximum light ( @xmath34 ) to be a good predictor of relative luminosity . sn 2005di is the object in that study with the largest value of @xmath34 , owing primarily to its red color . any possible effect of @xmath16580 emission near 642 nm either as a possible source signal or noise in the relationship between @xmath34 and hubble diagram residual was not discussed , since at the time of publication the presence of had not been noticed . we expect our @xmath16580 detection to have no implication for the flux ratio result , since that study was restricted to spectra obtained within 2.5 d of maximum , when the carbon signature is gone . nineteen snifs visits of sn 2005el were conducted between sep . 26 and nov . 20 . figure [ fig : spectra ] includes the spectra from the first 2 of these . the @xmath35 d observation is a co - addition of the 2 observations obtained that night . in this case , @xmath16580 is only positively detected by ` synapps ` on the first date of observation . however , it is clearly present in both high s / n spectra taken that night and the co - add . sn 2005el is a very well - observed sn @xcite . its spectroscopic normality has been remarked upon in the literature previously ( e.g. * ? ? ? * ; * ? ? ? initial classification reports indicated that sn 2005el was discovered a few to several days prior to maximum @xcite but until now no remarks have been made in the literature regarding the presence of in the spectra . these classification spectra were obtained the same night as the first snifs observations . the snifs quick - reduction pipeline in use during the sn 2005el campaign failed to properly correct for a cosmic - ray hit coincident with the @xmath16580 notch in one spatial element of the 2 spectra obtained on the first night . subsequent improvements to the pipeline later revealed the clear @xmath16580 notch in both spectra . snifs follow - up of sn 2005ki was triggered nov . six nights of observations were obtained in all , ending dec . 2 , around the time of maximum brightness . follow - up ended prematurely due to equipment failure at the uh 2.2 m telescope that forced an extended offline period for snifs . therefore , we adopt the date of maximum and light - curve fit parameters from a salt2 fit to more complete photometry published by @xcite and @xcite . the first 3 spectra of sn 2005ki are presented in figure [ fig : spectra ] . relatively weak @xmath16580 absorption is detected in the first spectrum , and very marginally in the second . returning to the first 2 snifs quick - reduction spectra of this object , the presence of @xmath16580 now seems obvious in retrospect . our classification announcement made no mention of it @xcite . twenty nights of spectroscopy of snf20080514 - 002 were obtained with snifs , starting may 16 and ending jul . the first 4 spectra of snf20080514 - 002 appear in figure [ fig : spectra ] . the @xmath36 d and @xmath37 d spectra are co - additions of 3 consecutive observations on each respective night . @xmath16580 is clearly detected in the first 2 spectra of this object , and probably in the third . though the emission peak visible at the position of the notch in the fourth spectrum looks flattened , we hesitate to posit a flux depression due to opacity in this case . at this point , ` synapps ` is able to convincingly fit the feature with or without a weak @xmath16580 line . as initially reported @xcite , the early spectra of snf20080514 - 002 are quite blue , indicative of a sn ia at or before a week prior to maximum light . both spectra exhibit absorption lines attributed to @xmath16580 , but @xmath17234 is not as readily apparent as in , say , sn 2005el . @xmath14745 was also mentioned , but is a much weaker and more blended line than the others . like sn 2005di , snf20080514 - 002 was used in the @xcite flux ratio study . in marked contrast to sn 2005di , this target has an extremely _ blue _ color but is well within the core of the hubble residual versus @xmath34 distribution ( their figure 2 ) . as with sn 2005di , the use of the at - maximum spectrum for the flux ratio study limits the influence of @xmath16580 on those results . snifs observed sn 2005cf on 12 nights from jun . 3 to jul . the first 2 spectra are shown at the bottom of figure [ fig : spectra ] . a small @xmath16580 absorption notch is visible in the @xmath31 d spectrum but becomes absent or nearly absent 2 days later . our selection guidelines require detection in 2 distinct snifs spectra , so we should technically disqualify sn 2005cf from further consideration . however , sn 2005cf is an object with extensive previously published spectroscopy @xcite . neither previous spectroscopic study nor the original classification announcement @xcite discussed the putative @xmath16580 notch visible in the spectra . when considering those observations , @xcite placed sn 2005cf into their `` uncertain '' detection category . but given 3 independent data sets showing the same feature at roughly the same phase , it now seems difficult to overlook the presence of in the case of sn 2005cf . . in ( b ) the red overlaid curve is the smoothed flux . , title="fig:",scaledwidth=48.0% ] . in ( b ) the red overlaid curve is the smoothed flux . , title="fig:",scaledwidth=48.0% ] the detailed evolution of the carbon features in the pre - maximum spectra of sne 2005di , 2005el , 2005ki , and snf20080514 - 002 is depicted in figure [ fig : carbon ] . sn 2005cf and a non - detection case for comparison ( snf20071021 - 000 ) appear in figure [ fig : others ] . the @xmath16580 notch is readily apparent , situated at or quite near the rest - frame emission peak of the @xmath16355 feature . the detailed evolution of the feature in each case is described in the appendix . in the first 4 cases , the notch is strongest in the earliest spectrum . it weakens gradually with time , and without appreciable wavelength shift . generally , the notch is centered on an apparent blueshift of @xmath38 km s@xmath3 relative to rest - frame @xmath16580 ( also roughly the same velocity as observed in sn 2006d ) . in sn 2005cf , the notch disappears between the first and second spectrum . excluding sn 2005cf , we see that while the notch dissipates , the shape of the spectrum just redward evolves as well . the slope of the decline in flux on the red side of the @xmath16355 emission feature becomes steeper while the notch weakens and the underlying @xmath16355 emission peak rises . in some cases ( sne 2005di , 2005ki , and snf20080514 - 002 ) the earlier spectra may even exhibit a mild `` flux plateau '' morphology . along with an absorption notch , such an emission plateau could be a signature of a `` detached '' distribution of line opacity in a shell separated somewhat from the photosphere @xcite . however , under spherical symmetry , such a plateau should be roughly symmetric about the line rest wavelength . visually , this does not seem to be the case here . the notch in the spectrum of sn 2005cf is morphologically distinct from the other cases , and appears to span a slightly smaller range in blueshift ( @xmath39 to @xmath40 km s@xmath3 ) . when the notch disappears , there is at most only a small amount of evolution in the flux slope just redward . confident identification of @xmath17234 is problematic . in almost none of the cases presented is there a clear absorption feature at the same blueshift as the @xmath16580 notch . at most , there is only an inflection visible around 69007000 and a slight emission `` bump '' around 7100 . visual inspection alone would seem to be insufficient to draw more firm conclusions about the presence or absence of @xmath17234 in general . to examine the extent to which the selected targets are spectroscopically normal sne ia , we again use ` synapps ` , this time to fit the full wavelength range of each pre - maximum spectrum of each selected sn ( detailed commentary is provided in the appendix ) . close examination of the features in the full fits may also enable some explosion - model independent statements about the distribution of carbon in the sn ejecta . two selected fits for each sn appear in figures [ fig : synapps_1 ] and [ fig : synapps_2 ] : one fit to the first available spectrum , and another fit to the first spectrum where is absent . in general the opacity profiles for lines decrease with time as expected . wavelengths of major ion contributors to the spectra are marked at the top of either figure . except for the presence of , the spectra of the first 4 carbon - positive sne ia are anything but extraordinary typical sn ia ions at normal pre - maximum ejection velocities . sn 2005cf is another matter , since hv and components are mandatory for a good fit , as discussed previously @xcite . in each case , we find compelling evidence for opacity from , , , and , , , singly - ionized ions of ti through co , and . for both and we enable 2-component fits : one for a photospheric ( undetached ) component , and another for a hv ( detached ) component . the opacity profiles of all other ions are undetached from the photosphere . optimal values for the velocity at the photosphere are found in the range @xmath38 to @xmath40 km s@xmath3 . including some opacity does seem to improve the fit morphology around the 5800 absorption feature however this identification is not unprecedented ( e.g. , * ? ? ? * ) . in all cases , we were able to reconstruct the basic structure redward of 3500 quite well , and at times had reasonable success fitting the ultraviolet ( uv , where lines from iron - peak elements dominate ) . we do not expect exceptional agreement in the uv , as the underlying assumptions of ` synapps ` ( in particular the treatment of line opacity ) begin to break down here . in general , the strengths of the features evolve smoothly with time , and no abrupt changes are observed as the becomes optically thin . ` synapps ` accounts for the @xmath16580 feature , including the region just redward of the notch , rather well ( see figure [ fig : closeup ] , red curve ) . variations with and without detached were tried , but again ` synapps ` tended to set the lower velocity of the region to the velocity at the photosphere , making the need for detachment uncompelling . analysis of the fit residuals indicates that the ` synapps ` model has insufficient fidelity for achieving a reduced @xmath41 of 1 , but we find no strong indications of persistent residual structure . in figure [ fig : closeup ] , the 58007500 region of the first spectrum of each carbon - positive sn is overlaid with the converged , all - ions , full - wavelength ` synapps ` fits ( shown in red ) . the blue curve results when the fully converged synthetic spectrum is recomputed with the opacity deactivated , while all other settings retain their converged values . the blue curves are thus _ not _ re - fits without included , in contrast to the example plots shown in figure [ fig : feature ] . thus , the change from the blue curve to the red one illustrates what role plays in the formation of the spectrum quite clearly . not only is @xmath16580 responsible for the absorption notch , it `` screens '' emission from bluer @xmath16355 effectively enough that some emission relative to the -free spectrum ( centered around 6600 ) is visible in the top 4 cases . finally , figure [ fig : closeup ] indicates that the flux depression around 69007000 may indeed be influenced by @xmath17234 to some extent . it is interesting that 3 of the 5 carbon - positive sne ia have relatively narrow light curves , with low salt2 @xmath42 values around @xmath43 ( @xmath44 , or @xmath45 , see table [ tab : list ] ) . these same targets also have bluer than normal colors . snf20080514 - 002 is also among the sne ia observed by swift / uvot @xcite with largest uv excesses @xcite , as shown in figure [ fig : uvot ] ( adapted from * ? ? ? swift / uvot photometry for sne 2008hv and snf20080514 - 002 will be presented in brown et al . 2011 , in preparation . figure [ fig : lc_params ] is a plot of salt2 fit parameters for all sne ia with snifs light curve fits and a spectrum before @xmath5 d , the latest phase for which any @xmath16580 is positively detected in this sample . sne 2005el , 2005ki , and snf20080514 - 002 form a relatively tight `` cluster '' at the fast - declining ( low @xmath42 ) but blue corner of this plot . other published results for sn 2005el have a somewhat larger value for @xmath42 @xcite , though all measurements indicate that @xmath42 is negative and this sn belongs in this group of relatively narrow light curves with blue colors . thus , the quantitative results for sn 2005el may be further refined , but our qualitative conclusions should remain unchanged . neither sn 2005cf nor sn 2005di fit into the fast - decliner / blue - color family . in the case of sn 2005cf , this may not be surprising since the and features quite robustly trace out hv ejecta more dramatically than in the other sne at the same phases , indicating real physical differences between sn 2005cf and the others . however , as described in [ sec : data ] there is evidence that sn 2005di is extrinsically reddened . since our data do not resolve the voigt profiles of the possible multiple velocity components comprising the d absorption line , we are unable to apply a quantitative correction for extinction using d equivalent width @xcite . instead , we use a different procedure to investigate whether sn 2005di could simply be a reddened member of the fast - decliner / blue - color subclass . to see if this is at least a plausible hypothesis , we warp the spectral time series of sn 2005di according to the ccm dust law , fitting for the optimal differential @xmath46 that brings the spectra into detailed agreement above 5000 with each of sne 2005el , 2005ki and snf20080514 - 002 . we make no attempt to constrain the differential @xmath47 and we assume a standard milky way value of 3.1 . if we assume that the main difference between sn 2005di and the other sne is the amount of dust extinction the former suffers , then correcting for it should bring the spectral time series into alignment . in the case of perfect `` twin '' sne , the agreement above 5000 should be exquisite on a feature - by - feature basis , and the extrapolated flux below 5000 a good match . the warping technique and a fully developed twin sn study are the subject of a future paper @xcite , here we briefly describe the mechanics for 2 given objects . spectral observations of the objects are paired based on phase , with similar phases being matched to each other . a linear interpolation correction to the fluxes accounts for overall flux difference arising from the small phase differences . a ccm dust law with fixed @xmath11 is fit to the ensemble of paired spectra , as well as an individual scale factor for each pair , to account for any small scatter . by fitting all of the pairs ( extending from early epochs to late epochs ) jointly for the appropriate dust law warping , we minimize the influence of any particular features on the fit at any given phase . figure [ fig : matches ] shows pairs of spectra that result when the time series of sn 2005di is warped to match the flux from another sn above 5000 , roughly matching by phase . the earliest pair of phases formed in each of the 3 cases is shown . the agreement above 5000 on a feature - by - feature basis is quite good , though there is a significant difference in the @xmath16355 absorption feature with respect to sn 2005ki clearly an `` intrinsic '' difference . blueward of the 5000 cut , the extrapolated color looks quite consistent . thus , while we can not prove definitively that sn 2005di is a dust - estranged member of the fast - decliner / blue - color family , there does seem to be a resemblance . figure [ fig : lc_params ] includes the position of sn 2005di when corrected for the median derived differential @xmath48 is removed . to summarize , our most important findings are : * definitive new detections of @xmath16580 absorption notches in the pre - maximum spectra of 5 largely spectroscopically normal sne ia , at apparent blueshifts around @xmath38 km s@xmath3 , consistent with the velocity of the ejecta at the photosphere . * the observed notches fade with time without appreciable shift in wavelength , disappearing before maximum light . detailed fitting shows that @xmath16580 influences the formation of the spectrum beyond the immediate vicinity of the absorption notch . * three of the 6 carbon - positive sne ia ( including sn 2006d ) in the full snfactory sample group together in light curve shape and color space at the fast - decliner / blue - color corner . a further sn may belong to this family , plausibly separated from it by the effect of extrinsic dust reddening . we consider 4 main questions in light of our results . first , factoring in the limitations of our sample and approach , what is the incidence of photospheric carbon in pre - maximum spectra of normal sne ia ? second , to what extent does carbon - positivity in sne ia predict a fast - declining light curves and blue color ( or vice - versa ) , and what is the physical significance of such a connection ? third , what inferences can we make about the geometry of the ejected carbon in the sne ia studied here ? finally , what impact do our findings have on our understanding of the sn ia explosion mechanism ? our data suggest that unburned carbon , while far from ubiquitous , is also not particularly rare in line with the result of @xcite . a detailed rate calculation is beyond the scope of this article , but we can try to estimate the frequency of photospheric - velocity carbon in normal sne ia based on the latest phase ( @xmath37 d ) and lowest median s / n per 2.4 bin in the wavelength range of interest ( 11.7 , that of sn 2005di at @xmath49 d ) where @xmath16580 is detected . 31 sne ia pass these cuts in the full snfactory sample ( which includes sn 2006d , * ? ? ? * ) , of which 2 are super - chandrasekhar candidates and a further 2 are sn 1991t - like . excluding these 4 sne places the fraction of sne iawith @xmath16580 that persists to @xmath5 d at @xmath4% ( bayesian binomial confidence interval , see * ? ? ? * ) . sn 2005el , sn 2005ki , snf20080514 - 002 , and perhaps sn 2005di could be members of a physically meaningful family of sne ia characterized by detectable carbon in their early spectra , relatively fast - declining light curves , and intrinsically blue colors . these 4 sne were selected solely based on the presence of carbon signatures in their spectra , so we now reverse the selection and re - examine the early spectra of other objects with similar light curve or color parameters . the most interesting results from our surveyed sample come from objects with similar light curve width . figure [ fig : other_carbon ] shows the 58007500 region of the pre - maximum spectra of sn 2006 dm @xcite and snf20080919 - 002 . neither case presents a robust notch , however the @xmath16355 emission peak looks somewhat `` squashed '' in the earliest spectra . the evidence is hardly decisively in favor of photospheric - velocity carbon in these 2 additional sne ia , but it is tantalizing . carbon - positive sn 2006d @xcite is also plotted in figure [ fig : lc_params ] . it also has a salt2 @xmath50 around @xmath43 but is redder than the objects uncovered by our survey here . returning to the swift / uvot color - curves in figure [ fig : uvot ] , we identify sn 2008q @xcite and sn 2008hv @xcite as having uv colors similar to those of snf20080514 - 002 . all 3 of these targets separate from the rest of the swift sample especially well in @xmath51 color space . while the spectroscopic classification announcements for neither sn 2008q @xcite nor sn 2008hv @xcite mention @xmath16580 , the pre - maximum spectra of both objects include the absorption notch signature ( v. stanishev and g. h. marion , private communication ) . both @xcite and @xcite report a faster - than - average decline rate ( corresponding roughly to @xmath42 of @xmath52 and @xmath53 respectively ) for sn 2008q , suggesting that it could also belong with the fast - decliner / blue - color family . optical light curve analysis for sn 2008hv is yet to be published , but a preliminary decline - rate estimate from the uvot photometry appears to be faster than average . next , we consider whether normal sne ia with previous @xmath16580 detections in the literature have a tendency towards narrow light curves and blue colors . a revisit of @xcite yields mixed results , partially because of the lack of consistent light - curve fits across all the data sets . of the 2 `` definite '' cases presented there , sn 1990n fails to fit the pattern ( @xmath54 ) ; the other is sn 2006d . the `` probable '' category ( where data are available ) is dominated by narrower light curves than average as well , but no strong tendency toward blue colors . the exact role that host galaxy dust extinction plays in the @xcite sample is unconstrained , however . overall , it would seem that the picture for unburned carbon in sne ia may be a complex one , where there are multiple subclasses and considerable variation between them . four of the 5 newly identified carbon - positive sne ia may indeed be representative of a meaningful subclass , but sn 2005cf stands apart . its light curve shape is less extreme and its colors more red . it also lacks the large uv excess of snf20080514 - 002 seen in figure [ fig : uvot ] . hv ejecta signatures in the spectra of sn 2005cf are much stronger than in the other carbon - positive sne ia considered here . though hv features may be `` ubiquitous '' @xcite , it is clear that there is considerable variation in the velocities and line strengths observed . cases with the strongest hv ir triplet also seem to exhibit strong @xmath16355 absorption early on @xcite . the origin of the hv features remains unclear ( is it a density or composition enhancement in the ejecta , or is it swept - up circumstellar material ? see e.g. , * ? ? ? ? * ; * ? ? ? evidence for a connection between carbon and the hv features is weak . the best candidate aside from sn 2005cf with hv features and a possible @xmath16580 notch seems to be sn 2003du @xcite . on the other hand , sn 2009ig ( also with hv features ) , appears to be a counter - example ( not a distinct notch , so * ? ? ? * label it at most a possible detection ) . snf20071021 - 000 , the non - detection example in figure [ fig : others ] is yet another . a number of `` peculiar '' sne ia ( or sn ia - like objects ) with carbon lines have been discussed in the literature . here we compare our sample of normal sne ia with those objects . features are a spectroscopic bellwether of candidate super - chandrasekhar sne ia , where they seem to persist to maximum light and even beyond @xcite . this has been interpreted as evidence supporting the super - chandrasekhar hypothesis , where more total carbon is present in the progenitor and/or much remains after disruption . these objects are quite overluminous , and have much broader than normal light curves . their spectra , while mostly consistent with one another , are quite different from those of objects considered here . overall , if the super - chandrasekhar candidates form a subclass of sne ia , it is distinct from the one proposed here . with varying degrees of certainty , carbon is also occasionally detected in low - luminosity , short time - scale , or otherwise peculiar sne ia . @xcite claim that @xmath16580 is present in the spectra of sn 2005bl , a low - luminosity sn 1991bg - like sn ia , but carbon is seldom remarked upon or observed in other members of this subclass . more unusual events like sn 2002cx @xcite , sn 2005hk @xcite , sn 2007qd @xcite , and sn 2008 ha @xcite have been conjectured to represent a distinct population of peculiar , low - luminosity sne ia . these sne are quite sub - luminous ( peak @xmath55 ) , have light curves that evolve more rapidly than normal sne ia , and have spectra that exhibit many narrow absorption lines that suggest lower kinetic energies than typical . pure white dwarf deflagrations or failed deflagrations may provide a plausible way to account for those observations , in which case carbon signatures may also be expected ( e.g. , * ? ? ? * ) . the carbon identifications in the spectra of these events are irregular , tentative in most cases , though sn 2008 ha presents the most clear signatures @xcite . spectra of candidate `` sne .ia '' @xcite sn 2002bj @xcite and sn 2010x @xcite exhibit robust detections , but so far detailed simulations of sne .ia @xcite do not predict observable carbon . all of the above cases are much dimmer than the sne considered here , and much more exotic spectroscopically . sn 2006bt @xcite , however , merits specific attention . an outlier on the sn ia light - curve - width / luminosity relation , sn 2006bt exhibits some spectroscopic peculiarities . most interestingly , its earliest spectra possess a unique @xmath16580 signature , a notch centered at @xmath56 km s@xmath3 , apparent in at least 2 consecutive observations . there is little , if any , evidence for emission to the red of the notch , and the vast difference between the apparent blueshift of the @xmath16580 notch and the velocities of other ejected elements ( @xmath57 km s@xmath3 for @xmath16355 ) is quite striking . @xcite suggest that the discrepancy can be accounted for by an ejected blob moving at an angle to the line of sight . sn 2006bt at the present seems to be the strongest evidence for carbon ejecta distributed nonspherically in sne ia . . , scaledwidth=49.0% ] inspired by the characteristics of multi - dimensional explosion simulations with strong asymmetries , observers have searched for spectroscopic and spectropolarimetric indications of deviation from spherical composition symmetry in sne ia . spectropolarimetry offers the best means for probing the geometry of sn ejecta , but opportunities for follow - up are relatively rare and the available sample is small . @xcite describe spectropolarimetric observations of 2 sne iain our carbon - positive sample , sne 2005cf and 2005el ( a single epoch each ) . in neither case is there a particularly strong case for deviation from basically spherical composition symmetry . a dominant axis in the stokes q / u diagram is at least identifiable for sn 2005el , suggestive of some combination of global and smaller scale asymmetries . while spectropolarimetry may be more definitive for constraining the geometry of the unburned material , it is interesting to consider the constraints we can set with our pure flux measurements . the comparison of synthetic spectra with and without shown in figure [ fig : closeup ] suggests that the spatial distribution of unburned carbon in the ejecta may be quite consistent with spherical symmetry . figure [ fig : blending ] demonstrates our reasoning simply , using toy model synthetic line profiles . the classic sn spectroscopic feature formed by a single line is plotted at the top of figure [ fig : blending ] : blueshifted absorption produced by material in front of the photosphere and expanding toward the observer , with an emission hump centered at the rest wavelength of the line arising from material adjacent to the photosphere in projection . for discussion purposes , we label this line `` line a '' and assign it a rest wavelength of 6355 . consider a second line , `` line b '' with rest wavelength 6580 , a doppler shift of @xmath38 km s@xmath3 , a shift quite similar to the cases seen in the data . if the ejecta bearing the parent ions of line b are distributed in a `` blob '' that only partially covers the photosphere and/or line a emissivity in projection , it is quite easy to construct a line b absorption notch blended with the line a emission component . an example of the resulting blend appears in the middle of figure [ fig : blending ] , overlaid on the line a profile . now , suppose the line b ions are distributed spherically symmetrically . in this case , a quite different blended profile results . line b notches can be produced , but not without additional disruption of the line a profile . flux redward of the notch is suppressed and there is even a mild red emission feature relative to the single unblended line a case . such a blend is depicted at the bottom of figure [ fig : blending ] , again overlaid on the single line a profile for comparison . because line b opacity does not encircle the photosphere in the blob case , its effect is restricted to wavelengths corresponding roughly to where the blob is optically thick in blueshift parallel to the line of sight . however , when line b ions are distributed under spherical symmetry , the photosphere is cocooned by them , and they scatter away continuum and line a photons alike over a larger range of projected velocity . this results in the notch , the flux suppression to the red of it , and the small emission further on . essentially , the driving factor is the projected area over which line b photons screen continuum but also line a emissivity , at a given wavelength . the fits in figure [ fig : closeup ] appear more consistent with the spherically symmetric case demonstrated in figure [ fig : blending ] . our toy models are quite simple , but illustrate the leading - order screening effect quite clearly . at the very least , our results highlight the critical need for detailed spectroscopy and spectropolarimetry of sne ia at early phases to make the most definitive conclusions about the geometry of the carbon ejecta . the possible existence of a family of sne ia with similar properties to those demonstrated by sn 2005el , sn 2005ki , snf20080514 - 002 , and perhaps sn 2005di , is particularly intriguing . the combination of a narrow light curve , large uv excess or at least blue colors , and the presence of photospheric velocity carbon suggests a very particular conceptual picture . a faster than average decline rate indicates lower than average luminosity for a sn ia , in turn suggesting a lower than average mass of freshly synthesized radioactive nickel . the blue colors suggest that the amount of iron - peak elements synthesized or mixed into the outer layers is low ( suppressing line - blanketing at shorter wavelengths ) , and these elements are limited to lower velocities . also , a relatively high carbon abundance in the material above the photosphere indicates less efficient processing of the white dwarf progenitor material . the above picture is somewhat consistent with a subset or region of delayed - detonation model space . estimates of the mass fraction of photospheric carbon in sne ia are generally low ( of order 1% , e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and are also generally consistent with most such models . of particular interest is the recent grid of multi - dimensional models presented by @xcite . in these models , a strong ignition ( arising from numerous initiating sparks ) raises the amount of burning that takes place during the initial sub - sonic deflagration phase . this allows the outer , unburned layers to pre - expand more relative to the weaker ignition case . when the later detonation wave passes through the pre - expanded material , nuclear burning is less efficient and the abundance of iron - peak elements is lower relative to the weak ignition case . this results in relatively lower radioactive nickel mass which is more centrally peaked in the ejecta . however , the total amount of unburned carbon left in these models is below 1% by mass , and the s / n of the synthesized monte carlo spectra is too low to attempt to detect any carbon signatures in them . this result is actually at variance to earlier ( one - dimensional ) delayed - detonation models ( e.g. , * ? ? ? * ) , where some amount of unburned carbon can be left behind . simply put , the multi - dimensional model grids may not yet be extensive enough . in most of the carbon - positive sne ia discussed here , there seems to be intriguing evidence that the carbon is distributed roughly spherically symmetrically , and at velocities overlapping silicon . there may still be room for deviation from spherical symmetry in the distribution of unburned material that we can not rule out here . for example , large scale asymmetries , such as proposed by @xcite to explain spectroscopic diversity , may allow for almost hemispherical covering of the photosphere by carbon . in that particular picture , an offset ignition in the white dwarf progenitor yields an asymmetric distribution of burning products . this scenario is proposed to explain the origin of `` high - velocity gradient '' ( hvg ) and `` low - velocity gradient '' ( lvg ) subclasses of sne ia , segregated by the rate of decrease in measured @xmath16355 blue - shift with time @xcite . viewed from the side closest to the ignition , one observes an lvg sn ia ; and viewed from the other an hvg sn ia . detonation products are distributed in the outer layers in all directions , but it is not yet clear how complete the burning is . interestingly , @xcite remarked that of 3 lvg sne ia they studied , all had marginally detected @xmath16580 , and blue colors . their analysis suggests that lvg sne ia undergo less intense burning than their counterparts . a systematic examination of @xmath16355 blueshift evolution of snfactory sne ia is in progress , but our preliminary results indicate that all 5 objects identified here are also lvg members . whether or not a @xcite scenario is capable of explaining all the observables , and can yield carbon amounts and distributions consistent with observations , remains to be seen through detailed multi - dimensional explosion modeling and radiative transfer . we have presented new detections of carbon signatures in the spectra of several sne ia . for the most part , the carbon - positive sne are normal spectroscopically , with the only significant deviation being the presence of hv and in one of them . four of the 5 objects discussed may form a robust subfamily of sne ia with a relatively fast decline rate and intrinsically blue colors . the carbon is detected at velocities consistent with the ejection velocity at the photosphere , and in the same velocity range as other freshly synthesized material . the distribution of carbon is apparently consistent with spherical or near - spherical symmetry . the behavior of most of the carbon - positive sne ia discussed here suggests strong radial abundance stratification , consistent with some delayed detonation models . only one of the objects uncovered in the course of our analysis has ever been previously identified in the literature as carbon - positive , either at the time of initial spectroscopic classification or in subsequent analysis . our intent in pointing this out is not to criticize ; indeed we have highlighted cases that earlier escaped our notice as well . rather , we suggest that the @xmath16580 notch signature is not well - known , easy to overlook , or easy to mistake for a reduction artifact . in fact , this is part of why we required 2 spectra for labeling detections as definitive . the examples provided here should help to sensitize observers to carbon signatures in early - phase sn ia spectra . understanding the frequency , distribution , and amount of unburned carbon in sne ia is indeed a useful discriminator among competing explosion models , so constraints on theory depend on reliable and efficient early triggering of spectroscopic ( and spectropolarimetric ) follow - up of sne ia for detailed study . as current sn ia cosmology efforts approach their limits in terms of calibration and empirical standardization , enhancing our understanding of sn ia physics becomes all the more important . wide - field and high - throughput robotic surveys have the power to grow the early - time data set , when coupled efficiently with ample spectroscopic follow - up resources . this opens new opportunities for fully exploiting the analysis of the outer layers of sn ia ejecta to constrain explosion models . we thank the anonymous referee for carefully reading our manuscript and for providing a thorough report that improved our paper s clarity . we are grateful for assistance from the technical / scientific staffs of the palomar observatory , the high performance wireless radio network ( hpwren ) , the national energy research scientific computing center ( nersc ) , and the university of hawaii 2.2 m telescope . we also wish to recognize and acknowledge the significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . this work was supported by the director , office of science , office of high energy physics , of the u.s . department of energy under contract no . de - ac02 - 05ch11231 ; the u.s . department of energy scientific discovery through advanced computing ( scidac ) program under contract no . de - fg02 - 06er06 - 04 ; by a grant from the gordon & betty moore foundation ; in france by support from cnrs / in2p3 , cnrs / insu , and pnc ; and in germany by the dfg through trr33 `` the dark universe . '' funding was also provided by a henri chretien international research grant administrated by the american astronomical society ; the france - berkeley fund ; and by an exploradoc grant by the region rhone alpes . much of this research was conducted with resources and support from nersc , which is supported by the director , office of science , office of advanced scientific computing research , of the u.s . department of energy under contract no . de - ac02 - 05ch11231 . hpwren is funded by national science foundation grant number ani-0087344 , and the university of california , san diego . we appreciated constructive discussions with d. kasen , j. parrent , d. poznanski and r. foley . we thank v. stanishev for sharing a new reduction of a spectrum of sn 2008q with us in advance of publication . we thank g. h. marion , r. p. kirshner , and p. challis ( harvard - smithsonian center for astrophysics ) for sharing unpublished spectra of sn 2008hv . the cfa supernova program is supported by nsf grant ast 09 - 07903 . a. m. acknowledges support from the nasa astrophysics data and analysis program grant nnx06ah85 g . r. c. t. thanks fellow participants at the aspen center for physics summer 2010 workshop `` taking supernova cosmology into the next decade , '' for their input . * the @xmath16580 absorption notch is clearly visible in the @xmath49 d spectrum , located between roughly @xmath58 and @xmath59 km s@xmath3 with a minimum around @xmath38 km s@xmath3 in apparent blueshift . almost 3 days later , the notch is barely discernible , and it is clearly absent by @xmath60 d. in the last spectrum , it is unlikely that the noise is obscuring a notch : the change in the shape of the spectrum between the last 2 spectra is overt . examining the @xmath17234 region , we detect an inflection in the shape of all 3 spectra , but its position is inconsistent across phase . this inconsistency may be a consequence of the low s / n of the observations . in the first spectrum , the flux depression is coincident with the @xmath16580 notch in blueshift , but this is not the case in the later spectra . however , in all 3 spectra , a bump in the flux centered at a @xmath17234 blueshift of @xmath61 km s@xmath3 is clearly visible . figure [ fig : carbon]a also clearly shows the narrow rest - frame d absorption , in the shaded rectangle . * synapps fits . * contributes to the spectrum on @xmath49 d and also in the next spectrum on @xmath32 d , but on @xmath60 d ` synapps ` finds no need for it . the ` synapps ` velocity at the photosphere ( @xmath62 ) fits are @xmath63 , @xmath64 , and @xmath65 km s@xmath3 respectively . hv opacity is invoked ( at @xmath66 km s@xmath3 ) though the signature is rather small it is more important for fitting the ir triplet than the uv h&k feature . ultimately , the case for hv is uncompelling , which is not surprising since the blue side of the @xmath16355 absorption only extends to @xmath67 km s@xmath3 , in contrast to instances where hv has been previously discussed ( e.g. , sn 2005cf , * ? ? ? * ) and found to extend upwards of @xmath68 km s@xmath3 . * . * a @xmath16580 absorption notch is readily seen in the @xmath35 d spectrum . as in sn 2005di , the notch is located in the apparent blueshift interval @xmath39 to @xmath59 km s@xmath3 . there is a clear depression in the flux around 6980 , possibly attributable to @xmath17234 absorption , though the blueshift is not completely consistent with that for the @xmath16580 notch . both features disappear from the spectrum by the time of maximum light . as in sn 2005di , the red side of the emission feature steepens as the sn ages and the @xmath16580 dissipates . there is also an emission bump to the red of the @xmath17234 notch , centered at a rest - frame blueshift of @xmath61 km s@xmath3 . like the 2 notches , this bump disappears by maximum light . * synapps fits . * ` synapps ` requires only in the @xmath35 d spectrum , and not at maximum light . the value of @xmath62 changes from @xmath57 to @xmath38 km s@xmath3 from the first spectrum to the second . hv and opacity components were enabled as was done with sn 2005di . the results were similar , with some hv contributing to the ir triplet and the bluer feature of the h&k absorption blend , and hv making only a minor contribution to the blue wing of the @xmath16355 absorption in the first spectrum . * . * in the @xmath69 d spectrum , the @xmath16580 notch appears slightly more blueshifted than in the first spectra of sne 2005di and 2005el , extending just beyond @xmath59 km s@xmath3 . the same is true for @xmath16355 absorptions . as before , the carbon signatures dissipate as the sn brightens : weaker by @xmath70 d and absent by @xmath71 d. also , in the first spectrum , an emission plateau just redward of the notch is visible , asymmetric with respect to the rest wavelength of @xmath16580 , and again it weakens in parallel with the notch . a clear association for @xmath17234 in this spectrum is quite problematic : the fading bump at 7100 is again visible , but there is no clear absorption notch . * synapps fits . * opacity is used in the @xmath69 and @xmath70 d spectra , but not in the @xmath71 d spectrum . ` synapps ` photospheric velocities for sn 2005ki are a bit higher than in the other sne considered here , with @xmath62 fit at @xmath72 , @xmath63 , and @xmath57 km s@xmath3 . one major difference is that a distinct hv opacity component is _ not _ required to reproduce the features , especially at @xmath69 d : a single opacity profile extending to high velocity is sufficient . * . * the @xmath16580 notch ( apparent blueshift interval again @xmath39 to @xmath59 km s@xmath3 ) is most apparent at @xmath73 and @xmath36 d , still discernable at @xmath37 d , and probably absent at @xmath74 d. the notch is accompanied by a modest and fading emission plateau to the red which is almost symmetric about the line rest wavelength . the @xmath17234 blueshift region is again characterized by an inflection and bump centered blueward of 7234 , and again the strength of this bump decreases with time . * synapps fits . * opacity is needed in the @xmath73 , @xmath36 , and @xmath37 d fits , but not on day @xmath74 . the values of @xmath62 fit are @xmath75 , @xmath76 , @xmath77 and @xmath77 km s@xmath3 , respectively . only a modest amount of hv opacity is used in these fits , and hv proves to be unnecessary . * . * the @xmath16580 absorption notch shape is significantly different from that seen in the other carbon - positive sne ia studied here . the notch is more narrow , extending in apparent blueshift from @xmath39 to @xmath40 km s@xmath3 in the @xmath31 d spectrum . the notch as faded away by @xmath78 d , a phase at which it is still present in the other carbon - positive sne studied here . * synapps fits . * ` synapps ` uses for the @xmath31 d spectrum but not for the spectrum obtained a week later . ` synapps ` determines @xmath79 and then @xmath38 km s@xmath3 , respectively . the @xmath31 d spectrum requires both hv and , though the hv feature becomes optically thin by @xmath80 d. the hv component is also clearly quite important for the formation of the h&k absorption in both spectra . the strength of the hv features in sn 2005cf distinguishes it from the other sne ia considered here .

the relative dispersion of pairs of inertial particles in incompressible , homogeneous , and isotropic turbulence is studied by means of direct numerical simulations at two values of the taylor - scale reynolds number and , corresponding to resolutions of and grid points , respectively . the evolution of both heavy and light particle pairs is analysed at varying the particle stokes number and the fluid - to - particle density ratio . for particles much heavier than the fluid , the range of available stokes numbers is $ ] , while for light particles the stokes numbers span the range $ ] and the density ratio is varied up to the limit of vanishing particle density . + for heavy particles , it is found that turbulent dispersion is schematically governed by two temporal regimes . the first is dominated by the presence , at large stokes numbers , of small - scale caustics in the particle velocity statistics , and it lasts until heavy particle velocities have relaxed towards the underlying flow velocities . at such large scales , a second regime starts where heavy particles separate as tracers particles would do . as a consequence , at increasing inertia , a larger transient stage is observed , and the richardson diffusion of simple tracers is recovered only at large times and large scales . these features also arise from a statistical closure of the equation of motion for heavy particle separation that is proposed , and which is supported by the numerical results . + in the case of light particles with high density ratios , strong small - scale clustering leads to a considerable fraction of pairs that do not separate at all , although the mean separation increases with time . this effect strongly alters the shape of the probability density function of light particle separations .

0904.2314

suspensions of dust , droplets , bubbles , and other finite - size particles advected by incompressible turbulent flows are commonly encountered in many natural phenomena ( see , e.g. , @xcite ) . understanding their statistical properties is thus of primary importance . from a theoretical point of view , the problem is more complicated than in the case of fluid tracers , i.e. point - like particles with the same density as the carrier fluid . indeed , when the suspended particles have a finite size and a density ratio different from that of the fluid , they have inertia and do not follow exactly the flow . as a consequence , correlations between particle positions and structures of the underlying flow appear . it is for instance well known that heavy particles are expelled from vortical structures , while light particles tend to concentrate in their cores . this results in the formation of strong inhomogeneities in the particle spatial distribution , an effect often refered to as _ preferential concentration _ ( see @xcite ) . this phenomenon has gathered much attention , as it is revealed by the amount of recently published theoretical work ( @xcite ) , and numerical studies ( @xcite ) . progresses in the statistical characterization of particle aggregates have been achieved by studying particles evolving in stochastic flows by @xcite and in two - dimensional turbulent flows by @xcite . also , single trajectory statistics have been addressed both numerically and experimentally for small heavy particles ( see , e.g. , @xcite ) , and for large particles ( @xcite ) . the reader is refered to @xcite for a review . in this paper we are concerned with particle pair dispersion , that is with the statistics , as a function of time , of the separation distance @xmath6 between two inertial particles , labelled by the subscripts 1 and 2 ( see @xcite for recent studies on that problem ) . in homogeneous turbulence , it is sufficient to consider the statistics of the instantaneous separation of the positions of the two particles . these are organised in different families according to the values of their stokes number @xmath7 , and of their density mismatch with the fluid , @xmath8 . for our purposes , the motion of particle pairs , with given @xmath9 values and with initial separations inside a given spherical shell , @xmath10 $ ] is followed until particle separation reaches the large scale of the flow . with respect to the case of simple tracers , the time evolution of the inertial particle pair separation @xmath11 becomes a function not only of the initial distance @xmath12 , and of the reynolds number of the flow , but also of the inertia parameters @xmath9 . a key question that naturally arises is how to choose the initial spatial and velocity distributions of inertial pairs . indeed , it is known that heavy ( resp . light ) particles tend to concentrate preferentially in hyperbolic ( resp . elliptic ) regions of the advecting flow , with spatial correlation effects that may extend up to the inertial range of scales , as shown in @xcite . moreover , when inertia is high enough , the particle pair velocity difference , @xmath13 , may not go smoothly to zero when the particle separations decreases , a phenomenon connected to the formation of _ caustics _ , see @xcite . in our numerical simulations , particles of different inertia are injected into the flow and let evolve until they reach a stationary statistics for both spatial and velocity distributions . only after this transient time , pairs of particles with fixed intial separation are selected and then followed in the spatial domain to study relative dispersion . by reason of the previous considerations , the main issue is to understand the role played by the spatial inhomogeneities of the inertial particle concentration field and by the presence of caustics on the pair separations , at changing the degree of inertia . we remark that these two effects can be treated as independent only in the limit of very small and very large inertia . in the former case , particles tend to behave like tracers and move with the underlying fluid velocity : preferential concentration may affect only their separation . in the opposite limit , particles distribute almost homogeneously in the flow : however , due to their ballistic motion , they can reach nearby positions with very different velocities ( @xcite ) . in any other case of intermediate inertia , both these effects are present and may play a role in the statistics of inertial pair separation . it is worth anticipating the two main results of this study : + ( _ i _ ) the separation between heavy particles can be described in terms of two time regimes : a first regime is dominated by inertia effects , and considerable deviations from the tracers case arise in the inertial relative dispersion ; in the second one , the tracers behaviour is recovered since inertia is weak and appears only in subdominant corrections that vanish as @xmath14 . the crossover between these two regimes defines a new characteristic spatial and temporal scale , connected to both the _ size of caustics _ and the stokes number , which influences the particle separation for not too long time - lags and not too large scale . + ( _ ii _ ) the strong clustering properties that are typical of light particles may lead to the fact that many pairs do not separate at all : their statistical weight is clear in the separation probability density function ( pdf ) , which develops a well defined power - law left tail . it would clearly be also interesting to investigate the dependence upon the reynolds number of the inertial particle pair separation . small - scale clustering seems to be poorly dependent on the degree of turbulence of the carrier flow ( @xcite ) , while much less is known about the reynolds number dependence of the caustics statistics . our numerical data do not allow to explore this question in detail , so that we will restrict ourselves to show data associated to the two reynolds numbers in all cases when differences are not significative . in the case of fluid tracers , the standard observables are the time evolutions of the mean square separation and of the separation probability density function , for which well established predictions exist since the pioneering work of @xcite . we contrast these observables obtained for tracers with the results for heavy and light inertial particles . the paper is organised as follows . in [ sec:0 ] , we briefly recall the basic equations of motion and describe the numerical simulations . in [ sec:1 ] , we analyse the stationary distribution of heavy particle velocity differences , conditioned on the particle initial separation , highlighting both the presence of small - scale caustics and the effects of particle inertia at those scales corresponding to the inertial range of turbulence . in [ sec:2 ] we study the behaviour of the mean separation distance of heavy pairs , at changing the stokes number @xmath7 ; we also analyse the influence of the caustics in the initial statistics on the subsequent pair separation evolution . mean - field _ model , which is able to capture the main numerical findings , is proposed in the same section . the time evolution of the separation probability density functions is discussed in [ sec : pdf ] and we present the data for light particles in [ sec : light ] . in [ sec : conc ] we summarise the main findings . we present results from direct numerical simulations of turbulent flows seeded with inertial particles . the flow phase is described by the navier - stokes equations for the velocity field @xmath15 @xmath16 the statistically homogeneous and isotropic external forcing @xmath17 injects energy in the first low wave number shells , by keeping constant their spectral content ( see @xcite ) . cccccccccc + & @xmath18 & @xmath19 & @xmath20 & @xmath21 & @xmath22 & @xmath23 & @xmath24 & @xmath25 & @xmath26 + run i & 512 & 185 & 0.01 & 0.012 & 0.9 & 0.002 & 0.047 & 0.004 & 0.0004 + run ii & 2048 & 400 & 0.0026 & 0.003 & 0.88 & 0.00035 & 0.02 & 0.00115 & 0.000115 + the kinematic viscosity @xmath23 is chosen such that the kolmogorov length scale @xmath27 , where @xmath21 is the grid spacing : this choice ensures a good resolution of the small - scale velocity dynamics . the numerical domain is cubic and @xmath28-periodic in the three directions of space . we use a fully dealiased pseudospectral algorithm with 2@xmath29 order adam - bashforth time - stepping ( for details see @xcite ) . we performed two series of dns : run i with numerical resolution of @xmath2 grid points , and the reynolds number at the taylor scale @xmath30 ; run ii with @xmath3 resolution and @xmath31 . details of the runs can be found in table [ table ] . the particle phase is constituted by millions of heavy and light particles the latter only for run i with different intrinsic characteristics . particles are assumed to be with size much smaller than the kolmogorov scale of the flow , @xmath20 , and with a negligible reynolds number relative to the particle size . in this limit , the equations ruling their dynamics take the particularly simple form : @xmath32 + \beta\ , d_t { \bm u}(\bm x , t ) \ ; , \label{eq:1}\ ] ] where the dots denote time derivatives . the particle position and velocity are @xmath33 , respectively ; @xmath34 is the eulerian fluid velocity evaluated at the particle position , and @xmath35 is the so - called added mass term , which measures the fluid acceleration along particle trajectory . the adimensional constant @xmath36 accounts for the added mass effect through the density contrast between particles @xmath37 and fluid @xmath38 . the particle response time , appearing in the stokes drag , is @xmath39 , where @xmath40 is the particle radius . particle inertia is quantified by the _ stokes number _ that is defined as @xmath41 , where @xmath42 is the flow kolmogorov timescale and @xmath22 the average rate of energy injection . equation ( [ eq:1 ] ) has been derived in @xcite under the assumption of very dilute suspensions , where particle - particle interactions ( collisions ) and hydrodynamic coupling to the flow can be neglected . for run i , we show results for the following set of @xmath9 families : ( _ i _ ) very heavy particles [ @xmath43 : @xmath44 ; ( _ ii _ ) light particles [ @xmath45 : @xmath46 . for each family the typical number of particle pairs that are followed is around @xmath47 . for run ii , we show results only for heavy particles but with a larger range of variation in the stokes number : @xmath48 . typical number of particle pairs for each family is @xmath49 . once injected particles have relaxed to their steady - state statistics , pairs have been selected with the following initial separations : @xmath50 and @xmath51 \eta$ ] for both run i and run ii , and @xmath52 \eta$ ] for run ii only . beside the time evolution of particle pairs , we also have instantaneous snapshots of the two phases ( fluid and dispersed ) , with a much higher particle statistics : around @xmath53 per family for run i , and @xmath54 per family for run ii . these are used to measure the stationary i.e . not along the trajectories distribution of particle velocity increments discussed in next section . turbulent pair dispersion for tracers is classically based on the application of similarity theory for eulerian velocity statistics : depending on the value of space and time scales , velocity increment statistics differently affect the way tracers separate . this results in different regimes for relative dispersion , see e.g. @xcite . in the case of inertial particles , the same reasoning holds , so that to analyse the way inertial pairs separate in time , the stationary statistics of particles velocity differences has to be investigated first . a stationary distribution for the typical velocity differences between two inertial particles is obtained by imposing periodic boundary conditions inside the physical volume and then measuring velocities on such a thermalised configuration . we are interested in the scaling behaviour of velocity increments at varying the degree of inertia and the distance between the particles ( in the dissipative or inertial range of the turbulent fluid flow ) . to fix the notation , we denote by @xmath55 the typical large - scale velocity of the fluid tracers and by @xmath56 the integral scale of the flow . moreover we define @xmath57 as the velocity difference at scale @xmath58 , conditioned on the presence of a pair of particles with stokes number @xmath7 , separated with a distance @xmath59 . since we are here interested in the case of heavy particles only , the stokes number is sufficient to identify a given particle family . for convenience , we introduce a specific notation for the tracer stationary velocity statistics : @xmath60 , which is exactly equal to the eulerian velocity increment at scale @xmath58 . recently , @xcite have shown that to describe inertial pair dispersion in synthetic flows it is useful to introduce the local or _ scale - dependent _ stokes number , using the ratio between the particle response time and the typical eddy turnover time @xmath61 of the underlying fluid at a given scale : @xmath62 . for real turbulent flow where different scaling ranges are present , we can equivalently define a scale dependent stokes number @xmath63 that recovers the usual definition of the stokes number @xmath64 when @xmath65 and behaves as @xmath66 when @xmath67 . the typical behaviour of @xmath63 is sketched in fig . [ fig : stokes ] , for two different values of the stokes number @xmath68 and using a batchelor - like parametrisation of the fluid velocity ( see @xcite ) : @xmath69 for stokes numbers , @xmath7 , order unity or larger , there always exists a typical scale where the local stokes number , @xmath63 , becomes order unity , @xmath70 such a scale , which is well in the inertial range if the stokes number @xmath7 is sufficiently large , can be considered a rough estimate of the upper bound for the region of scales where inertia plays an important role in the particle dynamics . we expect that two main features might be important in characterising the inertial particle stationary velocity statistics @xmath71 , with respect to that of tracers @xmath72 . the first concerns the small - scale behaviour of the particle velocity statistics . at small scales @xmath73 and for large - enough stokes numbers , the presence of caustics makes the particle velocity increments not differentiable . this feature can be accounted for by saying that @xmath74 where the @xmath75 is a constant prefactor and the function @xmath76 gives the typical scaling of caustic - like velocity increments . indeed we do expect that at changing the inertia of the particles , the statistical weight of caustics might monotonically vary as follows : at small @xmath7 , @xmath77 , i.e. the value for smooth , differentiable eulerian statistics of tracers ; at large values @xmath78 , it should approach the discontinuous limit @xmath79 , valid for particles that do not feel underlying fluid fluctuations at all . the right panel of fig . [ fig : stokes ] shows the typical shape of the function @xmath76 that is expected to be valid for turbulent flows . the second important feature concerns the particle velocity statistics at scales larger than the scale @xmath80 previously defined , but smaller than the integral scale of the fluid flow . for any fixed stokes number and for a large - enough reynolds number , we expect that inertia becomes weaker and weaker , by going to larger and larger scales @xmath81 . in such a case , particle velocity increments are expected to approach the underlying fluid velocity increments : @xmath82 where for simplicity we have neglected possible intermittent correction to the kolmogorov 1941 ( k41 ) scaling of the fluid velocity ( see @xcite for details ) . clearly , the reynolds number has to be sufficiently large to provide a well - developed scaling region @xmath83 , before approaching the large scale @xmath56 . we emphasise that in ( [ def : vtou ] ) , an adimensional normalisation factor @xmath84 has been introduced : it takes into account possible filtering effects induced by inertia at large scales . the normalisation is such that @xmath85 , while for any stokes larger than zero @xmath86 . + in fig . [ fig:1c ] we test the validity of the previous picture by analyzing the typical velocity fluctuation , @xmath87 , at changing stokes number and for data of run ii at reynolds number @xmath88 . at small scales one detect the presence of caustics in the velocity statistics , with a non smooth scaling behaviour below the kolmogorov scale @xmath20 . at scales within the inertial range and when the stokes number is sufficiently large , the effect of caustics affects also particle velocity statistics , up to a characteristic scale which becomes larger and larger by increasing particle inertia . beyond this scale , particle velocity increments tend to approach the scaling behaviour of the fluid tracers , but their amplitude is depleted of a factor @xmath89 , which increases with the stokes number , as shown in the inset of the right - hand panel of fig . [ fig:1c ] . a similar behaviour is expected for higher order fluctuations , if we neglect the role of intermittency . it is interesting to consider the scaling behaviour of particle velocity in terms of the underlying velocity statistics , not only at very small or very large separations , but for any value of the scale @xmath58 . this is not straigthforward , since we have to account not only of the fluid eulerian statistics at the dissipative and inertial range of scales , but also the modifications due to the inertia . this is responsible , as we have seen , for the appeareance of a new relevant scale , and for filtering effects in the velocity amplitude . to fully characterise particle velocity increments , we notice that the stokes scale , @xmath90 defines a typical _ stokes - velocity _ : this is the fluid velocity increment at the stokes scale , @xmath91 ( see left panel of fig . [ fig : stokes ] ) . previous reasonings can be summarised in the following interpolation formula for the heavy particle velocity increment : @xmath92^ { \left [ 1-\gamma(st(r))\right]/2}\,.\ ] ] the above expression is a batchelor - like parametrisation but in the velocity space , with a transient velocity given by the stokes velocity , @xmath93 . once known the large scale normalization function @xmath84 , the caustic exponent @xmath94 ( introduced in @xcite ) and the reference fluid velocity increment @xmath72 , then the formula has one free parameter only . it is the prefactor @xmath95 appearing in front of the stokes velocity @xmath93 , whose value depends again on the inertia of the particles . in fig . [ fig:3c ] , we show the result of the fit in terms of the expression ( [ eq : fit ] ) , where the caustics scaling exponent has been chosen as @xmath96 $ ] : this functional form provides a good fit to the numerical results . details of the small - scale caustic statistics will be reported elsewhere . the qualitative trend is very well captured by the interpolation function proposed . notice that in ( [ eq : fit ] ) , the argument of @xmath76 is not the simple stokes number at the kolmogorov scale , but the scale - dependent one @xmath63 : @xmath97 . this further ingredient is needed to take into account the fact that in presence of a rough underlying fluid velocity , as it happens in the inertial range of scales , no simple power law behaviour is expected for the scaling of particle velocity statistics . this was previously remarked in @xcite , in the study of heavy particle turbulent dispersion in random flows . equation ( [ eq : fit ] ) clearly matches the two limiting behaviours for very small and very large separations . in the former case , inertia dominates the small - scale velocity statistics with respect to the underlying smooth fluid velocity , and caustics lead to a pure power - law behaviour , @xmath98 where the local stokes number has attained its dissipative limit @xmath99 . in the latter case , at very large scales @xmath81 inertia is subleading , and the typical velocity difference between particles is close to the fluid velocity increment , @xmath100 at intermediate scales , for large stokes , @xmath101 , inertia brings a non - trivial dependency via the scale - dependent stokes number , @xmath63 , and we expect a pseudo power - law scaling : @xmath102 summarising , we propose that at changing the stokes and reynolds numbers , different regimes governing the particle velocity statistics can be distinguished . the relevance of such regimes of the particle velocity statistics for the associate relative dispersion dynamics can be easily explained with the help of the sketch reported in fig . [ fig : abc ] . in the parameter space of inertia and scale separation @xmath103 , we can distinguish three regions depending whether inertia is strong or weak , and whether particle velocity difference is large or not with respect of the fluid velocity difference at comparable scale . in agreement with what commented before , we pose that the curve @xmath104 distinguishes the region of weak ( @xmath105 ) and strong inertia ( @xmath106 ) . regime ( a ) is such that inertia is important since the scale @xmath107 , and moreover the typical particle velocity increments are larger than the fluid increments . in the region ( b ) , inertia is still important but particle velocity increments are depleted with respect to the fluid increments . this typically happens for large stokes numbers , and in our dns is visible only for very large separations @xmath11 of the highest stokes @xmath108 . finally , regime ( c ) is characterised by a weak inertia , which appears only in the filtering factor for the velocity large - scale amplitude and possibly in sub - leading corrections to the tracers relative dispersion . even for the largest value of the reynolds and stokes numbers achieved in our dns , it is very difficult to disentangle quantitatively the above mentioned regimes , because of the closeness of the three relevant scales , @xmath20 , @xmath80 , and @xmath56 . still , the quality of the fit shown in fig . [ fig:3c ] using the global functional dependence given by eqn . ( [ eq : fit ] ) makes us confident that the main physical features are correctly captured . before closing this section , we note that there is no reason to assume that the functional form entering in the pseudo - power law scaling in the inertial range , @xmath109 , in ( [ caseb ] ) is equal the one characterizing the scaling in the viscous range , @xmath76 , in eqn . ( [ casea ] ) . hint for this observation come from results obtained in @xcite for random flows , where a very high statistical accuracy can be achieved : there , depending if the underlying fluid velocity is spatially smooth or rough , a slightly different functional form has been found . the previous analysis gives us a clear quantitative picture of the scale and velocity ranges where caustics play a role in the particle dynamics . for example , for moderate stokes numbers , we have important departure from the tracers statistics only for very small scales , i.e. caustics gives a singular contributions to the particle velocity increments inside the viscous range ; then , at larger scales , the particle velocity scaling become indistinguishable from the tracer velocities . clearly , for such stokes , no important corrections for particle separation evolution is expected with respect to the usual richardson dispersion observed for tracers . this is because particle pairs tends to separate , and very soon all pairs will attain separations where their velocities are very close to the underlying fluid . on the other hand , for very heavy particles , those with stokes time falling inside the inertial range of fluid velocity statistics , the contribution from the caustics will be felt also at relatively large scales , up to @xmath110 . pair separations attain such scales when the initially large relative velocity difference has relaxed and become smaller than the corresponding fluid one crossing from region ( a ) to ( b ) . notice that at @xmath111 , we have that @xmath112 , i.e. there is a non - trivial effect from inertia . moreover , for large stokes , at scales @xmath113 , particle velocity increments are smaller than the fluid counterparts , indicating an important depletion induced by the stokes drag on the particle evolution . + it is clear from the above discussion that new physics should appear for the value of inertia and scales separation of region ( b ) . this regime that we can not access with the present data is the one where a new law of pair separation should appear as recently suggested by @xcite . a discussion of the dispersion regimes of inertial particle pairs follows in the next section in terms of the time behaviour of the mean square separation distance . in this section we analyse the effects of inertia on the mean square separation of heavy particle pairs with a given initial separation distance , @xmath12 at time @xmath114 , as a function of the stokes number : @xmath115 where in the left - hand side the average is performed over all pairs of particles such that @xmath116 . the study of the relative dispersion of small , neutrally buoyant tracer particles has recently been the subject of renewed interest . this has been motivated by the fact that very accurate highly resolved in time and space data have become available , experimentally ( @xcite ) and numerically ( @xcite ) . these studies have confirmed what was known since the works of @xcite and @xcite , i.e. the existence of different dispersive regimes for tracer pairs in turbulent flows , depending on the value of their initial distance and on the time scale considered . when released in a statistically homogeneous and isotropic , turbulent flow with an initial separation @xmath12 in the inertial range for fluid velocity , i.e. @xmath117 , tracer pairs initially separate according so the so - called batchelor regime , @xmath118 where @xmath119 is supposed to be a universal constant , and @xmath22 is the average kinetic energy dissipation of the flow . this _ ballistic _ regime appears because initially tracers separate _ as if _ the underlying velocity field were frozen , and it lasts for a time scale that is a function of the initial separation itself , @xmath120 ( see @xcite ) . after such a transient initial time , the relative separation dynamics forgets the initial conditions and tracers separate explosively with a power law behaviour given by the richardson law : @xmath121 where @xmath122 is known as the richardson constant . as set out in monin & yaglom @xcite , the tracer separation pdf that will be discussed later has a similar scaling behaviour in these ranges . a remarkable fact of the richardson dispersion ( [ eq : r2rich ] ) is the disappearance of the dependence on the initial separation @xmath12 , an effect also dubbed _ intrinsic stochasticity _ ( @xcite ) , which is just the signature of the non - lipschitz nature of the velocity field driving the separation between tracers , when their mutual distance is in the inertial range of fluid velocity statistics . the experimental and numerical validation of the previous prediction ( [ eq : r2rich ] ) has proved to be particularly difficult , the main reason being the strong contamination from viscous and large scale effects in the tracers dynamics . to overcome these problems , a series of techniques have been developed , including the study of _ doubling time statistics _ ; i.e. the probability distribution function of the time needed for a pair to double its separation ( @xcite ) . thanks to these techniques , a fairly good agreement on the value of the richardson constant has been achieved . here , we want to study how the tracer behaviour is modified by the presence of small - scale caustics in particular and by inertia effects in general , for the case of heavy particle pairs . standard direct measurements of the moments of separation as a function of time will be considered , while application of doubling time statistics is left for future studies . in fig . [ fig : m2 ] , we show the behaviour for the mean square separation at varying the stokes number , and for two values of the initial separation . we start with data at the lowest resolution , i.e. run i at @xmath123 , and for moderate stokes numbers , @xmath124 . initial distances are chosen equal to @xmath50 ( left panel ) and @xmath51\,\eta$ ] ( right panel ) . if the initial distance is small enough ( left panel ) , the presence of caustics in the particle velocity field at initial time gives a very remarkable departure from the tracer behaviour . at increasing the stokes number , such departure is more and more evident , and it lasts for a time lag which becomes longer and longer . for the highest value of the stokes number shown in the left panel of fig . [ fig : m2 ] ( @xmath125 ) , a sensible difference from the tracer behaviour is observed over almost two decades : @xmath126\,\tau_{\eta}$ ] . a way to better visualise the departure from the tracer statistics consists in plotting the mean square separation for heavy pairs of different stokes numbers , normalised to the tracer one , that is @xmath127 this quantity is shown in the insets of figure [ fig : m2 ] . for heavy pairs starting at @xmath128 and with @xmath125 , the relative difference is as large as @xmath129 at its maximum for @xmath130 . however , such effect becomes progressively less important if we start the separation experiment from larger initial distances as shown in the right panel of the same figure . this is because , at these same stokes numbers , the deviation of particle velocity difference with respect to the underlying fluid , due to caustics , has already decreased . this is equivalent to state that , for these stokes numbers , the typical size of caustics is smaller than the initial separation @xmath12 , and particle pairs therefore separate as fluid tracers do . at larger time lags , whatever the value of the initial separation , the richardson dispersion regime is recovered . + we now consider what happens for larger stokes numbers . in fig . [ fig : st70 ] , we show the results for the mean square separation of @xmath131 and @xmath132 and for the large reynolds number , @xmath133 . both initial distances , @xmath134 \eta$ ] and @xmath51\eta$ ] are displayed . as one can see , for the large value of @xmath108 , the tracer - like behaviour is never recovered , and even the separation of pairs starting with the largest distance @xmath12 is affected . the transient regime dominated by the caustics invades the whole inertial range : since particle pairs need a very long time to decrease their initial velocity difference to the value of the fluid increment at the corresponding scale , they separate with a quasi - ballistic behaviour:@xmath135 . + the above scenario can be interpreted in terms of caustic-_dimensions_. at any value of the inertia , there exist a spatial length , of the order of the scale @xmath80 , which identifies the typical spatial size of caustics , i.e. the range of scales where particle velocity increments are uncorrelated from the underlying fluid velocity field . if the initial pair separation @xmath12 is taken inside this region ( left panel of fig . [ fig : m2 ] ) , particle pair separation starts much faster than for fluid tracers , because of the much more intense velocity differences felt by the pairs inside the caustics . when particle pairs reach a separation larger than @xmath136 , they start to be synchronised with the underlying fluid velocity , recovering the typical richardson dispersion . however , if the initial separation is larger than the caustics size , the evolution of inertial particle pairs is almost indistinguishable from the tracers . finally , whether or not a richardson - like behaviour is recovered for very large inertia , may depend on the reynolds number also . in the limit of larger and larger reynolds , at fixed stokes number , one may expect a final recovery of the fluid tracers behaviour even for very heavy particle pairs . the turbulent relative dispersion of fluid tracers can be easily modelled by applying k41 scaling theory to the fluid velocity increments governing particle separation dynamics ( see , e.g. , @xcite ) . indeed , if @xmath137 is the tracer separation vector at a given time , its evolution is completely specified by the equation @xmath138 together with the initial condition @xmath139 . hence , we can directly write an equation for the root - mean - square separation @xmath140 @xmath141 we next assume the following mean - field closure for the right - hand side : @xmath142 where @xmath143 is the first - order eulerian longitudinal structure function of the underlying homogeneous and isotropic turbulent flow . according to k41 phenomenology , this structure function behaves in the inertial range as @xmath144 , where @xmath119 is an order - unity constant . this closure finally leads to @xmath145^{3/2}. \label{eq : mf - tracers}\ ] ] such an approximation gives a complete qualitative picture of the time evolution of the mean square separation between tracers . in particular , it encompasses the two important regimes of relative dispersion : when @xmath146 , a taylor expansion of the solution ( [ eq : mf - tracers ] ) gives the batchelor regime @xmath147 , while when @xmath148 , one recovers richardson s law @xmath149 . in the case of inertial particles , the number of degrees of freedom to describe the dynamics is obviously increased : the separation between two heavy particles obeys @xmath150 . \label{eq : heavydisp}\ ] ] in order to derive _ mean - field _ equations one has to track simultaneously the average distance and velocity difference between particles . for this we follow the same spirit as for tracers and introduce the particle velocity structure function @xmath151 , where @xmath152 is the velocity difference between the two particles . one can proceed as previously to write from ( [ eq : heavydisp ] ) exact equations for @xmath153 and @xmath154 : @xmath155 , \label{eq : rexact}\\ \dot v \,&=&\ , - \frac{1}{\tau_s}\left [ v - \frac{1}{v}\ , \langle \delta_r \bm v \cdot \delta_r \bm u \rangle \right],\label{eq : vexact}\end{aligned}\ ] ] where for the sake of a lighter notation the indication of conditional ensemble averages was dropped . it is worth noticing that the root - mean - square velocity difference @xmath154 evolves with a dynamics that resembles closely that of heavy particles . however , @xmath154 does not coincide with the time derivative of the mean distance @xmath153 . it is thus useful to rewrite the above equations introducing a sort of transverse particle velocity component @xmath156 defined as @xmath157 we can write an exact equation also for the evolution of @xmath156 @xmath158\ , . \label{eq : wexact}\ ] ] of course , equations ( [ eq : vexact ] ) , ( [ eq : v - w ] ) , and ( [ eq : wexact ] ) are not closed without supplying the correlation between the particle evolution and the underlying fluid . as in the case of tracers , the first unclosed term appearing in the right - hand side of ( [ eq : wexact ] ) is approximated by ( [ eq : c1 ] ) . the next unclosed term involving the correlation between fluid and particle velocity differences is approximated by @xmath159 where @xmath160 denotes the full second - order structure function of the fluid velocity field . when @xmath161 is in the inertial range , k41 phenomenology implies that @xmath162 . finally these approximations lead to a closed set of equations for the time evolution of the average separation and velocities @xmath161 , @xmath163 , and @xmath156 @xmath164 \,,\label{eq : mf2}\\ \dot w \,&=&\ , -\frac{1}{\tau_s}w -(2v - w)\,\frac{w}{r}+ \frac{1}{\tau_s } b\ , \varepsilon^{1/3}r^{1/3}\ , , \label{eq : mf3}\end{aligned}\ ] ] where @xmath165 and @xmath119 are positive order - unity dimensionless constants , reflecting the lack of control on the prefactors of the scaling laws in the closures ( [ eq : c1 ] ) and ( [ eq : c2 ] ) . this system of equations is supplemented by the initial conditions @xmath166 , @xmath167 and @xmath168 , which clearly depend on the dispersion experiment under consideration . it is worth noticing that this system of equations reduces to the mean - field equation ( [ eq : mf - tracers ] ) for tracers in the limit of vanishing inertia @xmath169 . similarly to the case of tracers , the crude approximation ( [ eq : mf1])-([eq : mf3 ] ) of the evolution of the root - mean - square distance between heavy inertial particles is able to capture the main features of the separation time behaviour . in fig . [ fig : jeremie1 - 2 ] , we show the result of the numerical integration of the set of equation ( [ eq : mf1])-([eq : mf3 ] ) obtained by an appropriate choice of the free parameters ( see figure caption ) , together with dns data from run ii , for two different large values of the stokes number . for fixed initial separation and at increasing the intensity of the caustics velocity increments in the initial condition ( i.e. at increasing inertia ) , the transient deviation from the richardson behaviour become more and more evident at intermediate times ( of the order of the stokes time @xmath170 , not shown ) . clearly such a simple approach can be valid only in a limited region of the phase space , where the initial conditions are , at least , at the edge of the inertial range so that k41 scaling is correct for the fluid velocity second - order increments . moreover , the matching scale where particle velocity increments become of the order of the fluid increments has to fall in the inertial range too : if this is not the case , then pairs enter the regime where inertia is important but particle relative velocity is small at scales of the dissipative range , and the mean - field closure proposed above becomes inadequate . a detailed quantitative comparison of the realm of applicability of the mean - field approach including effects of viscous scales and small - scale caustics will be the object of future work . despite its simplicity , the mean - field approach described above is able to correctly reproduce the pair dispersion of heavy particles with initial data in the inertial range . we might wonder if one can draw an even simpler qualitative picture of pair dispersion . for this , we consider the behaviour of particle pairs with moderately large stokes numbers , for which inertia plays an important role for the initial transient and the richardson behaviour is slowly recovered well inside the inertial range of scales . for simplicity , we assume that the scale where the fluid and particle velocity becomes of the same order , @xmath171 , and the scale @xmath80 , where inertia ceases to be important , are very close . as it is clear from the sketch of fig . [ fig : abc ] , this may not be always the case because of the effect of the normalisation factor @xmath84 for large stokes : in the picture , it corresponds to stokes number with a narrow transient region ( b ) . the general picture then goes as follows . initially particles separate almost ballistically during a time which is of the order of ( or larger than ) the time needed by their initial , caustics - dominated , velocities to relax to the fluid velocity . after that time , particles behave as tracers and reconcile with a standard richardson dispersion . this is a first order approximation since ( _ i _ ) the fluid flow actually correlates to the particle dynamics already at very small times , and ( _ ii _ ) inertia effects are present up to large times as previously discussed . nevertheless , such an approximation should give the two correct qualitative asymptotic behaviours , at small and large time scales . since we consider moderately large values of the stokes number , the initial typical particle velocity can be assumed to be much larger than the fluid velocity , i.e. @xmath172 [ see ( [ eq : vexact ] ) ] , and thus @xmath173 . as a consequence , the mean square separation between particles evolves initially as : @xmath174 this should be approximately valid up to a time scale , in the inertial range , where @xmath175 : it is easy to show that such a time scale is proportional to the particle response time @xmath170 . for larger times , inertia effects become subdominant and heavy pair dispersion suddenly gets synchronised to a richardson like regime . nevertheless , this richardson regime has started only after the previous relaxation has ended , that is at a distance much larger than the original separation @xmath12 of the particle pair . the combination of this initial exponential relaxation of heavy particles with moderately large inertia , plus the later standard richardson diffusion are the two main features due to inertia in the inertial pair dispersion . this is indeed confirmed by figure ( [ fig : jeremie3 ] ) , where we compare dns data for mean square separation , with the two phenomenological regimes just described , for which we have assumed that @xmath176 . as we can see , the main qualitative trends of the small and large time behaviours are very well captured . we have seen in previous subsections that the most noticeable effect of inertia on the mean pair dispersion is a long transient regime that takes place before reaching a richardson explosive separation ( [ eq : r2rich ] ) , and that this regime is due to the relaxation of particle velocities to those of the fluid . as we now argue , at larger times corresponding to regime ( c)there is still an effect of particle inertia that can be measured in terms of subleading corrections to the richardson law . to estimate these corrections , let us assume that in the mean - field equation ( [ eq : mf2 ] ) , the term stemming from the fluid velocity @xmath177 is much larger than the inertia term @xmath178 . this is true when @xmath179 , i.e. at times @xmath180 when @xmath181 . in this asymptotic , one can infer that the transverse velocity component @xmath156 is much smaller than the total velocity @xmath163 , so that @xmath182 ( see eq . ( [ eq : v - w ] ) ) . in the spirit of the weak inertia expansion derived in @xcite , we next write a taylor expansion of ( [ eq : mf2 ] ) to obtain @xmath183 where @xmath184 denotes the increment of the fluid acceleration over the separation @xmath161 . next we assume scaling invariance of the turbulent acceleration field , that is , according to dimensional arguments of k41 theory , @xmath185 . equation ( [ eq : mf_maxeyapprox ] ) can then be rewritten as @xmath186 where @xmath187 is an order - unity constant . the initial condition is given by @xmath188 where the initial separation has to be chosen such such that @xmath189 . we can next integrate the approximate dynamics perturbatively in terms of the small parameter @xmath190 by expanding the separation as @xmath191 . the leading order is @xmath192^{3/2}$ ] and corresponds to the relative dispersion of a pair of tracers . the first - order correction reads @xmath193 . at times much larger than the batchelor time associated to the initial separation @xmath194 , i.e. for @xmath195 , the leading term follows the richardson explosive law @xmath196 . this finally implies that in the asymptotics @xmath197 , one can write @xmath198 , \label{eq : correcrichard}\ ] ] where @xmath122 is the richardson constant introduced in [ sec:2 ] and @xmath199 is an order - unity factor , which a priori does not depend neither on the particle stokes number , nor on the initial particles separation . this behaviour is confirmed numerically as can be seen from fig . [ fig : devrichard ] that gives the behaviours at large times of the ratio @xmath200 between the mean square separation of heavy particles and that of tracers as defined by ( [ eq : defq ] ) . one can clearly see that data almost collapse on a line @xmath201 confirming the behaviour ( [ eq : correcrichard ] ) predicted above . only results from run i are displayed here . the reason is that the very large time statistics of tracer dispersion in run ii is not as well statistically converged , leading to more noisy data . the qualitative picture is however very similar . to conclude this section , let us stress that we have assumed above k41 scaling to hold for the acceleration field ( and thus for the pressure gradient ) . however it is well known that the scaling properties of pressure are still unclear : they might depend on the turbulent flow reynolds number and/or on the type of flow ( see , e.g. , @xcite ) . as stated in @xcite , rather than being dominated by k41 scaling , numerically estimated pressure increments of run i ( @xmath202 ) seem to be ruled by sweeping , so that @xmath203 . one can easily check that this difference in scaling leads to a behaviour similar to ( [ eq : correcrichard ] ) , except that this time logarithmic corrections are absent , and that the non - dimensional constant @xmath199 depends on the reynolds number of the flow . the present numerical data do not allow to distinguish between these two possible behaviors . we now discuss the shape of the probability density function for both light and heavy inertial particles . we focus on the time and scale behaviour of the non - stationary pdf @xmath204 defined as the probability to find a pair of inertial particles @xmath9 , with separation @xmath58 at time @xmath180 , given their initial separation @xmath12 at time @xmath205 . the case of tracers ( @xmath206 ) has been widely studied in the past , either experimentally , numerically and theoretically for two and three dimensional turbulent flows ( see @xcite ) . following the celebrated ideas of richardson , phenomenological modelling in terms of a diffusion equation for the pdf of pair separation leads to the well - known non - gaussian distribution , @xmath207}\,,\ ] ] which is valid for times within the inertial range @xmath208 , and is obtained assuming a small enough initial separation and statistical homogeneity and isotropy of the three - dimensional turbulent flow . here , @xmath187 is a normalization constant . this prediction is based on the simple assumption that , for inertial range distances , tracers undergo a diffusion dynamics with an _ effective _ , self - similar , turbulent diffusivity @xmath209 . moreover , it relies on the phenomenological assumption that tracers separate in a short - time correlated velocity field . indeed , it is only if the latter is true , that the diffusion equation for the pair separation becomes exact ( see @xcite ) . as mentioned before such a scenario may be strongly contaminated by particle inertia . the main modifications are expected to be due to the presence of small - scale caustics for small - to - large stokes numbers , and to preferential concentration . caustics make the small scale velocity field not differentiable and not self - similar , as if inertial particles were separating in a rough velocity field whose exponent were depending on distance . preferential concentration , leading to inhomogeneous spatial distribution of particles , manifests itself as a sort of _ effective compressibility _ in the particle velocity field . there exists a series of stochastic _ toy models _ for lagrangian motion of particles in incompressible / compressible velocity fields , where the statistics of pair separation can be addressed analytically . among these , the so - called kraichnan ensemble models , where tracer particles move in a compressible , short - time correlated , homogeneous and isotropic velocity field , with gaussian spatial correlations ( we refer the reader to the review @xcite for a description of this model ) . it is useful for the sequel to recall two main results obtained for relative dispersion in a kraichnan compressible flow . we denote with @xmath210 the velocity field compressibility degree is defined as the ratio @xmath211 , where @xmath212 and @xmath213 , and varies between @xmath214 for incompressible flows , and @xmath215 for potential flows . ] , and with @xmath216 the scaling exponent of the two - point velocity correlation function at the scale @xmath161 , in @xmath217-dimensions : @xmath218[u_j({\bm r})-u_j(0 ) ] { \rangle}\sim g_1 r^\xi [ ( d-1+\xi-\wp\xi ) \delta_{ij } + \xi(\wp\,d-1)r_ir_j / r^2]$ ] . for particles moving in such flows , it is possible to show that the pair separation pdf for tracer particles follows a richardson - like behaviour : @xmath219}\,.\ ] ] here @xmath220 and @xmath221 is the correlation dimension , characterizing the fractal spatial distribution of particles . + a different distribution emerges when the @xmath222dimensional kraichan flow is differentiable , i.e. for @xmath223 ; in such case , a log - normal pdf is expected : @xmath224}\,,\ ] ] with @xmath225 and @xmath226 . it is worth noticing that in the latter case , since the flow is differentiable , the large - time pdf depends on the initial data . the problem of inertial particle separation in a real turbulent flow presents some similarities with the previous toy cases but also important differences . + first , the effective degree of compressibility due to preferential concentration of inertial particles , is properly defined only in the dissipative range of scales . for @xmath227 , it is equal to the correlation dimension @xmath228 defined as @xmath229 , where @xmath230 is the probability to find two particles at distance smaller than @xmath161 , with @xmath231 . as it has been numerically shown in @xcite for three - dimensional turbulent flows , the correlation dimension depends only on the degree of inertia @xmath9 , while it does not seem to depend on the reynolds number of the flow . for @xmath232 , the effective degree of compressibility is no longer constant , but varies with the scale . + second , the underlying velocity field exhibits spatial and temporal correlations that are much more complex than in a gaussian short - correlated field . such correlations lead to non - trivial overlaps between particle dynamics and the carrying flow topology . as a result , it is not possible to simply translate the analytical findings obtained in the compressible kraichnan ensemble to the case of inertial particles : we may expect however that in some limits the compressible kraichnan results should give the leading behaviour also for the case studied here of inertial particles in real turbulent flows . with this purpose , we first notice that the separation probability density function that is valid in the rough case ( [ eq : rich_k ] ) has an asymptotic stretched - exponential decay that is independent on the compressibility degree . this suggests that inertial particle pdf ( [ eq : rich_st ] ) must recover the richardson behaviour ( [ eq : rich ] ) of tracers in the limit of large scales and large times . coherently with what discussed in previous sections , for large times and for scales larger than @xmath233 , we expect that the heavy pairs ( in the limit @xmath234 ) pdf recovers a tracer like distribution : @xmath235 } ; \qquad r \gg r^*_{st}. \label{eq : pdflight}\ ] ] for pairs of light particles , there is no straightforward formulation of such a prediction : as we shall see in the sequel , preferential concentration effects have a strong fingerprint on the separation pdf even at large times and large scales . in the opposite limit of very small separations , i.e. @xmath73 , one can correctly assume that the effective degree of compressibility is constant and therefore apply either the small - scale limit for rough flows ( [ eq : rich_k ] ) , or that for smooth flows ( [ eq : rich_lognormal ] ) , depending on the scaling properties of the particle velocity field entailed in the value of the exponent @xmath76 , defined from ( [ eq : gamma_st ] ) and related to the caustics . we thus expect @xmath236 here @xmath237 and @xmath238 are two different decaying functions of time @xmath180 , whose expression can be easily derived from ( [ eq : rich_k])-([eq : rich_lognormal ] ) . notice that for the smooth case , i.e. the small - scale limit of the log - normal distribution ( [ eq : rich_lognormal ] ) , we get for the spatial dependency a factor @xmath239 instead of the factor @xmath228 of the rough case . this will matter in the case of light particle separation , where , due to strong preferential concentration , the probability of finding pairs at a very small distances is large enough to allow for a detailed test of the prediction ( [ eq : log ] ) . the case of light particles will be discussed in [ sec : light ] , while we now turn to a discussion of the above scenario in the case of heavy particle pairs . we start by analyzing the qualitative evolution of @xmath240 at changing time , for different stokes numbers and in the limit @xmath241 . the four panels of fig . [ fig:1 ] show the evolution of the pdf at different times for pairs with @xmath242 ( tracers ) , and for heavy particles with @xmath243 , @xmath244 , and @xmath132 . initially , at @xmath114 , all selected pairs are separated by the same distance ( @xmath245\,\eta$ ] ) ; this initial distribution is represented in each figure by a grey area . as time elapses , particle separate and reach different scales , depending on their inertia . qualitatively , the pdf evolution is very similar for all moderate stokes numbers , and the pdfs at different moderate stokes numbers become more and more similar with time . however in the case of @xmath108for which the associate stokes time @xmath170 falls well inside the inertial range , the pdf shows a long exponential tail for intermediate separation , which tends to persist at all observed times . to better appreciate such differences , in fig . [ fig:2 ] we show the comparison between the different pdfs corresponding to various stokes numbers for two different times : at the beginning of the separation process , @xmath246 , and at a later time , @xmath247 . as one can see , it is only at early times that the pdfs for moderate - to - large stokes , @xmath68 differ in a sensible way from the tracers . in particular , one can clearly see that many pairs have separations much larger and much smaller than those observed for tracers or for heavy pairs with small stokes numbers . the right tails , describing pairs that are very far apart , are just the signature of the _ scrambling _ effect of caustics . such strong events are not captured by second order moments of the separation statistics that we discussed before , while they clearly affect higher - order moments . the left tails , associated to pairs much closer than tracers , are possibly due to particles that separate at a slower rate than tracers because of preferential concentration induced by inertia . later in the evolution , for @xmath248 , only the separation pdf for @xmath108 still shows important departure from the tracer case ; for all the other stokes numbers shown , pairs have had enough time to forget their initial distribution and have practically relaxed on the typical richardson - like distribution . in the inset , we also show the persistence in the exponential behaviour for the pdf at @xmath108 , by superposing the shapes measured at three times during the particles separation . with the present data the small scale asymptotic behaviour ( [ eq : log ] ) can not be validated for heavy particles . this is due to the limited statistics : very soon after the initial time @xmath205 , there are almost no pairs left with separations @xmath73 . at moderate to large stokes numbers the separation process of heavy particle pairs is largely influenced by the presence of large velocity differences at small scales , that is by the presence of caustics in the particles velocity field . in [ sec:1 ] , we have studied stationary statistics ( only first - order moment ) of velocity differences between heavy particles at changing the distance between particles and their inertia . however it is also informative to look at the non - stationary , time - dependent distribution of velocity differences , and more particularly to its distribution measured along heavy pairs separation . the relative velocity @xmath249 can be decomposed into the projection along the separation vector , and two transveral components , here equivalent since the system is statistically isotropic . for tracer particles , the statistics of relative velocity and the alignment properties of @xmath250 and @xmath251 have been discussed extensively ( see e.g. @xcite ) . here , we focus on the pdf of the relative longitudinal velocity only , which we denote by @xmath252 , where @xmath253 \cdot \hat{\bm r}(t)}$ ] . for pairs of tracers ( @xmath242 ) , the initial longitudinal velocity distribution is nothing else than the pdf of eulerian longitudinal velocity increments measured at the distance @xmath12 . for pairs of inertial particles , this initial pdf clearly coincides with the stationary distribution of velocity differences between particles that are at a distance @xmath254 $ ] . such a distribution has the signature of two mechanisms : ( _ i _ ) at small stokes numbers , only preferential concentration matters and particles probe only a sub - set of all possible fluid velocity fluctuations ; ( _ ii _ ) at large stokes numbers , particles are homogeneously distributed but with a velocity field which may be strongly different from the underlying fluid velocity . for what concerns heavy pairs , the first effect has not an important signature on small - scale quantities . however the second effect clearly becomes visible for moderate to high inertia as shown in fig . [ fig:4 ] . here we report the longitudinal velocity distributions for pairs with initial distance @xmath51\ , \eta$ ] , and with @xmath255 and @xmath132 ; the reynolds number of the underlying flow is @xmath1 . each panel contains the pdfs measured at different times spanning all turbulent timescales . at @xmath114 , the importance of caustics is manifest for the two largest stokes numbers , leading to fat tails towards both small and large velocity differences . interestingly enough , the left tail of @xmath252 , which describes approaching events of particle relative motion , is immediately dumped already at @xmath256 ; at the same time , however , the right tail continues to be quite fat for the two largest stokes numbers under consideration . at later stages of the separation process , the tendency of large - stokes - number pairs to wash out approaching events becomes even stronger . indeed , at time @xmath257 , the small velocity increments tail has almost disappeared for pairs with @xmath108 . it is worth noticing that at those times ( i.e. also at those typical scales ) , heavy particle velocity differences have already started to be smaller than the tracer velocity increments : the larger is the stokes number , the less pronounced are the pdf tails . summarising , because of the different effects of inertia , we observe a very complex evolution for the longitudinal relative velocity fluctuations along the trajectories of heavy particle pairs . this is certainly a key issue to be considered for stochastic modelling : here , as in a standard _ kinetic _ problem , both particle positions and velocities need to be modelled to quantitatively control the relative dispersion process . so far we have considered the relative motion of very heavy particle pairs , for which the density contrast @xmath8 with the underlying fluid is zero . in this section we present results on light particles dynamics as described by ( [ eq:1 ] ) , for different possible choices of the parameters @xmath9 . we discuss how the strong effect of preferential concentration typically observed in the case of light particles in turbulent flows might influence the intermediate and long - time behaviour of pair separations . in three - dimensional turbulent flows , as we consider here , light particles associated to different values of @xmath9 have been observed to always possess a positive largest lyapunov exponent @xcite : this implies that light pairs always separate in @xmath258 real turbulent flows . we recall , however , that this is not always true : for instance , in smooth two - dimensional random flows , there are values of @xmath9 for which the largest lyapunov exponent can become negative and particles form pointwise clusters ( see @xcite ) . light particles with moderate inertia and high density ratio ( order - unity @xmath7 , and @xmath259 ) , initially tend to separate much slower than heavy particles with the similar stokes number : this is evident from the much smaller values of the lyapunov exponents measured for light particles , with respect to those measured for heavy particles with equivalent stokes numbers but @xmath241 . moreover , finite - time lyapunov exponents show large fluctuations , indicating that there are pairs that do not separate even at long times . results on this issue will be reported elsewhere . clearly , pairs of light particles that do not separate do not influence the mean square distances : hence , we do not expect , and indeed do not measure , any large differences for the long - time behaviour of @xmath260 for light particles , with respect to the heavy case ( see fig . [ fig : rich.light ] ) . it is natural to ask if light particle strong preferential concentration affects high - order moments of relative separation of two initially close particles , and particularly the left tail of the separation pdf . figure [ fig : light1 ] shows the time evolution of the separation probability density function , @xmath261 . data refer to a case with very intense preferential concentration effects and minor influence of caustics @xmath262 , and a case with milder inhomogeneities in the spatial distribution @xmath263 . the initial separation pdf was chosen in both cases by selecting particle pairs with initial distance @xmath51\,\eta$ ] . a remarkable observation is a strong tendency to fill small separations . in other words there are many pairs that reduce their mutual distance even for a very long times . the development of the left tail for the strong clustering case ( @xmath264 ) is consistent with the estimate given by the long - time , small - scale asymptotic expansion of the log - normal distribution ( [ eq : log ] ) , @xmath265 , as shown by the straight line in the plot . this is the confirmation that the small - scale dynamics of the highly clustered light particles evolves as that of tracers moving in a smooth , compressible flow ( characterised by the same @xmath228 ) . we also remark that if there is high spatial preferential concentration , caustics can not be important . this may have important consequences for the estimation of collision kernel of light particles . the approaching events , shown by the left tail in figure [ fig : light1 ] , are clearly due to the preferential concentration inside vortex - like structures , typical of light particles . the right panel show a different case , where preferential concentration is less important , leading to a correlation dimension @xmath266 . of course , also in this latter case , there are events with approaching pairs , but these become less and less probable with time . the importance of preferential concentration can also be appreciated by looking at the pdfs of longitudinal velocity differences between light particles during the separation . we show such distributions for one of the pair family considered above , and we compare them with those of the tracers ( see fig . [ fig : light2 ] ) . the important difference between the two cases stems from the highly peaked nature of the relative velocity pdf for the strong clustered light particle case . the presence of many pairs with almost vanishing velocity differences is the signature of a coherent bunch of pairs moving on a strongly clustered set . we have studied the relative dispersion of inertial particles in homogeneous and isotropic turbulence from two dns at resolutions @xmath2 and @xmath3 , corresponding to @xmath267 and @xmath268 , respectively . we have analysed both heavy and light particle statistics at changing the stokes numbers . we have studied the evolution of mean separations and the whole pdfs shape , both for particle distance and velocity increments at changing time and for different typical initial distances . the main results that we have discussed can be summarised as follows . separations of very heavy particles , with stokes times falling in the inertial range of the underlying fluid , are strongly affected by the presence of caustics up to times , when the distance between particles reaches scales that are large enough for the separation dynamics to be again dominated by the underlying flow velocity . as a consequence , strong transient departure from the richardson diffusion , with a faster ballistic regime , is observed . a statistical closure of the equation of motions for heavy particle separation is also developed . this model is able to reproduce the main numerical findings . + for light particles , at high density ratio , we observe strong small - scale clustering properties , leading to a considerable fraction of pairs that do not separate at all although the maximum lyapunov exponent remains positive . in such a case , the non - stationary spatial concentration at small scales tends to be higher than the analogous case but with a stationary distribution of particles . such numerical findings open the way to experimental verifications and gives input to the community involved in modelling inertial particle diffusion in applied configurations . we thank massimo cencini , who collaborated at an early stage of this work , for very fruitful and continuous exchange of ideas . this study benefitted from constructive discussions with e. bodenschatz , e. calzavarini , l. collins and g. falkovich to whom we wish to express a warm gratitude . j. bec and a.s . lanotte aknowledge support from the national science foundation under grant no . phy05_51164 for their stay at the kavli institute for theoretical physics in the framework of the 2008 `` physics of the climate change '' program . part of this work was supported by agence nationale de la recherche under grant no . blan07 - 1_192604 . we thank the deisa consortium ( co - funded by the eu , fp6 project 508830 ) , for support for run ii within the deisa extreme computing initiative ( www.deisa.org ) . numerical simulations of run i were performed at caspur ( italy ) and cineca ( italy ) . numerical raw data of particle trajectories are freely available from the icfddatabase ( http://cfd.cineca.it ) kindly hosted by cineca ( italy ) .

this paper presents a scheme to deal accurately and efficiently with complex angular masks , such as occur typically in galaxy surveys . an angular mask is taken to be an arbitrary union of arbitrarily weighted angular regions bounded by arbitrary numbers of edges . the restrictions on the mask are ( i ) that each edge must be part of some circle on the sphere ( but not necessarily a great circle ) , and ( ii ) that the weight within each subregion of the mask must be constant . the scheme works by resolving a mask into disjoint polygons , convex angular regions bounded by arbitrary numbers of edges . the polygons may be regarded as the ` pixels ' of a mask , with the feature that the pixels are allowed to take a rather general shape , rather than following some predefined regular pattern . among other things , the scheme includes facilities to compute the spherical harmonics of the angular mask , and data - random and random - random angular integrals . a software package mangle which implements this scheme , along with complete software documentation , is available at http://casa.colorado.edu/~ajsh / mangle/. large - scale structure of universe methods : data analysis

astro-ph0306324

with massive new data sets shrinking the statistical error bars on cosmological quantities , it is becoming increasingly important to avoid inaccuracies in their modeling and analysis . for example , one of the less interesting aspects of modern studies of large scale structure is having to deal with complex angular masks . this humble but essential task is rendered more time - consuming by the fact that angular masks may require updating as a survey progresses . the purpose of this paper is to describe a scheme intended to remove much of the drudgery and scope for inadvertent error or unnecessary approximation involved in defining and using angular masks . the scheme is implemented in a publically available software package mangle , which can be obtained , along with complete documentation , from http://casa.colorado.edu/~ajsh / mangle/. the present paper is not a software manual : for that , visit the aforesaid website . rather , the purpose of this paper is to set forward the philosophy and to detail the algorithms upon which the software is built . angular masks of galaxy survey have grown progressively more complicated through the years . the first complete redshift surveys of galaxies , the first center for astrophysics redshift survey ( cfa1 ; @xcite ) , and the first southern sky redshift survey ( ssrs1 ; @xcite ) , had rather simple angular boundaries , defined by simple cuts in declination and in galactic latitude . the level of complexity increased with the _ iras _ redshift surveys . the first of these , the _ iras _ 2 jy redshift survey @xcite , in addition to a cut in galactic latitude , excluded 1465 lunes of high cirrus or contamination by local group galaxies , each lune being an approximately @xmath2 square with boundaries of constant ecliptic latitude and longitude . the angular masks of subsequent _ iras _ surveys followed a similar theme , leading up to the pscz survey @xcite , whose high - latitude mask ( the one most commonly used in large scale structure studies ) consisted of the whole sky less 11,477 @xmath2 ecliptic lunes . the automatic plate measuring ( apm ) survey ( @xcite , b , 1996 ) and associated surveys such as the apm - stromlo survey @xcite , consisted of a union of a couple of hundred photographic plates , cut to @xmath3 , each drilled with a smrgsbord of holes to avoid bright stars , satellite trails , plate defects , and the like . the edges of the excluded holes were straight lines on photographic plates , but unlike the _ iras _ surveys , the holes were not necessarily rectangles , their boundaries were not necessarily lines of constant latitude and longitude , and different holes could overlap . = 6.5 in the anglo - australian telescope 2 degree field survey ( 2df ; @xcite ; @xcite ) is a redshift survey of galaxies from the apm survey , and thus inherits the holey angular mask of the apm survey . superimposed on the apm backdrop , the 2df angular mask consists of several hundred overlapping @xmath4 diameter circular fields . the various overlaps of the circular fields have , at least in early releases of the data , various degrees of completeness . the sloan digital sky survey ( sdss ; @xcite ) has an angular mask comparable in complexity to that of the 2df survey . it consists of several stripes from the parent photometric survey , peppered with holes masked out for a variety of reasons . superimposed on the stripes are circular fields from the redshift survey . recently , the sdss team used the mangle scheme described in the present paper as part of the business of computing the 3d galaxy power spectrum @xcite . with both the 2df and sdss data going public , it has seemed sensible to publish the scheme so that others can use it too . the scheme described in the present paper began life in the delightful atmosphere of an aspen center for physics workshop in 1985 . the mathematics of harmonization ( [ harmonize ] ) and other aspects of the computation of angular integrals are written up in an appendix to @xcite . the methods described therein were first applied by @xcite , and have been used regularly by him since that time . the idea of adapting the methods to deal with angular masks in a rather general way , and in particular the concept of balkanization , is new to the present paper . the mangle software has been applied to the 2df 100k survey by @xcite , and to the sdss by @xcite . the figures in this paper were prepared from files generated by the mangle software . figure [ twoqzfig ] shows a zoom of a small piece of the northern angular mask of the 2df qso redshift survey ( 2qz ) 10k release @xcite . the angular mask of this survey is defined by files ( downloadable from http://www.2dfquasar.org/spec_cat/masks.html ) giving the boundaries of : ( 1 ) @xmath3 ukst plates , ( 2 ) holes in ukst plates , and ( 3 ) @xmath4 fields . these boundaries are illustrated in the top left panel of figure [ twoqzfig ] . the 2qz team provide the completeness of the angular mask in the @xmath5 pixelized form illustrated in the top right panel of figure [ twoqzfig ] . the 2qz mask is typical of the way that angular masks are defined in modern galaxy surveys . [ cols= " < , < " , ] motivated by common practice , an angular mask is defined in the present paper to be an arbitrary union of arbitrarily weighted angular regions bounded by arbitrary numbers of edges . the restrictions on the mask are 1 . that each edge must be part of some circle on the sphere ( but not necessarily a great circle ) , and 2 . that the weight within each subregion of the mask must be constant this definition of an angular mask by no means covers all theoretical possibilities , but it does reflect the actual practices of makers of galaxy surveys . the broad utility of spherical polygons to delineate angular regions is widely appreciated ; for instance , they play an integral part in the sdss database . the definition implies that an angular mask is a union of arbitrarily weighted non - overlapping polygons . a polygon is defined to be the intersection of an arbitrary number of caps , where a cap is defined to be a spherical disk , a region on the unit sphere above some line of constant latitude with respect to some arbitrary polar axis . for reference , table [ deftable ] collects definitions of mask , polygon , cap , and certain other terms used in this paper . the bottom left panel of figure [ twoqzfig ] shows the 2qz mask ` balkanized ' ( see [ balkanize ] below ) into non - overlapping polygons . the bottom right panel of figure [ twoqzfig ] shows the mask reconstructed from spherical harmonics up to @xmath6 ( see [ harmonize ] below ) . the information specifying a mask ( its angular boundaries and completeness ) is collected in files which we refer to as ` polygon files ' . typically a command in the mangle suite of software will : 1 . read in one or more polygon files , possibly in different formats ; 2 . do something to or with the polygons ; 3 . write an output file , possibly a polygon file , or files . the strategy adopted in the mangle software is to permit the most flexible possible input format for polygon files , the idea being to be able to read the files provided by the makers of a galaxy survey as far as possible in their original form , or perhaps mildly edited . mangle reads and writes several different formats of polygon files : * circle ; * vertices ; * edges ; * rectangle ; * polygon . for convenience , there are five additional formats that provide useful information about polygons , but that can only be written , not read , because the information they provide is too limited , or ambiguous , to specify polygons completely . the five output only formats are : area ; graphics ; i d ; midpoint ; weight . an abbreviated description of each format appears below ; see http://casa.colorado.edu/~ajsh/mangle/ for full details . the * circle * format is able to describe polygons in all generality . a circle is defined by the azimuth @xmath7 and elevation @xmath8 of its north polar axis , and by the angular radius , the polar angle @xmath9 , of the circle . each circle defines a cap . a polygon is an intersection of caps , and a line of the form @xmath10 containing @xmath11 angles defines a polygon with @xmath12 caps . the * vertices * format specifies polygons by a sequence of vertices , assumed to be joined by great circles . the general form of a line specifying a polygon in vertices format is @xmath13 which defines a polygon with @xmath12 caps . in vertices format , a line with @xmath14 angles defines a polygon with n caps . the * edges * format is a souped - up version of the vertices format . whereas the vertices format joins each pair of vertices with a great circle , the edges format uses an additional point ( or additional points ) between each pair of vertices to define the shape of the circle joining the vertices . although the edges format retains more information about a polygon than the vertices format , in general it does not retain all information about a polygon . a rectangle is a special kind of 4-cap polygon bounded by lines of constant azimuth and elevation . the * rectangle * format is offered not only because some masks are defined this way ( for example , the _ iras _ masks ) , but also because the symmetry of rectangles permits accelerated computation of their spherical harmonics ( [ harmonize ] ) . a line in rectangle format looks like @xmath15 with precisely 4 angles . the * polygon * format is the default output format for polygon files . besides the circle format , it is the only other format that is able to describe polygons in all generality without loss of information . it stores each cap , not as three angles as in the circle format , but rather as a unit vector along the north pole of the cap , together with a quantity @xmath16 , which is equal both to the area of the cap divided by @xmath17 , and to half the square of the 3-dimensional distance between the north pole and the cap boundary . it seems doubtful that one would want to create an original mask file in polygon format , since it is a bit peculiar , but it is the format used internally by the mangle software , and it specifies polygons in the manner expected for many years past by the fortran backend . the advantage of the format is that some computational operations are simpler and faster if the cap axis is stored as a unit vector rather than as an azimuth and elevation . of the purely output formats , one of the most useful is the * graphics * format , which is useful for making plots of polygons . the mangle software does not incorporate any plotting software : it is assumed that you have your own favourite plotting package . the graphics format is similar to edges format , but is generally more economical . whereas in edges format there is a specified number of points per edge , in graphics format there is a specified number of points per @xmath17 of azimuthal angle along each edge . thus in graphics format curvier edges get more points than straighter edges . the graphics format is implemented only as output , not as input , because of ambiguity in the interpretation of the format . another useful output format is the * midpoint * format , which returns a list containing the angular position of a point inside each polygon of a mask . this can be helpful in assigning weights to the polygons of a mask , if you have your own software that returns a weight given an angular position . [ midpoint ] for more about how midpoints of polygons are computed . the * area * , * i d * , and * weight * output formats give lists of , respectively , the areas , identity numbers , and weights ( completenesses ) of the polygons of a mask . one of the basic tasks that the mangle software does is resolve a mask into a set of non - overlapping polygons . this takes place in a sequence of four steps , elaborated in the subsections following . 1 . snap ; 2 . balkanize ; 3 . weight ; 4 . unify . resolving a mask into non - overlapping polygons greatly simplifies the logic of dealing with a mask , since it allows subsequent processing ( generation of random catalogues , computation of spherical harmonics , plotting , etc . ) to proceed without recourse to the intricate hierarchy of overlapping geometric entities and the associated complicated series of inclusion and exclusion rules that tend to characterize a survey mask . each individual polygon is by definition an intersection of caps . geometrically , this implies that a polygon is convex : the interior angles at the vertices of a polygon are all less than @xmath18 . the requirement that a polygon be an intersection of caps greatly simplifies the logic , since it means that a point lies inside a polygon if and only if it lies inside each of the caps of the polygon . the first thing that must be done on all the original polygon files of a mask is to ` snap ' them . this process identifies almost coincident cap boundaries and snaps them together . the problem is that the positions of the intersections of two almost but not exactly coincident circles ( cap boundaries ) on the unit sphere may be subject to significant numerical uncertainty . to avoid numerical problems , such circles must be made exactly coincident . you might think that that near - but - not - exactly - coincident circles would hardly ever happen , but in practice they occur often , because a mask designer tries to make two polygons abut , but imprecision or numerical roundoff defeats an exact abutment . the snap process adjusts the edges of each polygon , but it leaves the number and order of polygons the same as in the input file(s ) . edges that appear later in the input file(s ) are snapped to earlier edges . the snap process offers four tunable tolerances : * axis tolerance . are the axes of two caps within this angular tolerance of each other , either parallel or anti - parallel ? if so , change the axis of the second cap to equal that of the first cap . * latitude tolerance . if the axes of two caps coincide , are their latitude boundaries within this tolerance of each other ? if so , change the latitude of the second cap to equal that of the first cap . the two caps may lie either on the same or on opposite sides of the latitude boundary . * edge tolerance , and edge to length tolerance . are the two endpoints and midpoint of an edge closer to a cap boundary than the lesser of ( a ) the edge tolerance , and ( b ) the edge to length tolerance times the length of the edge ? in addition , does at least one of the two endpoints or midpoint of the edge lie inside all other caps of the polygon that owns the cap boundary ? if so , change the edge to align with the cap boundary . the purpose of the first two of these tolerances , the axis tolerance and the latitude tolerance , is obvious . the remaining two tolerances , the edge tolerance and the edge to length tolerance , are necessary because it is possible for two edges , if they are short enough , to almost coincide even though the axes and latitudes of their corresponding caps differ significantly . by default , the three angular tolerances ( axis , latitude , and edge ) are all two arcseconds , which is probably sufficient for typical large scale structure masks . the tolerances can be tightened considerably before numerical problems begin to occur , so it is fine to tighten the tolerance for a mask whose edges are more precisely defined . the default edge to length tolerance @xmath19 , should be fine in virtually all cases . the snap process accomplishes its work in two stages : * snap axes and latitudes of pairs of caps together , passing repeatedly through all pairs of caps until no more caps are snapped . * snap edges of polygons to other edges , again passing repeatedly through all pairs of caps until no more caps are snapped . as a finishing touch , snap prunes each of the snapped polygons in order to eliminate superfluous caps , those whose removal leaves the area of the polygon unchanged . the process of resolving a mask into disjoint polygons we dub ` balkanization ' , since it fragments an input set of possibly overlapping polygons into many non - overlapping connected polygons . the process involves two successive stages : 1 . fragment the polygons into non - overlapping polygons , some of which may be disconnected . identify disconnected polygons and subdivide them into connected parts . = 1.8 in the algorithm for the first stage of balkanization is simple and pretty : * is the intersection of two polygons neither empty nor equal to the first polygon ? if so , find a circle , a cap boundary , of the second polygon that divides the first polygon , and split the first polygon into two along that circle . * iterate . notice that only one of the two parts of the split polygon overlaps the second polygon , and that only the overlapping part needs iterating . for any pair of polygons , iteration ceases when the overlapping part lies entirely inside the second polygon . the final overlapping part is equal to the intersection of the original first polygon with the second polygon . all other fragments of the first polygon lie outside the second polygon . figure [ balkanizefig ] illustrates an example of the first stage of balkanization for two overlapping polygons a and b. first , a is split against b , which takes two iterations of the above cycle . then , b is split against a. again , this takes two iterations of the above cycle . the final system consists of 5 non - overlapping polygons . note that splitting the system shown in panel ( a ) of figure [ balkanizefig ] into its three connected parts ( the part of a that does not intersect b , the part of b that does not intersect a , and the intersection ab of a and b ) would not constitute a successful balkanization , since two of these regions are not convex and hence not polygons . one might ask , why not stop at panel ( c ) in figure [ balkanizefig ] ? do not the three polygons there already form a satisfactory set of non - overlapping polygons ? the answer is that the intersection polygon ab may well have a weight different from those of the non - overlapping parts of the parent a and b polygons ( this is typically true for example in the 2df and sdss surveys ) . to deal with this eventuality , balkanization must continue to completion , as illustrated in panel ( e ) . the question of whether two polygons overlap is determined by computing the area of the intersection of the polygons . the area is proportional to the monopole harmonic , computed as described in [ harmonize ] . the intersection of two polygons is itself a polygon , consisting of the intersection of the two sets of caps defining the polygons . stage 1 of the balkanization procedure yields polygons that can contain two or more connected parts , as illustrated in figure [ discpolfig ] . stage 2 attempts to subdivide such disconnected polygons into connected parts by computing the connected boundaries of the polygon , and lassoing ( see [ lasso ] ) each connected boundary with an extra circle . figure [ eyefig ] illustrates a polygon that has two distinct connected boundaries by virtue of being not simply - connected rather than not connected . a region is said to be * simply - connected * if , according to the usual mathematical definition , it is connected and any closed curve within it can be continuously shrunk to a single point . loosely speaking , this means that a simply - connected polygon has no holes . because it is connected , the polygon of figure [ eyefig ] need not be split . the strategy to deal with non - simply - connected polygons is based on the following theorem , proven in the appendix : * a connected part of a polygon is simply - connected if and only if all the boundaries of the connected part belong to a single group*. a group is defined here as follows : two circles are friends , belonging to the same group , if they intersect ( anywhere , not necessarily inside the polygon ) , and friends of friends are friends . according to this definition of group , the circles on a single connected boundary necessarily all belong to the same group . however , the circles on two distinct connected boundaries may or may not belong to the same group . this theorem implies that it is necessary to lasso only those boundaries of a polygon that belong to the same group . in figure [ discpolfig ] , for example , the two boundaries of the polygon belong to the same group of three intersecting circles , so these two boundaries must be lassoed , partitioning the polygon into two parts . in figure [ eyefig ] , on the other hand , the two boundaries belong to two separate groups , and need not be lassoed . figure [ discpolsfig ] illustrates a more complicated polygon , similar to the polygon of figure [ discpolfig ] but pierced with two circular holes . the polygon contains four boundaries belonging to three groups . the two original boundaries inherited from figure [ discpolfig ] belong to the same intersecting group of circles , but the additional two holes form two separate groups . here only the two boundaries belonging to the same group need lassoing . = 2.5 in figure [ fractalfig ] illustrates a yet more complicated multiply - connected polygon . the mangle software balkanizes this polygon correctly into seven polygons , a stringent test of the algorithms . a corollary of the theorem proven in the appendix is that the polygon formed by the intersection of the caps bounded by the circles of a single group must be a union of simply - connected parts . for example , the two parts of the polygon in figure [ discpolfig ] must be simply - connected which evidently they are because the circles of the polygon all belong to the same group . in the course of the proof in the appendix , it is shown that if the boundary of a polygon falls into two ( or more ) groups , then the circles of a second group must lie entirely inside exactly one of the simply - connected parts of the polygon bounded by the first group . for example , each of the two circles bounding the two holes in the polygon of figure [ discpolsfig ] must lie entirely inside exactly one of the two simply - connected parts of the original polygon from figure [ discpolfig ] , which again is evidently true . it follows from the statement of the previous paragraph that in lassoing the connected boundaries of a group , it is necessary to consider only the boundaries belong to the same group : any boundary belonging to another group can be ignored , because it must lie entirely inside one of the simply - connected parts bounded by the first group . thus each lassoing circle is required to enclose fully its connected boundary , while excluding fully all other connected boundaries belonging to the same group ; there is no constraint on the lasso from boundaries belonging to other groups . if a group of circles of a polygon defines a single boundary , then that boundary needs no lassoing , but stage 2 balkanization nevertheless attempts to lasso the boundary if the number of caps of the group exceeds the number of vertices . for example , in its original configuration the polygon shown in figure [ eggboxfig ] has a large number of caps . none of the caps can be discarded , since each excludes a small piece of the sky . here it is advantageous to lasso the polygon with an extra circle , allowing most of the original caps to be discarded as superfluous . a lasso that lassos the lone boundary of a group is discarded if the lasso completely encloses all the circles of the group to which the boundary belongs . for then either the lasso completely encloses the polygon , in which case it is superfluous , or else the lasso lies completely inside the simply - connected region bounded by the lone boundary , in which case the lasso , if kept , would divide the simply - connected region in two , which would be incorrect . if on the other hand the lasso of a lone boundary of a group intersects at least one of the circles of the group , then the lasso must completely enclose the simply - connected region bounded by the lone boundary , as in figure [ eggboxfig ] ; the lasso can not lie inside the simply - connected region because it is being assumed that the lasso intersects a circle of the group , whereas no such circle can exist within the simply - connected region . stage 2 balkanization may need more than one pass to succeed . figure [ chainfig ] shows an example of a polygon , bounded by one large cap punctuated by fifteen small caps , that contains four parts bounded by four boundaries all belonging to the same single group . the top and bottom boundaries can be lassoed successfully with single circles , but the middle two boundaries can not : any circle that encloses either of the middle boundaries necessarily intersects another boundary somewhere . here stage 2 balkanization succeeds by submitting the polygon to two passes . in the first pass , the polygon is split into three polygons , consisting of the top and bottom connected parts , plus a third polygon containing the two middle parts . in the second pass , the third polygon is split into two , completing the partitioning of the original polygon into its four parts . in certain convoluted cases , such as the polygon shown in figure [ vfig ] , it can be impossible to lasso any of the connected boundaries of the polygon with a circle that wholly encloses a connected boundary while wholly excluding all other connected boundaries in the same group . stage 2 balkanization gives up attempting to lasso a boundary after a certain maximum number of attempts , but it keeps a record of the best - attempt lasso , the one that encloses as much as possible of a boundary while wholly excluding all other boundaries in the same group . stage 2 balkanization proceeds to split the polygon into two parts with the best - attempt lasso , and then submits the two parts to a further pass . the polygon of figure [ vfig ] , for example , contains two parts bounded by two connected boundaries neither of which can be lassoed with a circle that completely encloses the boundary while completely excluding the other boundary . finding no satisfactory lasso , stage 2 balkanization splits the polygon into two polygons with a best - attempt lasso , shown as a dashed line in figure [ vfig ] . the two polygons are then submitted to further passes of stage 2 balkanization , which in this case succeeds with one pass . the upshot is that the original polygon is balkanized into four polygons . = 2.5 in it is conceivable that the algorithm of the above paragraph could continue for ever , continually splitting a polygon into two and continually failing to lasso successfully all the boundaries of the split polygons . however , polygons that defy lassoing have to be filamentary in character ( long , thin , and windy ) , such as that shown in figure [ wfig ] , and splitting such a polygon in two generally makes it less filamentary , like putting spaghetti in a blender . in the case of the polygon of figure [ wfig ] , stage 2 balkanization forcibly splits the polygon 8 times , eventually balkanizing the 13-part polygon into 34 disjoint parts . suffice to say that we know of no polygon that fails the algorithm , and it is possible that no such polygon exists . if the reader finds one , please tell us about it . in practice , the mangle software bails out if a polygon has to be split forcibly in two more than a certain maximum number ( 100 ) of times . even in this last gasp case , the set of polygons output by balkanization still constitutes a valid set of non - overlapping polygons that completely tile the mask . the only problem is that the ` failed ' polygons , those which could not be partitioned completely , may contain two or more disjoint parts with different weights . to finish , balkanization prunes each of the balkanized polygons in order to eliminate superfluous caps , those whose removal leaves the area of the polygon unchanged . for example , pruning discards the many superfluous caps of the polygon of figure [ eggboxfig ] . the caps are tested in order , with any new lassoing cap being tested last , so that the many superfluous caps are discarded , and the lassoing cap is kept . the algorithm to lasso a connected boundary of a polygon is to pick a point , initially taken to be the barycentre of the centres of the edges of the connected boundary ( or , if the connected boundary consists of a single circle , the centre of that circle ) , and find the circle centred on that point which most tightly encloses the boundary . the lassoing circle is enlarged slightly if possible , as a precaution against numerical problems that might potentially occur if the lasso just touched an edge or vertex of the boundary . a lasso that lassos , i.e. that encloses completely , one connected boundary of a polygon , is required to exclude completely all other connected boundaries belonging to the same group . a lasso attempt can sometimes fail if a polygon has two or more connected boundaries . a lasso attempt fails if the angular distance from the centre of the lasso to the farthest point @xmath20 on the to - be - lassoed connected boundary is greater than the angular distance from the centre of the lasso to the nearest point @xmath21 on all other connected boundaries belonging to the same group . if a lasso attempt fails , then the centre point of the lasso is shifted over the unit sphere along the vector direction from @xmath21 to @xmath20 , by an amount that puts the centre point just slightly closer to @xmath20 than @xmath21 . the lasso is then reattempted . the process of shifting the centre point and retrying a lasso is repeated until either the lasso succeeds , or until a certain maximum number of attempts has been made . multiple intersections occur where 3 or more circles ( cap boundaries ) intersect at a single point . multiple intersections pose a potential source of numerical problems , because the topology around multiple intersections may vary depending on numerics , as illustrated in figure [ multfig ] . circles kiss if they just touch . again , kissing circles pose a potential source of numerical problems , because whether two circles kiss may vary depending on numerics , as illustrated in figure [ kissfig ] . mangle is equipped to deal with both multiply - intersecting and kissing circles , and should cope in almost all cases , although it is possible to fool manglewith a sufficiently complicated polygon , for example a polygon whose vertices have a fractal distribution of separations . the strategy is as follows . circles are considered to be multiply intersecting , crossing at a single vertex , if the intersections are closer than a certain tolerance angle . similarly , circles are considered to kiss , touching at a single vertex , if their kissing distance is closer than the tolerance angle . the algorithm is friends - of - friends : two vertices closer than the tolerance are friends , and a friend of a friend is a friend . the position of each vertex @xmath22 of a polygon , where edge @xmath23 intersects ( or kisses ) edge @xmath24 , is computed two different ways , first as the intersection of edge @xmath23 with edge @xmath24 , then as the intersection of edge @xmath24 with edge @xmath23 . for each of the two ways of computing it , the intersection is tested against all other circles , to determine whether the intersection is or is not a vertex of the polygon , that is , whether the intersection lies on the edge of the polygon , or outside the polygon . for consistency , the test should give the same result in both computations : the intersection should be a vertex in both cases , or it should not be a vertex in both cases . if an inconsistency is detected , then the tolerance angle is doubled ( or set to a tiny number , if the tolerance is zero ) , and the computation is repeated for the inconsistent polygon , until consistency is achieved . by default , the initial tolerance angle for multiple intersections and kissings is @xmath25 arcseconds . in figure [ multfig ] , the intersection of the two diagonals is a vertex of the polygon in the left panel , but is not a vertex in the right panel , because it lies outside the polygon . in the case of an exact multiple intersection , as in the middle panel of figure [ multfig ] , the intersection @xmath22 is considered to be a vertex of the polygon only if @xmath23 and @xmath24 are both edges of the polygon . thus the intersection of the two diagonals in the middle panel of figure [ multfig ] is a vertex , because both diagonals are edges , but the intersection of the horizontal with either diagonal is not a vertex , because the horizontal is not an edge of the polygon . if @xmath23 is an edge , and it intersects multiply with a bunch of other circles , then the adjacent edge @xmath24 is formed by the circle which ` bends most tightly ' around the polygon , that is , the circle whose interior angle at the vertex is the smallest , or , if two circles subtend the same interior angle ( within the tolerance angle ) , then the circle whose polar angle @xmath9 is the smallest . in figure [ kissfig ] , the two circles @xmath23 and @xmath24 intersect at no vertices in the left panel , and at two vertices in the right panel . in the case of an exact kiss , as in the middle panel of figure [ kissfig ] , the kissing point @xmath22 is considered to be a vertex only if @xmath23 and @xmath24 are both edges of the polygon . thus the kissing point @xmath22 is _ not _ a vertex of the the upper and lower polygons , the two disks , but it is a vertex of the middle polygon , the pointy one . two polygons that just touch or kiss at a single isolated point ( or at a set of isolated points ) are considered to be disconnected from each other . thus for example the top and bottom polygons in the middle panel of figure [ kissfig ] are considered to be disconnected from each other ; and similarly the left and right polygons in the middle panel of figure [ kissfig ] are considered to be disconnected from each other . in practice , consistency of the topology of the distribution of vertices around a polygon is checked by means of a 64-bit check number . if the intersection @xmath22 of edge @xmath23 with edge @xmath24 ( two edges can intersect at two separate points , so @xmath22 is an ordered pair , going from edge @xmath23 to edge @xmath24 right - handedly around the boundary of the polygon ) is determined to be a vertex of the polygon , then a 64-bit pseudo - random integer is added to the check number , and if the same intersection @xmath22 of edge @xmath24 with edge @xmath23 is determined to be a vertex of the polygon , then the same 64-bit pseudo - random integer is subtracted from the check number . for consistency , the check number should be zero for the entire polygon . it is conceivable , with probability 1 in @xmath26 , or less than 1 in 10 billion billion , that the check number could evaluate to zero accidentally , but this seems small enough not to worry about , especially since inconsistency should be a rare occurrence in the first place . figure [ flowerfig ] illustrates a mask designed to be as ` difficult ' as possible : it contains many multiply intersecting and nearly multiply - intersecting circles , and many kissing and nearly kissing circles , including several simultaneously multiply - intersecting and multiply - kissing circles . the mangle software copes with this , a non - trivial accomplishment . the sum of the areas of the 332 polygons of the balkanized mask of figure [ flowerfig ] differ from the area , @xmath27 , of the overall bounding rectangle by @xmath28 , which is definitely satisfactory . given the algorithms , one could expect the numerical uncertainty in the area of a single polygon to be no better than machine precision times @xmath29 , which on the machine used for this computation was about @xmath30 . when balkanizing , does the order of the polygons in the input polygon files matter ? the answer is yes , if the input polygons overlap , and if the overlapping polygons carry different weights . as described in the following subsection [ weight ] , if two polygons overlap , then the weight of the polygon that appears later in the input file(s ) overrides the weight of the earlier polygon . if all polygons have the same weight ( say 1 ) , then the order of the input polygon files does not really matter . however , it may lead to a slightly smaller eventual polygon file ( after unifying , see [ unify ] ) if large , coarse polygons are put first , and small , finely detailed polygons are put last . each connected polygon of a mask may have a different weight . in galaxy surveys , the ` weight ' attached to a polygon is the completeness of the survey in that polygon . these weights must be supplied by the user . if no weights are supplied , then the weight defaults to 1 . if the input polygons of a mask overlap , then the policy adopted by the mangle software is to allow the weights of later polygons in polygon files to override the weights of earlier polygons . thus for example , to drill empty holes in a region , one would put the polygons of the parent region first ( with weight 1 , perhaps ) , and follow them with polygons specifying the holes ( with weight 0 ) . there are three ways to apply weights to polygons . the first way is simply to edit the polygon file or files specifying the mask . attached to each polygon in a polygon file is a line that includes a number for the weight ; one simply edits that number . the second way is to specify weights in a file . the mangle software contains a facility to read in these weights , and to apply them successively to the polygons of a polygon file . suppose that you have your own software that returns a weight given an angular position in a mask . the mangle software includes a utility ( [ midpoint ] ) to create a file giving the angular position of a point inside each polygon of a mask . this file of angular positions becomes input to your own software , which should create a file of weights , which in turn can be fed back to mangle . the third way to apply weights to polygons is to write a subroutine ( in either fortran or c ) that returns a weight given an angular position , and compile it into mangle . the mangle software includes some template examples of how to do this . in our experience , method two is the method of choice , except in cases that are simple enough that method one suffices . the set of non - overlapping polygons that emerges from balkanizing and weighting may be more complicated than necessary . the mangle software includes a facility for simplifying the polygons of a mask , which we call unification . unification is not strictly necessary , but it tidies things up , and it can save subsequent operations , such as harmonization ( see [ harmonize ] ) , a lot of computer time . = 1.8 in unification eliminates polygons with zero weight , and does its best to merge polygons with the same weight . the algorithm is to pass repeatedly through a set of polygons , merging a pair of polygons wherever the pair can be merged into a single polygon by eliminating a single abutting edge . figure [ unifyfig ] illustrates an example of the unification procedure . unification does not necessarily accomplish the most efficient unification , nor , as illustrated in figure [ nounifyfig ] , is unification necessarily exhaustive . the mangle software contains several utilities that do various things with a mask . one of the most important of these is a utility to take the spherical harmonic transform of a mask , a process we call harmonization . in particular , the area of a mask is proportional to the zeroth harmonic . computation of the area of a polygon is basic to several of the mangle algorithms . for example , whether two polygons intersect is determined by whether the area of their intersection is non - zero . the method for computing the spherical harmonics of a mask consisting of a union of polygons is described in the appendix of @xcite . the algorithm is recursive and stable , able to compute harmonics to machine precision to arbitrarily high order , limited only by computer power and patience . the recursion , as implemented in the mangle software , recovers correctly from underflow , which can occur at large harmonic number @xmath31 . while the recursive algorithm by itself is fast , there is a numerical penalty to be paid for allowing the polygons of a mask to have arbitrary shape : the computation time for harmonics up to @xmath31 increases as @xmath32 , a pretty steep penalty when @xmath31 is large . the computation time is proportional to the number of edges of the polygons of the mask , and on a 750mhz pentium iii it takes 3 cpu minutes per edge to compute harmonics up to @xmath6 . thus the method is slow compared to fast algorithms specially designed for regular pixelizations , such as healpix @xcite . the most time - consuming part of the computation is rotating the harmonics of an edge from its natural frame of reference into the final frame of reference : it is this rotation that takes @xmath32 time . the rotation is unnecessary if the edge is a line of constant latitude in the final reference frame , and the computation goes faster , as @xmath33 , in this case . another acceleration is possible if two edges are related by a rotation about the polar axis of the final frame . although computing the harmonics @xmath34 of a single edge @xmath35 still takes @xmath32 time , the harmonics @xmath36 of a second edge @xmath37 rotated right - handedly by azimuthal angle @xmath38 from edge @xmath35 are @xmath39 , which is fast to compute . in practice , the mangle software currently implements the latter acceleration only in the special case where ( some of ) the polygons of a mask are rectangles , polygons bounded by lines of constant azimuth and elevation . the acceleration applies only if at least two rectangles of the mask have the same minimum and maximum elevation . two such rectangles need not be adjacent in the polygon file : mangle reorders the computation of polygons so as to take advantage of acceleration where possible . for completeness , we give here an overview of the method detailed by @xcite . the spherical harmonic coefficients @xmath40 of a mask @xmath41 , a function of angular direction @xmath42 , are defined by @xmath43 @xmath44 where @xmath45 are the usual orthonormal spherical harmonics , and @xmath46 denotes an interval of solid angle about @xmath47 . the key mathematical trick is to convert the integral ( [ olm ] ) for @xmath40 from an integral over the solid angle of the mask to an integral over its edges . this is done by introducing the square @xmath48 of the angular momentum operator @xmath49 into the integrand @xmath50 which is valid except for the monopole harmonic @xmath51 , dealt with below . the hermitian character of the angular momentum operator @xmath52 allows equation ( [ olml ] ) to be rewritten @xmath53 by assumption , the mask is a sum over polygons @xmath54 , and @xmath41 is a constant @xmath55 within each polygon . it follows that @xmath56 in equation ( [ olmp ] ) is a sum over polygons , with the contribution from each polygon being a vector whose magnitude is @xmath55 times @xmath57 times a dirac delta - function , and whose direction is along the boundary of the polygon , winding right - handedly about the polygon . thus the integral ( [ olmp ] ) reduces to a sum of integrals over the boundaries @xmath58 of the polygons @xmath59 the boundary @xmath58 of a polygon is a set of edges , so the integral in equation ( [ olmo ] ) becomes a sum of integrals over each edge of each polygon . thus a harmonic @xmath40 is a sum of contributions @xmath60 from each edge @xmath35 of each polygon @xmath54 @xmath61 hence the analytic problem reduces to that of determining the harmonics @xmath34 of the edge @xmath35 of a polygon . the problem is well suited to computation , and could easily be parallelized if required ( currently , mangle is not parallelized ) . stable recursive formulae for computing the harmonics @xmath34 of a polygon edge are given by @xcite . first , the harmonics of an edge are computed in a special frame of reference where the axis of the edge cap is along the polar direction ( the @xmath62 direction ) of the spherical harmonics . the harmonics of the edge are then rotated into the actual frame of reference . the most time consuming part of the computation is the second part , the rotation , with the computational time going as @xmath32 . the above derivation of the harmonics @xmath40 of the mask fails for the monopole harmonic @xmath51 , for which equation ( [ olml ] ) is invalid . the monopole harmonic @xmath63 is @xmath64 where @xmath65 is the weighted area of the mask , a weighted sum of the areas @xmath66 of the polygons @xmath54 @xmath67 the general formula for the area @xmath66 of a polygon @xmath54 is @xmath68 where @xmath69 , an integer , is the euler characteristic (= faces minus edges plus vertices of any triangulation ) of the polygon , @xmath70 and @xmath71 are the lengths and polar angles of the edges @xmath35 of the polygon , and @xmath72 are the exterior angles ( @xmath73 @xmath18 minus interior angle ) at the vertices @xmath74 of the polygon . the euler characteristic @xmath69 of a polygon , a topological quantity , is calculable topologically it equals two plus the number of connected boundaries minus twice the number of groups to which the boundaries belong but in practice it is quicker to compute the euler characteristic as follows . first , in the trivial case that the polygon is the whole sky , its area is @xmath75 . second , in the special case that the polygon is a single cap with edge @xmath35 , its area is @xmath76 . otherwise , the polygon is an intersection of two or more caps . if the area of any one of the caps is less than or equal to @xmath17 , then the area of the polygon must be less than @xmath17 , so the euler characteristic @xmath69 must take that integral value which makes the area lie in the interval @xmath77 . this leaves the case where every one of the caps of the polygon has area greater than @xmath17 . the policy in this case is to introduce an extra cap which splits the polygon into two parts , each of whose areas is less than @xmath17 ; the area of the polygon is then the sum of the areas of the two parts . in practice , polygons which are intersections of caps all of whose areas exceed @xmath17 are fairly uncommon , so the slow down involved in splitting these particular polygons is not great . moreover the splitting is necessary only for the monopole harmonic : the higher order harmonics can be computed from the original polygon without splitting it . the mangle software contains a ` map ' utility to reconstruct the mask at arbitrary points @xmath42 from spherical harmonics up to a given maximum @xmath78 @xmath79 for example , the bottom right panel of figure [ twoqzfig ] was generated using this utility . another important feature provided in the suite of mangle software is a utility to compute precisely the angular cross - correlation @xmath0 at given angular separation @xmath9 between given ` data ' points and ` random ' points in the mask . the angular integral is done analytically and evaluated to machine precision , rather than by monte carlo integration with random points . if the ` data ' points are chosen randomly within the mask , as in [ random ] , then the cross - correlation becomes equivalent to the angular autocorrelation @xmath1 , for random - random , at given angular separation between pairs of points in the mask . the advantage of computing the @xmath0 angular integral analytically over the traditional monte carlo method is that it eliminates unnecessary shot noise . as discussed for example by @xcite , this unnecessary shot noise can adversely affect the performance of estimators of the correlation function at small scales . in the case of the pscz high latitude mask , which balkanizes into 744 polygons , it takes 5 cpu minutes per 1000 ` data ' points to compute @xmath0 at 1000 angular separations with a 1.2ghz pentium iii . the contribution of a data point at position @xmath42 to the angular correlation @xmath0 at angular separation @xmath9 is a weighted sum over the polygons of the mask of the azimuthal angle @xmath80 subtended within the polygon by a circle of radius @xmath9 centred at @xmath42 . the correlation @xmath0 at angular separation @xmath9 is an average over the contributions from each data point . the computation of @xmath0 is done at a specified angular separation or set of separations @xmath9 , not , as is common with monte carlo integration , over a bin of angular separations . for a finite number of data points , and a mask consisting , as in the present paper , of a union of weighted polygons , the correlation @xmath0 is a continuous function of angular separation @xmath9 , except that , if a data point @xmath42 happens to coincide with the axis of an edge of a polygon , then the function will be discontinous at a separation @xmath9 equal to the polar angle of the said edge . furthermore , the correlation @xmath0 is a differentiable function of separation @xmath9 except at a finite , possibly large , number of discontinuities where @xmath9 equals the separation between a data point @xmath42 and either a vertex of a polygon or a point on the edge of a polygon where the separation from @xmath42 is extremal . although @xmath0 , as a function of separation @xmath9 , is thus typically discontinuous in the derivative , and occasionally discontinuous in itself , in practical galaxy surveys @xmath0 tends to be relatively smooth , especially when the number of data points is large . thus in the practical case it is usually fine to sample @xmath0 at a suitably large number of angular separations , and to interpolate ( linearly ) on such a table . the dr utility loops in turn through each data point @xmath42 , so that two points take twice as much cpu time as one point . for each data point , the dr utility attempts to accelerate the computation with respect to the angular separations @xmath9 by first computing the minimum and maximum angles between the point @xmath42 and each polygon in the mask . the information about the minimum and maximum angles is used to decide whether the circle about @xmath42 lies entirely outside or entirely inside a polygon , in which case the ( unweighted ) angle @xmath80 subtended within the polygon is zero or @xmath17 . in practical cases the angle subtended is often zero for the great majority of polygons of a mask , especially when the mask is composed of many polygons . since calculation of the subtended angle can be skipped if the angle is zero , computation can be greatly speeded up . further acceleration comes from ordering the polygons in increasing order of the minimum angle from the given point @xmath42 to each polygon . this allows the computation to loop to the next value of @xmath9 as soon as it hits a polygon for which the subtended angle is zero , rather than checking through large numbers of polygons that all have zero subtended angle . the angular auto - correlation @xmath1 between pairs of points in a mask can be computed in various ways . the traditional method , as suggested by the random - random designation , is to count pairs of random points . however , the traditional method is certainly not the most precise method , and it may not be the most efficient , especially at small scales , if large numbers of random points are needed to reduce the shot noise to a subdominant level . an alternative method , mentioned above , is to compute @xmath1 using the @xmath0 algorithm with the ` data ' points chosen randomly within the mask . although this method is subject to some shot noise , the shot noise is liable to be substantially less than that of the traditional method at small scales . at the largest angular separations @xmath9 , a choice method is to compute @xmath1 from its spherical harmonic expansion , truncated at some suitably large harmonic @xmath78 : @xmath81 where @xmath82 is a legendre polynomial , which is accurate at angular scales @xmath83 . the exact expression for the autocorrelation @xmath1 at angular separation @xmath9 is @xmath84 for a mask of polygons as considered in this paper , this double integral over solid angles can be transformed into a double integral over the edges of the polygons @xcite @xmath85 where @xmath65 is the weighted area , equation ( [ a ] ) , of the mask , and @xmath86 is the green s function of the scalar product @xmath87 of angular momentum operators : @xmath88 } & ( x_{12 } \geq x ) \\ 0 & ( x_{12 } \leq x ) \ . \end{array } \right.\ ] ] unfortunately the integral ( [ rrexact ] ) can not be solved analytically , and we have not attempted to implement its numerical solution in the mangle software . @xcite gives a series expansion of the expression ( [ rrexact ] ) valid at small scales , but the series expansion is liable to break down already at tiny scales in the typically complex masks of modern surveys ( the coefficients of the series expansion change discontinuously wherever the angular separation equals the distance between distinct vertices , or more generally any extremal distance between pairs of edges ) , making the series expansion of limited applicability . the mangle software contains a utility for generating random points inside a mask . as indicated above , this can be useful for example in computing the angular auto - correlation @xmath1 of pairs of points in the mask . the algorithm , which is quite fast , is as follows : 1 . select randomly a polygon in the mask , with probability proportional to the product of the polygon s weight and area . lasso that polygon with a circle that is intended to be a tight fit , but is not necessarily minimal . generate a point randomly within the circle , test whether the point lies inside the polygon , and keep the point if it does . iterate . a lasso is computed for a polygon as needed , but is then recorded , so that a lasso is computed only once for any polygon . if the desired number of random points exceeds the number of polygons in the mask , then the computation starts by lassoing every polygon in the mask . the mangle software contains several other utilities , described below . one simple but oft - used utility is one that copies a polygon file or files into another polygon file in a different format see [ format ] for an abbreviated description of the possible formats . for example , most of the figures in this paper were produced from points generated by copying polygon files into graphics format . the utility for copying polygon files has some switches to copy polygons with weights only in some interval , or with areas only in some interval . this makes it easy for example to discard polygons with small weights or small areas . the mangle software contains routines that return the vertices of the polygons of a mask , and to return the positions of points along the edges of a mask . for example , when a polygon file is copied into graphics format , the copy utility invokes these routines . a single polygon can have more than one connected boundary , as illustrated in figures [ discpolfig][fractalfig ] and [ chainfig][wfig ] . here the routines return the vertices and edge points on distinct boundaries as distinct sets . the routines to determine the distinct connected boundaries of a polygon are used in stage 2 of the balkanization process , [ balkanize ] . the mangle software contains a routine to find a point inside each polygon of a mask . the aim is to find a point that is squarely inside the polygon , well away from its edges . for example , the weights attached to the balkanized set of polygons shown in the lower left panel of figure [ twoqzfig ] were obtained by picking a point inside each polygon , and evaluating the weight at that point from the @xmath5 pixelized map provided by the 2qz team . the pixelization means that the map is reliable only away from the edges of a polygon , so it is important to pick the point squarely inside the polygon . the algorithm finds one point for each distinct connected boundary of a polygon . if the polygon contains non - simply - connected parts , as in figures [ eyefig][fractalfig ] , that means that the algorithm will return more points than there are distinct connected parts of the polygon ; however , having more points than necessary is not a problem . the algorithm to find a point inside a polygon is mildly paranoid . for each connected boundary of a polygon , the algorithm first determines the barycentre of the midpoints of the edges of the connected boundary . this barycentre can not be guaranteed to lie inside the polygon , so it is not enough to stop here . instead , great circles are drawn from the barycentre to each of the midpoints of the edges of the connected boundary . on each of these great circles , the midpoint of the segment of the great circle lying within the polygon , with one end of the segment being the midpoint of an edge of the connected boundary , is determined . each of these segment midpoints inside the polygon is tested to see how far away it is from the nearest vertex or edge of the polygon , including edges and vertices other than on the connected boundary . the desired point inside the polygon is chosen to be that segment midpoint which is furthest away from any vertex or edge . the mangle software contains a utility to find inside which polygon or polygons of a polygon file a given point lies . a point may lie inside zero polygons , or one polygon , or more than one polygon . the utility takes no short cuts : it tests all points against all polygons . this can take time if there are large numbers of points and large numbers of polygons . if the polygon file has been produced by balkanization , then a point would normally lie inside at most one polygon . however , a point at the edge of two abutting polygons is considered to lie inside both polygons . the observing strategies of modern galaxy surveys typically produce angular masks with complex boundaries and variable completeness . the purpose of this paper has been to set forward a scheme that is able to deal accurately and efficiently with such angular masks , and thereby to reduce both the labour and the chance for inadvertent error . the fundamental idea is to resolve a mask into a union of non - overlapping polygons each of whose edges is part of a circle ( not necessarily a great circle ) on the sphere . the scheme has been implemented in a suite of software , mangle , downloadable from http://casa.colorado.edu/~ajsh / mangle/. the mangle software includes several utilities for accomplishing common tasks associated with angular masks of galaxy surveys . this includes generating random catalogues reflecting the angular selection function ( a tool employed in almost all galaxy survey analysis ) , measuring the @xmath0 and @xmath1 angular integrals ( needed for estimating the correlation function ) , and expanding the mask in spherical harmonics ( a key step in various techniques for measuring the power spectrum and redshift space distortions ) . the scheme was originally motivated by the nature of real angular masks of real galaxy surveys , and the underlying angular routines have been battle - tested over many years . the full apparatus of the mangle software has been used on the 2df survey by @xcite , and on the sdss survey by @xcite . ajsh was supported by nasa atp award nag5 - 10763 and by nsf grant ast-0205981 . mt was supported by nsf grants ast-0071213 & ast-0134999 , nasa grants nag5 - 9194 & nag5 - 11099 , and fellowships from the david and lucile packard foundation and the cottrell foundation . we thank yongzhong xu for his contributions to an early version of the balkanization software , and michael blanton for suggestions . colless m. et al . ( the 2dfgrs team ; 29 authors ) , 2001 , mnras , 328 , 1039 ( astro - ph/0106498 ) croom s. m. , smith r. j. , boyle b. j. , shanks t. , loaring n. s. , miller l. , lewis i. j. , 2000 , mnras , 322 , l29 ( 2qz , available at http://www.2dfquasar.org/spec_cat/ ) da costa l. n. , pellegrini p. s. , davis m. , meiksin a. , sargent w. l. w. , tonry j. l. , 1991 , apjs , 75 , 935 gorski k. m. , wandelt b. d. , hansen f. k. , hivon e. , banday a. j. , 1999 , astro - ph/9905275 hamilton a. j. s. , 1993a , apj , 406 , l47 hamilton a. j. s. , 1993b , apj , 417 , 19 huchra , j. , davis , m. , latham , d. and tonry , j. ( 1983 ) apjs , 52 , 89 kerscher m. , szapudi i. , szalay a. , 2000 , apj , 535 , l13 ( astro - ph/9912088 ) lewis i. j. et al . ( 2df team ; 23 authors ) , 2002 , mnras , 333 , 279 ( astro - ph/0202175 ) loveday j. , peterson b. a. , maddox s. j. , efstathiou g. , 1996 , apjs , 107 , 201 maddox s. j. , efstathiou g. , sutherland w. j. , 1990b , mnras , 246 , 433 maddox s. j. , efstathiou g. , sutherland w. j. , 1996 , mnras , 283 , 1227 maddox s. j. , sutherland w. , efstathiou g. , loveday j. , 1990a , mnras , 243 , 692 saunders w. , sutherland w. j. , maddox s. j. , keeble o. , oliver s. j. , rowan - robinson m. , mcmahon r. g. , efstathiou g. p. , tadros h. , white s. d. m. , frenk c. s. , carraminana a. , hawkins m. r. s. , 2000 , mnras , 317 , 55 ( pscz , available at http://www-astro.physics.ox.ac.uk/~wjs/pscz.html ) strauss m. a. , huchra j. p. , davis m. , yahil a. , fisher k. b. , tonry , j. , 1992 , apjs , 83 , 29 tegmark m. , hamilton a. j. s. , xu y. , 2002 , mnras , 335 , 887 ( astro - ph/0111575 ) tegmark m. et al . ( sdss collaboration ) , 2003 , apj , to be submitted york d. g. , et al . ( sdss collaboration , 144 authors ) , 2000 , aj , 120 , 1579 this appendix proves the following theorem , invoked in [ balkanize2 ] ( see table [ deftable ] for a definition of the term * group * ) : first , suppose that a polygon contains a region which is connected but not simply - connected . it is required to prove that the boundaries of this region belong to at least two distinct groups . by definition of simply - connectedness , a continuous line can be drawn entirely inside the non - simply - connected region such that one connected boundary of the region lies entirely on one side of the line , and another connected boundary of the region lies entirely on the other side of the line . in the polygon of figure [ eyefig ] , for example , such a line would circulate around the central boundary while remaining inside the outer boundary . the continuous line can not intersect any of the circles forming the caps of the polygon , because if the line did intersect a circle , then the line would be inside the polygon on one side of the intersection , and outside the polygon on the other side of the intersection , contradicting the assumption that the line lies entirely inside the polygon . hence the continuous line must partition the circles of the polygon into two non - intersecting groups . the two connected boundaries on either side of the continuous line must therefore belong to two distinct groups , as was to be demonstrated . conversely , suppose that a polygon has boundaries that belong to at least two distinct groups . it is required to prove that the two groups delineate a connected but non - simply - connected region of the polygon . consider the polygon , call it a , formed by the intersection of all the caps of one group . the polygon a so formed must consist of one or more simply - connected parts ; for if any connected part of a were not simply - connected , then according the previous paragraph the boundaries of that part would belong to different groups , contradicting the assumption that the caps of a all belong to a single group . now consider similarly the polygon , call it b , formed by the intersection of all the caps of a second group . the boundaries of polygon b must lie entirely inside one and only one of the simply - connected parts of a. for certainly b must lie inside at least one part of polygon a , since a must enclose the parent polygon , and the boundary of b lies inside ( at the border of ) the parent polygon . but the boundary of polygon b can not lie in more than one part of a , because if it did then , because the circles of b are all in the same group and therefore connected to each other , there would be a path lying along the circles of b traversing continuously from one part of a to another , and therefore necessarily intersecting one of the boundaries of polygon a , contradicting the assumption that the circles of a and b belong to distinct groups that nowhere intersect . similarly , the boundaries of polygon a must lie entirely inside one and only one of the simply - connected parts of b. this argument has identified two special boundaries : a boundary of a that entirely encloses b ; and a boundary of b that entirely encloses a. the region enclosed by these two special boundaries delineates a region of the polygon that is connected but not simply - connected . for consider a continuous line which is displaced slightly off the special boundary of a , towards the special boundary of b. by construction , the continuous line lies entirely inside both polygons a and b. the continuous line can not intersect any of the circles of a or b , because if it did then the line would lie inside a ( or b ) on one side of the intersection , and outside a ( or b ) on the other side , contradicting the fact that the line lies entirely inside a and b. the continuous line could possibly intersect circles belonging to a third group of circles of the parent polygon . however , if the continuous line is displaced off the special boundary of a by a sufficiently small amount that it does not encounter any third group , then the continuous line will lie entirely inside the parent polygon . this continous line forms a line inside the polygon that can not be shrunk continuously to a point , so the polygon must contain a region that is connected but not simply - connected , as was to be demonstrated .

physical layer network - coding for the-way wireless relaying scenario is dealt with , where each of the user nodes wishes to communicate its messages to all the other nodes with the help of the relay node r. the given scheme , based on the denoise - and - forward scheme proposed for two - way relaying by popovski et al . in , employs two phases : multiple access ( ma ) phase and broadcast ( bc ) phase with each phase utilizing one channel use and hence totally two channel uses . physical layer network - coding using the denoise - and - forward scheme was done for the two - way relaying scenario in , for three - way relaying scenario in , and for four - way relaying scenario in . this paper employs denoise - and - forward scheme for physical layer network coding of the-way relaying scenario illustrating with the help of the case not dealt with so far . it is observed that adaptively changing the network coding map used at the relay according to the channel conditions reduces the impact of multiple access interference which occurs at the relay during the ma phase . these network coding maps are chosen so that they satisfy a requirement called _ exclusive law_. we show that when the users transmit points from the same-psk constellation , every such network coding map that satisfies the exclusive law can be represented by a-fold latin hyper - cube of side . the singular fade subspaces resulting from the scheme are described and enumerated for general values of and and are classified based on their removability in the given scenario . a network code map to be used by the relay for the bc phase aiming at reducing the effect of interference at the ma stage is obtained .

1303.0229

the two - stage protocol for physical layer network coding for the two - way relay channel first introduced in @xcite , exploits the multiple access interference occurring at the relay so that the communication between the end nodes can be done using a two stage protocol . the works in @xcite , @xcite deal with the information theoretic studies for bidirectional relaying . in @xcite , modulation schemes to be used at the nodes for uncoded transmission for the two - way relaying were studied . the work done for the relay channels with three or more user nodes is given in @xcite . in @xcite , authors have proposed a two stage operation for three - way relaying called joint network and superposition coding , in which the three users transmit to the relay node one - by - one in the first phase , and the relay node makes two superimposed xor - ed packets and transmits back to the users in the bc phase . the packet from the node with the worst channel gain is xor - ed with the other two packets . the protocol employs four channel uses , three for the ma phase and one for the bc phase . it is claimed by the authors that this scheme can be extended to more than three users as well . in the work by pischella and ruyet in @xcite a lattice - based coding scheme combined with power control , composed of alternate ma and bc phases , consisting of four channel uses for three - way relaying is proposed . the relay receives an integer linear combination of the symbols transmitted by the user nodes . it is stated that the scheme can be extended to more number of users . these two works essentially deal with the information theoretic aspects of multi - way relaying . an ` opportunistic scheduling technique ' for physical network coding is proposed by authors jeon et al . in @xcite , where using a channel norm criterion and a minimum distance criterion , users in the ma as well as the bc phase are selected on the basis of instantaneous snr . this approach utilizes six channel uses in case of three - way relaying and it is mentioned that the approach can be extended to more number of users . in @xcite , a ` latin square - like condition ' for the three - way relay channel network code is proposed and cell swapping techniques on these latin cubes are suggested in order to improve upon these network codes . the protocol employs five channel uses , and the channel gains associated with the channels are not considered in the construction of this network coding map . -way relay channel , height=132 ] we consider the @xmath0-way wireless relaying scenario shown in fig . 1 , where @xmath0-way data transfer takes place among the nodes @xmath7 , @xmath8 , ... , @xmath9 with the help of the relay r assuming that the @xmath0 nodes operate in half - duplex mode . the relaying protocol consists of two phases , _ multiple access _ ( ma ) phase , consisting of one channel use during which @xmath7 , @xmath8 , ... , @xmath9 transmit to r ; and _ broadcast _ ( bc ) phase , in which r transmits to @xmath7 , @xmath8 , ... , @xmath9 in a single channel use . network coding is employed at r in such a way that @xmath10 can decode @xmath11 s message for @xmath12 and @xmath13 , given that @xmath10 knows its own message . latin cubes have been explored as a tool to find the network coding map used by the relay , depending on the channel gain in @xcite . the throughput performance of the two stage protocol for three - way relaying given in @xcite is better than the throughput performance of the ` opportunistic scheduling technique ' given in @xcite at high snr , as can be observed from the plots given in @xcite . the work in @xcite further extends the approach used in @xcite to four - way relaying and employs two channel uses for the entire information exchange amongst the four users , which makes the throughput performance of the scheme better than the other existing schemes . this scheme that utilizes two channel uses for the entire information exchange between three and four users using a relay in @xcite and @xcite respectively , is extended to @xmath0 users in this paper , for the case when @xmath5-psk is used at the end nodes . for our physical layer network coding strategy we use the mathematical structure called a latin hyper - cube defined as follows : an _ n - fold latin hyper - cube l of r - th order of side @xmath5 _ @xcite is an @xmath14 array containing @xmath15 entries , @xmath16 of each of @xmath17 kinds , such that each symbol occurs at most once for each value taken by each dimension of the hyper - cube . for our purposes , we use only @xmath0-fold latin hyper - cubes of side @xmath5 on the symbols from the set @xmath18 , @xmath19 . * _ multiple access ( ma ) phase : _ * + suppose user node @xmath20 for @xmath21 wants to send a @xmath22-bit binary tuple to all user nodes @xmath23 for @xmath24 and @xmath25 . the symmetric @xmath5-psk @xmath26 constellation , given by @xmath27 is used at @xmath7 , @xmath8 , ... , @xmath9 , and @xmath28 denotes the map from bits to complex symbols used at @xmath7 , @xmath8 , ... , @xmath9 where @xmath29 . let @xmath30 denote the complex symbols transmitted by @xmath7 , @xmath8 , ... , @xmath9 respectively , where @xmath31 . here , we assume that the channel state information ( csi ) is not available at the transmitting nodes and perfect csi is available at the receiving nodes . the received signal at r in the ma phase is given by , @xmath32where @xmath33 , @xmath34 , ... , @xmath35 are the fading coefficients associated with the @xmath7-r , @xmath8-r , ... , @xmath9-r link respectively . the additive noise @xmath36 , where @xmath37 denotes the circularly symmetric complex gaussian random variable with variance @xmath38 . the effective constellation seen at the relay during the ma phase , denoted by @xmath39 , is given by , + @xmath40 the minimum distance between the points in the constellation @xmath41 denoted by @xmath42 is given in ( [ dist ] ) on the next page . from ( [ dist ] ) , it is clear that there exists values of @xmath43 , for which @xmath44 . a fade state @xmath45 is defined to be a _ singular fade state _ for the ma phase of @xmath0-way relaying , if @xmath44 . let @xmath46 denote the set of all singular fade states . for singular fade states , @xmath47 . let the maximum likelihood ( ml ) estimate of @xmath48 be denoted by @xmath49 at r based on the received complex number @xmath50 , i.e. , @xmath51where @xmath52 $ ] and @xmath53^t.$ ] @xmath54 * _ broadcast ( bc ) phase : _ * + during the bc phase , the received signals at @xmath55 are respectively given by , @xmath56 where @xmath57 denotes the complex number transmitted by r and @xmath58 @xmath59 ... , @xmath60 respectively are the fading coefficients corresponding to the links r-@xmath7 , r-@xmath8 , ... , r-@xmath9 . the additive noises @xmath61 @xmath62 , ... , @xmath63 are @xmath64 . during the _ bc _ phase , r transmits a point from a signal set @xmath65 given by a many to one map @xmath66 chosen by r , depending on the values of @xmath67 , @xmath68 , ... , @xmath69 . the cardinality of @xmath70 , since @xmath71 bits about the other @xmath3 users needs to be conveyed to each of @xmath7 , @xmath8 , ... , @xmath9 . a _ cluster _ is the set of elements in @xmath72 which are mapped to the same signal point in @xmath65 by the map @xmath73 . let @xmath74 denote the set of all such clusters . the _ cluster distance _ between a pair of clusters @xmath75 , as given in ( [ cl1 ] ) on the next page , is the minimum among all the distances calculated between the points @xmath76 and @xmath77 in the effective constellation seen at the relay node r. the minimum among all the cluster distances among all pairs of clusters of a clustering @xmath78 is the _ minimum cluster distance _ of the clustering , as given in ( [ cl2 ] ) on the next page . during the ma phase , the performance depends on the minimum cluster distance , while during the bc phase , the performance is dependent on the minimum distance of the signal set @xmath65 . _ distance shortening _ , a phenomenon given in @xcite , is described as the significant reduction in the value of @xmath79 for values of @xmath80 in the neighborhood of the singular fade states . if the clustering used at the relay node r in the bc phase is chosen such that @xmath81 is non zero , then the effect of distance shortening can be avoided . a clustering @xmath82 is said to remove a singular fade state @xmath83 if @xmath84 , i.e. , any two message sequences @xmath85 that coincide in the effective constellation received at the relay during the ma phase is in the same cluster of @xmath82 . so , removing singular fade states for a @xmath0-way relay channel can alternatively be defined as : a clustering @xmath82 is said to _ remove the singular fade state _ @xmath86 , if any two possibilities of the messages sent by the users @xmath87 that satisfy @xmath88 are placed together in the same cluster by the clustering . we denote the clustering which removes the singular fade state @xmath89 by @xmath90 ( selecting one randomly if there are multiple clusterings which remove the same singular fade state @xmath89 ) . let the set of all such clusterings be denoted by @xmath91 , i.e. , @xmath92 . the minimum cluster distance of the clustering @xmath93 for @xmath94 , when the fade state @xmath43 occurs in the ma phase , denoted by @xmath95 , is the minimum among all its cluster distances . if @xmath96 the clustering @xmath97 is chosen to be @xmath98 , that satisfies , + @xmath99 + @xmath100 + @xmath101 . the clustering used by the relay is indicated to @xmath7 , @xmath8, ... ,@xmath9 using overhead bits . in order to ensure that @xmath102 is able to decode the message sent by @xmath103 , the clustering @xmath104 should satisfy the exclusive law , as given in ( [ mel ] ) . we explain exclusive law in more detail in the next section . the contributions of this paper are as follows : * we propose a scheme that enables the exchange of information in the wireless @xmath0-way relaying scenario when @xmath5-psk is used at the @xmath0 user nodes with totally two channel uses while attempting to remove the harmful effects of fading , extending the schemes given in @xcite , @xcite , @xcite for @xmath105 respectively . * for this scheme , the singular fade spaces are identified , enumerated and classified based on their removability in the given scenario . the remaining content is organized as follows : section iii demonstrates how a @xmath0-fold latin hyper - cube of side @xmath5 can be utilized to represent the network code that satisfies the exclusive law for @xmath0-way relaying when @xmath5-psk is used at the end nodes . in section iv we describe and enumerate the singular fade subspaces for the given scenario and in section v , focus in on the removal of such singular fade subspaces using @xmath0-fold latin hyper - cube of side @xmath5 . section vi provides some insights using simulations and section vii concludes the paper . the clustering @xmath104 that represents the map used at the relay should satisfy the exclusive law @xcite in order to ensure that @xmath106 is able to decode the message sent by @xmath107 , where we assume that the nodes @xmath108 transmit symbols from the @xmath5-psk constellation . consider a @xmath14 array , containing @xmath15 entries indexed by @xmath109 , i.e. , the @xmath0 symbols sent by @xmath108 in the ma phase . for @xmath21 , fixing the @xmath110 dimension of this @xmath14 array , the @xmath5 @xmath3 dimensional arrays obtained , denoted by say @xmath111 , are indexed by the @xmath5 values taken by @xmath112 . for fixed values of @xmath113 and @xmath114 , the repetition of a symbol in @xmath115 results in the failure of the @xmath110 exclusive law given by ( [ mel ] ) . thus , for the exclusive law to be satisfied , the cells of this array should be filled such that the @xmath14 array so obtained , is a @xmath0-fold latin hyper - cube of side @xmath5 , with entries from @xmath116 for @xmath117 ( definition 1 ) . the symbol @xmath118 of a particular clustering @xmath119 denotes the cluster obtained by putting together all the tuples @xmath120 such that the entry in the @xmath121-th slot is the same entry @xmath122 from @xmath123 . the adjoining figures fig . [ fig : three ] and fig . [ fig : four ] show the exclusive law condition for the three - way and four - way relaying scenario when 4-psk is used at end nodes . a set @xmath124 consisting of all the possibilities of @xmath125 that must be placed in the same cluster of the clustering used at relay node r in the bc phase in order to remove the singular fade state @xmath89 , is referred to as a _ singularity removal constraint _ for the singular fade state @xmath89 for @xmath0-way relaying scenario . at the end of the ma phase , the relay node receives a complex number , given by ( [ yr ] ) . using the ml estimate of this received complex number , r transmits a point from the constellation @xmath126 with cardinality at most @xmath127 . instead of r transmitting a point from the @xmath127 point constellation resulting from all the possibilities of @xmath109 , depending on the fade states , the relay r can choose to group these possibilities into clusters represented by a smaller constellation , so that the minimum cluster distance is non - zero , as well as all the users receive the messages from the other @xmath3 users , i.e. , the clustering satisfies the exclusive law . we provide one such clustering for the case of @xmath0-way relaying in the following . suppose the fade coefficient in the ma phase , @xmath128 , is a singular fade state , and @xmath129 is a singularity removal constraint corresponding to the singular fade state @xmath130 . then there exist @xmath131 , @xmath132 such that , @xmath133 \right\rangle ^{\bot } % \vspace{-.5 cm } \label{nullspacen } \vspace{-.2 cm}\end{aligned}\]]where for a @xmath134 non - zero vector @xmath135 over @xmath136 , @xmath137 note that @xmath138 is the dot product over @xmath136 ( and not an inner product over @xmath136 ) . for a @xmath134 non - zero vector @xmath135 over @xmath136 , @xmath139 is a @xmath3-dimensional vector subspace of @xmath140 over @xmath136 . let @xmath141 where @xmath142 . then , by definition , @xmath143 proving that the following subset of @xmath140 + @xmath144 + is a linearly independent set over @xmath136 of cardinality @xmath3 would be sufficient to prove that @xmath139 is a @xmath3-dimensional vector subspace of @xmath140 . let @xmath145 such that @xmath146comparing the @xmath147 components of the lhs and rhs , we get , @xmath148 . since @xmath149 , where @xmath150 is finite , there are only finitely many possibilities for the right - hand side of ( [ nullspacen ] ) . thus the uncountably infinite singular fade states @xmath43 , are points in a finite number of @xmath3-dimensional vector subspaces of @xmath140 over @xmath136 . we shall refer to these finite number of vector subspaces as the _ singular fade subspaces _ @xcite . we now give a detailed description of all the possibilities of singular fade subspaces for @xmath0-way relaying scenario when @xmath5-psk is used at the end nodes . for the proof of the following theorem , we extensively use the following lemmas given with proofs in @xcite . let @xmath151 denote the difference constellation of the signal set @xmath150 , i.e. , @xmath152 . then , for any @xmath5-psk signal set , @xmath151 is of the form , @xmath153 where @xmath154 and @xmath155 . as a result of the above lemma , the non - zero points in @xmath151 lie on @xmath156 circles of radius @xmath157 with each circle containing @xmath5 points . the phase angles of the @xmath5 points on each circle is @xmath158 , if @xmath114 is odd and @xmath159 if @xmath114 is even , where @xmath160 . @xcite let @xmath161 be the ordered indices corresponding to the non - zero components in @xmath162 and @xmath163 ( the location of non - zero components is the same in the vectors @xmath162 and @xmath163 ) . for @xmath5-psk signal set , @xmath164 , for some @xmath165 , if and only if the magnitudes of the non - zero components in @xmath166 are equal and the magnitudes of the non - zero components in @xmath163 are equal , i.e. , @xmath167 and @xmath168 . from ( [ nullspacen ] ) and lemma 2 , the singular fade subspaces are given by , @xmath169 \right\rangle ^{\bot } = \left\langle \left [ { \begin{array}{cc } 2 \sin ( \frac { \pi l_1}{m } ) e^{j m_1 } \\ 2 \sin ( \frac { \pi l_2}{m } ) e^{j m_2 } \\ . \\. \\ 2 \sin ( \frac { \pi l_n}{m } ) e^{j m_n } \\ \end{array } } \right ] \right\rangle ^{\bot}\ ] ] @xmath170 \right\rangle ^{\bot}\]]where @xmath171 if @xmath172 is odd and @xmath173 if @xmath172 is even , where @xmath174 for @xmath175 . there are @xmath176 m^{k-1}$ ] singular fade subspaces for @xmath0-way relaying when @xmath5-psk constellation is used at the end nodes . the singular fade subspaces are of the form @xmath177\right\rangle^{\bot } $ ] where @xmath178 . let @xmath113 be the number of non - zero @xmath179 . we fix the relative phase vector of the vector @xmath180 = w ~(say)$ ] . the points in @xmath181 lie on @xmath156 circles . so there are @xmath182 possibilities for absolute values of the non - zero components of @xmath183 . there are @xmath156 possibilities for the case that the absolute values of all the components of @xmath183 that are non - zero , are equal . from lemma 3 , the singular fade subspaces resulting from all of these @xmath156 cases are the same , and hence account for 1 out of the @xmath182 cases . so for a fixed relative phase vector , there are @xmath184 $ ] singular fade subspaces . from lemma 3 , fixing the absolute values of the non - zero components of @xmath183 , each distinct relative phase vector corresponds to a distinct singular fade subspace . there are @xmath185 distinct possibilities for the relative phase vector . so , there are @xmath184m^{k-1}$ ] singular fade subspaces when @xmath183 has @xmath113 non - zero components . here , @xmath113 can take values from 1 to @xmath0 . summing over all possible values of @xmath113 , we have @xmath176 m^{n-1}$ ] singular fade subspaces for @xmath0-way relaying when @xmath5-psk constellation is used at the end nodes . the above theorem coincides with the results given for @xmath186 in @xcite , and the results obtained using explicit enumeration for @xmath187 in @xcite and for @xmath188 in @xcite . for illustration , we discuss the case when @xmath189 . in five - way relaying , user nodes ( say ) a , b , c , d and e transmit @xmath190 , @xmath191 , @xmath192 , @xmath193 and @xmath194 @xmath195 respectively in the first channel use . suppose the fade coefficient in the ma phase , @xmath196 , is a singular fade state . then there exist @xmath197 , @xmath198 such that , @xmath199 \right\rangle ^{\bot}. % \vspace{-.5 cm } \label{nullspace } \vspace{-.2 cm}\end{aligned}\ ] ] the adaptive network coding for five - way relaying attempts at removing the singular fade subspaces for the case given by , @xmath200 \right\rangle ^{\bot } $ ] . in the second channel use , relay node r transmits @xmath201 using a network coding map that depends on the values of @xmath202 , @xmath203 , @xmath204 , @xmath205 and @xmath206 . as explained in section iii , using a network coding map represented by a 5-fold hyper latin - cube of side @xmath5 ( when @xmath5-psk is used at the end nodes a , b , c , d and e ) ensures that exclusive law is satisfied . it can be shown using explicit enumeration , that when the end nodes use 4-psk , there are @xmath207 singular fade subspaces for five - way relaying , which coincides with theorem 1 , for @xmath208 . we cluster the possibilities of @xmath209 into a clustering that can be represented by an @xmath0-fold latin hyper - cubes of side @xmath5 , to obtain a clustering that removes the singular fade subspaces , and also attempts to minimize the size of the constellation used by r. this clustering is represented by a constellation given by @xmath126 , which is utilized by the relay node r in the bc phase . this is done by first constraining the @xmath15 possibilities of @xmath209 transmitted at the ma phase , to remove the singular fade subspaces , and then using these constraints , filling the entries of an empty @xmath210 array representing the map to be used at the relay . this partially filled array is completed so as to form a @xmath0-fold latin hyper - cube of side @xmath5 . the mapping to be used at r can be obtained from the complete latin hyper - cube keeping in mind the equivalence between the relay map that satisfies the exclusive law with the @xmath0-fold latin hyper - cube of side @xmath5 as shown in section iii . start with the constrained @xmath211 array @xmath212 initialize all empty cells of @xmath212 to 0 the @xmath213 cell of @xmath212 is the @xmath214 transmission of @xmath7 , the @xmath215 transmission of @xmath8 , ... , the @xmath216 transmission of @xmath9 . during the ma phase for the @xmath0-way relaying scenario , nodes @xmath7 , @xmath8 , ... , @xmath9 transmit to the relay r. let the fade state @xmath43 denote a point in one of the @xmath176 m^{n-1}$ ] singular fade subspaces ( section iv ) . the constraints on the @xmath15 array representing the map at the relay node r during bc phase for a singular fade state , can be obtained using the vectors of differences , viz . , @xmath217 $ ] contributing to this particular singular fade state . so , if @xmath218 \right\rangle^{\bot}$ ] , then , for @xmath219 , @xmath220 . for a clustering to remove the singular fade state @xmath43 , i.e. , for the minimum distance of the clustering to be greater than 0 ( section ii ) , the pair @xmath221 must be kept in the same cluster . alternatively , we can say that the entry corresponding to @xmath109 in the @xmath15 array must be the same as the entry corresponding to @xmath222 . similarly , every other such pair in @xmath223 contributing to this same singular fade subspace must be kept in the same cluster . apart from all such pairs in @xmath223 being kept in the same cluster of the clustering , in order to remove this particular fade state , there are no other constraints . consider the ordering given as follows on the entries of the constrained @xmath15 array : @xmath224 if @xmath225 where @xmath122 is the first component among the @xmath0 components , where @xmath226 . this constrained @xmath15 array can then be completed by simply filling the first empty cell in this order , with @xmath227 in the increasing order of @xmath122 such that the completed array is an @xmath0-fold latin hyper - cube of side @xmath5 ( algorithm 1 ) . the above clustering scheme , however , can not be utilized to remove all the singular fade subspaces , as shown in the following lemma . ! width 1.ptc!width 0.9ptp0.3cm|p0.3cm|p0.3cm|p0.3cm!width 1.ptc!width 0.9ptp0.3cm|p0.32cm|p0.32cm|p0.32cm!width 1.ptc!width 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.32cm!width 1.ptc!width 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.32cm!width 1.pt @xmath228 & & & & & @xmath228 & & & & & @xmath228 & & & & & @xmath228 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath233 & @xmath234 & @xmath235 & @xmath236 & 0 & @xmath237 & @xmath238 & @xmath239 & @xmath240 & 0 & @xmath241 & @xmath242 & @xmath243 & @xmath244 & 0 & @xmath245 & @xmath246 & @xmath247 & @xmath248 + 1 & @xmath249 & @xmath250 & @xmath251 & @xmath252 & 1 & @xmath253 & @xmath254 & @xmath255 & @xmath256 & 1 & @xmath257 & @xmath258 & @xmath259 & @xmath260 & 1 & @xmath261 & @xmath262 & @xmath263 & @xmath264 + 2 & @xmath265 & @xmath266 & @xmath267 & @xmath268 & 2 & @xmath269 & @xmath270 & @xmath271 & @xmath272 & 2 & @xmath273 & @xmath274 & @xmath275 & @xmath276 & 2 & @xmath277 & @xmath278 & @xmath279 & @xmath280 + 3 & @xmath281 & @xmath282 & @xmath283 & @xmath284 & 3 & @xmath285 & @xmath286 & @xmath287 & @xmath288 & 3 & @xmath289 & @xmath290 & @xmath291 & @xmath292 & 3 & @xmath293 & @xmath294 & @xmath295 & @xmath296 + @xmath297 & & & & & @xmath297 & & & & & @xmath297 & & & & & @xmath297 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath298 & @xmath299 & @xmath300 & @xmath301 & 0 & @xmath302 & @xmath303 & @xmath304 & @xmath305 & 0 & @xmath306 & @xmath307 & @xmath308 & @xmath309 & 0 & @xmath310 & @xmath311 & @xmath312 & @xmath313 + 1 & @xmath314 & @xmath315 & @xmath316 & @xmath317 & 1 & @xmath318 & @xmath319 & @xmath320 & @xmath321 & 1 & @xmath322 & @xmath323 & @xmath324 & @xmath325 & 1 & @xmath326 & @xmath327 & @xmath328 & @xmath329 + 2 & @xmath330 & @xmath331 & @xmath332 & @xmath333 & 2 & @xmath334 & @xmath335 & @xmath336 & @xmath337 & 2 & @xmath338 & @xmath339 & @xmath340 & @xmath341 & 2 & @xmath342 & @xmath343 & @xmath344 & @xmath345 + 3 & @xmath346 & @xmath347 & @xmath348 & @xmath349 & 3 & @xmath350 & @xmath351 & @xmath352 & @xmath353 & 3 & @xmath354 & @xmath355 & @xmath356 & @xmath357 & 3 & @xmath358 & @xmath359 & @xmath360 & @xmath361 + @xmath362 & & & & & @xmath362 & & & & & @xmath362 & & & & & @xmath362 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath363 & @xmath364 & @xmath365 & @xmath366 & 0 & @xmath367 & @xmath368 & @xmath369 & @xmath370 & 0 & @xmath371 & @xmath372 & @xmath373 & @xmath374 & 0 & @xmath375 & @xmath376 & @xmath377 & @xmath378 + 1 & @xmath379 & @xmath380 & @xmath381 & @xmath382 & 1 & @xmath383 & @xmath384 & @xmath385 & @xmath386 & 1 & @xmath387 & @xmath388 & @xmath389 & @xmath390 & 1 & @xmath391 & @xmath392 & @xmath393 & @xmath394 + 2 & @xmath395 & @xmath396 & @xmath397 & @xmath398 & 2 & @xmath399 & @xmath400 & @xmath401 & @xmath402 & 2 & @xmath403 & @xmath388 & @xmath404 & @xmath405 & 2 & @xmath406 & @xmath407 & @xmath408 & @xmath409 + 3 & @xmath410 & @xmath411 & @xmath412 & @xmath413 & 3 & @xmath414 & @xmath415 & @xmath416 & @xmath417 & 3 & @xmath418 & @xmath419 & @xmath420 & @xmath421 & 3 & @xmath422 & @xmath423 & @xmath424 & @xmath425 + @xmath426 & & & & & @xmath426 & & & & & @xmath426 & & & & & @xmath426 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath427 & @xmath428 & @xmath429 & @xmath430 & 0 & @xmath431 & @xmath432 & @xmath433 & @xmath434 & 0 & @xmath435 & @xmath436 & @xmath437 & @xmath438 & 0 & @xmath439 & @xmath440 & @xmath441 & @xmath442 + 1 & @xmath443 & @xmath444 & @xmath445 & @xmath446 & 1 & @xmath447 & @xmath448 & @xmath449 & @xmath450 & 1 & @xmath451 & @xmath452 & @xmath453 & @xmath454 & 1 & @xmath455 & @xmath456 & @xmath457 & @xmath458 + 2 & @xmath459 & @xmath460 & @xmath461 & @xmath462 & 2 & @xmath463 & @xmath464 & @xmath465 & @xmath466 & 2 & @xmath467 & @xmath468 & @xmath469 & @xmath470 & 2 & @xmath471 & @xmath472 & @xmath473 & @xmath474 + 3 & @xmath475 & @xmath476 & @xmath477 & @xmath478 & 3 & @xmath479 & @xmath480 & @xmath481 & @xmath482 & 3 & @xmath483 & @xmath484 & @xmath485 & @xmath486 & 3 & @xmath487 & @xmath488 & @xmath489 & @xmath490 + ! width 1.ptc!width 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.32cm!width 1.ptc!width 0.9ptp0.3cm|p0.32cm|p0.32cm|p0.32cm!width 1.ptc!width 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.32cm!width 1.ptc!width 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.32cm!width 1.pt @xmath228 & & & & & @xmath228 & & & & & @xmath228 & & & & & @xmath228 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath491 & @xmath318 & @xmath492 & @xmath493 & 0 & @xmath494 & @xmath314 & @xmath317 & @xmath495 & 0 & @xmath496 & @xmath466 & @xmath342 & @xmath343 & 0 & @xmath323 & @xmath322 & @xmath497 & @xmath498 + 1 & @xmath499 & @xmath302 & @xmath500 & @xmath501 & 1 & @xmath347 & @xmath298 & @xmath502 & @xmath503 & 1 & @xmath504 & @xmath482 & @xmath505 & @xmath359 & 1 & @xmath307 & @xmath306 & @xmath506 & @xmath507 + 2 & @xmath319 & @xmath350 & @xmath508 & @xmath509 & 2 & @xmath315 & @xmath346 & @xmath510 & @xmath511 & 2 & @xmath416 & @xmath310 & @xmath512 & @xmath513 & 2 & @xmath355 & @xmath354 & @xmath514 & @xmath515 + 3 & @xmath516 & @xmath334 & @xmath517 & @xmath518 & 3 & @xmath519 & @xmath330 & @xmath333 & @xmath520 & 3 & @xmath401 & @xmath326 & @xmath327 & @xmath521 & 3 & @xmath339 & @xmath338 & @xmath522 & @xmath523 + @xmath297 & & & & & @xmath297 & & & & & @xmath297 & & & & & @xmath297 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath254 & @xmath253 & @xmath524 & @xmath525 & 0 & @xmath250 & @xmath265 & @xmath526 & @xmath527 & 0 & @xmath262 & @xmath261 & @xmath528 & @xmath529 & 0 & @xmath274 & @xmath530 & @xmath531 & @xmath403 + 1 & @xmath238 & @xmath237 & @xmath532 & @xmath239 & 1 & @xmath234 & @xmath281 & @xmath533 & @xmath534 & 1 & @xmath246 & @xmath245 & @xmath535 & @xmath536 & 1 & @xmath290 & @xmath291 & @xmath537 & @xmath418 + 2 & @xmath286 & @xmath285 & @xmath538 & @xmath287 & 2 & @xmath282 & @xmath539 & @xmath540 & @xmath363 & 2 & @xmath294 & @xmath293 & @xmath541 & @xmath542 & 2 & @xmath243 & @xmath244 & @xmath543 & @xmath372 + 3 & @xmath270 & @xmath269 & @xmath544 & @xmath545 & 3 & @xmath266 & @xmath546 & @xmath547 & @xmath379 & 3 & @xmath278 & @xmath277 & @xmath548 & @xmath549 & 3 & @xmath550 & @xmath260 & @xmath551 & @xmath388 + @xmath362 & & & & & @xmath362 & & & & & @xmath362 & & & & & @xmath362 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath337 & @xmath447 & @xmath552 & @xmath553 & 0 & @xmath444 & @xmath443 & @xmath554 & @xmath555 & 0 & @xmath556 & @xmath345 & @xmath557 & @xmath472 & 0 & @xmath452 & @xmath451 & @xmath558 & @xmath559 + 1 & @xmath353 & @xmath431 & @xmath560 & @xmath433 & 1 & @xmath428 & @xmath427 & @xmath561 & @xmath562 & 1 & @xmath563 & @xmath361 & @xmath564 & @xmath488 & 1 & @xmath436 & @xmath435 & @xmath565 & @xmath437 + 2 & @xmath432 & @xmath479 & @xmath434 & @xmath481 & 2 & @xmath476 & @xmath475 & @xmath289 & @xmath477 & 2 & @xmath313 & @xmath439 & @xmath566 & @xmath567 & 2 & @xmath484 & @xmath483 & @xmath568 & @xmath485 + 3 & @xmath448 & @xmath463 & @xmath450 & @xmath465 & 3 & @xmath460 & @xmath459 & @xmath273 & @xmath461 & 3 & @xmath329 & @xmath455 & @xmath569 & @xmath570 & 3 & @xmath468 & @xmath467 & @xmath571 & @xmath572 + @xmath426 & & & & & @xmath426 & & & & & @xmath426 & & & & & @xmath426 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath384 & @xmath383 & @xmath573 & @xmath574 & 0 & @xmath380 & @xmath395 & @xmath382 & @xmath397 & 0 & @xmath392 & @xmath391 & @xmath394 & @xmath575 & 0 & @xmath576 & @xmath577 & @xmath578 & @xmath579 + 1 & @xmath368 & @xmath367 & @xmath580 & @xmath581 & 1 & @xmath364 & @xmath410 & @xmath582 & @xmath412 & 1 & @xmath376 & @xmath375 & @xmath583 & @xmath303 & 1 & @xmath419 & @xmath405 & @xmath584 & @xmath299 + 2 & @xmath415 & @xmath414 & @xmath417 & @xmath585 & 2 & @xmath411 & @xmath586 & @xmath413 & @xmath242 & 2 & @xmath423 & @xmath422 & @xmath587 & @xmath588 & 2 & @xmath589 & @xmath390 & @xmath590 & @xmath591 + 3 & @xmath400 & @xmath399 & @xmath402 & @xmath592 & 3 & @xmath396 & @xmath593 & @xmath398 & @xmath258 & 3 & @xmath407 & @xmath406 & @xmath409 & @xmath335 & 3 & @xmath374 & @xmath594 & @xmath595 & @xmath596 + ! width 1.ptc!width 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.3cm!width 1.ptc!width 0.9ptp0.3cm|p0.32cm|p0.32cm|p0.32cm!width 1.ptc!width 0.9ptp0.3cm|p0.3cm|p0.3cm|p0.32cm!width 1.ptc!width 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.32cm!width 1.pt @xmath228 & & & & & @xmath228 & & & & & @xmath228 & & & & & @xmath228 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath335 & @xmath387 & @xmath597 & @xmath467 & 0 & @xmath331 & @xmath572 & @xmath468 & @xmath563 & 0 & @xmath593 & @xmath379 & @xmath410 & @xmath396 & 0 & @xmath461 & @xmath353 & @xmath598 & @xmath400 + 1 & @xmath372 & @xmath371 & @xmath599 & @xmath589 & 1 & @xmath421 & @xmath485 & @xmath600 & @xmath556 & 1 & @xmath586 & @xmath363 & @xmath601 & @xmath411 & 1 & @xmath477 & @xmath337 & @xmath602 & @xmath415 + 2 & @xmath388 & @xmath418 & @xmath603 & @xmath577 & 2 & @xmath559 & @xmath438 & @xmath604 & @xmath391 & 2 & @xmath412 & @xmath366 & @xmath605 & @xmath427 & 2 & @xmath581 & @xmath370 & @xmath606 & @xmath431 + 3 & @xmath303 & @xmath403 & @xmath607 & @xmath435 & 3 & @xmath299 & @xmath454 & @xmath313 & @xmath376 & 3 & @xmath397 & @xmath382 & @xmath380 & @xmath443 & 3 & @xmath574 & @xmath386 & @xmath608 & @xmath447 + @xmath297 & & & & & @xmath297 & & & & & @xmath297 & & & & & @xmath297 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath482 & @xmath519 & @xmath373 & @xmath416 & 0 & @xmath470 & @xmath592 & @xmath609 & @xmath516 & 0 & @xmath398 & @xmath596 & @xmath610 & @xmath460 & 0 & @xmath402 & @xmath465 & @xmath611 & @xmath612 + 1 & @xmath466 & @xmath401 & @xmath613 & @xmath273 & 1 & @xmath486 & @xmath585 & @xmath614 & @xmath407 & 1 & @xmath413 & @xmath591 & @xmath615 & @xmath476 & 1 & @xmath417 & @xmath481 & @xmath616 & @xmath617 + 2 & @xmath258 & @xmath494 & @xmath618 & @xmath452 & 2 & @xmath575 & @xmath491 & @xmath619 & @xmath455 & 2 & @xmath562 & @xmath430 & @xmath620 & @xmath367 & 2 & @xmath480 & @xmath433 & @xmath621 & @xmath622 + 3 & @xmath242 & @xmath437 & @xmath623 & @xmath375 & 3 & @xmath588 & @xmath378 & @xmath624 & @xmath471 & 3 & @xmath555 & @xmath446 & @xmath625 & @xmath383 & 3 & @xmath464 & @xmath553 & @xmath626 & @xmath627 + @xmath362 & & & & & @xmath362 & & & & & @xmath362 & & & & & @xmath362 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath291 & @xmath550 & @xmath628 & @xmath338 & 0 & @xmath357 & @xmath542 & @xmath629 & @xmath326 & 0 & @xmath546 & @xmath268 & @xmath630 & @xmath266 & 0 & @xmath545 & @xmath272 & @xmath631 & @xmath334 + 1 & @xmath530 & @xmath243 & @xmath632 & @xmath354 & 1 & @xmath549 & @xmath280 & @xmath633 & @xmath342 & 1 & @xmath539 & @xmath284 & @xmath634 & @xmath282 & 1 & @xmath287 & @xmath288 & @xmath635 & @xmath350 + 2 & @xmath260 & @xmath292 & @xmath306 & @xmath307 & 2 & @xmath536 & @xmath529 & @xmath499 & @xmath358 & 2 & @xmath534 & @xmath236 & @xmath636 & @xmath298 & 2 & @xmath239 & @xmath240 & @xmath637 & @xmath238 + 3 & @xmath244 & @xmath276 & @xmath322 & @xmath323 & 3 & @xmath638 & @xmath248 & @xmath310 & @xmath311 & 3 & @xmath527 & @xmath252 & @xmath639 & @xmath314 & 3 & @xmath525 & @xmath256 & @xmath640 & @xmath254 + @xmath426 & & & & & @xmath426 & & & & & @xmath426 & & & & & @xmath426 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath523 & @xmath289 & @xmath641 & @xmath277 & 0 & @xmath296 & @xmath521 & @xmath359 & @xmath327 & 0 & @xmath520 & @xmath333 & @xmath330 & @xmath269 & 0 & @xmath351 & @xmath509 & @xmath642 & @xmath643 + 1 & @xmath515 & @xmath341 & @xmath644 & @xmath293 & 1 & @xmath343 & @xmath645 & @xmath646 & @xmath647 & 1 & @xmath511 & @xmath349 & @xmath648 & @xmath285 & 1 & @xmath518 & @xmath305 & @xmath649 & @xmath650 + 2 & @xmath507 & @xmath309 & @xmath245 & @xmath246 & 2 & @xmath264 & @xmath651 & @xmath361 & @xmath652 & 2 & @xmath503 & @xmath301 & @xmath653 & @xmath237 & 2 & @xmath501 & @xmath321 & @xmath654 & @xmath655 + 3 & @xmath498 & @xmath325 & @xmath261 & @xmath262 & 3 & @xmath513 & @xmath656 & @xmath657 & @xmath658 & 3 & @xmath495 & @xmath317 & @xmath315 & @xmath253 & 3 & @xmath493 & @xmath659 & @xmath660 & @xmath661 + ! 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.3cm!width 1.ptc!width 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.32cm!width 1.ptc!width 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.32cm!width 1.ptc!width 0.9ptp0.32cm|p0.32cm|p0.32cm|p0.32cm!width 1.pt @xmath228 & & & & & @xmath228 & & & & & @xmath228 & & & & & @xmath228 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath361 & @xmath486 & @xmath662 & @xmath422 & 0 & @xmath570 & @xmath650 & @xmath663 & @xmath483 & 0 & @xmath481 & @xmath417 & @xmath647 & @xmath479 & 0 & @xmath664 & @xmath413 & @xmath665 & @xmath475 + 1 & @xmath409 & @xmath470 & @xmath666 & @xmath406 & 1 & @xmath627 & @xmath643 & @xmath667 & @xmath668 & 1 & @xmath465 & @xmath402 & @xmath669 & @xmath463 & 1 & @xmath670 & @xmath398 & @xmath671 & @xmath459 + 2 & @xmath378 & @xmath394 & @xmath672 & @xmath487 & 2 & @xmath437 & @xmath661 & @xmath673 & @xmath436 & 2 & @xmath553 & @xmath450 & @xmath674 & @xmath675 & 2 & @xmath477 & @xmath555 & @xmath676 & @xmath444 + 3 & @xmath345 & @xmath567 & @xmath677 & @xmath392 & 3 & @xmath405 & @xmath655 & @xmath452 & @xmath451 & 3 & @xmath433 & @xmath434 & @xmath678 & @xmath679 & 3 & @xmath430 & @xmath562 & @xmath680 & @xmath428 + @xmath297 & & & & & @xmath297 & & & & & @xmath297 & & & & & @xmath297 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath485 & @xmath425 & @xmath681 & @xmath484 & 0 & @xmath390 & @xmath478 & @xmath682 & @xmath683 & 0 & @xmath684 & @xmath490 & @xmath685 & @xmath414 & 0 & @xmath686 & @xmath412 & @xmath687 & @xmath688 + 1 & @xmath572 & @xmath689 & @xmath690 & @xmath468 & 1 & @xmath577 & @xmath462 & @xmath691 & @xmath692 & 1 & @xmath693 & @xmath474 & @xmath694 & @xmath399 & 1 & @xmath695 & @xmath397 & @xmath696 & @xmath289 + 2 & @xmath454 & @xmath421 & @xmath697 & @xmath456 & 2 & @xmath698 & @xmath699 & @xmath700 & @xmath419 & 2 & @xmath386 & @xmath574 & @xmath701 & @xmath384 & 2 & @xmath382 & @xmath702 & @xmath703 & @xmath410 + 3 & @xmath438 & @xmath559 & @xmath704 & @xmath440 & 3 & @xmath705 & @xmath589 & @xmath706 & @xmath576 & 3 & @xmath370 & @xmath581 & @xmath707 & @xmath368 & 3 & @xmath366 & @xmath708 & @xmath709 & @xmath395 + @xmath362 & & & & & @xmath362 & & & & & @xmath362 & & & & & @xmath362 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath710 & @xmath711 & @xmath712 & @xmath651 & 0 & @xmath341 & @xmath515 & @xmath713 & @xmath355 & 0 & @xmath321 & @xmath518 & @xmath714 & @xmath715 & 0 & @xmath349 & @xmath511 & @xmath716 & @xmath346 + 1 & @xmath594 & @xmath296 & @xmath717 & @xmath657 & 1 & @xmath718 & @xmath523 & @xmath719 & @xmath339 & 1 & @xmath509 & @xmath720 & @xmath721 & @xmath722 & 1 & @xmath333 & @xmath520 & @xmath723 & @xmath330 + 2 & @xmath327 & @xmath724 & @xmath656 & @xmath659 & 2 & @xmath325 & @xmath498 & @xmath249 & @xmath261 & 2 & @xmath725 & @xmath493 & @xmath726 & @xmath318 & 2 & @xmath317 & @xmath495 & @xmath727 & @xmath728 + 3 & @xmath729 & @xmath264 & @xmath343 & @xmath730 & 3 & @xmath309 & @xmath507 & @xmath233 & @xmath245 & 3 & @xmath305 & @xmath501 & @xmath731 & @xmath318 & 3 & @xmath301 & @xmath503 & @xmath732 & @xmath733 + @xmath426 & & & & & @xmath426 & & & & & @xmath426 & & & & & @xmath426 & & & & + @xmath229 & & & & & @xmath230 & & & & & @xmath231 & & & & & @xmath232 & & & & + 0 & @xmath280 & @xmath549 & @xmath734 & @xmath294 & 0 & @xmath292 & @xmath735 & @xmath260 & @xmath736 & 0 & @xmath288 & @xmath287 & @xmath350 & @xmath239 & 0 & @xmath284 & @xmath534 & @xmath737 & @xmath401 + 1 & @xmath542 & @xmath357 & @xmath738 & @xmath278 & 1 & @xmath276 & @xmath739 & @xmath740 & @xmath741 & 1 & @xmath272 & @xmath516 & @xmath742 & @xmath270 & 1 & @xmath268 & @xmath527 & @xmath743 & @xmath519 + 2 & @xmath248 & @xmath563 & @xmath326 & @xmath744 & 2 & @xmath745 & @xmath746 & @xmath323 & @xmath290 & 2 & @xmath256 & @xmath525 & @xmath747 & @xmath748 & 2 & @xmath252 & @xmath749 & @xmath750 & @xmath281 + 3 & @xmath529 & @xmath536 & @xmath622 & @xmath751 & 3 & @xmath752 & @xmath753 & @xmath307 & @xmath274 & 3 & @xmath240 & @xmath239 & @xmath238 & @xmath491 & 3 & @xmath236 & @xmath754 & @xmath755 & @xmath494 + the clustering map used at the relay node r can not remove the singular fade spaces @xmath756\right\rangle $ ] where at least one of @xmath757 for some @xmath21 and simultaneously satisfy the mutually exclusive law . let @xmath758\right\rangle^ { \bot } $ ] be a singular fade state where for some @xmath759 , @xmath760 . then , in order to remove @xmath150 , @xmath761 and @xmath762 that must be kept in the same cluster . this would imply user @xmath20 not being able to distinguish between the messages @xmath763 and @xmath764 for some @xmath765 sent by user @xmath23 . this will clearly violate the mutually exclusive law , since in order to satisfy the mutually exclusive law , for the same value of @xmath766 , all possible @xmath0-tuples of messages must be kept in different clusters . these two statements can not be satisfied at the same time , hence such a singular fade subspace can not be removed if the mutually exclusive law has to be satisfied by the relay map used in the bc phase . we refer to the singular fade subspaces whose harmful effects can not be removed by a proper choice of the clustering , as the _ non - removable singular fade subspaces _ also talked about in @xcite . there are @xmath767 m^{n-1}$ ] removable and @xmath768 m^{k-1}$ ] non - removable singular fade subspaces for @xmath0-way relaying when @xmath5-psk constellation is used at the end nodes . the number of non - removable singular subspaces is @xmath769 while the number of removable singular fade subspaces is @xmath770 . thus , the number of non - removable singular fade subspaces is a small fraction of the total number of singular fade subspaces . for the five - way relaying scenario described in the previous section , there are @xmath207 singular fade subspaces for five - way relaying , out of which the scheme given in this paper removes @xmath771 singular fade subspaces using @xmath772-fold latin hyper - cubes of side @xmath773 . this can be done , as described above , by first marking the singularity removal constraints in the empty @xmath774 array and then completing the array to form a @xmath772-fold latin hyper - cubes of side @xmath773 using algorithm 1 . we now illustrate the removal of a singular fade state for five - way relaying with the help of the following example . consider the case for five - way relaying scenario where 4-psk is used at end nodes a , b , c , d and e , and one of the singular fade subspace to be removed is given by , @xmath775\right\rangle ^{\bot}\hspace{-0.3 cm } ~=~\hspace{-0.1cm}\left\langle \left [ { \begin{array}{c } 1-j \\ 2 \\ 2 \\ 1+j \\ -1+j \\ \end{array } } \right]\right\rangle ^{\bot}\hspace{-0.3cm}\ ] ] @xmath776\right\rangle ^{\bot}\hspace{-0.3 cm } ~=~\hspace{-0.1cm}\left\langle \left [ { \begin{array}{c } -1+j \\ -2 \\ -2 \\ -1-j \\ 1-j \\ \end{array } } \right]\right\rangle ^{\bot}.\ ] ] the first vector is @xmath777 $ ] . now , @xmath778 can be obtained either as a difference of @xmath779 and @xmath780 or as a difference of @xmath781 and @xmath782 ; @xmath783 can be obtained only as a difference of @xmath784 and @xmath785 ; @xmath786 can be obtained as a difference of @xmath230 and @xmath787 or as a difference of @xmath788 and @xmath789 ; @xmath790 can be obtained as a difference of @xmath791 and @xmath792 or as a difference of @xmath793 and @xmath794 . thus , the entries corresponding to @xmath795 , @xmath796 , @xmath797 , @xmath798 , @xmath799 , @xmath800 , @xmath801 , @xmath802 must lie in the same clustering representing the network coding map used at the relay node in the bc phase . similarly the constraints resulting from the other three vectors above can be obtained . replacing @xmath803 with @xmath804 , we get constraints in the form of entries that must be kept the same in the @xmath774 array representing the clustering . for instance , corresponding to @xmath795 , the entries @xmath805 are kept in the same cluster @xmath496 as shown in bold in tables i , ii , iii , iv . the constrained hyper - cube of side 4 is completed using algorithm 1 to form a 4-fold latin hyper - cube of side 4 . the completed latin hyper cube is as shown in tables i , ii , iii and iv , where the constraints are marked in bold . similarly , a ( removable ) singular fade subspace can be removed by first constraining the array representing the relay map , and then completing the constrained array using the provided algorithm . simulation results presented in this section identify the cases where the proposed scheme outperforms the naive approach that uses the same map for all fade states and vice verse . here , the channel states are distributed according to rician distribution and channel variances equal to 0 db and the frame length is 256 bits . [ fig : comparison_plot ] compares the snr vs bit - error - rate curves for three - way , four - way and five - way relaying scenario for ( a ) the adaptive network coding scheme presented in this paper with ( b ) the non - adaptive network coding using two channel uses , in which the same array is used by the relay as an encoder for all channel conditions . the details of three - way and four - way relaying can be found in @xcite and @xcite . the non - adaptive network coding for three - way , four - way and five - way relaying utilizes the same @xmath806-fold latin hyper - cubes of side @xmath773 , @xmath773-fold latin hyper - cube of side @xmath773 and @xmath772-fold latin hyper - cube of side @xmath773 ( respectively ) for all channel conditions . for any communication system , the effects of additive noise are predominant at low snr and the effects of noise due to fading dominate at high snr . since adaptive network coding scheme attempts at reducing the effects of fading for the @xmath0-way relaying scenario by removing a part of singular fade states , ( a ) performs better than ( b ) at high snr . this also implies that the performance of the scheme in the ma phase ( removal of singular fade spaces takes place in the ma phase ) is predominant at higher snrs , and the performance of the scheme in the bc phase dominates at lower snrs . also , as the number of user nodes @xmath0 increases , the snr at which the performance of adaptive network coding improves over the performance of non - adaptive network coding increases , as can be seen in the plot , since the size of the received constellation in the ma phase and hence the size of the constellation used in the bc phase increases with increasing values of @xmath0 . we consider the @xmath0-way wireless relaying scenario , where @xmath0 nodes operate in half - duplex mode and transmit points from the same @xmath5-psk constellation . information exchange is made possible using just two channels uses , unlike the existing work done for the case , to the best of our knowledge . the relay node clusters the @xmath15 possible transmitted tuples @xmath209 into various clusters depending on the fade states such that _ the exclusive law _ is satisfied and some of the singular fade subspaces are removed . this necessary requirement of satisfying the exclusive law is shown to be the same as the clustering being represented by a @xmath0-fold latin hyper - cube of side @xmath5 . the size of the clustering utilizing modified clustering may not be the best that can be achieved , and it might be possible to fill the array with lesser symbols . the first author would like to thank her colleague vijayvaradharaj t muralidharan at the department of electrical and communication engineering , indian institute of science for his feedback and suggestions related to this paper . t. koike - akino , p. popovski and v. tarokh , `` optimized constellation for two - way wireless relaying with physical network coding '' , ieee journal on selected areas in comm . , vol.27 , pp . 773787 , june 2009 . srishti shukla , vijayvaradharaj t. muralidharan and b. sundar rajan , `` wireless network - coded three - way relaying using latin cubes '' , 23rd ieee personal indoor mobile radio communications ( pimrc)-2012 , sydney , australia . mylene pischella and didier le ruyet , `` lattice based coding scheme for mimo bi - directional relaying with three nodes '' , 22nd ieee personal indoor mobile radio communications , toronto , canada , pp . 14591463 , sept . 2011 . youngil jeon , young - tae kim , moonseo park , inkyu lee , `` opportunistic scheduling for three - way relay systems with physical layer network coding '' , ieee 73rd vehicular technology conference ( vtc - spring ) 2011 , budapest , hungary , pp . 15 , may 2011 . vijayvaradharaj t muralidharan , vishnu namboodiri and b. sundar rajan , `` channel quantization for physical layer network - coded two - way relaying '' , available online at arxiv:1109.6101v2 [ cs.it ] , 16 nov .

we investigate the star formation history of both the bright star clusters and the diffuse ` field star ' population in the dwarf starburst galaxy ngc 5253 using longslit ultraviolet spectroscopy obtained with the space telescope imaging spectrograph ( stis ) . the slit covers a physical area of 370.6 pc and includes 8 apparent clusters and several inter - cluster regions of diffuse light which we take to be the field . the diffuse light spectrum lacks the strong o - star wind features which are clearly visible in spectra of the brightest clusters . this discrepancy provides compelling evidence that the diffuse light is not reflected light from nearby clusters , but originates in a uv - bright field star population , and it raises the issue of whether the star formation process may be operating differently in the field than in clusters . we compare our spectra to starburst99 evolutionary synthesis models which incorporate a new low metallicity (/4 z ) atlas of o - star spectra . the clusters are well fit by instantaneous burst models with a salpeter initial mass function ( imf ) extending up to 100 m , and we derive ages for them ranging from 1 to 8 myrs . reasonable fits to the field spectrum are obtained by continuous star formation models with either an upper mass cut - off of m or an imf slope steeper than salpeter ( ) . we favor a scenario which accounts for the paucity of o - stars in the field without requiring the field to have a different imf than the clusters : stellar clusters form continuously and then dissolve on myr timescales and disperse their remaining stars into the field . we consider the probable contribution of an o - star deficient field population to the spatially unresolved spectra of high redshift galaxies .

astro-ph0103432

the energy produced in the star formation process is distributed over some 10 decades of frequency , from the x - ray to the radio ; however , the bulk of our knowledge about high redshift ( z @xmath4 2.5 ) star forming galaxies comes from a rather limited observational window : the rest - frame ultraviolet . this is a consequence of the simple fact that high redshift galaxies are faint ( at z@xmath13 @xmath5 corresponds to r @xmath6 24.5 , @xcite ) and thus accessible only in spectral regimes where the sensitivity is optimum . at present most observations of high - z galaxies are made by large ground - based telescopes working in the visible , which corresponds to the rest - frame ultraviolet for galaxies at redshifts larger than @xmath12.5 . fortuitously , the ultraviolet ( uv ) is an excellent regime in which to study star forming galaxies : it contains the direct spectroscopic signatures of the hot young massive stars which dominate the bolometric light , as well as an abundance of interstellar features that provide valuable probes of the gas . in fact , almost all of the relevant stellar and interstellar absorption lines have wavelengths below 3000 . the full diagnostic power of this rich spectral region is still being realized : a number of programs are underway to correlate uv spectral morphology with global galaxy parameters such as mass , metallicity , extinction , stellar content , and bolometric luminosity ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? these studies concentrate on local starburst galaxies whose global properties are well known from observations in a number of wavebands . nearby starburst galaxies are the most obvious local counterparts of the high redshift galaxies currently being found in large numbers through color selection based on the lyman break technique ( e.g. * ? ? ? * ; * ? ? ? while the extinction corrections to the rest - frame ultraviolet of high - z galaxies are still a matter of some debate , it is clear that the stellar surface mass densities and star formation rates per unit area of these galaxies are several orders of magnitude higher than ordinary late type galaxies , and are comparable to those of local starbursts @xcite . local starbursts and lyman - dropout galaxies share similar distributions of observed uv colors @xcite and have spectral morphologies which are characterized by blue continua with strong interstellar absorption , the broad p - cygni signatures of stellar winds , and weak ly-@xmath7 emission @xcite . high - resolution uv spectra of local starburst galaxies provide invaluable templates for comparison with spectra of high - z galaxies . however , _ such comparisons must be viewed with a note of caution because of differences in spatial resolution_. high redshift galaxies have small angular sizes and spectra which are spatially integrated , whereas local starbursts have such large angular scales that only select pieces of the burst can be sampled without using apertures so large as to substantially degrade the spectral resolution . differences in the spatial resolution of high and low - z observations are particularly significant because bursts of star formation are known to be highly inhomogeneous . space - based uv imaging has shown nearby star forming regions to be composed of compact , young , uv - bright stellar clusters embedded in a diffuse , irregular , uv background @xcite . historically , high - resolution uv spectroscopy of local starburst galaxies has focused on individual stellar clusters because of their high surface brightness . however , _ the uv - bright clusters are not the dominant contributors to the uv luminosity of a starburst galaxy : _ the diffuse uv light typically accounts for @xmath1 50% to 80% of the total @xcite . the nature of the diffuse uv component is hinted at by uv images of the nearest starbursts where individual high mass stars are resolved @xcite . these stars appear to trace the diffuse light isophotes , suggesting that the diffuse light is not simply reflected light from the bright clusters , but originates in a uv - bright field star population . this raises the issue of whether or not there are two distinct ` modes ' of star formation operating in a starburst : one which produces compact clusters and one which produces more diffuse star formation . what is clear is that a starburst is not the same thing as a star cluster ; nor is it the sum of its clusters . the diffuse light represents a large , if not dominant , part of the integrated light of a starburst , and may represent an important and distinct mode of star formation . understanding the nature of the diffuse uv component is one of the principal aims of our present investigation which utilizes stis longslit ultraviolet spectroscopy of the central starburst in the nearby galaxy ngc 5253 . by addressing the contribution that both the clusters and the diffuse light make to the integrated spectrum of a local galaxy , we hope to be able to more accurately interpret the spatially unresolved spectra of high - z star forming galaxies . ngc 5253 is a metal - poor dwarf galaxy in the centaurus group which contains an extremely young starburst . a re - calibration of existing cepheid data by the _ hst _ key project on the extragalactic distance scale team has produced an improved distance estimate to the galaxy of 3.33 @xmath8 mpc ( @xcite ; 1 @xmath9 16 pc ) . ngc 5253 has been the target of a number of observational programs at wavelengths ranging from the x - ray to the radio , the results of which are summarized by @xcite , @xcite , and @xcite . the metallicity of ngc 5253 is @xmath10 z@xmath2 @xcite and is relatively constant across the galaxy , with the exception of a few areas of enhanced nitrogen abundances @xcite . ngc 5253 s small size and modest metallicity make it an excellent analogue to galaxies in their early phases of formation , while its proximity makes it ideal for disentangling the diffuse uv light from that of the bright stellar clusters . our long slit traverses a relatively large portion of ngc 5253 , as can be seen in figure [ image ] . morphologically the galaxy is somewhat unusual ( termed ` amorphous ' by @xcite ) . various authors have suggested that ngc 5253 may have resembled a dwarf elliptical until an encounter with ngc 5236 ( m83 ) one to two gyrs ago triggered the formation of a new stellar population @xcite . star formation is ongoing in the central 20 ( @xmath1325 pc ) of the galaxy which contains a dozen uv - bright stellar clusters @xcite and is very blue except where dust produces patchy , heavy obscuration @xcite . a prominent dust lane with a@xmath11 mags girds the galaxy nearly perpendicular to its major axis @xcite . radio observations suggest that a large fraction of the most recent star formation is hidden by dust @xcite . the ionized gas in ngc 5253 is more extended by a factor of @xmath12 than the continuum emission , and it is circularly symmetric about a stellar cluster ( ngc5253 - 5 of @xcite ) which is located near the geometric center of the galaxy . the most active region of star formation is the starburst nucleus , which is 40 50 pc in size and centered at the position of the h@xmath7 peak . an age as young as 5 myr is suggested by the large h@xmath7 equivalent width @xcite , the presence of wolf - rayet stars @xcite , the small number of red supergiants @xcite , and the purely thermal component of the radio emission @xcite . stars in this nuclear region are still embedded in the parental cloud with a@xmath12 9 35 mag @xcite . recent observations at 1.3 and 2 cm have revealed the presence of a @xmath112 pc ` supernebula ' a very young and gas - rich counterpart of a super - star cluster which may be the youngest globular cluster known @xcite . the star formation rate density in the nuclear region is between @xmath13 and @xmath14m@xmath2 yr@xmath15 pc@xmath16 @xcite , which corresponds to the maximum levels observed in starburst galaxies @xcite . however , the bulk of the observed uv light is not produced by the starburst nucleus , but by the surrounding regions ( extending out to 250 pc ) , which are older , relatively unextincted , and producing stars at a rate 10 times lower than that of the nuclear starburst @xcite . we obtained longslit ultraviolet spectra of the central region of ngc 5253 with the space telescope imaging spectrograph ( stis ) aboard the _ hubble space telescope _ on 1999 july 27 . we utilized the 52@xmath0 0.1 slit and the g140l grating to obtain the best compromise between throughput and spectral resolution . the spectra span a wavelength range of 1150 1700 with a resolution of @xmath12 , which provides a good match to our stellar evolutionary synthesis models ( 1200 1600 ; @xmath17 ) . the position of the slit was chosen so as to intersect clusters 2 , 3 , and 5 of calzetti et al . 1997 ( see figure [ image ] ) . target acquisition was accomplished in a two phase process utilizing the stis ccd at visible wavelengths . first , a bright foreground star 14 from cluster 5 was acquired in a 5 field of view and centered on the ccd to an accuracy of @xmath18 001 . an offset was then performed to cluster 5 ( ra = 13395602 , dec = -3138250 , j2000 ) and a peakup was done to refine the centering . the spacecraft moved the slit across the target parallel to the dispersion direction in five 0075 steps . the optimal spacecraft location ( accurate to 5% of the slit , @xmath10005 ) was determined from a flux weighted centroid algorithm . the slit orientation ( pa = 56246 ) was dictated by clusters 2 and 3 , which are separated by @xmath19 and are @xmath18 distant from cluster 5 . following the successful target acquisition , the g140l grating was moved into the beam and four frames of duration 1410 , 2945 , 2945 , and 2945 seconds were obtained with the stis fuv - mama detector . the data were processed with the space telescope science institute s _ calstis _ pipeline , which provides automated rebinning , global detector linearity correction , dark subtraction , flat fielding , wavelength calibration , and conversion to absolute flux units . for our subsequent analysis we utilized the fully calibrated two - dimensional images produced by the pipeline , which are rectified to be linear in both the wavelength and spatial dimension . after examining the four calibrated frames for spatial shifts , we combined them into a single two - dimensional image using the iraf task mscombine . the slit traverses roughly 370 pc of ngc 5253 , including both the intense regions of star formation near the nucleus and the more quiescent outer regions ( see figure [ image ] ) . the spatial dimension of the data extends 23 with an image scale of 00244 pixel@xmath15 ( @xmath10.4 pc pixel@xmath15 ) , which is sufficient to resolve stellar clusters . figure [ spatial ] shows the integrated 1150 1710 flux as a function of slit position . a number of bright point - like sources are visible , embedded within a @xmath1300 pc plateau of diffuse light . for our subsequent analysis , we define the 8 brightest peaks along the spatial axis to be clusters , and take regions between those peaks to be the field or diffuse component . ( we use the terms ` cluster ' and ` field ' quite liberally here . ) data within these subregions were binned into one - dimensional spectra . the spatial pixels binned up to form a given cluster s spectrum were chosen by eye from an examination of the spectrally integrated data shown in figure [ spatial ] . the spectral properties of the fainter clusters were fairly sensitive to the boundaries chosen . we made a concerted effort to balance the competing demands of signal - to - noise ( s / n ) and spectral purity by examining the effects that broadening the extraction had on the resultant spectrum . some of our so - called clusters may nevertheless include more than one coeval population ( note the two component blended profile of cluster 5 in figure [ spatial ] ) . the extracted spectra of the four brightest clusters are shown in figure [ bright_clusters ] . because of their greater extent , the field regions were not nearly as sensitive to their spatial boundaries . however , subtle differences were apparent when large areas were considered . we therefore defined three distinct field regions ( shown in figure [ spatial ] ) , each with a s / n comparable to that of the brightest clusters . the extracted field spectra are shown in figure [ fields ] . the low redshift of ngc 5253 ( @xmath20 km s@xmath15 ) precluded direct measurement of our spectral resolution because the strong interstellar features of the starburst are blended with intervening milky way absorption lines . we therefore estimate our resolution based on the known properties of stis . at the distance of ngc 5253 , a typical cluster is neither a point source nor a fully extended source , so we expect our spectral resolution to be intermediate between point and extended source values ( 0.9 and 2.4 respectively ) . we estimate a value of @xmath11.8 as an appropriate resolution for our cluster spectra . our field spectra should have the resolution of a fully extended source . we measure a 2.2 fwhm for the lyman-@xmath7 airglow line . because the clusters are embedded in a diffuse background and are resolved , we treated all of our spectra as extended continuum sources when converting from units of surface brightness in the two - dimensional image to units of flux ( ergs s@xmath15 cm@xmath16 @xmath15 ) in the one - dimensional extracted spectra . treating the clusters as point sources would increase their measured fluxes by approximately a factor of two due to corrections for the extended wings of the psf . the extracted spectra were corrected for slight relative wavelength shifts ( 0.02 0.2 ) based on the peak geocoronal ly-@xmath7 emission , de - redshifted , and then rebinned to the model resolution of 0.75 per pixel . we then corrected the spectra for the strong geocoronal emission at ly-@xmath7 and @xmath22 . data in a region at the edge of the diffuse light plateau ( the region labeled `` sky '' in figure [ spatial ] ) were extracted to produce a one - dimensional spectrum . while this spectrum included substantially more pixels than any of the field regions , the stellar features were extremely faint , making it a suitable `` sky '' spectrum . we subtracted a weak continuum from the sky spectrum and set all the pixels to zero except for a 40 window around ly-@xmath7 and a 20 window around . taking this to be a purely geocoronal spectrum , we then subtracted it from the 1-dimensional spectra of the cluster and field regions in proportion to the number of pixels covered by each . the intrinsic interstellar absorption lines of and near @xmath23 appear to be fairly well recovered . extinction corrections were applied individually to each of the unrectified spectra . we removed the foreground galactic extinction with the standard milky way extinction law of @xcite , using a reddening of e ( ) = 0.05 @xcite for all of the spectra . the extinction intrinsic to the starburst was removed using the starburst obscuration curve ( * ? ? ? * and references therein ) . assuming a foreground dust geometry with a standard extinction curve ( e.g. the milky way or lmc curve ) would generally give a similar or lower color excess than the values derived with the starburst obscuration curve . however , the differences would be small in the present case , owing to the modest reddening affecting the stellar clusters ( with the exception of cluster 5 which is discussed further in [ clusters ] , and possibly cluster 8 . ) we used @xmath24 , the power law slope of the uv continuum ( @xmath25 ) , as a diagnostic of the reddening internal to the starburst . because of the plethora of weak lines in the spectra , the value of @xmath24 measured was fairly sensitive to the method of continuum fitting . we used 10 iterations of a sigma rejection routine with a 2.0@xmath26 lower bound and a 3.5@xmath26 upper bound to perform a power law fit over the wavelength range 1240 1600 . this method produces systematically shallower slopes than the method of @xcite in which a fit is performed over 10 line - free windows ( only 5 of which are in our wavelength range . ) however , in this instance we are not concerned with the absolute value of @xmath24 , merely the difference in @xmath24 between the data and the models , which are free of reddening . for a wide range of model parameters ( detailed in [ modeling ] ) , we measure @xmath24 values of -2.50 to -2.75 . since our measurements are probably not accurate to more than @xmath27 , we take -2.6 to be the intrinsic slope of the uv continuum for all of our spectra . we derive e ( ) values for each of the cluster and field regions by de - reddening the observed spectrum using the starburst obscuration law until the measured value of @xmath24 is -2.6 . the observed @xmath24 values for each of the cluster and field regions are listed in tables [ table_cluster_data ] and [ table_field_data ] . table [ table_cluster_results ] contains the @xmath24 values of the clusters corrected for galactic foreground reddening and the derived color excess . we compare the spectra of the clusters and the field to the _ starburst99 _ stellar evolutionary synthesis models @xcite . the _ starburst99 _ models ( hereafter sb99 ) have been optimized to reproduce many spectrophotometric properties of galaxies with active star formation . we briefly describe those aspects of the models which we utilize in [ sb99 ] ; in [ modelpars ] we describe our input model parameters ; and in [ modelfitting ] we describe our fitting routine . the sb99 models use two limiting cases of the star - formation law : an instantaneous burst with no subsequent star formation and star formation proceeding continuously at a constant rate . stellar populations are parameterized by their stellar initial mass function ( imf ) , which is assumed to be a power law ( @xmath28 ) between a lower mass cut - off ( m@xmath29 ) and an upper mass cut - off ( m@xmath30 ) . the stellar population is evolved from the zams using the evolutionary models of the geneva group . luminosity , effective temperature , and radius are calculated for each star during each evolutionary time step ( 10@xmath31 years ) . spectral types are determined , and an appropriate ultraviolet spectrum ( 1200 1600 ) is assigned to each star from a stellar library . the resultant model spectra are simply a combination of the constituent stellar spectra at each time step . the library spectra utilized by sb99 are composite stellar spectra with signal - to - noise ratios of @xmath120 . the original library consisted of spectra created from high dispersion ( 0.75 ) iue observations of solar or slightly sub - solar metallicity milky way stars , with o and wolf - rayet star spectra drawn from the atlas of @xcite and b - star spectra from @xcite . in this study we take advantage of an important and very recent addition to the sb99 library : an atlas of metal poor ( @xmath11/4 z@xmath2 ) o - star spectra created from hst observations of stars in the large ( lmc ) and small ( smc ) magellanic clouds @xcite . we compare our data to sb99 models with a variety of star formation scenarios . the high mass stellar content of a starburst depends upon both the slope of the initial mass function , @xmath7 , and the upper mass cut - off of the imf , m@xmath30 . populations which form in an instantaneous burst will also have a dependence on age , since the most massive stars are the shortest lived . we construct three grids of models where two of these three parameters ( @xmath7 , m@xmath30 , and age ) are held fixed and the third is allowed to vary . our benchmark model is an instantaneous burst with a salpeter imf slope @xmath7 = 2.35 , a lower mass cut - off of m@xmath29 = 1 m@xmath2 , and an upper mass cut - off of m@xmath30 = 100 m@xmath2 . we allow this model to evolve from 0 50 myrs and create a grid of spectra with ages in 1 myr increments . the parameters of our instantaneous burst model are in line with results from recent studies that have shown the high - mass end ( m@xmath325 m@xmath2 ) of the imf to be remarkably uniform in both resolved regions of active star formation @xcite and in starbursts @xcite . the high - mass end slope in stellar clusters is measured to be @xmath33 2.32.5 ( e.g. , * ? ? ? * ) , which is , for all practical purposes , a ` salpeter slope ' , and the upper mass cut - off is @xmath3470100 m@xmath2 @xcite . no systematic variations in the imf slope are observed over about a factor 10 in metallicity and a factor @xmath11000 in stellar density @xcite . in this study we are not particularly concerned with the low end of the imf , except for cluster mass determinations ( see [ clusters ] ) , as our uv spectroscopy is insensitive to the low - mass cut - off . we have tested that low - end - truncated imfs up to 10 m@xmath2 do not induce a significant variation in the uv spectral features of our models , and , thus , in the cluster age determinations . we also consider continuous star formation models . continuous models take approximately 10 myr to equilibrate , after which the uv spectrum changes very little as a function of age . we therefore fix the age at 50 myr and let the free parameters be the imf slope and the upper mass cut - off . we create 2 grids of continuous models : one with @xmath35 , m@xmath29 = 1 , and m@xmath30 = 10 , 20 , 30 , 40 , 50 , 100 m@xmath2 and the other with m@xmath29 = 1 m@xmath2 , m@xmath30 = 100 m@xmath2 , and @xmath7 = 2.35 , 3 , 3.5 , 4 , 5 . continuous star formation models should be more appropriate to star formation occurring in spatially extended regions such as the stellar field , where the crossing times are long , typically of order 10 myr . for all of our sb99 models , we used an evolutionary track metallicity of 0.2 z@xmath2 ( z@xmath2 = 0.02 , by mass ) and the new lmc / smc stellar libraries . this provides a good match to the metallicity of ngc 5253 s gas which is known to be @xmath10 z@xmath2 . the good agreement between the metallicity of our data and that of the the stellar libraries is particularly important because the age and imf sensitive features also have a z - dependence , with the general trend being towards weaker lines at lower metallicity @xcite . in particular , the strength of the p - cygni features depends upon the mass loss rate which is a function of metallicity ( e.g. * ? ? ? * ) . this dependence is evident in comparisons of stars of the same spectral type and luminosity class in the milky way , lmc , and smc ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? one important caveat is that the low metallicity library employed by sb99 extends only from o3 through b0 , with all other spectral groups being drawn from the original solar metallicity library . this metallicity mismatch between the high mass stars and intermediate and low mass stars can affect the computed synthetic spectrum of a composite population depending on its age or imf . the line profiles of the stellar wind lines , which are produced primarily in very massive stars ( @xmath36 m@xmath2 ) , are not much affected , while the continuum and many photospheric lines are more sensitive to contributions from b stars . @xcite conclude that for populations with a standard salpeter imf up to 100 m@xmath2 the effect of the solar metallicity b - stars is relatively minor for continuous star formation models or instantaneous bursts with ages less than @xmath37 myr . this is discussed further in [ field ] . the ultraviolet spectrum of a star forming region provides a sensitive measure of its high - mass star content because the most massive stars are also the most luminous , and they dominate the integrated light . figure [ field_cluster ] contrasts the o - star dominated composite cluster spectrum with the b - star dominated spectrum of the field . an o - star dominated spectrum is characterized by the broad ( @xmath12000 km s@xmath15 ) stellar wind p - cygni profiles of @xmath38 , @xmath39 , and @xmath40 . these features are notably absent in b - star dominated spectra , which exhibit increasingly strong narrow ( @xmath1350 km s@xmath15 ) photospheric features , such as @xmath41 and @xmath42 . our data have sufficient s / n ( @xmath110 ) and spectral resolution ( @xmath12 ) to allow us to utilize these stellar features as probes of the massive star content in various regions of ngc 5253 . we determined which models provided the best fit to our cluster and field spectra in the 1200 1600 regime using an automated routine that takes as input the extinction corrected spectrum of a cluster or field region , a grid of sb99 model spectra , and an array of values corresponding to the free variable in the grid of models ( i.e. the age , imf slope , or m@xmath30 . ) for each of the model spectra in the grid the goodness of the fit is characterized by a @xmath43 value computed in the following manner : @xmath44 , where @xmath45 represents the observed data for the @xmath46th pixel , @xmath47 the model data , @xmath48 the error in the observed spectrum , and @xmath49 the assigned weight . weights were applied in order to maximize the sensitivity of the @xmath43 to regions of the spectrum which are most influenced by the age / stellar content of the burst , namely the stellar wind lines . on the basis of the detailed line analysis of de mello et al . ( 2000 ) , we classified each pixel in our spectra as belonging to either the continuum , an interstellar line , or a stellar wind line , and assigned weights accordingly . the interstellar lines were given a weight of zero ; the 470 pixels of continuum a weight of 1 ; and the 69 pixels corresponding to stellar wind lines a weight of 10 . the result of this weighting scheme was to eliminate the interstellar lines from consideration , and to give the wind lines a weight @xmath50 times greater than that of the continuum . it was necessary to eliminate the interstellar absorption lines from our @xmath43 because sb99 is designed to model the integrated uv light of a purely _ stellar _ population ; no consideration is given to the gaseous environment of the star - forming region . interstellar absorption lines are present in the models only in so much as they are present in the library of stellar spectra . unfortunately , nearly all of the features which are sensitive diagnostics of the stellar population have some interstellar contamination . in the case of the broad stellar wind features , we attempt to mask out the narrow interstellar core ( see figure [ field_cluster ] ) . however , the separation of stellar and interstellar features remains a source of uncertainty . our routine returns the best fit model ( minimum @xmath43 ) and plots the evolution of the @xmath43 with the free parameter , as shown in figure [ chisq ] . the best fits proved to be relatively robust to the method of @xmath43 evaluation : small changes in the weighting scheme employed had virtually no effect , especially in cases where the s / n was good or o - star wind lines were still present . in order to quantify the error associated with our best fit models , we utilized the bootstrap method . this entailed randomly resampling the residuals of the best fit , adding these to the model spectrum , and running the resultant spectrum through our automated fitting routine . this procedure was repeated 1000 times for each spectrum . the error bars associated with the 90% confidence interval were derived from a histogram of the results . the regions of the spectrum that we define as clusters are shown in figure [ spatial ] . our definition of what constitutes a ` cluster ' is clearly somewhat arbitrary : we have taken the eight brightest peaks in the spectrum . our clusters are not necessarily bound entities , but we retain this labeling for simplicity . for consistency we use the cluster naming convention of @xcite for clusters 2 , 3 , and 5 , and number the remaining clusters ( starting with 7 ) in order of decreasing observed flux . the extracted spectra of the four brightest clusters are shown in figure [ bright_clusters ] . the measured properties of the clusters are reported in table [ table_cluster_data ] . all of our clusters , except for cluster 8 , have spatial profiles that are clearly resolved ( their guassian fwhm values are well in excess of the 1.6 pixels expected for a point source at 1400 ) and relatively symmetric . cluster 8 is actually a collection of a few peaks of uv emission covering 13 pixels ( 5.2 pc ) . the most prominent of these peaks extends no more than four pixels and is compatible with a point source . its uv luminosity , l@xmath51 ergs s@xmath15 , could be produced by a single o - supergiant , but the observed spectral features are more consistent with those of a b - star . we calculate that 34 b supergiants could produce the observed uv luminosity , implying that cluster 8 is actually an association with a modest number of stars . we compared our dereddened cluster spectra to a grid of instantaneous burst models with a salpeter imf slope ( @xmath35 ) , an upper mass cut - off of 100 m@xmath2 , and ages ranging from 0 to 50 myr in 1 myr increments . an instantaneous burst model is appropriate for the clusters since they have typical half light radii of 0.6 parsecs , and thus have negligibly short crossing times , of order 0.1 myr . all eight cluster spectra were remarkably well fit by the models . the four brightest clusters are shown with their best fitting models in figure [ clustermodels ] . we assign ages to the clusters on the basis of their best fits and error bars on these ages which represent the 90% confidence interval ( see [ modelfitting ] for details ) . the robustness of our age determination can also be assessed by examining the shape of the @xmath43 evolution , shown in figure [ chisq ] for clusters 2 , 3 , 5 , and 7 . the ages of the other clusters are expected to be more uncertain than the formal error bars reflect since they are less luminous and potentially subject to strong stochastic effects if only a handful of o - stars are present . the ages and other model dependent properties of the clusters are summarized in table [ table_cluster_results ] . we derive cluster masses from the dereddened luminosity of the clusters . for the 4 brightest clusters ( 2 , 3 , 5 , 7 ) we derive fluxes from photometry of an archival wfpc2 image in the f170w filter . the signal - to - noise of the image ( a combination of two 400 second images ) is too low for the remaining clusters to be clearly detected . we have corrected our f170w image for warm pixels , decontamination events , geometric distortion , and charge transfer efficiency . the photometry we obtain provides a more accurate estimate of the total luminosity of the clusters since they are extended objects and their exact position relative to the slit is unknown . in the most extreme case , cluster 3 , the luminosity derived from the photometry is a factor of @xmath52 greater than that obtained from our stis spectrum . the masses of the four clusters too faint to be detected in the f170w image ( 8 , 9 , 10 , 11 ) are therefore lower limits . all of the clusters have been de - reddened as described in [ observations ] using the starburst obscuration curve @xcite . this assumption provides a lower limit to the actual dust obscuration ( and hence a lower limit to the dereddened uv luminosity and mass ) if more complex geometries , such as dust mixed with stars , are present . to clarify the magnitude of the effect we use as an example the most reddened cluster in our sample , cluster 5 . cluster 5 is located in the heart of what @xcite refer to as the `` starburst nucleus '' of ngc 5253 . its measured uv slope is @xmath53 ( after correction for the small amount of reddening from our own galaxy ) , much redder than the expected slope of a 2 myr old cluster , @xmath54 . using the starburst obscuration curve , the color excess implied by the difference between the slopes is e ( ) = 0.42 . the resulting intrinsic uv luminosity is l@xmath55 ergs s@xmath15 @xmath15 , taking the flux measured from the f170w image ( to avoid slit losses ) and assuming a distance of 3.33 mpc . we use the number of ionizing photons reported by sb99 for a 2 myr old cluster scaled by l@xmath56/l@xmath57 and the transformations of @xcite to predict an extinction corrected h@xmath7 luminosity of l@xmath58 erg s@xmath15 . this value is in reasonable agreement with the measured value of @xcite , which is l@xmath59 erg s@xmath15 after applying the same obscuration correction used above and taking into account the fact that the nebular gas is more reddened than the stellar flux by a factor of @xmath60 @xcite ) . unfortunately , this `` dust scenario '' is degenerate with a number of others , in the sense that multiple dust geometries can reproduce the observed quantities ( uv slope and h@xmath7 flux ) , the only difference being the prediction of larger intrinsic luminosities , and hence the cluster masses . for instance , if we embed cluster 5 in a dust cloud with a@xmath61=35 mag @xcite with a@xmath620.2 of foreground dust , we recover the observed uv slope and h@xmath7 luminosity while predicting a cluster mass 10 times greater than that implied by the previous scenario . the dust cloud geometry can not be constrained by diagnostics at uv - optical wavelengths ; for example , the maximum value of the color excess that is measurable from the balmer decrement ( h@xmath7/h@xmath24 ) in these cases is e(b@xmath63v)@xmath640.35 @xcite . only by going to much longer wavelengths , such as the radio , can the degeneracy be broken @xcite . cluster 5 has recently been mapped with the nrao very large array in the a configuration at wavelengths of 1.3 and 2 cm @xcite . the authors claim no optical counterpart for their source ; however , the position of the dominant radio source is only 1.17 distant from our pre - peak up coordinates for cluster 5 , which is within the error bars for absolute position measurements from _ the radio source is partially optically thick at 2 cm , implying a lyman continuum rate of 3 @xmath65 s@xmath15 , which is about 3 times higher than that predicted on the basis of the observed dereddened h@xmath7 flux . we estimate a mass of 4 @xmath66 m@xmath2 for cluster 5 by comparing the measured lyman continuum rate with that reported by sb99 for a 2 myr old 10@xmath67 m@xmath2 cluster with a salpeter imf extending from 1100 m@xmath2 . this value is a factor of @xmath110 above the mass we derive solely on the basis of our uv measurements . however , it should be kept in mind that cluster 5 is an extreme case : we expect luminosity and mass uncertainties to be substantially lower for most of the other clusters in our sample , since they are considerably less reddened . another factor which can contribute to underestimated masses for the stellar clusters is the adopted lower mass limit to the imf . direct measurements @xcite and morphological and dynamical information @xcite favor a significant low - mass ( @xmath681 m@xmath2 ) population in stellar clusters . however , due to the observed flattening of the imf at low masses , the total mass in stars should be similar to that derived using a salpeter imf extending down to 1 m@xmath2 . if we take a more extreme assumption , for instance , if the imf follows a salpeter slope down to @xmath10.3 m@xmath2 , the clusters will be a factor @xmath11.6 more massive than what is reported in table [ table_cluster_results ] . we investigated the diffuse light of ngc 5253 by creating spectra of three ` field ' regions . we choose these regions such that they excluded any apparent clusters , but fell within the plateau of diffuse light ( see figure [ spatial ] ) . the resultant one - dimensional field spectra have signal - to - noise values of 12 18 , comparable the brightest clusters . the three field spectra are shown in figure [ fields ] and their measured properties are reported in table [ table_field_data ] . since the three regions have similar spectral morphologies and amounts of reddening , we combine them into a single field spectrum which we use for all of our subsequent analysis . the field spectrum does not show any evidence of the broad line profiles of , , and which are characteristic of o - star winds . this is in sharp contrast to the spectra of the brightest clusters . the obvious disparity between the cluster and field spectra ( see figure [ field_cluster ] ) is clear evidence that the diffuse light is not simply reflected light from the clusters , but must originate in a uv bright field star population . we devote our subsequent analysis to the nature of the field star population . what is immediately apparent from the weakness of the o - star wind features is that the field is deficient in the most massive stars . we consider whether this could be a statistical effect due to the small area and modest star formation rate of the region we are sampling . if we assume that the light in the field is the result of a continuous star formation episode , sb99 predicts that @xmath37 o - stars should be present for a salpeter imf extending from 1 to 100 m@xmath2 . ( we have assumed a burst duration of 50 myr , although the result is largely insensitive to this value . ) the salient point is that a non - negligible number of o - stars is expected to be present . the fact that we do not see the spectral signatures of these high mass stars in our field spectrum indicates that either the imf slope is steeper than salpeter , the upper mass cut - off is lower than 100 m@xmath2 , or that the stellar population has aged past @xmath110 myr . we create models to explore each of these scenarios . we compare our various field models to the data using the same automated routines that we used for the clusters . however , we do not quote formal error bars on our best fits because the field spectrum has such high s / n that systematic errors dominate , as is evident in figure [ fieldmodels ] . the probable sources of systematic error are discussed subsequently . the power law slopes of models which are very deficient in massive stars are much shallower than our assumed value of -2.6 used for the clusters . in order not to bias our fits against these models , we rectify both the data and the models by fitting and dividing by the appropriate power law before comparison . shallower intrinsic values of @xmath24 imply less reddening in the data . this provides a useful constraint because models can not be redder that the extreme case of zero reddening for the field , which corresponds to @xmath69 , after correction for foreground galactic reddening . the first model we consider is the fiducial model that we use for the clusters : an instantaneous burst with a salpeter imf slope ( @xmath7 = 2.35 ) and an upper mass cut - off of m@xmath30 = 100 m@xmath2 . this model produces a best fit to the composite field spectrum at an age of 8 myr . younger ages are strongly ruled out by the increasing strength of o - star wind lines . at ages older than about 8 myr , the primary age discriminator becomes the strength of the b - star photospheric lines . however , the strength of these lines is expected to increase with metallicity , so the use of the solar metallicity b - star library could bias our fit towards younger ages . thus , 8 myr should be viewed as a lower bound to the age of the field . a strict upper bound of 45 myr is set by the reddest model slope allowable , corresponding to the rather unlikely case of e ( ) = 0 . ages of 8 - 18 myr imply an e ( ) value of @xmath70 , in good agreement with the reddening of most of the clusters flanking the field regions . however , it is not implausible that the field could be less reddened than the stellar clusters . the second model we consider is a continuous star formation model with a fixed age of 50 myr , a salpeter imf slope , and variable upper mass cut - off . we test values of m@xmath30 of 10 , 20 , 30 , 40 , 50 , and 100 m@xmath2 , and find an upper mass cut - off of 30 m@xmath2 is preferred . higher upper mass cut - offs are strongly ruled out by the increasing presence of the o - star wind lines , particularly . values as low as 10 m@xmath2 are unlikely because of the extreme strength of the b star photospheric lines ( which are barely evident in the field spectrum ) . however , such models can not be conclusively eliminated because of the unknown impact of the solar metallicity b - star library on these features . the third model we consider has continuous star formation with the age fixed at 50 myr , the upper mass cut - off fixed at 100 m@xmath2 , and a variable imf slope . we examine values of @xmath7 = 2.35 , 3 , 3.5 , 4 , and 5 . the steepest slope we consider , @xmath71 , corresponds to the ` field star imf ' derived by @xcite for the lmc and smc . this model is ruled out a priori by the fact that it has an intrinsic slope of @xmath72 , far redder than the value measured for the field after correction for foreground reddening ( @xmath73 . the best fitting model , @xmath74 , has a slope of @xmath75 , implying a very modest reddening of e ( ) = 0.08 . imf slopes shallower than this are ruled out by the increasing width of the stellar wind line . the field spectrum is shown with each of the three model spectra overplotted in figure [ fieldmodels ] ( a c ) . none of the models provides a satisfactory fit . in each case , the continuum level of the data is above that of the models at wavelengths shortward of 1270 and longward of 1530 . in addition , photospheric lines like [ ] @xmath761294,1296,1298 are grossly overpredicted . one consideration is that our subtraction of the geocoronal emission at ly-@xmath7 and [ ] @xmath22 could be inducing errors in these regions of the spectrum . a residual emission feature is clearly observed in the field spectrum near 1300 . however , other photospheric features such as @xmath771260 , which are uncontaminated by geocoronal lines , are poorly matched with the models as well . it is also worth noting that the cluster data seem to be reasonably well fit in the regions surrounding the geocoronal lines , suggesting that sky subtraction is not a major source of error . another possibility is that the discrepancies between the data and the models are the result of a metallicity mismatch , as the stars with masses less than 20 m@xmath2 ( the predominant contributors to the continuum and photospheric lines in these models ) have empirical spectra drawn from the solar metallicity library . at solar metallicity , the broad wing of lyman-@xmath7 is more extended , and increased line blanketing is expected near 1550 , which could account for discrepancies in the continuum level . also , photospheric lines are observed to be substantially weaker at lmc / smc metallicity than at solar metallicity @xcite , consistent with what is seen in the field . however , until low metallicity b - stars spectra are added to the sb99 library , we can not be sure if this fully accounts for the discrepancy . an intriguing possibility is that the field spectrum might contain contributions from an older underlying stellar population . adding a burst with an age of @xmath78 gyr to the models can eliminate the continuum mismatch problems at both ends ; however , the already strong photospheric features become even stronger , and an unrealistically large stellar mass is implied . until the metallicity issues are resolved , conclusions about an underlying older population are premature . the parameters which provided the best fit for each of the three models are listed in table [ table_fieldmodels ] along with the measured power law slope of the model , and the implied reddening . the reduced @xmath43 values of the best fits are also given for reference . we do not use these values to discriminate among the models , since the differences are relatively minor and the overall quality of the fits is not high . we consider the plausibility of each of the models further in [ fielddiscussion ] . the eight clusters subtended by our slit all have ages below @xmath110 myr , with three having ages @xmath683 myr , for a salpeter imf in the range 1100 m@xmath2 . among the three youngest is cluster 5 , the primary target of our stis observations , whose very young age is in agreement with that inferred by @xcite based on the large observed equivalent width of the h@xmath7 emission line ( ew(h@xmath7)=1650 ) . notably young are clusters 2 and 3 , with ages 8 myr and 3 myr , respectively . for these two clusters , age estimates based on colors give values of @xmath160 myr and @xmath130 myr , respectively @xcite , much larger than what is inferred from uv spectroscopy . age determinations based on colors alone are indeed bound to be relatively uncertain unless additional constraints ( e.g. , h@xmath7 emission ) are available . for instance , consider cluster 2 : the difference in color between a 60 and a 8 myr old instantaneous burst is only @xmath79 mag . the narrowness of this range makes accurate reddening corrections imperative . @xcite deredden their observed colors based on the ratio of h@xmath7 to h@xmath24 ; however the measured equivalent widths are very small ( @xmath80 ) and therefore bound to be uncertain . after correction for galactic foreground reddening and underlying stellar absorption , they find zero magnitudes of foreground attenuation between h@xmath7 and h@xmath24 . by contrast , the reddening derived from our uv spectroscopy is e ( ) = 0.16 , very nearly what is needed to bring the broad - band color age estimate into line . similar results are found for cluster 3 , although the color difference between the two age estimates is greater ( @xmath10.3 ) and can not be entirely reconciled by applying our reddening correction . additional corroboration for the young ages of clusters 2 and 3 comes from an analysis of the soft x - ray emission near the center of ngc 5253 @xcite . the two clusters appear to be located within a superbubble of expanding hot gas . energy limitations constrain the age of the superbubble to be @xmath6410 myr @xcite , in better agreement with the age estimates from uv spectroscopy . the measured and derived properties of the clusters are summarized in tables [ table_cluster_data ] and [ table_cluster_results ] . the ages of the clusters show no particular spatial correlation ( which could indicate propagating star formation . ) their half - light radii span a wide range of values ( 0.6 2.4 pc ) , but visual examination of the profiles suggests that the largest clusters are an unresolved blend of smaller ones . the clusters display varying amounts of reddening , ranging from e ( ) = 0.42 0.05 . the reddening values show a broad spatial correlation ; that is , clusters which are physically close show similar amounts of reddening . the extinction - corrected 1500 luminosities are in the range l@xmath81@xmath82 ergs s@xmath15 @xmath15 for the eight clusters , implying masses in the 10@xmath8310@xmath84 m@xmath2 range , for a low - end imf mass of 1 m@xmath2 . however , both luminosities and masses listed can generally be considered lower limits to the actual values , as discussed in [ clusters ] . we note that the lowest mass clusters ( @xmath85 m@xmath2 or less ) might be subject to strong stochastic effects since only a handful of o - stars are expected to be present for a salpeter imf . these effects could potentially bias our age and mass determinations for clusters 9 , 10 , and 11 . the uv emission from the field regions surrounding the clusters has a clearly stellar origin and is not due to reflected light from the clusters themselves , since , as already mentioned in [ field ] , the spectral features of clusters and field are notably different . in contrast to the clusters spectrum , the field spectrum lacks the strong o - star wind lines of , , and which are signatures of the most massive stars . we stress that this deficiency is not an artifact of undersampling of the field , as our field spectrum contains a considerable amount of light . in fact , the field has a flux nearly equivalent to the integrated cluster spectrum , which contains over @xmath86 m@xmath2 of stars . a similar discrepancy between the cluster and field population has been observed in the starburst galaxy ngc 1569 . using wfpc2 photometry , @xcite find that the stellar population of ngc 1569 is composed of recently formed super star clusters and resolved field stars with ages greater than 10 myr . in order to understand why the field might be deficient in massive stars we have compared the field spectrum of ngc 5253 to three different families of model spectra . the first is an instantaneous burst model with a salpeter imf in the 1100 m@xmath2 range . our best fit model has an age of 8 myr , although this should be considered a lower limit for the reasons discussed in [ field ] . while this model provides a reasonable fit to the data , we rule it out on physical grounds : our field regions are spatially extended ( @xmath87 pc ) and therefore unlikely to be forming stars in an instantaneous burst . the subsequent models we consider have continuous star formation which is probably more appropriate given the galaxy s @xmath37 myr crossing time . since the age of the burst is largely irrelevant in continuous models , we allow the free parameters to be the slope and upper mass cut - off of the imf . our second model uses a fixed salpeter slope and a variable upper mass cut - off . a reasonable fit is produced for an upper mass cut - off of @xmath130 m@xmath2 . the third model fixes the upper mass cut - off at 100 and allows the imf slope to vary . the best fit is achieved by a value of @xmath74 , which is steeper than salpeter , but shallower than the @xmath885 value found by @xcite for the stellar field of the magellanic clouds . an imf slope of @xmath89 is strongly ruled out by the observed blue color of the field s uv continuum ( although it might be argued that our definition of the ` field ' is not as rigorous as massey s . ) we note that our fits are sensitive to stars with masses @xmath420 m@xmath2 which is comparable to the stellar mass range @xmath11560 m@xmath2 sampled by @xcite . the basic result is that the field of ngc 5253 is generally lacking very massive stars , either because of aging or because very massive stars do not form in large numbers . @xcite found that the field accounts for a significant fraction of the uv light from starbursts , between 50% and 80% , and proposed that the field is the outcome of diffuse star formation , a mode different from the one which produces stellar clusters . if the hypothesis that two modes of star formation exist in starbursts is correct , then our observations show that the two modes produce drastically different imfs : a ` regular ' salpeter imf up to @xmath1100 m@xmath2 in clusters , and a massive - star - deficient imf in the field . a scenario other than bi - modal star formation can be suggested to explain the properties of the stellar field . the eight clusters along our stis slit are all younger than @xmath110 myr ; the other three bright clusters in the center of ngc 5253 are younger than @xmath120 myr , although for two of the clusters ages have been derived from photometry alone and can therefore be fairly uncertain @xcite . if the clusters along our slit are representative of the general cluster population in the center of ngc 5253 , we can conclude that the field appears generally older than the clusters or as old as the oldest among them . the field thus contains stars at a more advanced evolutionary stage than the clusters . one way to obtain such a population mix is to hypothesize that stellar clusters dissolve over timescales of @xmath11020 myr and their surviving stars disperse into the field . thus , there is only one star formation mode : stars form in clusters and a fraction of them continue their life in the field . to test this hypothesis , we have constructed a model for the field by subtracting a 10 myr constant star formation population from a 100 myr constant star formation population . star formation has been going on for a while in the center of ngc 5253 ( e.g. , * ? ? ? * ) , and if clusters dissolve over a timescale of @xmath110 myr , the field will consist of stars with ages between @xmath110 myr and the maximum age of the star formation episode , which we take to be 100 myr , although the resulting uv spectrum is largely insensitive to this value . figure [ fieldmodels ] d ) compares the field to the @xmath90 myr model spectrum . the similarity between this model spectrum and the other models shown in figure [ fieldmodels ] is not surprising : the uv spectrum of the @xmath90 myr model is dominated by its youngest , most luminous contributors , i.e. , stars with ages @xmath110 myr which have just come out of dissolving stellar clusters . one prediction of the @xmath90 myr model is that about 40% of the uv light at 1500 is in the field . this fraction becomes slightly higher as one goes to longer wavelengths , for instance , at 2200 ( the wavelength of the @xcite study ) the fraction becomes 45% . however , this value is still somewhat below the 5080% value derived for nearby starbursts @xcite . pushing the dissolution timescale to an age as young as 6 myr , brings the fraction of light in the field to @xmath3450% . however , the differences between the measured and predicted fraction of light in the field may simply reflect the observational difficulty of disentangling small clusters from the field population . star clusters are known to dynamically evolve , with a number of mechanisms contributing to stellar depletion eventual cluster dissolution . however , the dissolution timescales implied by our model for ngc 5253 are very short . we consider the feasibility of these timescales using the model results of @xcite . @xcite calculated the evaporation timescales for compact ( half - mass radius @xmath911 pc ) stellar clusters near the center of the milky way , including stellar evolution among the ingredients . with the possible exception of 5 , our clusters appear fairly compact , with half - light radii @xmath682 pc ; we assume that the half - light radii are a reasonable approximation of the half - mass radii , although mass segregation may have already acted on the oldest clusters to make the first radius smaller than the second . the environment surrounding our clusters is also quite different from that of the milky way . ngc 5253 has about 100 times less baryonic mass than the galaxy , spread over about 20 times less volume ( adopting flattened mass distributions for both galaxies , with similar scale heights ) ; this implies that the central density in ngc 5253 is about 5 times lower than the central density in the milky way . @xcite identify the tidal radius as one of the crucial parameters which determines the evaporation timescale of the stellar clusters ; in ngc 5253 the stellar clusters have tidal radii about 70% bigger than in the milky way , for constant cluster mass , and , thus , evaporation timescales about 2.2 times longer . a 5@xmath010@xmath92 m@xmath2 cluster placed 100 pc away from the dynamical center of ngc 5253 will evaporate in @xmath115 myr for a 1150 m@xmath2 salpeter imf or in @xmath125 myr for a 0.1150 m@xmath2 salpeter imf . we are therefore in the expected ballpark timescale for the less massive among our clusters to agree with the dissolving clusters scenario ; heavier clusters , like 2 , will have lifetimes in the range of 50100 myr and will not contribute significantly to the uv emission of the field . the most massive of our clusters , cluster 5 , has a relaxation time of 400 myr for a 1150 m@xmath2 salpeter imf , and is potentially bound if the imf extends to lower stellar masses . a final consideration is the validity of our definitions of ` cluster ' and ` field ' . the first issue of concern is that some of the less massive objects which we have defined as clusters may simply be random superpositions of a few field o - stars . in this scenario our finding of an massive star deficient field population would be merely circular logic since we would have selectively eliminated many of the o - stars from the field . the brightness of clusters 9 , 10 , and 11 , while lower limits , could be consistent with being one to a few o - stars or b supergiants . however , all of these clusters are well resolved and have relatively smooth spatial profiles . ( cluster 8 , which may include a bright point source , was previously discussed in [ clusters ] . ) an additional check is performed by adding the light from clusters 8 , 9 , 10 , and 11 to our field spectrum . these four clusters contribute about 20% of the total field light , but produce a negligible difference in the integrated spectrum . thus , we can be relatively secure that our choice of clusters is not robbing the field of o - stars . the other issue of concern is the converse : that what we call the field may in fact include a number of faint unresolved clusters . taking the flux of our faintest cluster as our detection limit , we find that a 10@xmath84 m@xmath2 cluster with an e ( ) of 0.2 would be detectable with an age as old as 50 myr , while a 10@xmath92 m@xmath2 cluster with a similar amount of reddening would only be seen at ages less than 15 myr . clusters with masses less than 10@xmath92m@xmath2 probably do not contribute much uv light since they are unlikely to contain many very massive stars due to statistical effects . thus , clusters which are both relatively old and not very massive could contribute somewhat to our field population . however , such light - weight clusters would be subject to a number of destruction mechanisms , and would tend to dissolve on timescales shorter than those discussed above @xcite . most importantly , the addition of faded clusters to the field would not alter the basic notion that there is only one ` mode ' of star formation . the differences seen in the spectra of the field and clusters of ngc 5253 have potentially important implications for restframe uv observations of high redshift galaxies . because of the small angular sizes of high redshift objects , the entire star - forming region is typically observed , with no discrimination between clusters and field . since the field s uv light may represent between 50% and 80% of the uv output from a starburst galaxy @xcite , the common tendency to compare spatially integrated spectra of distant galaxies with the uv spectra of local stellar clusters neglects the important contribution of the field . to quantify this statement , we have created two ` integrated ' spectra , the first combining a 50% ` cluster ' spectrum with 50% ` field ' spectrum , and the second with a 20%80% cluster field mix ( figure [ fiftyfifty ] ) . the ` cluster ' spectrum is the unweighted sum of the eight clusters along our slit prior to dereddening . the 5050 and 20 - 80 template spectra are available for download from http://www.stsci.edu/starburst/templ.html . the spectral features of the 5050 ` integrated ' spectrum closely resemble those of a continuous starburst with a standard salpeter imf extending up to 100 m@xmath2 . the 2080 integrated spectrum has weaker wind lines than the 5050 mix , and may be more appropriate to populations with declining star formation rates . in both cases the stellar wind lines , most notably @xmath771550 , are substantially weaker than in a pure young population . thus , the generally observed weakness or even absence of stellar wind lines in high - z galaxies ( e.g. , * ? ? ? * ) may not be entirely a metallicity effect , but rather a result of time - extended star - forming episodes . the utility of our template spectra for comparisons with high redshift galaxies is illustrated in figure [ cb58 ] , where we compare our 50% cluster50% field spectrum to the spectrum of the gravitationally lensed z = 2.73 starburst galaxy , ms 1512-cb58 @xcite . cb58 appears to be a relatively typical l@xmath93 lyman break galaxy , with z @xmath9 1/3 z@xmath2 @xcite . our empirical template provides an exceptionally good fit overall to the stellar features , particularly to the stellar wind lines . the widths of the interstellar features appear to be fairly well matched , but it must be kept in mind that the interstellar features of the ngc 5253 template spectrum are artificially broadened by blending with intervening milky way absorption features . the intrinsically broader and deeper interstellar features of cb58 are probably a result of the larger covering factor of the gas , and the hydrodynamical consequences of the galaxy s high star formation rate . thanks to the exceptional s / n of both cb58 ( @xmath94 ) and our template spectrum ( @xmath52 ) , figure [ cb58 ] provides one of the best examples to date of the striking similarity between the spectral morphology of a high redshift star - forming galaxy and a local starburst . we have obtained a stis longslit uv spectrum of the central star forming region of the dwarf starburst galaxy ngc 5253 . we extracted spectra of 8 clusters and three field regions along the slit , and compared these spectra to the starburst99 stellar evolutionary synthesis models . the clusters are well fit by instantaneous burst models with a standard salpeter imf extending from 1100 m@xmath2 . we derive ages for the clusters in the range 110 myr and masses in the range 10@xmath9510@xmath31 m@xmath2 , although the latter are probably underestimated by factors of a few . the spectral features of the field and the brightest clusters exhibit pronounced differences , clear proof that the diffuse uv light is not simply reflected cluster light . the main difference is that the field of ngc 5253 is found to be lacking in the most massive stars , either as a consequence of aging or because massive stars are not forming in the field in large numbers . one possibility is that star formation in starbursts is bimodal , with compact clusters forming stars with a standard salpeter imf extending up to 100 m@xmath2 , and more diffuse star formation occurring with a massive - star deficient imf . we suggest an alternate hypothesis which requires only one ` mode ' of star formation . in this scenario , stellar clusters form continuously and dissolve on @xmath110 myr time scales , dispersing their remaining stars into the field . we find this hypothesis to be consistent with the evaporation timescales that we derive for the majority of our clusters . our dissolving clusters scenario suggests that continuous star formation models with a standard salpeter imf in the range 1100 m@xmath2 should be appropriate for the integrated spectra of high - z galaxies , which will contain both young clusters and their dissolved by - products , field stars . the commonly observed absence of very broad o - star wind lines in high redshift galaxies may indicate that we are observing star formation episodes of sufficient duration to populate the field with substantial numbers of older stars . care must be taken when comparing high resolution rest - frame uv spectra of distant galaxies to local starburst templates because the presently available template spectra are primarily of individual star clusters . we have constructed a 50% cluster50% field and a 20% cluster80% field spectrum of ngc 5253 which we present as new benchmarks for comparison with spectra of high - z galaxies . the authors would like to thank gerhardt meurer for insightful and stimulating discussions which helped shape this paper ; duilia de mello for her expert advice about b star lines ; and max pettini and chuck steidel for contributing the spectrum of ms 1512-cb58 . we also wish to thank our anonymous referee for his or her helpful comments . we are grateful for support we received from nasa through grant no . go-08232.01.97a from the space telescope science institute which is operated by the association of universities for research in astronomy , inc . for nasa under contract nas5 - 26555 . 2 & 21 & 8 & 0.68 & 1.69 & @xmath96 & @xmath97 & -1.11 + 3 & 23 & 7 & 0.58 & 0.96 & @xmath98 & @xmath99 & -1.27 + 5 & 27 & 14 & 2.38 & 4.43 & @xmath100 & @xmath101 & 0.50 + 7 & 29 & 15 & 0.57 & 0.73 & @xmath102 & @xmath103 & -1.11 + 8 & 13 & 8 & 0.17 & & @xmath104 & & 0.13 + 9 & 23 & 9 & 0.59 & & @xmath105 & & -0.62 + 10 & 25 & 7 & 0.55 & & @xmath106 & & -1.14 + 11 & 27 & 6 & 0.70 & & @xmath107 & & -1.87 + cccccccc 2 & -1.55 & 0.16 & @xmath108 & @xmath109 & @xmath110 & @xmath111 & 0.94 + 3 & -1.71 & 0.14 & @xmath112 & @xmath113 & @xmath114 & @xmath115 & 1.20 + 5 & @xmath1160.06 & 0.42 & @xmath117 & @xmath118 & @xmath119 & @xmath120 & 1.49 + 7 & -1.55 & 0.16 & @xmath121 & @xmath122 & @xmath123 & @xmath124 & 2.36 + 8 & -0.29 & 0.36 & @xmath125 & & @xmath126 & @xmath127 & 1.46 + 9 & -1.07 & 0.26 & @xmath128 & & @xmath129 & @xmath130 & 1.27 + 10 & -1.48 & 0.18 & @xmath131 & & @xmath132 & @xmath133 & 1.33 + 11 & -2.31 & 0.05 & @xmath134 & & @xmath135 & @xmath136 & 1.57 + field 1 & 127 & 12 & @xmath11680.8 & @xmath137 & -1.08 + field 2 & 161 & 16 & 102.4 & @xmath138 & -1.20 + field 3 & 167 & 18 & 106.2 & @xmath139 & -1.09 + field & 455 & 27 & 289.4 & @xmath140 & -1.12 +

the millimeter wave ( mmwave ) band , which is a prime candidate for 5 g cellular networks , seems attractive for wireless energy harvesting . this is because it will feature large antenna arrays as well as extremely dense base station ( bs ) deployments . the viability of mmwave for energy harvesting though is unclear , due to the differences in propagation characteristics such as extreme sensitivity to building blockages . this paper considers a scenario where low - power devices extract energy and/or information from the mmwave signals . using stochastic geometry , analytical expressions are derived for the energy coverage probability , the average harvested power , and the overall ( energy - and - information ) coverage probability at a typical wireless - powered device in terms of the bs density , the antenna geometry parameters , and the channel parameters . numerical results reveal several network and device level design insights . at the bss , optimizing the antenna geometry parameters such as beamwidth can maximize the network - wide energy coverage for a given user population . at the device level , the performance can be substantially improved by optimally splitting the received signal for energy and information extraction , and by deploying multi - antenna arrays . for the latter , an efficient low - power multi - antenna mmwave receiver architecture is proposed for simultaneous energy and information transfer . overall , simulation results suggest that mmwave energy harvesting generally outperforms lower frequency solutions .

1509.01653

millimeter wave ( mmwave ) communications is a key candidate technology for future 5 g cellular networks . this is mainly due to the availability of large spectrum resources at higher frequencies , which leads to much higher data rates . recent research suggests that mmwave systems will typically feature ( i ) large - dimensional antenna arrays with directional beamforming at the transmitter / receiver which is motivated by the small wavelength that allows packing a large number of antenna elements into small form - factors ; and ( ii ) a dense deployment of base stations ( bss ) to ensure comparable coverage to ultra high frequency ( uhf ) networks@xcite . these mmwave design features are also attractive for rf ( radio frequency ) energy harvesting where a harvesting device may extract energy from the incident rf signals@xcite . this could potentially power the massive number of low - power wireless devices in future paradigms such as the internet of things@xcite . the signal propagation at mmwave frequencies , however , suffers from poor penetration and diffraction characteristics , making it sensitive to blockage by buildings@xcite . it is , therefore , unclear if mmwave cellular networks will be more favorable for rf energy harvesting compared to the conventional ( below 6 ghz ) frequencies . further , the network level design principles for mmwave energy harvesting systems are not well understood . this motivates a network view of energy harvesting in a mmwave cellular network . in this paper , we provide a tractable framework to characterize the performance of wireless energy and information transfer aided by a large - scale mmwave cellular network . our analysis accounts for the key distinguishing features of mmwave systems , namely the sensitivity to blockage and the use of potentially large antenna arrays at the transmitter / receiver . we first consider mmwave energy harvesting , where devices only extract energy from the incident mmwave signals . our analysis models two operating scenarios , one where devices have their beams aligned to that of a mmwave bs , and the other where no such beam alignment is assumed . for both operating modes , we derive simple analytical expressions for metrics such as the energy coverage probability and the average harvested power using tools from stochastic geometry . we then extend the analysis to characterize the overall ( energy - and - information ) coverage probability for the general case where a device extracts both energy and information from the mmwave signals . to get design insights , we examine the network level performance trends in terms of key parameters such as the mmwave network density and the antenna geometry parameters for both operating modes of the energy harvesting devices . numerical results suggest that narrower antenna beams are preferred when the users are aligned with a bs , whereas wider beams are favorable when no beam alignment is assumed . our findings also suggest that there typically exists an optimum transmit antenna beamwidth that maximizes the network - wide energy coverage for a given user population . this implies that the mmwave bss will need to adapt the antenna beam patterns depending on the fraction of the users operating in each mode . similar to the bs - related parameters , we also investigate the role of device - related parameters on system performance . for example , the overall ( energy - and - information ) coverage probability can be improved by optimizing over a design parameter ( power splitting ratio ) to optimally portion the received signal between the energy harvesting and the information decoding modules . another important design feature at the user is the receive antenna array . similar to the bss , the mmwave users can , in principle , benefit from using large antenna arrays . for low - power energy harvesting devices , however , the associated antenna circuity could increase the power consumption , offsetting the potential gains of large antenna arrays . to leverage multiple antennas at the receiver without resorting to power - hungry circuit components , we propose a simple switch - based receiver architecture for simultaneous energy and information transfer . simulation results reveal that the proposed low - power solution performs reasonably well compared to more advanced but power - hungry receiver architectures . wireless energy harvesting is becoming increasingly feasible due to the reduction in the power consumption requirements of wireless sensors and the improvements in energy harvesting technologies @xcite . this has also led to considerable research in advancing the theoretical understanding of wireless powered systems ( see @xcite for a comprehensive overview ) . for example , wireless energy and information transfer has been studied for different information - theoretic setups such as a broadcast channel@xcite , a fading channel@xcite , and an interference channel @xcite . many of these papers highlight the fundamental trade - off between energy and information transfer efficiency and characterize the achievable rate - energy regions for different practical receiver architectures@xcite . wireless energy and/or information transfer in large - scale networks has also been investigated@xcite . in @xcite , the performance of ambient rf energy harvesting was characterized using tools from stochastic geometry . using a repulsive point process to model rf transmitters , it was shown that more repulsion helps improve the performance at an energy harvester for a given transmitter density . in @xcite , cognitive radio networks were considered , and opportunistic wireless energy harvesting was proposed and analyzed . in @xcite , a hybrid cellular network architecture was proposed to enable wireless power transfer for mobiles . in particular , an uplink cellular network was overlaid with power beacons and trade - offs between transmit power and deployment densities were investigated under an outage constraint on the data links . a broadband wireless network with transmit beamforming was considered in @xcite , where optimal power control algorithms were devised for improving the throughput and power transfer efficiency . simultaneous information and energy transfer in a relay - aided network was considered in @xcite . under a random relay selection strategy , the network - level performance was characterized in terms of the relay density and the relay selection area . our work differs from the prior work in that we investigate wireless energy and information transfer in a large - scale _ mmwave _ cellular network . due to different physical characteristics and design features at mmwave , prior work on energy / information transfer in lower frequency networks does not directly apply to mmwave networks . in another line of work , the performance of mmwave cellular networks in terms of signal - to - interference - and - noise ratio ( sinr ) coverage and rate has also been analyzed using stochastic geometry@xcite . none of this work on mmwave networks , however , provides a performance characterization from the perspective of wireless energy and information transfer . the paper is organized as follows . in section [ secsys ] , we introduce the system model . section [ seceh ] presents the analytical results for mmwave energy harvesting . the case with simultaneous information and energy transfer is treated in section [ secswipt ] . we conclude the paper in section [ secconc ] . in this section , we introduce the network and channel models , followed by a description of the antenna model . the parameters defined in this section are summarized in table [ table1 ] . we consider a large - scale cellular network consisting of mmwave bss and a population of wireless - powered devices ( or users ) that operate by extracting energy and/or information in the mmwave band . the mmwave bss are located according to a homogeneous poisson point process ( ppp ) @xmath0 of density @xmath1 . the user population is drawn from another homogeneous ppp @xmath2 of density @xmath3 , independently of @xmath4 . in general , mmwave bss and users may be located outdoors or indoors . empirical evidence suggests that mmwave signals exhibit high penetration losses for many common building materials@xcite . assuming the building blockages to be impenetrable , we focus on the case where the bss and users are located outdoors . we say that a link is line - of - sight ( los ) or non - line - of - sight ( nlos ) depending on whether or not it is intersected by a building blockage . channel measurement campaigns have reported markedly different propagation characteristics for los / nlos links@xcite . to model blockage due to buildings , we leverage the results in @xcite where the buildings are drawn from a boolean stochastic point process . we define a line - of - sight ( los ) probability function @xmath5 for a link of length @xmath6 , where @xmath7 is a constant that depends on the geometry and density of the building blockage process : a bs - receiver link of length @xmath6 is declared los with a probability @xmath8 , independently of other links . while conducting stochastic geometry analysis , we will apply this result to split the bs ppp into two independent but non - homogeneous ppps consisting of los and nlos bss . we allow the user population to consist of two types of users , namely _ connected _ and _ nonconnected_. a connected user is assumed to be tagged with the bs , either los or nlos , that maximizes the average received power at that user . moreover , for the connected case , we assume perfect beam alignment between a bs and its tagged user , i.e. , the bs and user point their beams so as to have the maximum directivity gain . further , we assume that a bs serves only one connected user at a given time . for a nonconnected user , we do not assume any prior beam alignment with a bs , i.e. , it is not tagged with any bs . this allows us to model a wide range of scenarios . for instance , due to limited resources , the mmwave network may ( directly ) serve only a fraction of the user population as connected users , leaving the rest in the nonconnected mode . another interpretation could be that due to the challenges associated with channel acquisition , not all the users could be simultaneously served in the connected mode . we let @xmath9 be the probability that a randomly selected node is a connected user , independently of other nodes . with this assumption , we can thin the user ppp @xmath10 into two independent ppps @xmath11 and @xmath12 , with respective densities @xmath13 and @xmath14 . note that an arbitrary user , either connected or nonconnected , may experience an energy outage if the received power falls short of a required threshold @xmath15 . this threshold would depend on the power consumption requirements of the receiver . to capture the sensitivity requirements of the harvesting circuit , we define @xmath16 to be the harvester activation threshold , i.e. , the minimum received energy needed to activate the harvesting circuit ( the energy outage threshold @xmath15 would typically be greater than @xmath16 ) . we use @xmath17 to denote the rectifier efficiency . we define @xmath18 to be the energy coverage probability given an outage threshold @xmath19 for a connected user , while @xmath20 denotes the same for the nonconnected case . with these definitions , we can define the overall energy coverage probability @xmath21 of the network as @xmath22 where the energy coverage probability is a function of several parameters such as the bs density , the channel propagation parameters , as well as the antenna beam patterns at the transmitter / receiver . for cleaner exposition , we drop the subscript in @xmath19 or @xmath23 , using the notation @xmath24 when the context is clear . in section [ secstoch ] , we provide analytical expressions to compute the energy coverage probability in a mmwave network . we now describe the channel model for an arbitrary user without losing generality . empirical evidence suggests that mmwave frequencies exhibit different propagation characteristics for the los / nlos links@xcite . while the los mmwave signals propagate as if in free space , the nlos mmwave signals typically exhibit a higher path loss exponent ( and additional shadowing)@xcite . we let @xmath25 and @xmath26 be the path loss exponents for the los and nlos links respectively . we define the distance - dependent path loss for a user located a distance @xmath27 from the @xmath28-th bs : @xmath29 when the link is los , where the constant @xmath30 is the path loss intercept ; and @xmath31 for the nlos case . note that by including blockages in our model ( section [ secnet ] ) , we capture the distance - dependent signal attenuation due to buildings . to simplify the analysis , we do not include additional forms of shadowing in our model . we further define @xmath32 to be the small - scale fading coefficient corresponding to a bs @xmath33 . assuming independent nakagami fading for each link , the small - scale fading power @xmath34 can be modeled as a normalized gamma random variable , i.e. , @xmath35.17ex@xmath36 when the link is los and @xmath37.17ex@xmath38 for the nlos case , where the fading parameters @xmath39 and @xmath40 are assumed to be integers for simplicity . to compensate for higher propagation loss , mmwave bss will use large directional antennas arrays . we assume that the bss and users are equipped with @xmath41 and @xmath42 antenna elements each . to simplify the analysis while capturing the key antenna characteristics , we use a sectored antenna model of fig . [ fig : ant_pat ] ( except for section vi ) , similar to the one considered in @xcite . we use @xmath43 to characterize the antenna beam pattern , where @xmath44 gives the angle from the boresight direction , @xmath45 denotes the directivity gain and @xmath46 the half power beamwidth for the main lobe , while @xmath47 and @xmath48 give the corresponding parameters for the side lobe . with this notation , @xmath49 denotes the antenna beam pattern at an arbitrary bs in @xmath4 , and @xmath50 denotes the same for an energy harvesting user in @xmath51 . we further define @xmath52 , the total directivity gain for the link between the @xmath28-th bs and the typical user ; @xmath53 and @xmath54 give the angle - of - arrival and angle - of - departure of the signal . [ t ] ) and side lobe ( @xmath47 ) , and the half power beamwidths for the main lobe ( @xmath46 ) and side lobe ( @xmath48 ) . ] without any further assumptions about the beam alignment , we model the directivity gain @xmath55 as a random variable . we assume the angles @xmath53 and @xmath54 are uniformly distributed in @xmath56 . due to the sectored antenna model , the random variable @xmath57 with a probability @xmath58 ( @xmath59 ) , where @xmath60 with corresponding probabilities @xmath61 ; the constants @xmath62 , @xmath63 , @xmath64 , @xmath65 , and @xmath66 . note that @xmath67 models the extreme case where the bs and user beams have no alignment at all . note that for the connected mode , since we assume perfect beam alignment between the typical user and its serving bs ( hereby denoted by subscript @xmath68 ) , the directivity gain @xmath69 due to the sectored antenna model . .model parameters [ cols="<,<",options="header " , ] [ table1 ] in this section , we assume that each user is equipped with an energy harvesting circuit , and attempts to extract energy from the incident mmwave signals . no decoding of information is considered in this section . the case with simultaneous information and power transfer is treated in section [ secswipt ] . we first provide analytical expressions to evaluate the energy coverage probabilities for both connected and nonconnected users . we then validate the analytical model , and conclude the section by providing network level design insights . we first provide some lemmas before stating the main analytical results for this section . the probability density function ( pdf ) of the distance from an energy harvesting user to its nearest los bs , given that the user observes at least one los bs , is given by @xmath70 , where @xmath71 and @xmath72 is the probability that the receiver observes at least one los bs . similarly , the distance distribution of the link between the user and its nearest nlos bs , given that the user observes at least one nlos bs , is given by @xmath73 , where @xmath71 and @xmath74 is the probability that the user observes at least one nlos bs . let @xmath75 and @xmath76 denote the probability that the energy harvesting user is connected to a los and a nlos bs respectively , then @xmath77 is given by @xmath78 where @xmath79 and @xmath80 . given that the energy harvesting user is connected to a los mmwave bs , the pdf of the link distance is given by the expression @xmath81 where @xmath71 . given that the user is connected to a nlos mmwave bs , the pdf of the link distance is given by @xmath82 for @xmath71 and @xmath83 . leveraging slivnyak s theorem@xcite , we conduct the analysis at a typical energy harvesting user located at the origin without losing generality . we let @xmath84 be the bs transmit power , and @xmath85 be the power received at the user . recall that @xmath16 denotes the harvester activation threshold defined in section [ secnet ] . the energy harvested at a typical receiver ( in unit time ) can be expressed as @xmath86 where @xmath87 $ ] is the rectifier efficiency . note that we have neglected the noise term since it is extremely small relative to the aggregate received signal . the remaining parameters follow from section [ secsys ] . recall that given a bs @xmath88 , the corresponding fading parameters will be distinct depending on whether the link is los or nlos , which in turn depends on the los probability function ( section [ secnet ] ) . further note that for the connected case , it follows from section [ secant ] that @xmath69 for the link from serving bs ( denoted by subscript @xmath68 ) . the following theorem provides an analytical expression for the energy coverage probability @xmath89 at a connected user , where the random variable @xmath90 is given in ( [ eq1 ] ) and @xmath15 is the energy outage threshold . note that @xmath91 can also be interpreted as the complementary cumulative distribution function ( ccdf ) of the harvested energy . [ thm1 ] in a mmwave network with density @xmath1 , the energy coverage probability @xmath91 for the connected case given an energy outage threshold @xmath15 can be evaluated as @xmath92 where @xmath93 , @xmath94 is given in lemma 2 , while @xmath95 and @xmath96 are the conditional energy coverage probabilities given the serving bs is los or nlos . these terms can be tightly approximated as @xmath97 where @xmath98 , the approximation constant @xmath99 where @xmath100 denotes the number of terms in the approximation , while @xmath101 defines the minimum link distance and is included to avoid unbounded path loss at the receiver . similarly , @xmath102 where @xmath103 , @xmath104^{-n_{{\mathrm{l}}}}}\right ) p(t)t\text{d}t,\end{aligned}\ ] ] @xmath105^{-n_{{\mathrm{n}}}}}\right)\left(1-p(t)\right)t\text{d}t,\end{aligned}\ ] ] and the distance distributions @xmath106 and @xmath107 follow from lemma 3 . see appendix a. recall that @xmath108 is the los probability function defined in section [ secnet ] , and captures the effect of building blockages . in ( [ los ] ) , the term @xmath109 models the contribution from the los serving link , @xmath110 accounts for other los links , and @xmath111 captures the effect of the nlos links . note that the @xmath112th term in ( [ fac1 ] ) , ( [ fac2 ] ) corresponds to the contributions from the bs links with directivity gain @xmath113 . similarly , @xmath114 in ( [ nlos ] ) models the case where the serving bs is nlos . note that these terms further depend on the channel propagation conditions ( @xmath25 , @xmath26 , @xmath39 , @xmath40 , @xmath30 , @xmath115 ) , the network density @xmath1 as well as the antenna geometry parameters ( via @xmath113 , @xmath58 ) , which are summarized in table [ table1 ] . furthermore , the outage threshold @xmath15 will depend on the power requirements at a particular user , and would typically be greater than the sensitivity of the harvesting circuit ( i.e. , @xmath116 such that @xmath117 ) . though the expressions in theorem 1 can be numerically evaluated using tools such as matlab , this could be tedious due to the presence of multiple integrals . to address this , we simplify the analysis by approximating the los probability function with a step function , and ignoring small - scale fading . this results in a much simpler expression for the energy coverage . let @xmath118 , @xmath119 , and @xmath120 for @xmath121 , @xmath122 . the energy coverage probability @xmath123 @xmath124 where @xmath125 is the generalized incomplete gamma function . see appendix b. we note that proposition 1 is remarkably efficient to compute as it involves integration over a finite interval only , and because gamma function can be readily evaluated using most numerical tools . we also observe that the coverage is mainly influenced by the los bss . for example , a key term in ( [ eq : prop1 ] ) is @xmath126 which represents the average number of los bss seen by the user . we now provide analytical expressions for the average harvested power at a connected user . the average harvested power for the connected case @xmath127 for an energy outage threshold @xmath128 is given by @xmath129 the proof follows by noting that @xmath90 has nonnegative support , and by treating @xmath130 as the ccdf of @xmath90 . here , @xmath130 follows from theorem 1 or proposition 1 . @xmath127 can be interpreted as the _ useful _ average harvested power . this is because only those incident signals that meet the activation threshold can be harvested . to get further insights , we now analyze the limiting case @xmath131 of proposition 2 . this provides an upper bound on the average harvested power . the average harvested power for the limiting case @xmath132 , where @xmath133 @xmath134 @xmath135 @xmath136 and @xmath137 . see appendix c. @xmath138 and @xmath139 denote the average harvested power given the user is tagged to an los or an nlos bs . note that the average harvested power is independent of the small - scale fading parameters . to reveal further insights , we provide the following approximation for the average harvested power ( which is validated in section [ secresults1 ] ) . the average harvested power @xmath140 can be further approximated as @xmath141 the proof follows by using the simplifying assumptions of appendix b , and by further ignoring the contributions from other los bss . this approximation suggests that the average power is mainly determined by the los serving link . note that @xmath142 grows linearly with the transmit power @xmath143 . depending on the path loss exponent @xmath144 , it exhibits a sublinear to approximately - linear scaling with the bs density @xmath1 . when @xmath144 is large , the denominator @xmath145 in ( a ) overshadows the impact of @xmath1 on the numerator @xmath146 . therefore , the scaling behavior is essentially determined by @xmath147 . this suggests that increasing the transmit power or bs density has almost the same effect on the average harvested power when @xmath144 is large ( e.g. , when @xmath148 ) . also note that ( [ approx2 ] ) is remarkably simple as it is expressed in terms of the incomplete gamma function only . having discussed the connected case , we now consider the case where a user operates in the nonconnected mode . the following theorem characterizes the energy coverage probability at a typical user for the nonconnected case . in a mmwave network of density @xmath1 , the energy coverage probability for the nonconnected case @xmath149 given an outage threshold @xmath15 can be evaluated using @xmath150 where @xmath110 and @xmath111 are given by ( [ fac1 ] ) and ( [ fac2 ] ) respectively , @xmath151 , and @xmath101 is the minimum link distance . the proof follows from that of theorem 1 and is therefore omitted . similar to the connected case , the energy coverage probability for this case is also a function of the propagation conditions , network density and antenna geometry parameters . we note that the expressions in theorem 2 are very efficient to compute , obviating the need of further simplification . we now consider the average harvested power for the nonconnected case . the average harvested power for the nonconnected case @xmath152 for an energy outage threshold @xmath128 is given by @xmath153 the proof follows from that of proposition 2 and is therefore omitted . the average harvested power for the limiting case @xmath154 is given by @xmath155 where @xmath156 and @xmath157 are given in ( [ fac3 ] ) and ( [ fac4 ] ) respectively . the proof follows from that of corollary 1 and is therefore omitted . the average harvested power for the nonconnected case scales linearly with the transmit power and bs density . this follows from ( [ equation19 ] ) since @xmath156 and @xmath157 scale linearly with @xmath147 . this also suggests that increasing transmit power or density has the same effect on the average harvested power . note that this is different from the connected case where the path loss exponent affects how average harvested power scales with the bs density . we first verify the accuracy of the analytical expressions provided in section [ secstoch ] using simulations . we then study how key design parameters such as the antenna beam pattern affects the energy coverage probability in purely connected ( @xmath158 ) and nonconnected ( @xmath159 ) networks . we also compare the performance of mmwave energy harvesting with lower frequency solutions . after developing key insights for purely connected / nonconnected scenarios , we provide energy coverage results for the general case ( @xmath160 ) , where the network serves both types of users . in the following plots , the users are assumed to be equipped with a single omnidirectional receive antenna , the mmwave carrier frequency is set to 28 ghz , the blockage constant @xmath161 @xcite , and @xmath162 . without loss of generality , we set the rectifier efficiency @xmath163 since this parameter does not impact the shape of the results , i.e. , setting @xmath164 results in shifting all the curves to the left by the same amount . we assume the rectifier efficiency to be the same when comparing mmwave and uhf . note that there are no standard values for @xmath17 since prior work has reported widely varying values@xcite depending on the device technology , operating frequency , etc . for example , @xcite suggests that mmwave energy harvesting circuit may have better overall performance than its lower frequency counterparts . [ fig : con_ant ] and [ fig : con_den ] plot the energy coverage probability for the connected case using different model parameters . the analytical results based on theorem 1 are obtained using @xmath165 terms in the approximation . the simulation results are generated using monte carlo simulations with 10,000 runs . similarly , using theorem 2 , fig . [ fig : uncon_ant ] and [ fig : uncon_den ] plot the energy coverage probability for the nonconnected case . we observe agreement between analytical and simulation results . in fig . [ fig : con_ant ] , we plot the energy coverage probability with three distinct transmit beam patterns for a given network density . we observe that the energy harvesting performance improves with narrower beams ( i.e. , smaller beamwidths and larger directivity gains ) . as the beamwidth decreases , relatively fewer beams from the neighboring bss would be incident on a typical user . but the beams that do reach , will have larger directivity gains , which results in an overall performance improvement . this is possible due to the use of potentially large antenna arrays at the mmwave bss . note that this performance boost will possibly be limited due to the ensuing eirp ( equivalent isotropically radiated power ) or other safety regulations on future mmwave systems @xcite . for the purpose of comparison , we also plot the energy coverage probability for uhf energy harvesting under realistic assumptions . given the current state - of - the - art @xcite , the uhf bss are assumed to have 8 transmit antennas each . further , they are assumed to employ maximal ratio transmit beamforming to serve a connected user . for the channel model , we assume an iid rayleigh fading environment and a path loss exponent of @xmath166 ( no blockage is considered ) . the network density is set to @xmath167 nodes/@xmath168 , which corresponds to an _ average _ distance of about @xmath169 to the closest uhf bs . the carrier frequency is set to @xmath170 ghz and the transmission bandwidth is 100 mhz . as can be seen from fig . [ fig : con_ant ] , mmwave energy harvesting could provide considerable performance gain over its lower frequency counterpart . moreover , the anticipated dense deployments of mmwave networks would further widen this gap . this effect is illustrated in fig . [ fig : con_den ] , where we plot the energy coverage probability for different mmwave network densities for a given transmit antenna beam pattern . in fig . [ fig : con_rate ] , we use proposition 2 to plot the average harvested power at a typical mmwave user against the transmit array size . the plots based on corollary 1 are also included . this figure confirms our earlier intuition that mmwave energy harvesting can benefit from ( i ) potentially large antenna arrays at the bss , and ( ii ) high bs density , which would be the key ingredients of future mmwave cellular systems . [ fig : con_rate2 ] shows how the path loss exponent impacts the scaling behavior of average harvested power with bs density ( see discussion following corollary 2 ) . we now analyze the energy harvesting performance when the harvesting devices operate in the nonconnected mode . in a stark contrast to the connected case , fig . [ fig : uncon_ant ] shows that for the nonconnected case , mmwave energy harvesting could benefit from using wider beams . this is because bs connectivity is critical for the nonconnected case . with wider beams , it is more likely that a mmwave bs gets aligned with a receiver , though sacrificing the beamforming gains . moreover , a comparison with uhf energy harvesting shows that mmwave energy harvesting gives a comparable performance to uhf solutions . similarly , fig . [ fig : uncon_den ] plots the energy coverage probability for different deployment densities . we can observe that performance can be further improved with denser deployments , which would be a key feature in future mmwave cellular systems . having presented the energy coverage trends for the two extreme network scenarios , we now consider the general case where the user population consists of both connected and nonconnected users . we expect this to be the likely scenario for reasons explained in the network model ( section [ secnet ] ) . as described in section [ secant ] , an antenna beam pattern can be characterized by the half power beamwidth and directivity gain for both the main and side lobes . by tuning these parameters , the beam pattern can be particularized to a given antenna array . as an example , we assume that uniform linear arrays ( ula ) are deployed at the mmwave bss . we use the following relations to approximate the main and side lobe beamwidths as a function of the transmit array size : @xmath171 and @xmath172 @xcite . we use @xmath173 and @xmath174 for the directivity gains of the main and side lobes@xcite . to ensure the power normalization , the constant @xmath175 is chosen to satisfy @xmath176 . in fig . [ fig : tradeoff ] , we plot the overall energy coverage probability @xmath177 against transmit array size @xmath178 for different values of parameter @xmath9 . we find that the optimal transmit array size depends on the type of user population . for example , when @xmath9 is large , it is desirable to use large antenna arrays at the bss . when @xmath9 is small , it is favorable to use small antenna arrays to improve the overall energy coverage probability . depending on the network load ( or the user population _ mix _ ) captured via @xmath9 , the energy coverage probability can be substantially improved by intelligent antenna switching schemes . since the parameter @xmath9 would typically vary over large time - scales , such schemes would be practically feasible . [ thb ] for different values of @xmath9 . depending on the fraction of users operating in connected / nonconnnected modes , the transmit array size ( which controls the beamforming beamwidth in this example ) can be optimized to maximize the network - wide energy coverage . this could translate into massive gains given that the number of served devices would be potentially large . the users are assumed to be equipped with a single omnidirectional receive antenna . the energy outage threshold @xmath15 is @xmath179 db for @xmath11 and @xmath180 db for @xmath12 . @xmath181 db , @xmath182 . channel parameters are same as used in fig . [ fig : con_ant ] . ] having presented the energy coverage trends for mmwave energy harvesting , we now consider the scenario where the user attempts to extract both power and information from the incident mmwave signals . in this section , we consider the case where the energy harvesting device also attempts to decode information from the received signals , in what is known as simultaneous wireless information and power transfer ( swipt)@xcite . we now assume that the energy harvesting receiver is also equipped with an information decoding circuit . we focus on the case where a given user is already aligned with its serving bs , i.e. , @xmath183 for this section . further , we consider a power splitting receiver architecture@xcite where the received signal is split using factors @xmath184 and @xmath185 , @xmath186 $ ] . a fraction @xmath185 of received signal is available for energy harvesting , while the remaining signal is used for information decoding . with this notation , the signal - to - interference - plus - noise ratio ( sinr ) at a typical receiver can be expressed as @xmath187 , where @xmath188 denotes the useful signal power and @xmath189 gives the aggregate interference power from the neighboring bss . @xmath190 is the thermal noise power before splitting , while @xmath191 captures possible signal degradation after power splitting . similarly , @xmath192 denotes the received signal power fed to the energy harvester . note that a user will be in outage if the harvested energy and/or the sinr fall below their respective thresholds . we now define @xmath193 $ ] to be the probability of successful reception given the sinr outage threshold @xmath194 , the energy outage threshold @xmath15 , and the power splitting ratio @xmath195 . extending the results from the previous sections , we now provide an analytical expression to characterize the system performance with swipt . before stating the main result of this section , we first provide a lemma for the sinr coverage probability at a mmwave receiver @xcite . in a mmwave network of density @xmath1 , the conditional sinr coverage probability @xmath196 at a swipt device that is not in energy outage , given an sinr outage threshold @xmath194 and a power splitting ratio @xmath195 , is given by @xmath197 where @xmath94 is defined in lemma 2 , and @xmath198 gives the conditional sinr coverage probability given the receiver is served by a los bs , and can be approximated as @xmath199 similarly , the conditional sinr coverage probability for the nlos case @xmath200 is given by @xmath201 where @xmath202^{-n_{{\mathrm{l}}}}}\right ) p(t)t\text{d}t,\end{aligned}\ ] ] @xmath203^{-n_{{\mathrm{n}}}}}\right ) \left(1-p(t)\right)t\text{d}t,\end{aligned}\ ] ] @xmath204^{-n_{{\mathrm{l}}}}}\right ) p(t)t\text{d}t,\end{aligned}\ ] ] @xmath205^{-n_{{\mathrm{n}}}}}\right ) \left(1-p(t)\right)t\text{d}t,\end{aligned}\ ] ] @xmath206 for @xmath59 , @xmath207 and @xmath208 . [ lem4 ] the following theorem provides the main result of this section . in a mmwave network of density @xmath1 , the success probability @xmath209 given the sinr outage threshold @xmath194 , the energy outage threshold @xmath15 , and the power splitting ratio @xmath195 is given by @xmath210 \ ] ] where the sinr coverage probability @xmath211 can be evaluated using the expressions in lemma [ lem4 ] , while the energy coverage probability @xmath212 follows from theorem 1 . we further define @xmath213 , where @xmath214 and @xmath215 can be specified by ( [ los ] ) and ( [ nlos ] ) respectively with @xmath216 . moreover , @xmath217 and @xmath218 are a function of several parameters including the power splitting ratio @xmath195 , the sinr outage threshold @xmath194 , the energy outage threshold @xmath15 , the harvester activation threshold @xmath16 , the rectifier efficiency @xmath17 and the noise parameters . further , @xmath219 @xmath220 and @xmath93 . see appendix d. note that @xmath221 in ( [ eq : thm3 ] ) is the interference ccdf evaluated at parameter @xmath217 . it plays a key role in determining the operating mode of the system . though the interference is harmful for information decoding , it can be beneficial for energy harvesting . when the interference is high , the sinr coverage probability will typically dominate the success probability . in the other extreme , the energy coverage probability will play the dominating role . also note that the success probability can be optimized over the design paramter @xmath195 , given other parameters . moreover , we can recover theorem 1 and lemma 4 from ( [ eq : thm3 ] ) by letting @xmath222 and @xmath223 respectively . note that , in principle , the success probability at a connected mmwave energy harvesting or swipt device can be further improved by leveraging large antenna arrays at the receiver , thanks to smaller wavelengths . though our analytical model allows the users to have receive antenna arrays , it implicitly assumes the presence of ideal rf combining circuitry consisting of power - hungry components such as phase shifters , multiple rf chains , etc . when large antenna arrays are used at the receiver , the power consumption due to additional antenna circuitry may get prohibitively high , overshadowing the array gains . as swipt typically targets low - power devices , we present a simple low - power receiver architecture in the next section . note that the analytical results based on theorem 3 can be interpreted as an upper bound on performance when the receiver consists of suboptimal components ( as is the case in the following section ) . we now propose a novel architecture for a mmwave swipt receiver with multi - antenna array , as depicted in fig . [ fig : model_switches_v1 ] . in this architecture , we assume per - antenna power splitting with parameter @xmath195 ( as defined earlier ) . after power splitting , the input signal at each antenna passes through a rectifier , followed by a dc combiner that yields the harvested energy . for the information path , after passing through power - splitters , the received signals are first combined in the rf domain using a combining vector @xmath224 . the resulting signal is then decoded in the baseband . because they require extremely small power , the combining vector is assumed to be implemented using switches @xcite , i.e. , @xmath225\in[0,1]^{n_\textrm{r}}$ ] . note that both the signals for information decoding and energy harvesting are in the order of @xmath217w ( fig . [ fig : cov_3d ] ) . it is worth mentioning here that recent results have shown that mmwave energy harvesting circuits can run with 1.5 @xmath217w @xcite . [ t ] we now derive the combining gain expression for the proposed swipt receiver architecture in fig . [ fig : model_switches_v1 ] . let @xmath226 be the signal output at the rf combiner . if a bs applies a beamforming vector @xmath227 to send data symbol @xmath228 @xmath229=p_\textrm{t}\right)$ ] to a target user , it follows that @xmath230 , \label{eq : received_signal}\ ] ] where @xmath231 is the channel between the user and its serving bs , and @xmath232 is the received signal due to the interfering bss . since the channel between each user and its bs is assumed to be single - path , the channel matrix @xmath233 , where @xmath234 and @xmath235 are the array response vectors at the bs and user , respectively . recall that @xmath236 denotes the path gain from the serving bs , while @xmath237 and @xmath238 respectively denote the channel angle of arrival and angle of departure at the user and bs . if the channel is known at the bs , and given the antenna model in section [ secant ] , the bs will design the beamforming vector @xmath239 to maximize the beamforming gain , i.e. , to have @xmath240 . denoting @xmath241 , the received signal in can be written as @xmath242 the post - combining sinr can then be expressed as @xmath243 where @xmath244 represents the combining gain at the receiver , and @xmath245 denotes the aggregate interference power . the sinr in ( [ eq : maxsinr ] ) can be maximized if the receiver designs the optimum combining vector , which can be implemented by activating certain antennas on or off . this requires the receiver to have global channel knowledge , which is often challenging in practice . we relax this condition by assuming that the receiver has the angle of arrival information for the serving bs only . ignoring the interference , we propose to design the combining vector by maximizing the @xmath246 instead , i.e. , the receiver designs its combining vector @xmath224 such that @xmath247^{n_\textrm{r } } } \frac{\left|\bw^ * \ba_\mathrm{r}\left(\phi_\mathrm{r}\right)\right|^2 } { \bw^ * \bw } . \label{eq : opt_arch1}\end{aligned}\ ] ] the optimal solution in can be found by an exhaustive search over all possible combinations of @xmath224 . for large receive antenna arrays , this could entail high computational costs , which would further increase the power consumption . therefore , it is important to consider computationally efficient approaches for designing the combining vector . as outlined in algorithm [ alg : alg_arch1 ] , we propose a greedy solution for designing @xmath224 by ( step - wise ) activating only those antennas that boost the received snr . with @xmath248 denoting the combining vector designed using algorithm 1 , the combining gain for the switch - based architecture can be defined as @xmath249 where @xmath250 denotes the wavenumber and @xmath251 is the antenna element spacing . despite its low - complexity , numerical simulations in the next section reveal that our low - power greedy approach could give a good combining gain , without losing substantial performance compared to more advanced but power - hungry solutions . t ] * input * @xmath252 , @xmath237 * initialization * @xmath253 , @xmath254 @xmath255 using theorem 3 , fig . [ fig : cov_3d ] plots the overall success probability for a given transmit antenna beam pattern . the users are equipped with a single - antenna receiver , similar to the one in fig . [ fig : model_switches_v1 ] with @xmath256 . first , fig . [ fig : cov_3d ] shows that a good success probability can be obtained with mmwave swipt system for typical mmwave propagation and system parameters . further , this plot illustrates that the power splitting ratio @xmath195 needs to be optimized for a given sinr outage threshold to maximize the overall success probability . matching the intuition , the figure shows that in the low sinr outage regime ( when @xmath194 is large ) , it is desirable to divert more power to the information decoding module , while a larger fraction of power needs to be portioned for the energy harvesting system in the high sinr outage regime ( when @xmath194 is small ) . this trend is consistent with prior studies on swipt architectures@xcite . we now evaluate the performance of the proposed low - power receiver architecture for different number of receive antennas . in fig . [ fig : cov_switch ] , the success probability @xmath257 is plotted for a fixed transmit antenna beam pattern . for the proposed architecture , the combining vector is obtained using algorithm 1 , and the curves are averaged over the angle of arrival parameter . for comparison , we also plot the success probability for ( fully digital ) maximal ratio combining ( mrc ) receivers . we observe that the success probability improves with the receive antenna array size . further , when the sinr outage threshold @xmath194 is small , the success probability is mainly limited by the energy outage . this also explains why the success probability converges to a limit ( determined by the energy outage threshold ) as @xmath194 decreases . moreover , there are diminishing returns as the number of antennas are increased . a comparison with power - hungry mrc receivers shows that the proposed switch - based architecture performs reasonably well . this is particularly desirable for future mmwave swipt devices . [ t ] , and the sinr outage threshold @xmath194 for a given energy outage threshold @xmath15 and network density @xmath1 . as @xmath194 gets large , the system becomes sinr - limited , and the optimum value of @xmath195 increases , suggesting that a larger fraction of received signal should be used for information extraction to optimize the overall success probability . the transmit antenna beam pattern is set to @xmath258 . other parameters include @xmath259 db , @xmath260 db , @xmath261 , and @xmath262 dbm . ] [ t ] at the user given a fixed transmit beam pattern @xmath258 at the bss . proposed low - power architecture achieves good performance compared to superior receiver architectures . other parameters include @xmath263 , @xmath259 db , @xmath260 db , @xmath261 , and @xmath262 dbm . ] in this paper , we analyzed the energy harvesting performance at low - power devices powered by a mmwave cellular network . using a stochastic geometry framework , we derived analytical expressions characterizing the performance of mmwave energy and information transfer in terms of system , channel and network parameters . simulations results were used to validate the accuracy of the derived expressions . leveraging the analytical framework , we also provided useful network and device level design insights . for the connected case when the transmitter and receiver beams are aligned , results show that the energy coverage improves with narrower beams . in contrast , wider beams provide better energy coverage when the receivers are not aligned with a particular transmitter . this trade - off is evident in the more general scenario having both types of receivers , where there typically exists an optimal beamforming beamwidth that maximizes the network - wide energy coverage . moreover , we found that several device - related parameters can significantly impact the system performance . for example , the performance can be substantially improved by optimizing over the power splitting ratio and by leveraging large antenna arrays . to allow using multiple antennas at the mmwave receivers while keeping the power consumption low , we proposed a low - power receiver architecture for mmwave energy and information transfer using antenna switches . simulation results show that the proposed architecture can provide good gains for the overall mmwave energy harvesting performance . simulation results also reveal that mmwave cellular networks could potentially provide better energy coverage than lower frequency solutions . the following inequality approximates the tail probability of a normalized gamma distribution . for a normalized gamma random variable @xmath264 with parameter @xmath100 , the probability @xmath265 can be tightly upper - bounded by @xmath266 , where the constant @xmath71 and @xmath99 . we write @xmath267=\pr\left[s+i>\hat{\psi}\right]$ ] , where @xmath268 is the received signal power from the serving bs , and @xmath269 is the received signal power from all the other bss . we can derive the result in theorem [ thm1 ] by finding the conditional distributions @xmath270 and @xmath271 . to proceed , first consider the conditional distribution @xmath272 given the receiver is aligned with a los bs ( which is indicated by the subscript @xmath273 in the following notation ) . @xmath274 \overset{(a)}{\approx}\mathbb{e}_{s , i|{{\mathrm{l}}}}\left[\left(1-e^{-a\frac{s+i}{\psi}}\right)^{{n}}\right ] \nonumber \\ & \overset{}{=}\mathbb{e}_{s , i|{{\mathrm{l}}}}\left[\sum\limits_{k=0}^{n}\left(-1\right)^{k}\binom{n}{k}e^{-a k\frac{s+i}{\psi}}\right ] \overset{}{=}\sum\limits_{k=0}^{n}\left(-1\right)^{k}\binom{n}{k}\mathbb{e}_{s , i|{{\mathrm{l}}}}\left[e^{-\hat{a } \left(s+i\right)}\right]\end{aligned}\ ] ] where we have included a dummy random variable @xmath275.17ex@xmath276 in the first equation . note that @xmath264 converges to @xmath277 as @xmath278 . therefore , this substitution is in fact an approximation when @xmath100 is finite . the introduction of @xmath264 allows leveraging the inequality in lemma 5 , which leads to @xmath279 , where the constant @xmath99 . the last equation follows from the binomial series expansion of @xmath280 , and by further substituting @xmath281 . to evaluate the expectation in ( [ temp ] ) , consider @xmath282=\mathbb{e}_{s|{{\mathrm{l}}}}\left[e^{-\hat{a}s } \mathbb{e}_{i|{s,{{\mathrm{l}}}}}\left[e^{-\hat{a}i}\right]\right ] . \ ] ] the inner expectation in ( [ temp2 ] ) can be simplified by applying the thinning theorem for a ppp @xcite . note that @xmath4 can be independently thinned into two ppps @xmath283 and @xmath284 , where the former comprises the los bss whereas the latter consists of nlos bss . therefore , we can interpret @xmath285 and @xmath286 as two independent tiers of bss . the user will be tagged with either the closest bs in @xmath285 or in @xmath286 , whichever maximizes the average received power at the user . we can further thin @xmath283 into four independent ppps @xmath287 , where each resulting ppp @xmath288 contains bss that correspond to a nonzero directivity gain @xmath113 with @xmath58 being the thinning probability . this follows because the beam orientations are assumed to be independent across links . thus , a link can have a directivity gain of @xmath113 with probability @xmath58 independently of other links . we let the received power due to the transmission from the bss in @xmath288 be @xmath289 . likewise , @xmath284 can be split into @xmath290 with the corresponding received powers denoted by @xmath291 . since the resulting ppps are independent , ( [ temp2 ] ) can be simplified as @xmath292=\prod\limits_{i=1}^{4 } \mathbb{e}_{i|{s,{{\mathrm{l}}}}}\left[e^{-\hat{a}i^{i}_{{\mathrm{l}}}}\right ] \prod\limits_{j=1}^{4}\mathbb{e}_{i|{s,{{\mathrm{l}}}}}\left[e^{-\hat{a}i^{j}_{{{\mathrm{n}}}}}\right]\end{aligned}\ ] ] where @xmath293 \overset{(a)}{=}\mathbb{e}_{\phi_{\mathrm{l}}^i|r_o}\left[\prod\limits_{\ell\in\phi_{\mathrm{l}}^i\setminus\mathbb{b}\left(r_o\right ) } \mathbb{e}_{h_{\ell}}\left[e^{-\hat{a}p_\textrm{t}h_{\ell}d_ic_{\mathrm{l}}{r_{\ell}}^{-{\alpha}_{\mathrm{l}}}}\right]\right ] $ ] @xmath294 = e^{-2\pi\lambda p_i\int\limits_{r_o}^{\infty}\left(1-\left(\frac{1}{1+\hat{a}p_\textrm{t}d_ic_{\mathrm{l } } { t}^{-{\alpha}_{\mathrm{l}}}{n_\mathrm{l}}^{-1 } } \right)^{n_{\mathrm{l}}}\right)p(t)t\text{d}t } \end{aligned}\ ] ] where @xmath279 follows by conditioning on the length @xmath295 of the serving los link , and by further noting that small - scale fading is independent across links . here , @xmath296 denotes a circular disc of radius @xmath295 centered at the typical user . @xmath280 is obtained by using the moment generating function of a normalized gamma random variable , while the last equation follows by invoking the probability generating functional@xcite of the ppp @xmath297 . substituting ( [ temp3 ] ) in the first ( left ) product term of ( [ temp4 ] ) yields ( [ fac1 ] ) . similarly , @xmath298 $ ] is given by @xmath299 = e^{-2\pi\lambda p_i\int\limits_{\rho_{\mathrm{l}}(r_o)}^{\infty}\left(1-\left(\frac{1}{1+\hat{a}p_\textrm{t}d_ic_{\mathrm{n } } { x}^{-{\alpha}_{\mathrm{n}}}{n_\mathrm{n}}^{-1 } } \right)^{n_{\mathrm{n}}}\right)\left(1-p(t)\right)t\text{d}t } .\end{aligned}\ ] ] by substituting ( [ temp5 ] ) in the second ( right ) product term of ( [ temp4 ] ) yields ( [ fac2 ] ) . using the expressions in ( [ temp4])([temp5 ] ) in ( [ temp2 ] ) , and by further evaluating the expectation of the resulting expression with respect to @xmath300 , we obtain @xmath301 where we have again used definition of the moment generating function of a normalized gamma distribution . @xmath110 and @xmath111 are given in ( [ fac1 ] ) and ( [ fac2 ] ) respectively , @xmath101 denotes the minimum link distance , while the distance distribution is provided in lemma 3 . using ( [ temp ] ) and ( [ temp2 ] ) , we can thus retrieve the expression in ( [ los ] ) . we can similarly derive the conditional distribution @xmath302 in ( [ nlos ] ) for the nlos case . to simplify the analysis , we approximate the los probability function @xmath8 by a step function @xmath303 , i.e. , the bss within a los ball of radius @xmath304 are marked los with probability 1 , while the rest as nlos@xcite . the radius @xmath304 is chosen such that the los association probability @xmath305 remains the same ( as in our original model ) . using a step function for @xmath8 , it follows from lemma 2 that @xmath306 and @xmath307 . moreover , we neglect small scale fading for all the links except for the serving bs . we also ignore the nlos signals in the analysis . this effectively leads to a scenario where the user receives signals from the bss within the los ball only . intuitively , this would be the likely scenario in sufficiently dense networks . we only list the key steps since the rest of the proof follows from appendix a. ignoring the nlos signals , we approximate ( [ temp4 ] ) as @xmath308\approx\prod\limits_{i=1}^{4 } \mathbb{e}_{i|{s,{{\mathrm{l}}}}}\left[e^{-\hat{a}i^{i}_{{\mathrm{l}}}}\right]$ ] where @xmath309\overset{}{=}\mathbb{e}_{\phi_{\mathrm{l}}^i|r_o}\left[\prod\limits_{\ell\in\phi_{\mathrm{l}}^i\cap[\mathbb{b}\left(r_b\right)\setminus\mathbb{b}\left(r_o\right ) ] } e^{-\hat{a}p_\textrm{t}d_ic_{\mathrm{l}}{r_{\ell}}^{-{\alpha}_{\mathrm{l}}}}\right ] \overset{(a)}{=}e^{-2\pi\lambda p_i\int\limits_{r_o}^{r_b}\left(1- e^{-\hat{a}p_\textrm{t}d_ic_{\mathrm{l}}{t}^{-{\alpha}_{\mathrm{l } } } } \right)t\text{d}t } \nonumber\\ & \overset{(b)}{=}e^{-\frac{2\pi\lambda p_i}{\alpha_{{\mathrm{l}}}}\int_{w_{ik}r_o^{-\alpha_{{\mathrm{l}}}}}^{w_{ik}r_b^{-\alpha_{{\mathrm{l}}}}}\frac{1-e^{-v}}{v^{1+\frac{2}{\alpha_{{\mathrm{l}}}}}}\text{d}v } \overset{(c)}{=}e^{-{\pi\lambda p_i\left(r_b^2-r_o^2\right ) } } e^{-\frac{2\pi\lambda p_iw_{ik}^{\frac{2}{\alpha_{{\mathrm{l}}}}}}{\alpha_{{\mathrm{l}}}}\gamma\left(\frac{-2}{\alpha_{{\mathrm{l } } } } ; w_{ik}r_o^{-\alpha_{{\mathrm{l } } } } , w_{ik}r_b^{-\alpha_{{\mathrm{l}}}}\right)}\end{aligned}\ ] ] where ( a ) follows by ignoring the small scale fading and invoking the probability generating functional@xcite of ppp , ( b ) by a change of variables , and ( c ) by the definition of the generalized incomplete gamma function . the result in proposition 1 is obtained by assuming @xmath310 , and further noting that the distance distribution simplifies to @xmath311 due to the los ball approximation . we can derive corollary 1 by finding the conditional means @xmath312 $ ] and @xmath313 $ ] . @xmath138 can be evaluated by conditioning on the link distance @xmath314 from the serving bs as follows . @xmath315&= \mathbb{e}_{}\left[s|r_0,{{\mathrm{l}}}\right ] + \sum_{i=1}^{4}\mathbb{e}_{}\left[i_{{\mathrm{l}}}^i+i_{{\mathrm{n}}}^i|r_0,{{\mathrm{l}}}\right ] \overset{(a)}{=}p_\textrm{t } m_{\textrm{t}}m_{\textrm{r}}c_{\mathrm{l } } { r_0}^{-{\alpha}_{\mathrm{l}}}+\sum_{i=1}^{4}2\pi\lambda p_\textrm{t}p_id_i c_{\mathrm{l}}\int\limits_{r_0}^{\infty}t^{-\left({\alpha}_{\mathrm{l}}-1\right)}p(t)\text{d}t\nonumber\\ & \qquad\qquad\qquad\qquad\qquad+\sum_{i=1}^{4}2\pi\lambda p_\textrm{t } p_i d_i c_{{\mathrm{n}}}\left[\displaystyle\frac{\left(\rho_{{\mathrm{l}}}(r_0)\right)^{-(\alpha_{{\mathrm{n}}}-2)}}{\alpha_{{\mathrm{n}}}-2 } -\int\limits_{{\rho}_{{\mathrm{l}}}(r_0)}^{\infty } t^{-\left({\alpha}_{{\mathrm{n}}}-1\right)}p(t)\text{d}t\right ] \nonumber\\ & = p_\textrm{t } m_{\mathrm{t } } m_{\mathrm{r } } c_{{\mathrm{l}}}{r_0}^{-\alpha_{{\mathrm{l}}}}+{\psi}_{{\mathrm{l}}}\left(r_0\right)+{\psi}_{{\mathrm{n}}}\left(\rho_{{\mathrm{l}}}(r_0)\right)\end{aligned}\ ] ] where @xmath279 is obtained by averaging over the fading distribution , followed by invoking campbell s theorem@xcite , while ( [ appb ] ) follows from the definitions of @xmath316 and @xmath317 provided in ( [ fac3 ] ) and ( [ fac4 ] ) respectively . taking expectation of @xmath318 $ ] with respect to @xmath295 using lemma 3 yields ( [ alos ] ) . the expression for @xmath139 is ( [ anlos ] ) can be derived using similar steps . recall that @xmath319 . let @xmath320 where @xmath188 and @xmath189 . to find @xmath321 $ ] , consider @xmath322 \nonumber \\ & = { \mathbb{e}}_{i}\left[\pr\left[s > t\left(i+\sigma^2+\frac{\sigma^2_c}{\nu}\right ) , s > \frac{\hat{\psi}}{\left(1-\nu\right)}-i-\sigma^2\right]\right ] \nonumber \\ & \overset{}{=}{\mathbb{e}}_{i}\left[\pr\left[s > t\left(i+\sigma^2+\frac{\sigma^2_c}{\nu}\right)\right]\right]\pr\left[i>\mu\right ] + { \mathbb{e}}_{i}\left[\pr\left[s > \frac{\hat{\psi}}{\left(1-\nu\right)}-i-\sigma^2\right]\right ] \pr\left[i\leq\mu\right ] \nonumber \\ & = p_{\text{cov}}(\lambda , t,\nu)\tilde{p}_{\textrm{con}}(\lambda,\mu ) + { p}_{\text{con}}(\lambda,\varphi)\left[1-\tilde{p}_{\textrm{con}}(\lambda,\mu)\right]\end{aligned}\ ] ] where the sinr coverage probability @xmath323 follows from lemma [ lem4 ] , and the energy coverage probability @xmath324 from theorem [ thm1 ] . @xmath221 is the interference ccdf evaluated at @xmath217 , where parameter @xmath217 follows from the inequality @xmath325 . c. valenta and g. durgin , `` harvesting wireless power : survey of energy - harvester conversion efficiency in far - field , wireless power transfer systems , '' _ ieee microw . _ , vol . 15 , pp . 108120 , jun 2014 . a. sakr and e. hossain , `` cognitive and energy harvesting - based d2d communication in cellular networks : stochastic geometry modeling and analysis , '' _ ieee trans . _ , vol . 63 , pp . 18671880 , may 2015 .

in this work i study the problem of e / b - mode separation with binned cosmic shear two - point correlation function data . motivated by previous work on e / b - mode separation with shear two - point correlation functions and the practical considerations of data analysis , i consider e / b - mode estimators which are linear combinations of the binned shear correlation function data points . i demonstrate that these estimators mix e- and b - modes generally . i then show how to define estimators which minimize this e / b - mode mixing and give practical recipes for their construction and use . using these optimal estimators , i demonstrate that the vector space composed of the binned shear correlation function data points can be decomposed into approximately ambiguous , e- and b - mode subspaces . with simple fisher information estimates , i show that a non - trivial amount of information on typical cosmological parameters is contained in the ambiguous mode subspace computed in this formalism . next , i give two examples which apply these practical estimators and recipes to generic problems in cosmic shear data analysis : data compression and spatially locating b - mode contamination . in particular , by using wavelet - like estimators with the shear correlation functions directly , one can pinpoint b - mode contamination to specific angular scales and extract information on its shape . finally , i discuss how these estimators can be used as part of blinded or closed - box cosmic shear data analyses in order to assess and find b - mode contamination at high - precision while avoiding observer biases . = 0 [ firstpage ] gravitational lensing : weak ; cosmology : theory ; methods : data analysis

1208.0068

cosmic shear , or the weak gravitational lensing of background galaxies by cosmological density fields , is one of the most important techniques for probing the properties of dark energy ( see , e.g. , @xcite @xcite for a recent review ) and also the growth of structure predicted by general relativity ( gr ) or its possible modifications ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? cosmic shear measurements can also help constrain other cosmologically interesting signals , such as primordial non - gaussianity ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the properties of various hot and warm dark matter models ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , or the properties of neutrinos ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? to this end , ongoing and planned wide - field optical surveys , such as the des , lsst , euclid , wfirst , hsc , kids , and pan - starrs surveys , will measure the shapes of hundreds of millions to billions of galaxies and thus cosmic shear signals with unprecedented statistical precision . given this incredible statistical power , understanding and mitigating potential systematic errors in these measurements will be very important . systematic contamination to cosmic shear signals can arise from a variety of sources , including the process of observing and estimating galaxy shapes from pixelated images ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) or estimating photometric redshifts ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? ? * ; * ? ? ? there are also astrophysical sources of systematic errors , like intrinsic alignments ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , source galaxy clustering @xcite , or the effects of baryons and galaxy formation on the matter power spectrum ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . these systematic errors , if left uncontrolled , can bias and/or degrade constraints on the properties of dark energy or modifications to gr from future surveys ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? besides direct image and structure formation simulations to study cosmic shear data analysis , systematics , and theoretical modeling in detail ( e.g. , step1 @xcite ; step2 @xcite ; great08 @xcite ; great10 @xcite ; @xcite ) , it is important to distinguish between observational signals which can arise from gr and those which can not @xcite . at first order in the gravitational potential , gr will only produce cosmic shear patterns known as e - modes ( see , e.g. , @xcite @xcite for a pedagogical introduction ) . the complementary patterns , know as b - modes , are not produced by gr at first order , though they can be produced in small amounts at higher order ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? many of the sources of systematic contamination , though not all , can produce b - modes in addition to e - modes ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . therefore assessing b - mode contamination in cosmic shear signals can test for systematic errors throughout the various steps of cosmic shear data analysis , from observing the galaxies with a telescope and imaging camera , all the way through to the theoretical modeling and constraints on cosmological parameters . methods for separating e- and b - modes in cosmic shear data have been studied extensively by previous authors . broadly , these methods either operate directly on the shear field @xcite or on the shear two - point correlation functions @xcite . e / b - mode separation in the context of higher - order correlation functions has been studied as well @xcite . additionally , techniques originally designed for the analysis of cosmic microwave background polarization signals ( e.g. , * ? ? ? * ; * ? ? ? * ) can also be applied to cosmic shear @xcite . importantly , the details of the implementation of these methods can effect their performance significantly ( e.g. , * ? ? ? * ; * ? ? ? * ) . for the shear two - point correlation functions , @xmath0 defined below , @xcite have shown that a broad class of e / b - mode statistics , can be written in the following form @xmath1\nonumber\\ b_{c } & = & \frac{1}{2}\int_{l}^{h}d\theta\,\theta\big[t_{+}(\theta)\xi_{+}(\theta ) - t_{-}(\theta)\xi_{-}(\theta)\big]\nonumber\ .\end{aligned}\ ] ] by choosing the range of integration @xmath2 $ ] and the forms of @xmath3 properly , one can show that @xmath4 will contain only e - mode information and @xmath5 will contain only b - mode information either over an infinite interval or finite interval with @xmath6 @xcite . note that these statistics assume one has continuous shear correlation function data . practical implementations of the @xmath7 and @xmath8 statistics in cosmic shear data analysis are constrained in several ways . the shear correlation functions are usually estimated in bins of angle , say @xmath9 bins , with some effective binning weight @xmath10 . in particular , the expectation value of the estimated shear correlation function data point for the @xmath11th bin is ( e.g. , * ? ? ? * ) @xmath12 the form of these binning functions and this result is discussed below and in appendix [ app : srcclust ] . thus the statistics @xmath4 and @xmath5 in this case must be estimated from the binned shear correlation function data , @xmath13 . also , in order for the statistics @xmath4 and @xmath5 to remain pure two - point statistics , any procedure for estimating them from cosmic shear data can only consider linear combinations of the binned shear correlation function data , @xmath14\ , \ ] ] where the @xmath15 are constants which describe the statistics ( see e.g. , * ? ? ? * ) . in this work , i study the use of these linear combinations for e / b - mode separation and thus the effects of these constraints on cosmic shear data analysis . in particular , after covering the basics of cosmic shear in section [ sec : cosmodefs ] , i demonstrate in section [ sec : ebmix ] that _ even if the statistics @xmath4 and @xmath5 contain pure e- and b - mode information , the statistics @xmath16 generally exhibit e / b - mode mixing due to the binning_. in section [ sec : buildebest ] , i show how to define the statistics @xmath16 through the choice of the @xmath15 , so that they suppress the e / b - mode mixing below detectable levels for any current or upcoming cosmic shear survey . in this section i also give practical recipes for computing the @xmath15 . other potential and ultimately equivalent optimal estimator definitions are discussed in section [ sec : otherest ] . i then show how to use these optimal estimators to divide the vector space of correlation function data points up into approximately ambiguous , e- and b - mode subspaces in section [ sec : ebmodedec ] . i compute the gaussian covariances of these statistics in section [ sec : ebcov ] . in sections [ sec : cosebi ] and [ sec : wvstats ] , i provide two examples of these statistics to illustrate how to build and use them in practice . i provide an example of how to decompose the shear correlation functions into ambiguous , e- and b - modes and also discuss the fisher information content of these subspaces in section [ sec : fishinfo ] . i find that the ambiguous mode subspace has a non - trivial amount of information about typical cosmological parameters . finally , i conclude and discuss how these statistics are applicable to blinded or closed - box cosmic shear data analyses in section [ sec : conc ] . the basic equations describing weak lensing by cosmological density fields are covered in detail in other works ( see e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the discussion presented here is merely a summary of the relevant results needed for this work . given the 3d density matter power spectrum @xmath17 as a function of wave number @xmath18 and scale factor @xmath19 , the 2d convergence power spectrum as a function of 2d wave number @xmath20 is defined in the limber approximation as ( cf . * ) @xmath21 where @xmath22 is the redshift , @xmath23 is the comoving distance , and @xmath24 is the redshift distribution of the lensing sources for source set @xmath11 normalized to the total source density , @xmath25 . these expressions assume straight - line photon paths from the sources to the observer , commonly called the born approximation , and a spatially flat universe . i use the non - linear power spectrum fitting formula of @xcite to compute @xmath26 and the fitting formula of @xcite to evaluate the linear power spectrum . additionally , in this work all lensing sources are at a single redshift , @xmath27 , such that @xmath28 , where @xmath29 is the dirac delta function . in cosmological weak lensing , one observes the shear field ( neglecting reduced shear effects , see e.g. , @xcite @xcite ; @xcite @xcite ) along with an assumed to be random contribution from galaxy shapes and orientations . this last effect is commonly called shape noise and is characterized by the shape noise per component , @xmath30 . the breaking of the assumption of random galaxy orientations is generically referred to as intrinsic alignments and is a primary source of systematic error in cosmic shear measurements ( see the references given above ) . given the complex shear field , @xmath31 , one can define the @xmath32 correlation functions as ( cf . * ) @xmath33 where @xmath34 , @xmath35 , and @xmath36 is the polar angle of the vector connecting the two points . in appendix [ app : srcclust ] , i present the standard expressions for galaxy pair - wise shear correlation function estimators and their expectation values ( see , e.g. , * ? ? ? * ; * ? ? ? * ) . in fourier - space , the shear field is typically separated into a component with no net handedness , the e - mode part , and a handed component , the b - mode part . in terms of the power spectra of the e- and b - mode parts , these correlation functions are ( cf . * ) @xmath37\label{eqn : xipell}\\ \xi_{-}(\theta ) & = & \int_{0}^{\infty}\frac{d\ell\,\ell}{2\pi}j_{4}(\ell\theta)\left[p_{e}(\ell ) - p_{b}(\ell)\right]\label{eqn : ximell}\end{aligned}\ ] ] where the @xmath38 are cylindrical bessel functions . in the born approximation , the b - mode power is identically zero , @xmath39 , and the e - mode power is equal to the convergence power spectrum , @xmath40 . corrections to the born approximation are very small ( e.g. , * ? ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the shape noise contribution to the power spectra , @xmath41 , has been purposefully left out of these expressions because pair - wise estimators of the shear correlation functions ( see appendix [ app : srcclust ] ) do not exhibit noise biases @xcite . finally , the covariance of the shear - shear correlation functions can be computed under the assumption that the shear fields are gaussian with the following expressions from @xcite @xmath42\right\ } + \delta_{\theta_{1}\theta_{2}}\frac{4\sigma_{e}^{4}}{\bar{n}^{2}2\pi\theta_{1}\delta\theta_{1}\omega_{s}}\nonumber\end{aligned}\ ] ] @xmath43\bigg\}\label{eqn : shearcov}\end{aligned}\ ] ] where @xmath44 is the bin width and @xmath45 is the survey area . note that the shape noise contributes to cross terms in braces in addition to the diagonal terms given by the kronecker delta function . the covariance between @xmath46 and @xmath47 is given by second of the above expressions . the expressions for the correlation function covariance matrix will be useful below for computing the fisher information content of the shear correlation functions under the assumption the errors are gaussian . the fisher information matrix is ( e.g. , * ? ? ? * ) @xmath48\ ] ] where @xmath49 is the covariance matrix of the observations , @xmath50 , and @xmath51 . here @xmath52 is the vector of mean values of the data and the notation @xmath53 indicates a partial derivative with respect to parameter @xmath54 . below i neglect the information in the covariance matrix so that the fisher information is computed only using the last term in the equation above . there are several conventions in the literature for comparing the fisher information content of various analyses . i roughly follow @xcite and simply compare various analyses by computing @xmath55 for a fiducial set of parameters for each analysis . i use @xmath56 , the normalization of the linear matter power spectrum today filtered in 8 @xmath57mpc spheres , and @xmath58 , the mean matter density today in units of the critical density , as the fiducial set of parameters for comparison . these fisher information estimates are not meant to be realistic survey projections , but to give a sense of how much of the total fisher information is retained by the e / b - mode statistics presented in this paper . throughout this work i assume a flat @xmath59cdm universe with @xmath60 , @xmath61 kms@xmath62mpc@xmath62 , @xmath63 , @xmath64 , @xmath65 , and @xmath66 . also , i consider two different prototypical weak lensing surveys with different areas , @xmath45 , and lensing source number densities , @xmath67 . the first survey has @xmath68 , typical of the des and the second has @xmath69 , typical of the lsst survey . i set the lensing shape noise per component to @xmath70 for both the des- and lsst - like surveys . in this section i study the problem of e / b - mode estimation with binned cosmic shear data in detail . i show that generally estimators which are linear combinations of the binned cosmic shear data points mix e- and b - mode signals . i then demonstrate how to define optimal estimators which minimize this e / b - mode mixing and give practical recipes for constructing and using these estimators . next i discuss other potential optimal estimator definitions , demonstrating that they are approximately equivalent to the definition used throughout this work . then i discuss how to decompose the vector space composed of the binned shear correlation functions into approximately ambiguous , e- and b - mode subspaces . finally , i compute the variance and covariance of these estimators under the assumption the shear power spectra are gaussian . as stated above , shear correlation functions are typically measured in a bins of angle so that @xmath71 $ ] for the @xmath11th bin . additionally , for the standard shear two - point function estimator , the shear correlation function measurements are weighted by the bin weighting function @xmath10 , as in equation ( [ eqn : xpmdef ] ) . generally these bin weighting functions are normalized to unity so that @xmath72 . throughout this work i assume the bin weighting functions are normalized properly . as detailed in appendix [ app : srcclust ] , for pair - wise estimators of the shear correlation functions this bin weighting function is in general quite complicated because of the survey window function and source galaxy clustering . note however that for unclustered sources and neglecting the survey window function , @xmath73 , so that the dominant effect is a purely geometric weighting . this effect arises from the increase in the number of galaxy pairs due to the increase in the area in the outer part of the bin relative to the inner part . i show in appendix [ app : srcclust ] that by using small enough bins , the effects of the source clustering can be made negligible . assessing the form and magnitude of the weightings from the survey window function is beyond the scope of this work . thus when needed , i only use the geometric weightings given above . given the bin window functions , i now consider statistics which are general linear combinations of the shear correlation function data points described by equation ( [ eqn : xpmlinstat ] ) . the form of these statistics has been chosen for easy comparison with work for e / b - mode statistics for unbinned shear correlation function data presented by @xcite . it is important to realize however that i have included a bin size dependent weight through the normalization of the @xmath10 in the definition of the statistics unlike @xcite . the correspondence between the @xmath74 coefficients , which define the statistics @xmath16 , and the @xmath75 statistics presented in @xcite is @xmath76 for each bin @xmath11 . i will now show that _ the statistics @xmath16 always mix e- and b - mode information _ , barring a very special choice of the bin window functions @xmath10 . this derivation follows closely a derivation presented in @xcite , but accounts for the binning explicitly . in order to proceed , i substitute the definitions of the shear correlation functions in terms of the e- and b - mode power spectra from equations ( [ eqn : xipell ] ) and ( [ eqn : ximell ] ) into the definition of @xmath16 to get the following expression for the expectation value of the statistics @xmath77 these equations with @xmath39 were presented by @xcite . the statistic @xmath78 will have no b - mode contribution if @xmath79 for all @xmath20 and thus would be an e - mode statistic . similarly , @xmath80 would be a b - mode statistic . setting @xmath79 , i can integrate over @xmath20 against @xmath81 to get @xmath82 this equation must hold for arbitrary @xmath36 . now let @xmath83 $ ] . using the closure relationship for the bessel functions @xmath84 and the following result from @xcite @xmath85 i get the following expression for @xmath86 in terms of the @xmath87 @xmath88 here @xmath89 is the heaviside step function and @xmath90 is the dirac delta function . this equation is the discrete analog of the results presented in @xcite . in fact , it can be obtained by setting the quantities @xmath91 from @xcite to @xmath92 in each bin @xmath93 $ ] . originally i assumed that the @xmath15 were constants not functions of @xmath36 , but the equation derived above in general has dependence on @xmath36 . therefore generally @xmath94 will not be identically zero for all @xmath20 . the only way to ensure @xmath95 is to pick the bin window functions @xmath96 so that they exactly cancel the @xmath36 dependence in the equation above and then use this equation to compute the @xmath97 in terms of the @xmath87 . apart from this caveat , statistics of the form @xmath16 will mix e- and b - mode information in general . this mixing arises directly from the binning and is a general feature of e / b - mode separation with pixelized data as well ( see e.g. , * ? ? ? * ; * ? ? ? these results have important implications for the theoretical analysis of binned cosmic shear correlation function data . for binned shear correlation functions with @xmath9 bins , it would be nice to divide the space of the 2@xmath9 data points into pure e - mode , pure b - mode , and ambiguous modes , similar to the decomposition achieved by @xcite with the cosebi statistics over the continuous function space of the shear correlation functions . however , i have shown that for general window functions @xmath10 , this decomposition is impossible because there do not exist pure e- and b - mode linear combinations . in the next section , i will define approximately pure e- and b - mode linear combinations , along with approximately ambiguous modes . thus i will show that there does in fact exist a division of the space of @xmath98 data points into _ approximately _ ambiguous , e- and b - mode subspaces . this approximate decomposition is explored fully in section [ sec : ebmodedec ] . these results are also quite useful for cosmic shear data analysis . suppose one did in fact use a statistic which is a linear combination of the shear correlation function data points . then from equations ( [ eqn : xipmell ] ) and ( [ eqn : wpmdef ] ) one can compute how the statistic mixes e- and b - modes due to the binning . additionally as i will show below , as the shear correlation function bins are made smaller , the magnitude of @xmath94 will decrease , so that the e / b - mode mixing decreases as well . if the bins are made small enough , the bias in @xmath80 , the b - mode statistic , due to e / b - mode mixing , can be made smaller than the statistical errors . thus these results give a quantitative criterion by which to decide the number and size of the bins used to compute the shear correlation functions . finally , equation ( [ eqn : fmnobin ] ) suggests two ways to minimize the e / b - mode mixing . the first is to reweight the data in each bin by choosing the @xmath96 to cancel the @xmath36 dependence in equation ( [ eqn : fmnobin ] ) and then use this equation to compute the @xmath97 . i will not consider this possibility here . the second is to pick the @xmath97 in order to minimize the e / b - mode mixing without adjusting the bin window functions . given a fiducial choice for the @xmath87 , one can define @xmath97 in some way ( roughly similar to equation ( [ eqn : fmnobin ] ) ) in order to minimize the mode mixing by minimizing @xmath94 . the details of this definition along with a general procedure for computing the @xmath15 are described in the next section . consider now binned shear correlation function data over the range @xmath2 $ ] in angle with @xmath9 bins . each bin is described by the bin window functions introduced above so that the bins cover the range @xmath2 $ ] without any overlaps between the ranges of each bin , @xmath93 $ ] . in order to define the @xmath86 , i minimize the square amplitude of the window function @xmath94 with respect to the coefficients @xmath86 , @xmath99\ .\ ] ] the solution to this equation is @xmath100 this solution is equivalent to multiplying equation ( [ eqn : fmnobin ] ) by @xmath101 and then integrating over @xmath36 with the weight @xmath102 . similarly to the unbinned estimators described in @xcite , the binned estimators described in this work must satisfy two integral constraints in order to be non - zero only over a finite range in angle . these constraints can be understood as follows . suppose that the @xmath87 are non - zero only in the interval @xmath103 $ ] contained in @xmath2 $ ] . now consider @xmath104 . then according to equation ( [ eqn : fmdef ] ) , the @xmath97 will be non - zero in the interval @xmath103 $ ] only if the @xmath87 satisfy the following constraints @xmath105}f_{+i}\int_{l_{i}}^{h_{i}}d\theta\,w_{i}(\theta)\label{eqn : sumconta}\\ 0 & = & \sum_{\theta_{i}\in[l_{s},h_{s}]}f_{+i}\int_{l_{i}}^{h_{i}}d\theta\,w_{i}(\theta)\,\theta^{2}\label{eqn : sumcontb}\end{aligned}\ ] ] where the sums run only over bins in the interval @xmath103 $ ] . given data in some fiducial angular interval , one can always increase the angular range considered as long as the @xmath15 are zero outside the fiducial angular range . thus in order for the @xmath16 statistics to self - consistently consider only data in a finite angular range and minimize e / b - mode mixing by minimizing @xmath94 , they must satisfy the constraints given above . as discussed in @xcite , these constraints serve to project out ambiguous modes in the continuous correlation functions @xmath32 which can not be uniquely classified as either e- or b - modes . these modes are @xmath106 and so these constraints simply project out the binned versions of these modes . barring the caveat discussed above , in the discrete case no modes can be uniquely classified as pure e- or b - modes . thus in this sense all modes are ambiguous in the discrete case , due to the binning . however , e / b - mode separation for the discrete modes along @xmath106 is not possible even in the limit of infinitely small bins , so these discrete modes are the analogues of the ambiguous continuous modes . importantly , these ambiguous modes can still potentially carry cosmological information , it is just that one can not uniquely determine if the modes are sourced by the e- or b - mode power spectrum . assuming one has determined that the cosmic shear data is free of systematic contamination , it may be advantageous to keep and use the information in the ambiguous modes ( see @xcite for a similar point in the context of the pseudo-@xmath107 formalism and also @xcite ) . this point is illustrated below with a simple fisher information analysis . given the linear relation between the @xmath15 and the linear constraints on the @xmath87 , it is natural to treat them as vectors of length @xmath9 . thus the two integral constraints in equations ( [ eqn : sumconta ] ) and ( [ eqn : sumcontb ] ) mean that the @xmath108 vector can not have any component along the directions @xmath109 similarly , equation ( [ eqn : fmdef ] ) defines a matrix , which i will denote as @xmath110 , which relates the @xmath108 to the @xmath111 via @xmath112 . i give explicit forms for these quantities assuming geometric bin weightings in appendix [ app : estgeomwgt ] . the overall process to create a set of @xmath108 and @xmath111 vectors and measure the e- and b - mode signals in this scheme is quite straight forward . 1 . choose an initial shape for the @xmath108 vector . usually one has some external motivation for this choice , like for example , statistics which form a complete set of functions and exhibit efficient data compression @xcite , statistics that are compact and non - oscillatory in fourier space @xcite , or statistics which maximize the signal - to - noise in the measurement of cosmological parameters , like @xmath56 or @xmath58 @xcite . 2 . make @xmath108 is orthogonal to the @xmath113 and @xmath114 vectors defined above . in practice , an efficient way to do this is to first make @xmath113 and @xmath114 orthogonal to one another by modifying @xmath114 . then it is easy to make @xmath108 orthogonal to both @xmath113 and the modified @xmath114 by projecting out the components of @xmath108 along the modified constraint directions . 3 . use the matrix @xmath110 defined in equation ( [ eqn : fmdef ] ) to compute the corresponding @xmath111 filter via @xmath112 . 4 . compute the e- and b - mode signals by forming the vector inner product of the binned @xmath115 shear correlation function data with the @xmath74 filters . then @xmath116 and @xmath117 , where @xmath118 . below i use both the procedure just described and also gram - schmidt orthogonalization to define @xmath108 functions . in the case of gram - schmidt orthogonalization , one simply uses @xmath113 and @xmath114 first in the gram - schmidt process , followed by the rest of the initial estimator shapes for @xmath108 . the result of applying the gram - schmidt procedure is a set of orthogonal @xmath108 filters contained in the @xmath119 elements of the orthogonal basis . the first two elements of this basis span the space generated by the constraint directions @xmath113 and @xmath114 . the @xmath111 filters are again obtained from the matrix defined in equation ( [ eqn : fmdef ] ) . in this section i will explore two other potential optimal estimators definitions . the first definition simply consists of swapping the roles of @xmath108 and @xmath111 in the previous section . thus in this scheme the optimal estimator is now defined by @xmath120\ .\ ] ] the full estimators for this case are presented in appendix [ app : fminusest ] . in this case , all of the same results presented above hold and in particular there exists a matrix @xmath121 which defines the @xmath108 in terms of the @xmath111 via @xmath122 . there also exist ambiguous modes , @xmath123 and @xmath124 , in analogy to the ones defined above . finally , the procedure just given for e / b - mode separation can be carried out in exactly the same way as described above by first defining fiducial @xmath111 filters , projecting out the @xmath123 and @xmath124 modes , computing the @xmath108 filters from the matrix @xmath121 , and then computing @xmath16 . the second potential optimal estimator definition consists of minimizing the power in the @xmath94 window function by varying both @xmath108 and @xmath111 simultaneously . the two equations that must be solved for this kind of estimator are @xmath125\nonumber\\ 0 & = & \frac{\partial}{\partial f_{+k}}\left[\int_{0}^{\infty}d\ell\,\ell|w_{-}(\ell)|^{2}\right]\nonumber\ .\end{aligned}\ ] ] using the results for the two optimal estimators already presented , the two equations the @xmath74 vectors must satisfy are @xmath126 combing these two relations one gets that the optimal @xmath108 and @xmath111 vectors computed in this way must satisfy @xmath127 where @xmath128 is the identity matrix . thus in this case the @xmath74 vectors must be in the null space or kernels of the matrices @xmath129 . a straight forward way to investigate the kernels of these matrices is via computing their singular value decomposition ( svd ) ( e.g. , * ? ? ? the svd of a matrix @xmath130 is defined as @xmath131 . here @xmath132 and @xmath133 are orthogonal matrices , assuming @xmath130 is real , and @xmath134 is a diagonal matrix with the singular values along the diagonal . one can show that a basis for the kernel of a square matrix can be found by using all of the columns of @xmath133 which have zero singular values in @xmath134 . similar results hold for non - square matrices but they are of no use here . additionally , the rank of the matrix is number of non - zero singular values so that a square matrix with all non - zero singular values has full rank and is invertible . using logarithmic binning and the geometric bin weighting @xmath10 , i find that the matrix @xmath135 has two large singular values which exceed all of the others by several orders of magnitude . the @xmath119 other singular values are of similar magnitude to each other , but much smaller than the largest two . the difference between the two largest and the other @xmath119 singular values increases as the number of bins is increased in a fixed angular range . also , the columns of @xmath133 along the two largest singular values are approximately along the ambiguous directions @xmath113 and @xmath114 . analogous properties hold for the singular values of the matrix @xmath136 and the ambiguous directions @xmath123 and @xmath124 . unfortunately , all of the singular values of these matrices are non - zero so that no vector except the zero vector lies in their null spaces . note that one must be careful to account for rounding errors when considering whether or not singular values are zero . i find however that the @xmath119 smaller singular values are well above rounding errors . thus optimal estimators defined by varying both the @xmath108 and @xmath111 coefficients simultaneously do not exist in this case . however , this analysis does provide useful insight into the relationship between the estimators defined by varying @xmath111 and those defined by varying the @xmath108 coefficients . in particular , given that the singular values along the directions orthogonal to the ambiguous directions @xmath137 and @xmath138 are very small , we have that @xmath108 and @xmath111 vectors orthogonal to these directions span subspaces which are close to being kernels of the matrices @xmath129 . thus @xmath139 one can verify this property empirically as well . therefore the matrix @xmath110 generates @xmath111 vectors which are approximately in the approximate null space of @xmath136 and vice versa . ( to see this , simply multiply the two equations above by @xmath110 and @xmath121 respectively and use the relationship @xmath140 . ) thus @xmath111 vectors generated from @xmath108 vectors are very close to being orthogonal to the constraint directions @xmath138 . the analogous property holds for @xmath108 vectors generated from @xmath111 vectors and the constraint directions @xmath137 . additionally , due to these approximate equalities , the operation of `` composing '' the two different estimator definitions approximately returns the identity . in other words , one can first compute optimal estimators by varying the @xmath111 and supplying fiducial guesses for the @xmath108 . then the computed @xmath111 vectors can be used to supply fiducial guesses for the set of optimal estimators defined by varying the @xmath108 . the new set of estimators that result from varying the @xmath108 will be very close to the first set of estimators defined by varying the @xmath111 . this fact can be verified empirically . in this sense , the two optimal estimator definitions are consistent . it is now straight forward to build the approximate decomposition of the space of @xmath98 data points into e - mode , b - mode , and ambiguous modes discussed above . to do this properly , i must specify a basis for the vector space of @xmath98 data points which can be divided into directions approximately along ambiguous , e- and b - modes . the subspaces spanned by each of these three sets of modes need not be mutually orthogonal , though in practice the e- and b - mode subspaces are approximately orthogonal to the ambiguous mode subspace defined below . first , i specify the basis vectors for the ambiguous directions . these vectors are @xmath141 where the @xmath142 are zero vectors of length @xmath9 . next i specify the e- and b - mode directions . this is done by building a set of @xmath119 e- and b - mode estimators from a set of @xmath119 @xmath108 vectors . additionally these @xmath119 @xmath108 vectors together with @xmath113 and @xmath114 should form a basis for the @xmath46 subspace of the total vector space composed of both @xmath46 and @xmath47 . the construction of the optimal estimators enforces that the @xmath119 @xmath108 vectors are exactly orthogonal to the @xmath113 and @xmath114 vectors . additionally , as shown above , the resulting @xmath111 vectors will be approximately orthogonal to the @xmath123 and @xmath124 vectors . so using the @xmath119 @xmath108 vectors along with their @xmath111 counterparts , one can define the final @xmath143 potential basis vectors of this space as @xmath144 where this construction is repeated for each of the @xmath119 @xmath108 vectors . the first of these vectors is simply the direction of @xmath78 and the second is in the direction of @xmath80 so that they are approximate e- and b - mode directions . finally , one must verify that this set of vectors combined with the ambiguous mode directions is in fact a basis for the vectors space of dimension @xmath98 by , for example , determining the rank of the matrix composed of these vectors as rows . note that this construction would work just as well by first building a basis with the @xmath111 vectors and the @xmath138 vectors , and then computing the @xmath108 vectors . below , i will investigate this construction for the cosebi - like @xcite basis vectors defined in section [ sec : cosebi ] . finally , it is important to note that the exact properties of the matrices @xmath145 influence the properties of the basis constructed above nontrivially . specifically , if one uses the equal area binning ( i.e. @xmath146 is the same for all bins @xmath11 ) , then one can show using the expressions for these matrices given in appendices [ app : estgeomwgt ] and [ app : fminusest ] for geometric binning functions that @xmath147 . thus for vectors along the @xmath74 directions , the matrices @xmath145 in this case are approximately orthogonal . a quick calculation shows that if these matrices were exactly orthogonal and one started out with an orthonormal basis for the @xmath46 subspace , then then the @xmath111 vectors would also be orthonormal . in this case the e- and b - mode subspaces would then be exactly mutually orthogonal . in practice this exact orthogonality is not realized , but the e- and b - mode subspaces do retain some degree of orthogonality even for logarithmic binning for the cosebi - like statistics described below in section [ sec : cosebi ] . with equation ( [ eqn : xipmell ] ) it is possible to compute the covariance of these statistics . i follow the same method outlined in @xcite and get , assuming that the eb cross - power @xmath148 is zero and the power spectra are gaussian , @xmath149\end{aligned}\ ] ] @xmath150\end{aligned}\ ] ] @xmath151\ .\end{aligned}\ ] ] these expressions reduce to those in @xcite when @xmath152 . also note that the squared amplitude of the @xmath94 window function controls the amount of excess variance in the e- and b - mode statistics due to e / b - mode mixing . the procedure for defining @xmath111 introduced above serves to minimize this excess variance . similarly , by minimizing @xmath94 , the covariance between the e- and b - mode statistics is minimized . in this section , i will illustrate the ideas discussed above by constructing and evaluating the performance of two different e / b - mode statistics . i first present cosebi - like ( * ? ? ? * hereafter sek10 ) statistics in section [ sec : cosebi ] . these statistics have @xmath153 data compression ratios for upcoming surveys which can greatly simply data analysis . using these estimators i will also evaluate the level of e / b - mode mixing as function of the number of shear correlation function bins . then i present wavelet - like b - mode statistics which allow one to determine the scale and location of b - mode contamination in shear correlation function data in section [ sec : wvstats ] . finally in section [ sec : fishinfo ] , i explore the properties of the approximate ambiguous , e- and b - mode decomposition discussed above and the fisher information content of these subspaces using the coseb - like statistics . as demonstrated in sek10 , filter functions which are polynomials in @xmath154 have very optimal data compression ratios . additionally , sek10 built a set of filter functions which are orthonormal over a finite interval ( using a slightly different definition of orthonormality than the vector inner product used above ) . in order to generate a similar set of filters for correlation functions measured with @xmath9 bins , i take as an initial shape for the @xmath108 filters , @xmath155 $ ] , where @xmath156 is the legendre polynomial of order @xmath20 and @xmath157 is simply the logarithm of the angle @xmath158 remapped to the appropriate interval , @xmath159 . and @xmath74 filter functions , @xmath160 where @xmath54 is a representative point in bin @xmath11 . however , i have found that this procedure produces statistics which do not exhibit the data compression properties of the sek10 statistics . ] i increase @xmath20 for each vector in the basis starting at @xmath161 and i use the logarithmic mean of each bin interval to evaluate the polynomial . the first two vectors in the basis are the constraint directions @xmath113 and @xmath114 so that the third vector is started with @xmath161 . as the number of roots in the legendre polynomials increases , they become harder to represent properly in the discrete binned space . thus when the index of the legendre polynomial @xmath20 is greater than @xmath162 , where @xmath9 is the number of bins in the interval , i switch to generating the fiducial basis by producing @xmath108 vectors at order @xmath20 divided into @xmath163 sections alternating between @xmath164 and @xmath165 . this construction generates as many roots in the interval as the legendre polynomial of order @xmath20 would have , but these roots are now resolved properly . note that the maximum value of @xmath20 is @xmath166 , so that one gets exactly @xmath166 roots when @xmath9 intervals alternate between @xmath164 and @xmath167 . then the gram - schmidt procedure is applied to the constraint vectors and the @xmath119 fiducial @xmath108 vectors as described above in order to produce the final @xmath108 filters . the factor of @xmath168 $ ] has been inserted to roughly account for the difference between the vector inner product definition and that in sek10 . in figure [ fig : cosebisbasis ] , i show the cosebi - like vector basis for @xmath169 and @xmath170 for 50 shear correlation function data points between 1 and 400 arcmin . in figure [ fig : cosebiss2n ] , i show the level of e / b - mode mixing for these statistics as a function of the number of shear correlation function bins . i have plotted the ratio of the first 12 e- and b - mode statistics to their errors for both des- and lsst - like surveys . note that in general these statistics are correlated , but to make this plot i only use the diagonal contributions to the error covariance matrix . the statistics in this plot were computed with @xmath39 , so any nonzero b - mode statistic amplitude is completely due to e / b - mode mixing . additionally , the e / b - mode mixing in the estimators is negligible if the nonzero amplitude of the b - mode statistic is well below the error bar on the statistic . this condition ensures that any statistically significant systematic effect is detectable and is not confused with e - modes due to e / b - mode mixing . in general , the level of e / b - mode mixing decreases as the number of bins used for the shear correlation function increases . additionally , it is clear that for a des - like survey fewer shear correlation function bins can be used than for an lsst - like survey . if one uses a threshold of the b - mode statistic being no more than 10% of the 1-sigma error bar , then a des - like survey will need of order @xmath171 bins whereas an lsst - like survey will need @xmath172 bins . this figure also demonstrates the data compression properties of these statistics . in the case of a des - like survey of order 50 shear correlation function data points can be compressed into only @xmath173 e - mode statistics which are measurably non - zero . an lsst - like survey has increased statistical power , compressing of order 100 data points into only @xmath174 statistically significant e - mode statistics . due to correlations between these remaining e - mode statistics , further data compression may be possible , but i will not explore this issue further in this work . in this section , i give an example of the use of spatially - local basis functions to build sets of b - mode estimators which can pinpoint the size and location of b - mode systematics in shear correlation function data . consider the following function , know as the ricker wavelet @xcite , @xmath175\ .\ ] ] i take as a set of starting functions the following functions , indexed by @xmath176 , @xmath177\end{aligned}\ ] ] where @xmath178 and @xmath179 . this definition shifts and scales the location and size of the base wavelet so that an integer number @xmath9 of them fit in the interval @xmath180 $ ] and they have most of their support in each width @xmath181 subinterval of the base interval @xmath182 . given this starting set of smooth functions , they are turned into e- and b - mode estimators by discretizing the functions over the interval @xmath180 $ ] and then projecting out the @xmath113 and @xmath114 modes as described above . the filters which result from this process will be denoted as @xmath183 and @xmath184 . figure [ fig : hatwavebasis ] shows an example set of these filters for 100 correlation function data points between 1 and 400 arcmin . in order to gain intuition into how these b - mode estimators work , i show a simple , but somewhat contrived example ( however see e.g. , * ? ? ? * ; * ? ? ? * ) of two different kinds of b - mode systematics in figure [ fig : wvtest ] . the top row displays a smooth b - mode systematic in the @xmath47 correlation function , while the bottom row shows a b - mode systematic with much more fine scale variations . these b - mode systematics were generated by setting @xmath185 to be non - zero using simple gaussian kernels , the form of which is uninteresting , and then computing the shear correlation functions using equations ( [ eqn : xipell ] ) and ( [ eqn : ximell ] ) . the contour plots show for each value of @xmath9 on the vertical axis , the amplitude of the b - mode statistics for each @xmath181 sized interval across the horizontal axis . in this type of analysis , a b - mode signal with more fine scale variation has a different signature than the smoother b - mode signal , with more b - mode signal detected in higher @xmath9 and thus smaller sized filters . for data with shape noise , a similar contour plot can be made , except that the deviation from zero measured in units of the standard deviation should be used to produce the color scaling . the overall level of b - mode contamination can then be assessed via a @xmath186 statistic computed over all of the different estimators indexed by @xmath187 , accounting for the correlations between the different estimators . here i have chosen to layout the @xmath188 filters in a simple geometric pattern . however , when a b - mode signal crosses between two filters , the significance drops . thus it might be more advantageous to consider sets of filters which shift and scale in size more continuously . i leave this generalization to future work . note however that the filters i have presented here are offset between different levels @xmath9 so that in practice , no b - mode signal ever goes completely undetected . this general point can be clearly illustrated by examining the fourier - space filters @xmath189 introduced above for the wavelet - like @xmath188 statistics . these filters for @xmath190 are shown in figure [ fig : waveellwin ] . the peaks and troughs for each @xmath9 correspond to a single location of the filter in real space , with @xmath9 peaks and troughs present for the @xmath9-indexed filters . as the filters shift in real space over the shear correlation function range , they cover different , but overlapping ranges in fourier space . thus by combining filters of different size and location , any b - mode signals in fourier space will produce a nonzero amplitude in these filters . for the @xmath190 wavelet - like statistics computed for @xmath191 $ ] arcmin with 100 shear correlation function data points . the different colors show each of the filters . the @xmath94 filters are approximately three orders of magnitude smaller than the @xmath189 filters . the different filters in size ( the overall width of the peaks and troughs ) and location ( the mean location in @xmath20 for each set of peaks and troughs ) approximately cover all of @xmath20-space accessible given the angular range of the data.[fig : waveellwin ] ] in this section i present the full ambiguous , e- and b - mode decompositions discussed in section [ sec : ebmodedec ] using the cosebi - like statistics defined above . i also compute the fisher information content of the approximate ambiguous , e- and b - mode subspaces for the parameters @xmath56 and @xmath58 . i use a basis composed of 50 correlation function bins logarithmically spaced from 1 to 400 arcmin . i find that the full basis composed of these different spaces completely spans the space of @xmath98 data points , demonstrating that at least one such approximate decomposition exists . due to the fact that this decomposition is only approximate and that the matrices @xmath145 are not exactly orthogonal for the vectors @xmath74 , the different subspaces will in general not be perfectly orthogonal to one another . the degree of orthogonality of the ambiguous , e- and b - mode subspaces can be characterized by the maximum absolute normalized projection of a basis vector of any one of the spaces onto the basis vectors of the other spaces , @xmath192 where @xmath193 and @xmath194 are vectors of length @xmath98 with the normal inner product definition . this quantity is just the cosine of the minimum angle between the vectors of any of the subspaces with the other . for subspaces which nearly share a basis vector , this maximum absolute normalized projection will be @xmath195 , but it can not be greater than one or else two of the vectors in the basis would be linearly dependent . for spaces which are mutually orthogonal , this maximum absolute normalized projection will be zero . for the set of basis vectors just computed , i find that the maximum projection between the e- and b - mode subspaces is @xmath196 . similarly , the maximum projection between the e - mode and ambiguous subspaces is @xmath197 and for the b - mode and ambiguous subspaces is @xmath198 . consistent with the argument made above , with equal area binning the e- and b - mode subspaces are more orthogonal with the maximum absolute normalized projection between them being @xmath199 . however in this case , the ambiguous modes are less orthogonal to the e- and b - mode subspaces with the maximum absolute normalized projections being @xmath200 and @xmath201 respectively . the particularly poor orthogonality of the b - mode subspace to the ambiguous mode space in this case is reflected in the svd of the matrix @xmath136 computed with equal area binning . as mentioned above , the cost of e- and b - mode separation is that one must potentially throw away information in the ambiguous modes ( e.g. , * ? ? ? * ; * ? ? ? * ) . using the approximate mode decomposition described in this section , this point can be illustrated clearly and approached quantitatively . i compute the fisher information content of the ambiguous , e- and b - mode subspaces and combinations of them assuming gaussian statistics and using the expressions presented above for the covariances of the two - point correlation functions and fisher information matrices . for the e- and b - mode statistics , i simply transform the covariance matrix of the two - point correlation functions , since the statistics are linear combinations of the two - point data . one can obtain similar results computing their covariance matrices directly from their fourier space window functions and the expressions presented in section [ sec : ebcov ] . the covariance matrix of the ambiguous modes is computed by direct transformation as well . the fisher information content @xmath202 of these subspaces for the parameters @xmath56 and @xmath58 using the binning scheme and angular range for the basis constructed above and also assuming des - like errors is @xmath203 , @xmath204 , and @xmath205 for the ambiguous , e- and b - mode subspaces respectively . by comparison , the full fisher information for the entire two - point function data set , @xmath206 for @xmath207 $ ] , is @xmath208 . note that these figures are not additive and so can only be qualitatively compared . however the basic trends are clear . the b - mode subspace has very little information on the parameters as expected . any residual information is either a numerical artifact or is due to e / b - mode mixing . the e - mode subspace has most of the information , but the ambiguous modes carry a non - trivial fraction of the total information . by retaining the ambiguous modes in the analysis with the e - modes , most of the information can be restored in comparison with the full two - point function with the e- plus ambiguous mode fisher information being @xmath209 . i have tested several other fiducial choices besides the cosebi - like statistics for constructing the full mode decomposition of the correlation function data and found similar conclusions regarding the relative fisher information content of the various subspaces . similar conclusions hold for lsst - like surveys as well . while these fisher information calculations are not particularly realistic as survey projections , they illustrate nicely the cost of e / b - mode separation in cosmic shear . finally , these fisher information calculations can also illustrate the data compression properties of the cosebi - like statistics . in particular , for the des - like survey , the fisher information in the first eight e - mode statistics is @xmath210 as opposed to @xmath204 for all of the e - mode statistics . similarly , for the lsst - like survey using 100 correlation function bins in the same angular range as the des - like survey , the first ten e - mode statistics have an information content of @xmath211 as opposed to @xmath212 for all of the e - mode statistics . these results are in qualitative agreement with those of @xcite . in this work , i have demonstrated that the use of two - point e / b - mode estimators with binned shear correlation function data will generally result in non - trivial e / b - mode mixing . i computed the amount of this mixing and provided new e / b - mode estimators which minimize the unwanted mode mixing . i also gave practical recipes for building and using these estimators with binned cosmic shear data . using these optimal estimators , i demonstrated that approximate decompositions of the vector space of binned correlation function points into ambiguous , e- and b - modes do exist . using one of these decompositions , i found that the ambiguous mode subspace contains a non - trivial amount of information on typical cosmological parameters . i also gave two example applications of these new estimators to generic problems in cosmic shear data analysis , data compression ( section [ sec : cosebi ] ) and b - mode localization with wavelets ( section [ sec : wvstats ] ) . the estimators presented here have several nice features adapted to practical cosmic shear data analysis . * they are linear combinations of the binned shear correlation function data , defined in equation ( [ eqn : xpmlinstat ] ) , and thus treat the binning explicitly . this property makes their computation and also error propagation / analysis with them trivial once the shear correlation function and its errors are known . * the level of e / b - mode mixing due to the binning can be computed exactly , up to the knowledge of the binning window functions , for these estimators using equations ( [ eqn : xipmell ] ) and ( [ eqn : wpmdef ] ) . in the limit of small bins , which is needed to suppress the e / b - mode mixing , the binning window functions are expected to be close to the geometric approximation used throughout this work . * they give quantitative criteria in terms of the e / b - mode mixing by which to decide the number of shear correlation function bins to use , demonstrated in figure [ fig : cosebiss2n ] for the cosebi - like estimators . * the design of new e / b - mode estimators with specific purposes using the formalism presented in this work reduces to simple linear algebraic manipulations , presented in section [ sec : buildebest ] . the optimal statistics presented in this work are well - suited to blinded or closed - box , high - precision cosmic shear data analyses . for example , before any shear correlation functions are computed from the data , one can estimate given the expected statistical accuracy of the data , the exact amount of e / b - mode mixing for a given binning scheme and set of estimators . one can then choose a fiducial binning scheme and estimator choice that properly minimizes the e / b - mode mixing and retains all of the cosmological information . then these choices can be fixed throughout the data analysis process in order to avoid any observer biases in detecting or assessing b - mode contamination arising from how exactly the e / b - mode separation was done . additionally , in a blinded or close - box analysis one might not look at plots of the shear correlation function data until the entire analysis is complete . unfortunately in this case , one might miss crucial information about potential systematics in the _ shape _ of the shear correlation function data . the wavelet - like b - mode estimators presented in this work can be used as a substitute for and quantitative measure of the information gained by looking at the shape of the shear correlation function data . additionally , they can be applied in automated way to large data sets in order to pinpoint areas of potential systematic contamination for further investigation . importantly , because these estimators have a large degree of spatial locality , they can potentially provide crucial information on where any b - mode contamination is coming from and not just indicate its existence . the ability to easily and quickly design e / b - mode estimators tailored to a specific purpose can potentially be very useful in practice as well . for instance , the problem of computing an e / b - mode statistic which maximizes the signal - to - noise or cosmological information content , as considered by @xcite , could now be reformulated with the linear estimators presented here . additionally , one could build estimators which are along the normal modes of the correlation function data computed from the correlation function covariance matrix . also , one could attempt to build direct estimators for the e / b - mode correlation functions @xcite over a finite interval . alternatively , one could attempt to build filters which are localized in fourier space in order to exclude certain wave modes , for example to mitigate the effects of uncertainty in the matter power spectrum at small scales ( see e.g. , * ? ? ? * ) , or to be sensitive to b - modes from only a range of @xmath20 . as simple linear combinations of cosmic shear data points , the estimators presented in this work are well - suited to these applications and to practical cosmic shear data analysis in general i thank tim eifler , eduardo rozo , scott dodelson , bhuvnesh jain , aaron roodman , and the des weak lensing working group for the many discussions which inspired aspects of this work . this work was supported by the kavli institute for cosmological physics at the university of chicago through grants nsf phy-0114422 and nsf phy-0551142 and an endowment from the kavli foundation and its founder fred kavli . this work made extensive use of the nasa astrophysics data system and arxiv.org preprint server . i use the formalism presented in appendix a of @xcite to derive the leading order bin weighting terms in the pair - wise estimator for the shear correlation functions . this estimator is ( e.g. , * ? ? ? * ) @xmath213 where @xmath214 is the binning function which is unity inside the bin and zero outside with the bin centered at @xmath215 with some bin width @xmath216 , @xmath217 is the component of the galaxy shape parallel or crossed with respect to the great circle connecting the two galaxies , and the @xmath218 are weights applied to each galaxy . below i include the weights @xmath218 in the survey window function . starting with equation a15 of @xcite and assuming that the source galaxy density field is uncorrelated with the shear field and also neglecting lensing bias / reduced shear effects , the expectation value of this estimator can be written as @xmath219\end{aligned}\ ] ] where @xmath220 is the shear correlation function for @xmath221 , @xmath222 is the survey averaged bin area , and @xmath223 is the source galaxy angular correlation function . @xmath224 is the survey window function including the weights @xmath218 . the second term in the brackets arises directly from the weighting over the bin by the sampling density of the source galaxies and the third term in the brackets is the first non - trivial term in the power - series expansion of the denominator of the estimator ( i.e. , the total number of observed galaxies in the bin ) . the last two terms in this integral do not exactly cancel as stated in @xcite because @xmath223 is not exactly equal to its average over the bin for all @xmath158 in the bin . in addition to this source clustering weighting term , the survey window function contributes an additional weight over the bin . the form of this weighting function is highly non - trivial , but for small enough bins , this weighting should be negligible . however , quantitative results describing how small the bins need to be in order to suppress this weighting require detailed survey simulations , which are beyond the scope of this work . thus for simplicity i neglect the survey window function weights . a simple estimate of the magnitude of the effect of source galaxy clustering using typical galaxy angular correlation functions ( e.g. , * ? ? ? * ) shows that this effect ( i.e. , the difference of the last two terms in the bracket above ) is @xmath225 when using at least five bins per decade . to get this number i assumed the galaxy - galaxy angular correlation function was a power law over the angular range of @xmath226 $ ] arcmin with @xmath227 , consistent with the results of @xcite in this appendix , i give the exact form of the constraint directions , @xmath113 and @xmath114 , and the matrix to compute the @xmath97 from the @xmath87 under the assumption of geometric bin weightings . consider @xmath9 bins in angle @xmath158 from @xmath228 to @xmath229 and let @xmath93 $ ] be the angular range of the @xmath11th bin . also , assume the bins are non - overlapping . then using the geometric bin weighting function , @xmath73 , i get for the constraint direction vectors ( see section [ sec : buildebest ] ) @xmath230 the matrix to compute the @xmath86 in terms of the @xmath87 from equation ( [ eqn : fmdef ] ) is @xmath231 where @xmath11 and @xmath18 run over 1 to @xmath9 . then the @xmath86 are computed as @xmath232 one can easily define optimal e / b - mode estimators with @xmath111 fixed instead of @xmath108 fixed . in this case one fixes the @xmath108 by minimizing @xmath233\ .\ ] ] the solution to this equation is ( c.f . equation [ eqn : fmdef ] ) @xmath234 in this case the constraint directions can be derived by similar arguments to the those in section [ sec : buildebest ] and are @xmath235 and equation ( [ eqn : fm2fpmat ] ) again defines a matrix which is used to compute the @xmath108 in terms of the @xmath111 . finally , for the geometric bin weighting functions the constraint directions and matrix relating the @xmath108 to @xmath111 are @xmath236}{h_{1}^{2}-l_{1}^{2}},\frac{2\log[h_{2}/l_{2}]}{h_{2}^{2}-l_{2}^{2}},\right.}\nonumber\\ & & \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \left . ... , \frac{2\log[h_{n}/l_{n}]}{h_{n}^{2}-l_{n}^{2}}\right)_{n}\\ \lefteqn{f_{-b } = \left(\frac{1}{h_{1}^{2}-l_{1}^{2}}\left(\frac{1}{l_{1}^2 } - \frac{1}{h_{1}^2}\right),\frac{1}{h_{2}^{2}-l_{2}^{2}}\left(\frac{1}{l_{2}^2 } - \frac{1}{h_{2}^2}\right),\right.}\nonumber\\ & & \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \left . ... , \frac{1}{h_{n}^{2}-l_{n}^{2}}\left(\frac{1}{l_{n}^2 } - \frac{1}{h_{n}^2}\right)\right)_{n}\end{aligned}\ ] ] and @xmath237 & \mbox{if $ i = k$}\\ 2(h_{k}^{2}-l_{k}^{2})\ln\left(h_{i}/l_{i}\right ) & \\ \ \ \ \ \ \ \ \ + \frac{3}{2}\left(h_{k}^{4 } - l_{k}^{4}\right)\left(\frac{1}{h_{i}^{2 } } - \frac{1}{l_{i}^{2}}\right ) & \mbox{if $ i > k$}\\ \end{array } \right.\end{aligned}\ ] ] where @xmath11 and @xmath18 run over 1 to @xmath9 . then the @xmath238 are computed as @xmath239

bilinear restriction estimates have been appeared in work of bourgain , klainerman , and machedon . in this paper we develop the theory of these estimates ( together with the analogues for kakeya estimates ) . as a consequence we improve the spherical restriction theorem of wolff from to , and also obtain a sharp spherical restriction theorem for . # 1_#1 _ # 1*#1 * [ subsection]theorem [ subsection]proposition [ subsection]lemma [ subsection]corollary [ subsection]conjecture [ subsection]remark [ subsection]remarks [ subsection]definition

math9807163

the purpose of this paper is to investigate bilinear variants of the restriction and kakeya conjectures , to relate them to the standard formulations of these conjectures , and to give applications of this bilinear approach to existing conjectures . the methods used are based on several observations and results of bourgain ( see @xcite-@xcite ) , together with some refinements by moyua , vargas , and vega @xcite . this paper is organized as follows . in the first section we discuss bilinear restriction estimates , and show how one can pass back and forth between these estimates and the standard restriction estimates . we also generalize the @xmath5 bilinear restriction estimate of @xcite to higher dimensions . in the second section we give analogues of the above results for the kakeya operator . in particular we give a bilinear improvement to wolff s kakeya theorem in arbitrary dimension . in the third section we give applications of these bilinear estimates in three dimensions . for example , we are able to improve the @xmath6 exponent in wolff s restriction theorem to @xmath7 . we are also able to prove a sharp @xmath3 restriction theorem which improves on the classical @xmath8 tomas - stein theorem , and also give some concrete progress on a bilinear restriction conjecture of klainerman and machedon . we also give a non - bilinear approach to these estimates , which gives weaker results but is more direct and probably has a wider range of application . finally , we collect some standard harmonic analysis estimates in an appendix for easy reference . this work was conducted at msri ( nsf grant 9701955 ) . the authors wish to thank tony carbery , adela moyua , and wilhelm schlag for many helpful discussions . the second author was partially supported by the spanish dgicyt ( grant number pb94 - 149 ) and the european comission via the tmr network ( harmonic analysis ) . fix and @xmath9 . ] @xmath10 and @xmath11 , and let @xmath12 be the cube @xmath13^{n-1}$ ] in @xmath14 . let @xmath15 be a phase function satisfying the following conditions : * @xmath16 for all @xmath17 , where @xmath18 is a large constant . * @xmath19 . * for all @xmath20 , the eigenvalues of the hessian @xmath21 all lie in @xmath22 $ ] , where @xmath23 is a small constant . we will call such a phase _ elliptic_. the model example of an elliptic phase function is of course the quadratic phase @xmath24 , but any smooth compact convex surface can be decomposed into finitely many graphs whose graphing function ( after an affine transformation ) obeys the above properties . in particular , the unit sphere can be decomposed in this manner . we will consider linear and bilinear bounds for the operator @xmath25 defined by @xmath26 this operator can be thought of as an adjoint restriction operator associated to the surface @xmath27 . for @xmath28 , we use @xmath29 to denote the estimate @xmath30 for all test functions @xmath31 , with the constant depending only on @xmath32 and @xmath9 . similarly , we use @xmath33 to denote the estimate @xmath34 for all test functions @xmath31 , @xmath35 supported on @xmath36 , @xmath37 respectively , where @xmath36 , @xmath37 are any sub - cubes of @xmath12 whose size and separation are comparable to 1 . ( we will call such cubes @xmath38-separated in the sequel ) . estimates of the form @xmath29 are adjoint restriction estimates and have attracted wide interest . the ( sharp ) restriction conjecture states that [ rest ] @xmath29 holds whenever @xmath39 and @xmath40 . these conditions are well known to be best possible ( see e.g. @xcite ) . this conjecture has been verified for @xmath41 @xcite , but remains open in higher dimensions . the main difficulty lies in making the @xmath42 exponent as low as possible ; the estimate is trivial for @xmath43 , hlder s inequality can be used to raise @xmath44 , and in certain cases factorization theory can be used to lower @xmath44 . when @xmath45 lie on the sharp line @xmath46 we abbreviate the estimate @xmath29 to @xmath47 . we summarize the known results in @xmath48 in table 1 . the classical theorem of tomas and stein states that @xmath49 for any @xmath10 . later improvements have been made on this result @xcite,@xcite,@xcite ; in particular , moyua , vargas , and vega @xcite have recently observed that one has the estimate @xmath50 in three dimensions . however , none of these improvements to the tomas - stein theorem lies on the sharp line @xmath51 . as one of the applications of this paper we will prove a new restriction theorem on this sharp line . .known restriction theorems for @xmath48 . @xmath52 denotes an arbitrary positive number . [ cols= " < , < , < " , ] our improvements will be based on the bilinear restriction estimates @xmath53 defined earlier , which we will now discuss . these estimates have appeared implicitly in many works ( e.g. @xcite , @xcite ) , and are closely related to null form estimates for the wave equation ( see @xcite-@xcite ; related ideas also appear in @xcite ) , but do not appear to have been explicitly studied until very recently . when @xmath54 lie in the range predicted by conjecture [ rest ] then @xmath55 and @xmath53 are almost equivalent . indeed , in section [ iff ] we will prove [ imply ] let @xmath10 and @xmath56 be such that @xmath57 and @xmath58 . then @xmath59 implies @xmath53 . furthermore , if @xmath60 holds for all @xmath61 in a neighbourhood of @xmath62 , then @xmath59 holds . however , the bilinear estimate @xmath53 can hold for exponents which are not covered by the above theorem . for instance , when @xmath41 an easy computation using plancherel s theorem and a change of variables shows that @xmath63 holds , even though the knapp example shows that @xmath64 fails completely . thus one expects the range of exponents for the bilinear restriction estimate to be larger than that of conjecture [ rest ] . for @xmath48 the first results in this direction were by bourgain @xcite ( although the theorem @xmath65 implicitly appeared in @xcite ) ; more recently , moyua , vargas , and vega @xcite showed that @xmath66 for @xmath48 . we modestly generalize this result to higher dimensions as [ modest ] suppose that @xmath10 . then @xmath67 holds if and only if @xmath68 . recently klainerman and machedon conjectured that @xmath69 for all @xmath10 . by interpolating with what is implied by conjecture [ rest ] , one is led to the following [ conj ] if @xmath10 , then @xmath53 holds whenever @xmath70 by theorem [ modest ] and interpolation the conjecture is verified for @xmath71 ( and thus for @xmath72 ) . the exponents in the above conjecture are best possible ; we will sketch the proof of this statement in section [ nec ] . from theorem [ imply ] we see that conjecture [ conj ] implies conjecture [ rest ] . we depict the conjectured ranges for the estimates @xmath73 and @xmath53 in figure 1 . the restriction conjecture states that @xmath73 holds for all @xmath45 in the trapezoidal region bounded by @xmath74 , @xmath75 , @xmath76 , and @xmath77 , except for the upper line between @xmath75 and @xmath76 inclusive ; by the above theorem , this is almost equivalent to @xmath78 holding in this region . klainerman s conjecture asserts that @xmath53 holds at the endpoint @xmath79 . the combined conjecture [ conj ] states that @xmath53 holds in the pentagonal region bounded by @xmath74 , @xmath79 , @xmath75 , @xmath76 , and @xmath77 , including the upper line mentioned previously ; this region is best possible . by theorem [ imply ] the standard restriction estimate @xmath80 and the bilinear estimate @xmath53 are essentially equivalent in the line between @xmath75 and @xmath74 . the points @xmath81 correspond to the standard restriction results , while the point @xmath82 corresponds to the bilinear restriction theorem . from theorem [ imply ] and bilinear interpolation it is possible to obtain new linear and bilinear restriction theorems ; for instance , by interpolating between the bilinear form . ] of the result in @xcite and and using theorem [ imply ] , one may obtain the sharp restriction theorem @xmath47 for all @xmath83 . we will improve on these results in in section [ apply ] . the first implication is a trivial consequence of hlder s inequality , so we concentrate on the latter . in view of the known results for @xmath41 we may take @xmath84 . from the tomas - stein theorem ( see e.g. @xcite ) and the necessity of it suffices to consider the case @xmath85 . in particular we may assume that @xmath86 . the bilinear hypothesis @xmath60 allows us to control @xmath87 if @xmath31 and @xmath35 have @xmath38-separated supports . by a parabolic rescaling argument this will imply a similar estimate when @xmath31 and @xmath35 have @xmath88-separated supports for any @xmath89 . piecing these estimates together one may obtain an estimate on @xmath87 for arbitrary @xmath31 , @xmath35 , from which the conclusion @xmath29 will follow . we now turn to the details . assume that the hypotheses of theorem [ imply ] hold . we have to show that @xmath90 by marcinkeiwicz interpolation it suffices to show the restricted estimate @xmath91 for a slightly better value of @xmath54 , where @xmath92 is some arbitrary subset of @xmath12 . let @xmath93 be the positive integer such that @xmath94 . then by squaring the above estimate , we reduce ourselves to @xmath95 the next step is a whitney decomposition . for each @xmath89 , we dyadically decompose @xmath12 into @xmath96 dyadic subcubes @xmath97 of sidelength @xmath98 in the usual manner . if @xmath97 , @xmath99 are two cubes with the same sidelength which are not adjacent but have adjacent parents , we say that these cubes are _ close _ and write @xmath100 . for almost every @xmath101 there exists a unique pair of close cubes @xmath97 , @xmath99 containing @xmath102 and @xmath103 respectively . thus we have @xmath104 thus to prove it suffices to show that @xmath105 for all @xmath89 and some @xmath106 , since follows from the triangle inequality . informally , the above estimate asserts that the most significant separation scale is of the order of @xmath107 ; this is already evident from the knapp example . our next reduction will be to exploit some quasi - orthogonality between the functions @xmath108 . from the definition of @xmath109 we see that the fourier transform @xmath110 is supported on the infinite tube @xmath111 . thus , the fourier transform of @xmath108 is supported in the tube @xmath112 where @xmath113 is a cube of sidelength @xmath114 whose center is twice that of @xmath97 . from lemma [ quasi - lemma ] in the appendix and the assumption @xmath115 , we thus have @xmath116 thus will be proven if we can show that @xmath117 this will follow from the following estimate . [ rescale ] for all @xmath118 in a neighbourhood of @xmath44 , we have @xmath119 and subtracting a harmless affine factor by a constant factor to do this ; we will gloss over this technicality . ] we may assume that @xmath97 is centered at the origin . we now observe that since @xmath120 is an elliptic phase , the function @xmath121 is also elliptic . since @xmath122 holds for all @xmath118 in a neighbourhood of @xmath44 by assumption , we have @xmath123 whenever @xmath31 and @xmath35 are supported on disjoint @xmath38-separated cubes , where @xmath124 is the adjoint restriction operator corresponding to @xmath125 . applying a parabolic scaling @xmath126 to this estimate one obtains @xmath127 whenever @xmath31 and @xmath35 are supported on @xmath97 and @xmath128 respectively , and follows . it remains to obtain from the proposition . let @xmath129 be such that holds . if we apply and the triangle inequality , we see that reduces to @xmath130 by polarization and the fact that for each @xmath131 there are only finitely many cubes @xmath99 close to @xmath97 , this is in turn reduces to @xmath132 and @xmath133 . if @xmath134 then we use from lemma [ xr - est ] in the appendix with @xmath135 to obtain @xmath136 thus reduces to @xmath137 by convexity it suffices to verify this inequality for the values @xmath138 , @xmath139 , and @xmath140 . when @xmath138 becomes @xmath141 which is true for some @xmath106 since @xmath129 . when @xmath139 becomes @xmath142 which holds since @xmath58 . finally , when @xmath143 becomes @xmath144 which holds for some @xmath106 since @xmath57 . it remains to treat the case @xmath133 . by repeating the above procedure but with replaced by , we see that reduces to @xmath145 since the left - hand side is completely linear it suffices to verify this when @xmath138 and when @xmath140 . but in these two cases reduces , as before , and so the argument proceeds as in the previous case . the fact that this theorem requires knowledge of @xmath53 for all elliptic phase functions is a defect of the argument . when restricted to the quadratic phase @xmath24 however , no other phase functions are required in the proof , due to the algebraic properties of @xmath120 . the quadratic phase is the simplest of all the elliptic phases ; indeed , a parabolic scaling and limiting argument shows that any sharp restriction theorem for an elliptic phase implies the corresponding estimate for the quadratic phase . ( see @xcite ) . in this section we sketch the proof of the assertion that the conditions in conjecture [ conj ] are necessary . for simplicity we take @xmath120 to be a graphing function for a small portion of a sphere ; one can easily modify the arguments below for more general phases . the estimate @xmath53 can then be rewritten as @xmath146 where @xmath147 is surface measure on the unit sphere @xmath148 , and @xmath31 and @xmath35 are functions on fixed disjoint caps @xmath149 , @xmath150 in @xmath148 whose size and separation are comparable to a small quantity @xmath151 . to prove , we take @xmath152 on @xmath149 , and @xmath153 on @xmath150 , where @xmath154 is a point to be determined later . from standard stationary phase estimates , we see that for any @xmath155 one can find a cube @xmath12 of sidelength @xmath156 such that @xmath157 on @xmath12 . by choosing @xmath158 appropriately , one can also arrange matters so that @xmath159 on the same cube @xmath12 . by inserting these estimates into one obtains @xmath160 if one now uses the fact that @xmath161 and takes @xmath162 the condition follows . the proof of the necessity of and is based on modifications of the standard knapp example . we note in passing that without modification the knapp example only gives the weaker condition @xmath163 to prove , we will take @xmath31 and @xmath35 to be `` squashed caps '' . we factor @xmath164 as @xmath165 , and use @xmath166 to denote the great circle @xmath167 . we may assume that @xmath166 intersects @xmath149 and @xmath150 . fix @xmath168 . we take @xmath31 and @xmath35 to be the characteristic functions of the sets @xmath169 respectively , where @xmath170 denotes the ball in @xmath171 of radius @xmath156 centered at @xmath102 , and @xmath172 are arbitrary elements of @xmath173 for @xmath174 . then the fourier transforms of @xmath175 , @xmath176 exhibit essentially no cancellation on the box @xmath177 indeed , we have @xmath178 on this set . inserting this estimate into one obtains @xmath179 and by taking @xmath180 one obtains . the estimate is proven by taking @xmath31 and @xmath35 to be `` stretched caps '' . with the notation as before we take @xmath31 and @xmath35 to be the characteristic functions of @xmath181 respectively to begin with , although we will later need to multiply @xmath31 and @xmath35 by a phase as in the proof of . when restricted to the slab @xmath182 , the functions @xmath183 behave essentially like fourier transforms of measures on @xmath166 . indeed , a stationary phase computation shows that @xmath184 on a large portion of this slab , and similarly for @xmath185 . thus , multiplying by a phase to translate @xmath186 and @xmath185 as necessary , one can arrange matters so that @xmath187 on the box . inserting this into one obtains @xmath188 and follows by taking @xmath180 . unlike the situation with the disc multiplier problem @xcite , it appears that the besicovitch set construction does not give any further restrictions on @xmath44 , @xmath42 . indeed , for @xmath41 the conditions - are sufficient as well as necessary . our argument will be a routine modification of the one in @xcite . the necessity of the condition on @xmath44 follows from section [ nec ] , so we will only show the sufficiency of this condition . by hlder s inequality it suffices to show that @xmath189 by symmetry and interpolation this will follow from @xmath190 it suffices to show that @xmath191 for all @xmath192 , @xmath193 , @xmath194 , @xmath195 , supported on @xmath36 , @xmath36 , @xmath37 , @xmath37 respectively , where @xmath36 and @xmath37 are @xmath38-separated cubes . by applying the symmetry @xmath196 , @xmath197 to and applying multi - linear interpolation one obtains @xmath198 and the desired estimate follows from substituting @xmath199 , @xmath200 . it remains to prove . by plancherel s theorem the left - hand side can be written as @xmath201 where @xmath202 is the dirac distribution . from the positivity of the kernel in the above expression we may reduce to @xmath203 for arbitrary functions @xmath192,@xmath193,@xmath194,@xmath195 on @xmath14 , where @xmath204 and @xmath205 are smooth cutoff functions adapted to ( a slight thickening of ) @xmath36 and @xmath37 respectively . since @xmath192 and @xmath193 are controlled in @xmath206 and @xmath207 respectively , we may assume that @xmath208 and @xmath209 for some @xmath158 ; we may take @xmath210 by translating @xmath120 and subtracting off a harmless affine factor . in particular , we may assume that @xmath77 is in ( a slight thickening of ) @xmath36 . the estimate thus reduces to @xmath211 which by duality becomes @xmath212 where @xmath213 is the averaging operator @xmath214 it is well known ( see below ) that the estimate will hold if the defining function @xmath215 satisfies the rotational curvature condition @xmath216 uniformly on the support of @xmath217 . however , since @xmath120 is elliptic , we have the estimates @xmath218 where @xmath219 is the kronecker delta . inserting these estimates into the definition of @xmath220 , one can estimate the above determinant as @xmath221 however , from the support of @xmath222 and the assumption that @xmath77 is in a thickening of @xmath36 we see that @xmath223 . thus follows , if @xmath224 is sufficiently small . this finishes the proof . the above proof shows that there exist asymmetrical bilinear restriction theorems in addition to the symmetrical ones . in particular , one may conjecture that @xmath225 which is a strengthening of . non - symmetrical versions of the counterexamples in the previous section show that this conjecture is best possible . for @xmath226 theorem [ modest ] is an improvement on the classical tomas - stein theorem . however for @xmath227 the two estimates are not directly comparable . because of this , we have no significant improvements to wolff s restriction theorem @xcite in four and higher dimensions . for completeness we sketch a proof of the following standard fact which was used in the above proof . if @xmath220 satisfies the rotational curvature condition on the support of a cutoff function @xmath228 , then the operator @xmath229 obeys . we imbed this operator in the analytic family @xmath230 defined by @xmath231 where @xmath232 is defined for @xmath233 by @xmath234 and @xmath235 is a cutoff function adapted to @xmath236 $ ] for some small @xmath106 ; for @xmath237 @xmath232 ( and thus @xmath230 ) is defined by analytic continuation . since @xmath238 , will follow from complex interpolation between the estimates @xmath239 for all real @xmath240 and some fixed @xmath241 . the former estimate follows immediately from the observation that the kernel of @xmath242 is uniformly bounded in @xmath240 ( indeed , the @xmath243 term makes it rapidly decreasing in @xmath240 ) . to prove the latter estimate , it suffices to show that @xmath244 is a fourier integral operator of order @xmath77 uniformly in @xmath240 ( see e.g. @xcite ) . accordingly , we write @xmath244 as @xmath245 where @xmath246 ranges over @xmath247 . from the rotational curvature hypothesis we see that the phase is non - degenerate in the sense of @xcite . since the amplitude is a symbol of order @xmath248 , @xmath103 , @xmath249 range over a @xmath250 dimensional space , and @xmath246 ranges over a @xmath74-dimensional space , the reduction - of - variables theorem ( see e.g. @xcite ) states that @xmath244 will be a fourier integral operator of order @xmath77 , as desired ; the uniformity in @xmath240 follows from the rapid decrease of @xmath251 with respect to @xmath240 , caused by the @xmath243 factor . we now show that the assumptions and in conjecture [ kakeya - conj ] are necessary . to show the necessity of , we take @xmath486 where @xmath487 denotes the kronecker delta function , and @xmath488 is a suitable point . a routine calculation using shows that ( if @xmath488 is chosen properly ) @xmath489 on a ball of radius @xmath490 . inserting this into @xmath288 yields @xmath491 and by taking @xmath180 one obtains . to show the necessity of , we adapt the `` stretched caps '' example used to show . we consider a tube @xmath492 in @xmath14 , and take @xmath493 if @xmath324 and @xmath325 are chosen appropriately , then @xmath494 are both comparable to @xmath74 on a slab which looks roughly like @xmath495 . inserting this into @xmath288 yields @xmath496 and by taking @xmath180 one obtains . we now begin the second part of this paper , in which we give analogues of the previous results for the kakeya operator . throughout this section @xmath168 will be a small parameter , and we will use @xmath252 to denote the estimate @xmath253 for all @xmath106 , otherwise we write @xmath254 . we say that a quantity @xmath9 has _ logarithmic size _ if @xmath255 , while we say it has _ polynomial size _ if @xmath256 for some constant @xmath257 . finally , all functions and quantities in this section are assumed to be non - negative . let @xmath258 be a @xmath202-net of the unit cube @xmath12 in @xmath14 . we give two measures on @xmath258 , counting measure @xmath259 and normalized counting measure @xmath260 . for @xmath261 , define the @xmath262 tube @xmath263 by @xmath264 and @xmath265 the direction and base of @xmath266 respectively . note that for fixed @xmath267 the tubes @xmath266 essentially form a partition of the unit ball @xmath268 . this discretization is not essential to the statements and estimates , but it allows for some technical simplification to the argument . for any function @xmath31 on @xmath164 , define the discretized x - ray transform @xmath269 on @xmath270 by @xmath271 for @xmath272 , let @xmath273 denote the estimate @xmath274 where @xmath270 is understood to be endowed with the measure @xmath275 . by taking @xmath31 to be the characteristic function of a @xmath202-ball we see that the factor @xmath276 is best possible . the kakeya conjecture asserts that @xmath273 holds if and only if @xmath277 and @xmath278 . in particular , it is conjectured that @xmath279 holds . it is easy to see that these conditions on @xmath44 , @xmath42 are necessary . the conjecture is trivial for @xmath280 ; the difficulty is in making @xmath44 ( and to a lesser extent @xmath42 ) as large as possible . so far the best result on this conjecture is due to wolff @xcite , who showed that @xmath281 in particular , for @xmath48 we have @xmath282 . this estimate is sharp in the sense that the @xmath283 exponent can not be raised without decreasing the @xmath284 exponent . the adjoint estimate @xmath285 to @xmath273 will be denoted @xmath286 ; note that @xmath287 following the philosophy of the previous sections , we define the bilinear version @xmath288 of the above estimate by @xmath289 for all @xmath31 , @xmath35 supported on @xmath290 , @xmath291 , where @xmath292 and @xmath293 are @xmath38-separated subsets of @xmath258 . we have the following analogue of theorem [ imply ] , which we will prove in section [ kakeya - iff ] . for technical reasons we will restrict ourselves to the case @xmath294 , which is the case of most interest . it is likely that one can use factorization theory and affine invariance to extend these results to the case @xmath295 . [ kakeya - imply ] suppose that @xmath296 . then @xmath273 and @xmath288 are equivalent . as with the restriction conjecture , it is possible to have bilinear kakeya estimates which are outside the range of the usual kakeya conjecture . for instance , one has the easy estimate [ 111 ] for any @xmath10 we have @xmath297 . we defer the simple proof of this proposition to section [ cordoba ] . interpolating this estimate with the estimate @xmath298 which by theorem [ kakeya - imply ] is the bilinear form of the kakeya conjecture , we see that the kakeya conjecture is equivalent to [ kakeya - conj ] if @xmath10 and @xmath299 , then @xmath288 holds if and only if @xmath300 we will show the necessity of and in section [ kakeya - nec ] . these two conditions correspond to and respectively ; the analogue of is the degenerate condition @xmath301 . wolff s theorem @xcite is equivalent to @xmath302 in particular , we have @xmath303 for @xmath48 . in section [ wolff - sec ] we improve the above estimate to [ b - wolff ] for all @xmath10 we have @xmath304 in particular , we have @xmath305 in three dimensions . the result can be thought of as a bilinear version of the ( false ) estimate @xmath306 , and is sharp in the sense that is obeyed with equality . we display the known kakeya and bilinear kakeya results in figure 2 . the trapezoidal region represents the conjectured range of @xmath45 for which @xmath273 should hold , and the pentagonal enlargement represents the range on which the bilinear version @xmath288 should hold . by theorem [ kakeya - imply ] the two estimates are equivalent in the triangular region below the dashed line . the point @xmath74 is the trivial @xmath307 estimate , while the point @xmath308 represents the higher - dimensional analogue of cordoba s argument ( @xcite , @xcite ) , while @xmath309 is the `` bush '' argument as given by bourgain @xcite ( see also @xcite , @xcite ) . proposition [ 111 ] , the bilinear improvement to cordoba s argument , is the point @xmath310 . the point @xmath311 is bourgain s kakeya maximal theorem @xcite ( see also @xcite ) , while @xmath312 is wolff s theorem @xcite , which we improve in theorem [ b - wolff ] to the point @xmath313 . the region to the right of the dotted line thus represents the best results known to date ( excepting the results in @xcite , which are not directly representable on this figure ) . we will need the following geometric observation of cordoba : [ cord - geom ] for any @xmath314 , one has @xmath315 in particular , if @xmath316 and @xmath317 have unit separation , then the intersection between the two tubes has measure at most @xmath318 . we leave the easy proof of this lemma to the reader . from this observation we easily see that @xmath319 which is @xmath297 , as desired . apart from several technical changes , this theorem will be proven using the geometric and combinatorial arguments of wolff @xcite , namely cordoba s observation ( lemma [ cord - geom ] ) and the `` brush '' argument . the bilinear setting allows for some simplification since the average angular separation @xmath320 between two tubes , as defined in @xcite , may be ( heuristically at least ) taken to be @xmath74 . in fact , this result informally follows by setting @xmath321 in lemma 2.1 of @xcite , and removing the `` two ends '' condition as in that paper . we will also take advantage of some simplifications noted by later authors ( notably @xcite , @xcite , @xcite ) . of course , due to the fact that we are in a bilinearized adjoint setting , there are some technical difficulties , most notably defining the analogue of the quantity @xmath322 in @xcite . also , since the target space @xmath323 is not a banach space , certain reductions and techniques ( e.g. duality , elimination of the @xmath324 , @xmath325 variables ) become unavailable . in particular , the lebesgue space approach of @xcite becomes technically very difficult , and we will use restricted weak - type methods instead . in other words , we will use the pigeon - hole principle to reduce as many functions as possible to characteristic functions . we first make the trivial observation that since @xmath326 is discretized , the operator boundedness of @xmath326 on lebesgue spaces is automatic with some large power of @xmath327 ; the issue is to control the dependence on @xmath202 efficiently . let us normalize @xmath31 and @xmath35 so that @xmath328 we have to show that @xmath329 it will suffice to show the weak - type bound @xmath330 for all @xmath331 , since the strong - type estimate can be recovered ( with only a logarithmic loss ) by integrating this over all @xmath332 of polynomial size ; the contribution of @xmath333 or @xmath334 can be easily controlled using trivial estimates . we now make the assumption that there exists sets @xmath335 of cardinality @xmath336 for @xmath337 such that @xmath338 this assumption is justified as any @xmath339-normalized @xmath31 , @xmath35 can be majorized by a sum of at most logarithmically many functions of this type . we may assume that the @xmath340 have polynomial size . from the pigeon - hole principle will follow from the estimate @xmath341 is any set such that @xmath342 and @xmath343 are arbitrary . we may assume that @xmath344 , @xmath345 have polynomial size , since this estimate is easily obtainable ( with a large gain ) otherwise . the @xmath346 , @xmath337 represent a normalized multiplicity of the tubes in the supports of @xmath31 and @xmath35 ; roughly speaking , they are related to the quantity @xmath18 defined in @xcite by the informal relationship @xmath347 define the quantity @xmath9 by @xmath348 . we may assume without loss of generality that @xmath9 is essentially minimal in the sense that @xmath349 , @xmath350 , @xmath351 , @xmath31 , @xmath35 which obey , . to copy wolff s argument in @xcite we will need some control on the quantity @xmath352 if @xmath353 is a `` typical '' tube in a direction in @xmath292 . ( in @xcite such control is automatic as one is not working in the adjoint setting ) . from and we have the pointwise estimate @xmath354 integrating this on @xmath351 we obtain @xmath355 where @xmath356 is defined for @xmath337 by @xmath357 from our assumptions we see that the @xmath356 are of polynomial size . the @xmath356 are the analogues of the quantity @xmath322 in @xcite . indeed , from and we expect @xmath358 on the average . in fact , because we are considering only an extremal configuration , a more precise statement is possible . we say that a tube @xmath266 is _ good _ if @xmath359 . let @xmath360 be the set of all @xmath361 in the support of @xmath31 associated to good tubes . the following improvement of states that most tubes are good . [ ext ] we have @xmath362 in particular , we have that @xmath360 is non - empty , so that @xmath363 . the upper bound follows immediately from , so it suffices to show the lower bound . let @xmath364 be a small number of logarithmic size to be chosen later . if the lower bound failed , then we would have @xmath365 the idea is to then replace @xmath31 by @xmath366 , and contradict the extremality of @xmath9 in . of course , we must modify @xmath367 further , as well as @xmath351 , @xmath368 and @xmath369 , in order to retain and . we replace @xmath351 by @xmath370 note that @xmath367 obeys if @xmath351 is replaced by @xmath371 and @xmath368 is replaced by @xmath372 . the next step is to show that @xmath371 is comparable to @xmath351 in size . from the definition of @xmath371 we see that @xmath373 however , we have from and the definition of @xmath367 that @xmath374 combining the two estimates and using the definition of @xmath360 we obtain @xmath375 using and this simplifies to @xmath376 so that @xmath377 as desired . we now have to modify @xmath367 , @xmath368 , @xmath371 , and @xmath369 further so that is restored . from hypothesis we have @xmath378 however , from we have @xmath379 thus by hlder s inequality this implies that @xmath380 thus , as before , we can find a logarithmic number of functions @xmath381 which each obey for some @xmath382 , and such that @xmath383 this implies that @xmath384 on @xmath371 . thus , by reducing @xmath371 by a logarithmic factor one can find a @xmath131 such that @xmath385 on the reduced set ( which we will still call @xmath371 ) . thus is satisfied with @xmath31 replaced by @xmath381 , @xmath351 replaced by @xmath371 , and @xmath368 replaced by @xmath386 . but from the definition of @xmath9 this implies that @xmath387 comparing this with and our estimates for @xmath371 and @xmath388 we thus obtain a contradiction , if @xmath75 is sufficiently small . from the above proposition , the definition of @xmath360 and the identity @xmath389 we obtain @xmath390 from this becomes @xmath391 from we thus have @xmath392 expanding out @xmath393 using this becomes @xmath394 on the other hand , from we have @xmath395 thus there must exist @xmath396 in the support of @xmath35 such that @xmath397 the tube @xmath398 plays the role of the central tube of a `` brush '' . unlike wolff s argument in @xcite ( which considered more general angular separations @xmath320 than the unit separation ) , we will be able to obtain our estimate using only a single brush . on the other hand , by utilizing the extremality hypothesis as in proposition [ ext ] , one could certainly obtain a large number of brushes if desired . by affine invariance we may take @xmath399 , so that the central tube is the vertical tube through the origin . in particular , @xmath77 is in @xmath292 , so every @xmath267 in @xmath293 has roughly unit separation from the origin . let @xmath400 be the collection of all good @xmath361 in the support of @xmath31 such that @xmath266 intersects the central tube @xmath401 . then expanding out @xmath402 in , we thus obtain @xmath403 from lemma [ cord - geom ] we have @xmath404 , so that @xmath405 let @xmath406 be the collection of all @xmath267 such that @xmath361 is in @xmath407 for at least one @xmath265 . from we see that @xmath408 so that @xmath409 for each @xmath410 we choose a tube @xmath411 from @xmath407 which is in the direction of @xmath267 . these tubes form the `` bristles '' of the brush . from construction , @xmath412 , @xmath411 intersects @xmath401 , and @xmath413 as in wolff @xcite , we will use to obtain a lower bound on the size of @xmath351 . more precisely , we will show that @xmath414 by a completely symmetrical argument one also has @xmath415 multiplying these estimates together one obtains @xmath416 applying this reduces to @xmath417 which simplifies to , as desired . it remains to prove . we use the argument in @xcite ; we adopt the observation in @xcite ( see also @xcite ) that one does not need to utilize the `` two ends '' reduction in @xcite to achieve . we need some notation . for all dyadic numbers @xmath418 let @xmath419 be the cylindrical region @xmath420 from the properties of @xmath411 we see that @xmath421 for all @xmath410 . by the pigeonhole principle , one can refine @xmath422 by a logarithmic factor so that @xmath423 for all @xmath267 in the refined @xmath422 , and some @xmath424 independent of the choice of @xmath267 ; henceforth this @xmath425 is considered fixed . the directions in @xmath422 are @xmath202-separated . it will be more convenient to work with a sparser set of directions , so we take @xmath426 to be any @xmath427-net of @xmath422 . from the estimates @xmath428 and @xmath429 we see that will follow from @xmath430 be a @xmath427-net of the unit sphere @xmath431 in @xmath14 . for each @xmath432 , we associate an ( essentially unique ) element @xmath433 of @xmath434 by requiring that @xmath435 recall that @xmath412 for all @xmath432 . furthermore , from elementary geometry and the fact that @xmath411 intersects @xmath436 we see that @xmath437 is contained in the slab @xmath438 given by @xmath439 as the @xmath438 are essentially disjoint , will follow from the estimate @xmath440 , where @xmath441 for the remainder of the argument @xmath267 ( and later @xmath442 ) are always assumed to range over @xmath443 . we now estimate the quantity @xmath444 in two different ways . firstly , from the above geometrical considerations and we have @xmath445 for all @xmath267 in @xmath443 . summing the above estimate we obtain @xmath446 we now obtain a different estimate for @xmath12 . from the cauchy - schwarz inequality we have latexmath:[\[q \lesssim squaring both sides and expanding out the integrand into the diagonal and off - diagonal term , this reduces to latexmath:[\[\label{cord } \frac{q^2}{|e \cap \pi_\theta| } \lesssim ( \int_{|e \cap \pi_\theta| } \sum_{\omega } \chi_{t_\omega } ) + \sum\sum_{\omega \neq \tilde\omega } term on the right - hand side is just @xmath12 . the second term we may estimate by lemma . thus becomes @xmath448 however , @xmath267 , @xmath442 range over a @xmath427-separated set whose elements are within @xmath427 of the ray @xmath449 . thus for each @xmath267 , the number of @xmath442 such that @xmath450 is at most @xmath451 , for any @xmath452 . thus the above estimate reduces to @xmath453 since the number of such @xmath452 is only logarithmic , we may simplify the above to @xmath454 combining this with and using the hypothesis @xmath429 we obtain @xmath455 which is . this finishes the proof . the proof will be a reprise of the argument in theorem [ imply ] . the main difference is that the quasi - orthgonality estimate is replaced by a quasi - triangle inequality , namely lemma [ young ] in the appendix . also the argument is technically simpler as we allow a logarithmic loss in the estimates . the case @xmath280 is trivial , so we will assume @xmath456 . the implication of @xmath288 from @xmath273 is immediate from duality and hlder s inequality . now suppose that @xmath288 holds for some @xmath457 obeying @xmath277 , @xmath458 . we have to show that @xmath459 holds . since the kakeya conjecture is known to hold for @xmath460 ( see e.g. @xcite , @xcite ) we may assume that @xmath461 . let @xmath31 be an arbitrary function on @xmath270 . we have to show that @xmath462 for each integer @xmath89 such that @xmath463 , we divide @xmath258 into @xmath96 dyadic `` subcubes '' @xmath464 of sidelength @xmath98 , and define the notion of closeness @xmath100 as in section [ iff ] . we partition @xmath465 as @xmath466 we now observe the geometric fact that the summand in the above expression is only non - zero when @xmath467 and @xmath468 are within @xmath88 of each other ; we will implicitly assume this in our summation . by inserting the above identity into and applying lemma [ young ] from the appendix we reduce ourselves to @xmath469 this will follow from the following analogue of proposition [ rescale ] . [ k - rescale ] we have @xmath470 by an affine transformation we may take @xmath97 , @xmath467 to be centered at the origin . applying the hypothesis @xmath288 to tubes of eccentricity @xmath471 we see that @xmath472 for all @xmath31 , @xmath35 whose @xmath267-supports are on disjoint cubes . applying the rescaling @xmath473 , @xmath474 to this estimate we obtain @xmath475 whenever @xmath31 and @xmath35 are supported on @xmath476 and @xmath477 respectively , and the proposition follows from substitution and some algebra . from this proposition reduces to @xmath478 since there are only logarithmically many @xmath452 s it suffices to show this for a fixed @xmath452 . by polarization it suffices to show that @xmath479 which we rewrite as @xmath480 where @xmath481 . it suffices to verify this for the case @xmath280 and for the endpoint @xmath482 , since the general case @xmath483 follows by interpolation . in these two cases becomes @xmath484 respectively . the estimate is trivial , while follows from a further interpolation between the trivial estimates @xmath485 we note that if one inserts the result of theorem [ b - wolff ] into the above line of reasoning , then one not only recovers wolff s kakeya estimate , but also the entropy estimate improvement proven in lemma 2.1 of @xcite . in this section we use the bilinear estimates above to prove the following restriction theorems . [ main ] if @xmath48 , then @xmath497 holds whenever @xmath498 and @xmath499 . furthermore , @xmath500 holds for all @xmath501 . the proof of this theorem will be based on bilinear versions of certain arguments of bourgain ( @xcite , @xcite ; see also @xcite ) . the first step will be to obtain localized linear and bilinear restriction theorems . if @xmath502 and @xmath503 , then we use @xmath504 to denote the estimate @xmath505 and @xmath506 to denote the estimate @xmath507 where @xmath508 are as in section [ rest - sec ] and @xmath509 is a ball of radius @xmath156 in @xmath164 ( the center of @xmath509 is irrelevant by translation symmetry ) . it will be convenient to recast these estimates as a restricted bilinear estimate on the fourier transform . [ rest - equiv ] @xmath510 is true if and only if one has @xmath511 for all @xmath155 , @xmath512 and all functions @xmath31 , @xmath35 supported on @xmath513 , @xmath514 , where @xmath515 if @xmath510 holds , then follows by translating @xmath120 by @xmath516 and averaging using hlder s inequality . now suppose that holds . to show @xmath510 it suffices to show that @xmath517 where @xmath518 is a real radial @xmath206-normalized bump function adapted to @xmath519 , such that @xmath520 is non - negative on @xmath521 . but this follows from , young s inequality , and the identity @xmath522 where @xmath523 is the lift of @xmath31 to the surface @xmath524 . from this proposition and the trivial estimate @xmath525 we obtain the bilinear trace lemma @xmath526 thus by interpolating this with other estimates ( such as ) we may obtain estimates of the form @xmath527 with a large value of @xmath332 . to lower the value of @xmath332 we will use a bilinear form of an argument of bourgain @xcite ( see also @xcite ) : [ b1 ] if @xmath528 and @xmath331 are such that @xmath529 and @xmath530 hold , then @xmath531 holds for all @xmath106 . from proposition [ rest - equiv ] it suffices to show that @xmath532 for all @xmath31 , @xmath35 supported on @xmath533 , @xmath534 respectively , and all @xmath106 . ( the implicit constants will depend on @xmath52 ) . let @xmath518 be as in proposition [ rest - equiv ] , and define @xmath535 for all @xmath512 . then from the hypothesis @xmath536 and proposition [ rest - equiv ] we have @xmath537 for all @xmath102 . averaging this over all @xmath538 we obtain @xmath539 thus to show it suffices to show that @xmath540 this will be accomplished by repeated use of the uncertainty principle and plancherel s theorem , together with the kakeya hypothesis . let @xmath258 , @xmath292 , @xmath293 be as in section [ b - kakeya ] with @xmath541 . we partition the annuli @xmath542 into caps @xmath543 for @xmath544 , defined by @xmath545 from the ellipticity of @xmath120 and some elementary geometry we see that the @xmath543 are essentially disks of diameter @xmath546 and thickness @xmath547 oriented in the direction @xmath548 , which form a finitely overlapping cover of @xmath542 . we decompose @xmath549 where @xmath550 , @xmath551 are adapted restrictions of @xmath31 , @xmath35 respectively to ( a suitable dilate of ) @xmath543 . from the support conditions on @xmath550 , @xmath551 and @xmath552 we see that reduces to @xmath553 the function @xmath554 is rapidly decreasing outside of the ball @xmath521 . thus by plancherel s theorem the left - hand side of is majorized by @xmath555 since the portions of @xmath554 on translates of @xmath521 can be handled by translation symmetry . let @xmath556 be a schwarz function which is comparable to @xmath74 on @xmath543 and rapidly decreasing away from this cap , and whose fourier transform satisfies the pointwise estimate @xmath557 where @xmath558 is a thickening of @xmath559 , and @xmath560 . if we define @xmath561 , we have the estimate @xmath562 from hlder s inequality and we thus obtain @xmath563 where @xmath564 and @xmath565 is the kronecker delta . since @xmath566 is essentially constant on balls of radius @xmath546 we essentially have @xmath567 from this ( and similar considerations for @xmath35 ) we see that is majorized by @xmath568 where @xmath569 is defined in analogy to @xmath570 . we simplify this as @xmath571 where @xmath572 on the other hand , from the definition of the hypothesis @xmath573 we have @xmath574 comparing this with and , we see that we will be done once we show that @xmath575 after some algebraic manipulation we see that it suffices to show that @xmath576 together with the completely analogous estimate for @xmath35 , @xmath360 . from the definition of @xmath550 and the measure @xmath577 we have @xmath578 and so it suffices to show that @xmath579 uniformly in @xmath267 . from hlder s inequality norms on @xmath31 , @xmath35 by the @xmath580 norms as used in @xcite , lemma 2.2 . ] , the hypothesis @xmath581 and the support conditions on @xmath550 we have @xmath582 and so after some algebra we reduce ourselves to @xmath583 however , the left - hand side is majorized by @xmath584 and the claim follows from plancherel s theorem and the pointwise comparability of @xmath550 and @xmath585 . applying lemma [ b1 ] with @xmath586 and @xmath587 and using , we see that @xmath588 for all @xmath589 . on the other hand , from interpolating with we obtain @xmath590 so by another interpolation we obtain the implication @xmath591 for all @xmath592 . combining these two implications we see that @xmath593 the map @xmath594 is a contraction with fixed point @xmath595 . since the estimate @xmath596 holds for at least one value of @xmath332 , we thus see that @xmath597 for all @xmath106 . applying lemma [ b1 ] one more time , we obtain @xmath598 an inspection of the proof of theorem [ imply ] shows that the statement of the theorem still holds when @xmath599 and @xmath600 are replaced by their local analogues @xmath506 and @xmath601 . applying this to we obtain @xmath602 we now remove the @xmath332 completely , borrowing the following argument of bourgain @xcite ( for the concrete case @xmath48 , @xmath603 , @xmath604 , @xmath605 , @xmath606 , @xmath607 , see @xcite ) : [ alpha]@xcite if @xmath608 are such that @xmath609 , then @xmath504 implies @xmath610 whenever @xmath611 applying this to we obtain the first conclusion of theorem [ main ] . using theorem [ imply ] to return to the bilinear setting , we thus obtain @xmath612 interpolating this with and using theorem [ imply ] , one obtains the second conclusion of theorem [ main ] . we summarize the various estimates used in figure [ fig : rest ] , which is an expanded version of figure [ fig : bilinear ] . for comparison , the previously known results are also displayed . the dotted line thus represents the best global restriction theorems ( both linear and bilinear ) known to date . ( it is possible to improve on these results slightly ; see @xcite ) . by interpolating between the main result and we also obtain some progress on klainerman s conjecture for the sphere in @xmath613 : [ beta ] if @xmath48 , then @xmath614 whenever @xmath615 . these techniques are certainly not best possible . for instance , one can use the techniques in @xcite to obtain better versions of corollary [ beta ] . see @xcite . the sharp restriction theorem @xmath47 is scale - invariant under parabolic scaling . thus , the compact support condition on @xmath120 can be removed . in particular , one has a sharp restriction theorem for the entire paraboloid @xmath616 for @xmath501 . one can extend the above results to bochner - riesz multipliers , so that the bochner - riesz conjecture holds for @xmath48 and @xmath617 . we sketch the argument very briefly as follows . by the usual techniques of carleson - sjlin reduction and factorization theory ( see @xcite ) it suffices to show that @xmath618 for all @xmath2 , @xmath619 and @xmath620 , where @xmath621 @xmath12 is thought of as imbedded in @xmath164 , and @xmath82 is a bump function on @xmath622 which is supported away from the diagonal @xmath623 . by the analogue of lemma [ alpha ] for bochner - riesz multipliers ( see @xcite ) it suffices to show that @xmath624 for all @xmath106 . by a modification of theorem [ imply ] it suffices to show that @xmath625 for all @xmath31 , @xmath35 with @xmath38 separated supports , together with variants of this estimate in which the phase function @xmath626 is replaced by a parabolically scaled ( but essentially equivalent ) version . however , from the analogue of lemma [ b1 ] for bochner - riesz operators ( which is proven by a bilinear modification of the arguments in @xcite ) this will follow from the restriction estimate and the analogue of theore [ b - wolff ] for the nikodym maximal operator ( see e.g. @xcite ) , which is proven similarly . the required nikodym estimate also follows formally from the original formulation of theorem [ b - wolff ] : see the argument in @xcite . in higher dimensions @xmath627 theorem [ modest ] becomes too weak to be of much use , and we can only achieve a minor improvement on known results . by interpolating between and the bilinear form @xmath628 of the tomas - stein theorem , we obtain @xmath629 applying this and theorem [ b - wolff ] to lemma [ b1 ] we obtain @xmath630 applying theorem [ imply ] this becomes @xmath631 applying lemma [ alpha ] this becomes @xmath632 this is only a slight improvement on the result in wolff @xcite , which showed @xmath633 for the same range of @xmath42 . for @xmath627 the results obtained by interpolating these estimates with theorem [ modest ] are inferior to the tomas - stein theorem . in the previous sections we obtained a non - trivial sharp restriction theorem @xmath634 from an ordinary restriction theorem @xmath29 ( in this case @xmath635 ) and the bilinear estimate . the original formulation of in @xcite was stated in terms of @xmath636 spaces . in this section we show how one can use these estimates instead of the bilinear estimate to obtain non - trivial sharp restriction theorems . despite the fact that these estimates can be extended ( for characteristic functions ) from @xmath637 to @xmath638 , the methods we will use do not appear to be as efficient as the bilinear techniques . however , they seem to be more robust and applicable to a wider range of situations . [ x - imply ] let @xmath48 . suppose that @xmath29 holds for some @xmath639 , and suppose that the quantity @xmath640 satisfies @xmath641 . then we have @xmath642 for all @xmath643 . note that the points @xmath644 are collinear when @xmath645 . it suffices to show the restricted weak - type estimate @xmath646 and @xmath647 , where @xmath93 is the integer such that @xmath648 . we may assume that @xmath649 for some constant @xmath257 , since is trivial ( by e.g. the tomas - stein restriction theorem ) otherwise . the idea of the proof will be to decompose @xmath92 into a sparse set and a collection of sets concentrated on caps . on the sparse set the tomas - stein estimate @xmath650 can be improved using the @xmath636 estimates of @xcite . the sets on caps can be rescaled parabolically to become sets of measure comparable to 1 , in which case the estimate @xmath29 is equivalent to the scale - invariant estimate @xmath47 . combining the two estimates one expects to obtain @xmath634 for some @xmath106 . we now turn to the details . for each @xmath452 let @xmath651 be a quantity to be chosen later . by the usual caldern - zygmund stopping time arguments , we may partition @xmath92 as @xmath652 where the `` good '' set @xmath653 satisfies @xmath654 for all @xmath655 , and @xmath656 is a collection of disjoint dyadic cubes such that @xmath657 for all @xmath658 . we decompose @xmath659 as @xmath660 to control the contribution of @xmath661 we use the results of @xcite and the hypothesis @xmath662 to obtain @xmath663 in order for @xmath29 to hold we must have @xmath664 , so that @xmath665 . applying of lemma [ xr - est ] with @xmath666 and @xmath667 we obtain ( after some algebra ) @xmath668 for each @xmath452 ; informally , this shows that the most significant scales occur when @xmath452 is near @xmath93 . inserting these estimates into we obtain @xmath669 let @xmath670 be a positive integer to be chosen later . we now choose @xmath346 so that @xmath671 when @xmath672 , and @xmath673 otherwise . the estimate then becomes @xmath674 from tchebyshev s inequality we thus obtain @xmath675 it remains to estimate the quantity @xmath676 for each @xmath452 . note that this quantity vanishes when @xmath677 by , so we may assume that @xmath678 . we will estimate in @xmath679 norm . from lemma [ young ] in the appendix we have @xmath680 on the other hand , by parabolically rescaling the hypothesis @xmath29 as in proposition [ rescale ] we obtain @xmath681 on the other hand , from we have @xmath682 using this and the fact that @xmath683 we obtain @xmath684 combining this with and the definition of @xmath685 one eventually obtains @xmath686 so by the triangle inequality and tchebyshev s inequality we have @xmath687 combining this with we obtain @xmath688 the claim then follows by choosing @xmath689 . in practice proposition [ x - imply ] is inferior to the implications obtained by theorem [ imply ] and interpolation with . for instance , if we insert the first conclusion of theorem [ main ] into proposition [ x - imply ] one obtains @xmath642 for @xmath690 , which is inferior to the second conclusion of theorem [ main ] . in this section we state some elementary results which were used repeatedly in the paper . we begin with a well - known quasi - orthogonality property of functions with disjoint frequency support . rectangle _ to be the product of @xmath32 ( possibly half - infinite or infinite ) intervals in @xmath164 . [ quasi - lemma ] let @xmath691 be a collection of rectangles in frequency space such that the dilates @xmath692 are almost disjoint , and suppose that @xmath693 are a collection of functions whose fourier transforms are supported on @xmath691 . then for all @xmath694 we have @xmath695 where @xmath696 . let @xmath697 be a smooth fourier multiplier adapted to @xmath692 which equals @xmath74 on @xmath691 . we claim that @xmath698 for arbitrary functions @xmath699 ; the lemma then follows by setting @xmath700 . by interpolation it suffices to prove this estimate for @xmath701 , @xmath702 , and @xmath703 . when @xmath702 the estimate is immediate from plancherel s theorem . when @xmath280 or @xmath703 the lemma follows from the triangle inequality and the estimates @xmath704 which follow from young s inequality and standard estimates on the kernel of @xmath697 . [ xr - est ] let @xmath647 be a set such that @xmath705 for some @xmath706 , and let @xmath97 be defined as in section [ iff ] . let @xmath707 be such that @xmath708 for all @xmath131 . if @xmath709 , then we have the estimate @xmath710 if @xmath711 , then we have the estimate @xmath712 by the log - convexity of @xmath713 norms , @xmath714 , it suffices to prove the three bounds @xmath715 but these bounds follow trivially from the estimates @xmath716 @xmath717 and the cardinality of the @xmath97 .

we construct small - amplitude solitary traveling gravity - capillary water waves with a finite number of point vortices along a vertical line , on finite depth . this is done using a local bifurcation argument . the properties of the resulting waves are also examined : we find that they depend significantly on the position of the point vortices in the water column .

1503.06143

the steady water - wave problem concerns two - dimensional water waves propagating with constant velocity and without change of shape . historically , the focus has mainly been on irrotational waves , which are waves where the vorticity @xmath0 of the velocity field @xmath1 is identically zero . one reason for this is kelvin s circulation theorem @xcite , which says that a flow which is initially irrotational will remain so for all time , as long as it is only affected by conservative body forces ( e.g. gravity ) . another reason is mathematical , as the velocity field can then be written as the gradient of a harmonic function ; the velocity potential . this enables the use of powerful tools from complex- and harmonic analysis , and the problem can be reduced to one on the boundary in a number of different ways @xcite . an important class of such waves are the stokes waves , which are periodic waves that rise and fall exactly once every minimal period . the stokes conjecture on the nature of the so - called stokes wave of greatest height fueled research on waves throughout the 20th century , and would not be fully resolved until 2004 ( see the survey @xcite and @xcite , which settled the convexity of this wave ) . more recently , however , there has been renewed interest in rotational waves . there are several situations where such waves are appropriate , as effects like wind , temperature or salinity gradients can all induce rotation @xcite . rotational waves can be markedly different from irrotational waves : for instance , in rotational waves it is possible to have internal stagnation points and critical layers of closed streamlines known as cat s eye vortices @xcite . the first result on rotational waves came surprisingly early , in the beginning of the 1800s with @xcite ( for a more modern exposition , see @xcite ) . there , gerstner gave the first , and still the only known , explicit ( nontrivial ) gravity - wave solution to the euler equations on infinite depth . although significant because it is an exact solution , it is viewed as more of a mathematical curiosity , even today ( see ( * ? ? ? * chapter 4.3 ) ) . much later , in @xcite , came the first existence result for small - amplitude waves with quite general vorticity distributions . a vorticity distribution is a function @xmath2 such that @xmath3 where @xmath4 is the relative stream function ( which , unlike the velocity potential , is still available for rotational waves , but is not harmonic ) . a sufficient , but not necessary , condition for such a vorticity distribution to exist is that the wave has no stagnation points . several improvements have been made to the existence result of dubreil - jacotin , but it was not until the pioneering article @xcite that large waves were constructed , using global bifurcation theory . this article sparked mathematical research into rotational waves . the use of a semi - hodograph transform in @xcite and @xcite , and the corresponding deep - water result in @xcite , means that the resulting waves can not exhibit critical layers . since then , small - amplitude waves with constant vorticity and a critical layer have been constructed in @xcite , and later in @xcite with a different approach that allows for waves with overhanging surface profiles ( there is numerical evidence for the existence of such waves , e.g. @xcite , but this is still an open problem ) . a reasonable next step is that of waves with an affine vorticity distribution , whose existence was shown in @xcite . spurred by the above results there has also been interest in studying the properties and dynamics of these waves below the surface @xcite . this had been done for linear waves in @xcite . several other avenues have also been considered : we mention heterogeneous waves both with @xcite and without @xcite surface tension , waves with discontinuous vorticity @xcite , a variational approach @xcite and hamiltonian formulation with center manifold reduction @xcite . existence of large amplitude waves with constant vorticity and a critical layer was established in @xcite , in the presence of capillary effects . there is also a forthcoming result for pure gravity waves @xcite , using an entirely different approach . common for all the previously mentioned works on rotational waves is the feature that the vorticity is supported on the entire fluid domain ( due to the assumption of the existence of a vorticity distribution ) . recently , gravity - capillary waves with compactly supported vorticity were constructed in @xcite , on infinite depth . this includes small- and large - amplitude periodic waves with a point vortex , and small - amplitude solitary waves with either a point vortex or vortex patch . by a point vortex we mean that the vorticity is given by a @xmath5-function , while we use vortex patch to mean that the vorticity is locally integrable and compactly supported . the waves with a point vortex are the simplest form of waves with compactly supported vorticity , and are in a sense `` almost irrotational '' . in this paper , which is based on @xcite , we extend the existence result for solitary small - amplitude waves with a point vortex to finite depth , and also give both qualitative and quantitative properties for these waves . the main approach to showing existence follows that of @xcite , but we also treat the natural generalization of waves with several point vortices along a vertical line , and show existence for all but exceptional configurations of vortices . finally , by finding an explicit expression for the rotational part of the stream function , we give some explicit expressions for the small - amplitude periodic waves with a point vortex on infinite depth that were constructed in @xcite . an outline of the paper is as follows : in we formulate the problem , and in we give the functional - analytic setting for this formulation . then , in we prove existence of small solutions , and give some properties for these . treats the extension to several point vortices . the final section , , contains the explicit expressions for periodic waves on infinite depth . under the assumption of inviscid ( absence of viscosity ) and incompressible ( constant fluid density ) flow , the governing equations of motion are the so - called incompressible euler equations . for describing water waves on the open sea , these are realistic assumptions @xcite , and standard . we will further assume two - dimensional flow under the influence of gravity , where the cartesian coordinates @xmath6 describe the horizontal and vertical direction , respectively . then the equations read @xmath7 where @xmath1 is the velocity of the fluid , @xmath8 is the pressure distribution and @xmath9 is the constant gravitational acceleration is approximately @xmath10 , varying by less than @xmath11 on the earth s suface ( see @xcite ) . ] . for convenience we place , at time @xmath12 , the flat bottom at @xmath13 and the surface at @xmath14 where @xmath15 describes the deviation of the free boundary . we assume that @xmath16 is bounded , continuous and strictly bounded below by @xmath17 . it should be emphasized that , due to the free boundary assumption , the function @xmath15 is a priori unknown ; determining it is part of the problem . in addition to , we require boundary conditions to match our domain . in order to model the bottom being impermeable , we will demand that @xmath18 where @xmath19 describes the surface tension and @xmath20 is the nonlinear differential operator @xmath21 yielding the curvature of the surface . the symbol @xmath22 denotes the japanese bracket defined through @xmath23 . is known to physicists as the young laplace equation , and states that the pressure difference across a fluid interface ( in this case water / air ) is proportional to its curvature . note that in the lower limit @xmath24 , the dynamic boundary condition in corresponds to the assumption of constant pressure on the surface , but we will require that @xmath25 be strictly positive . the proof of , for example , relies upon the assumption that @xmath26 . by taking the curl of , one obtains after some simple calculations that @xmath27 which states that the vorticity @xmath28 is transported by the vector field @xmath29 . due to this , it is natural to expect that if the vorticity consists of a point vortex at some time , then it will remain a point vortex at all future times , and be transported with the flow . it should be emphasized that , for now , this is not justified by ; the multiplication of @xmath28 with @xmath29 is not well defined , as @xmath29 will not be smooth at the point vortex . thus , we will have need of a weaker form of the equation . we remind the reader of the fundamental solution of the poisson equation . [ prop : delta_vorticity ] the distribution @xmath30 defined by @xmath31 satisfies @xmath32 and @xmath33 if @xmath28 is of the form @xmath34 then we deduce from that @xmath29 is of the form @xmath35 where @xmath36 satisfies @xmath37 and @xmath38 , and is therefore smooth in space ( see the discussion before ) . as the first term , which we may think of as the part of @xmath29 generated by the point vortex , is singular , divergence free and odd around @xmath39 , it is not unreasonable to think that the dynamics of the point vortex should depend only on @xmath36 . in other words , that the path @xmath40 along which the point vortex moves should satisfy @xmath41 this can indeed be made rigorous . in ( * theorems 4.1 and 4.2 ) it is proved that if one considers initial data consisting of a vortex patch converging in the sense of distributions to a point vortex , then the weak solutions of the vorticity transport equation converge to a moving point vortex in an appropriate sense . moreover , the position of this point vortex satisfies . thus , we will allow for point vortices , as long as they are propagated in the fluid as in . we now assume that there are functions @xmath42 , depending only on space , and a constant velocity @xmath43 such that @xmath44 for all relevant @xmath45 and @xmath12 . positive and negative @xmath46 then correspond to waves moving in the positive and negative @xmath47-directions , respectively . in the new _ steady _ variables @xmath48 , after dropping the tildes , our equations read @xmath49 with boundary conditions @xmath50 on the now time - independent domain @xmath51 we call the problem of finding @xmath52 and @xmath15 such that these equations are satisfied the _ steady water - wave problem_. note also that the vorticity equation given in reduces to @xmath53 for a point vortex centered at @xmath54 . it turns out that it is possible to reduce the water - wave problem to an entirely one - dimensional one on the surface in a clever way . this is known as the zakharov sulem formulation , and was first introduced by zakharov in @xcite , and then later put on a firmer mathematical basis in @xcite . the original formulation relies on the fluid being irrotational , but it is in fact sufficient that this holds near the surface . this is where the compact support of the vorticity comes in . suppose that we have solved the steady water wave problem for some @xmath55 . it is then convenient to split the velocity @xmath29 as @xmath56 where @xmath36 is irrotational , that is , @xmath57 and @xmath58 . we also assume that both @xmath36 and @xmath59 are divergence free . for us , @xmath59 will be known . although we will allow for @xmath28 to be a singular distribution , the vector field @xmath60 will be assumed to at least be in the sobolev space @xmath61 , and @xmath62 to be at least @xmath63 . by the assumption of @xmath64 , the differentials @xmath65 on @xmath66 are closed . hence , as @xmath66 is simply connected , these differentials are exact by generalizations of the poincar lemma ( * ? ? ? * theorems 2.1 and 3.1 ) . thus , there are stream functions @xmath67 ( by ( * ? ? ? * corollary 2.1 ) ) , determined uniquely modulo constants by @xmath36 and @xmath59 , such that @xmath68 moreover , by the assumed curl of these vector fields , the function @xmath69 is harmonic , and @xmath70 satisfies @xmath71 . in particular , @xmath69 is smooth , and so is @xmath59 outside the support of @xmath28 . by the above , we thus have that @xmath72 holds for the velocity field @xmath29 . suppose now that @xmath59 and @xmath70 are chosen such that @xmath73 at the bottom . then the boundary condition in translates to @xmath69 being constant along the bottom . since @xmath69 is unique modulo constants , we may as well take this condition to be @xmath74 instead . we will now apply the assumption that @xmath75 is compact . this means that a velocity potential for @xmath29 exists on any simply connected domain not containing @xmath76 . can then in turn be used to show that @xmath77 holds on any such domain . hence , in particular , we obtain the surface bernoulli equation @xmath78 where @xmath79 is a real constant . here we have inserted the boundary condition for the pressure at the surface , . we will set @xmath80 , since we are looking for localized waves . this can be seen by letting @xmath81 in . we now need the following formal definitions to proceed , which will be specified later on . [ def : harmonic_extension_operator ] given @xmath15 , we define the harmonic extension operator @xmath82 as the operator mapping a function @xmath83 to the harmonic function @xmath84 satisfying @xmath85 [ def : dirichlet_to_neumann_operator ] given @xmath15 , we define the dirichlet - to - neumann operator @xmath86 as the operator mapping dirichlet data to non - normalized neumann data ; that is , the operator defined by @xmath87\right\rvert_{y=\eta}}\ ] ] for functions @xmath83 . with in mind , define @xmath83 by @xmath88 that is , the trace of @xmath69 on the surface . by our assumptions , then , we have @xmath89 and we will use this to reformulate in a way that only involves @xmath90 and @xmath70 . note that @xmath91 where the right - hand side is evaluated at @xmath92 . inverting these relations and inserting them in yields @xmath93\\+\frac{(\zeta ' + ( 1,\eta')\cdot\nabla\psi)^2 + ( g(\eta)\zeta+(-\eta',1)\cdot\nabla\psi)^2}{2\jb{\eta'}^2}\\+ g\eta -\alpha^2\kappa(\eta)=0 , \end{multlined}\ ] ] and in a similar fashion , one obtains @xmath94 from the kinematic boundary condition in . can be integrated once to yield @xmath95 where we have assumed decay at infinity . we emphasize that the function @xmath70 and its derivatives are evaluated at @xmath92 in , which we suppress for readability . form the zakharov craig sulem formulation , which we will combine with a suitable vorticity equation . one may note that the pressure , @xmath8 , has been eliminated from the formulation entirely . is slightly different than the equation used in @xcite . the equation in @xcite can be obtained by inserting the kinematic boundary condition , , into . we now focus on proving the existence of a family of small amplitude and small velocity traveling waves with vorticity consisting of a point vortex situated on the @xmath96-axis . in other words , solutions with vorticity of the form @xmath97 where @xmath98 , @xmath99 and where we have defined @xmath100 the constant @xmath101 then corresponds to the relative position of the point vortex above the bottom , and the parameter @xmath102 , describing the strength of the vortex , will be used as the bifurcation parameter . we will from here on always assume that @xmath15 is such that @xmath103 , which prevents the surface from touching the point vortex . furthermore , we will also assume that @xmath104 . the reason for this is purely technical ( as we will see after ) . for the purpose of accounting for these assumptions , define the set @xmath105 whose intersection @xmath106 is open in @xmath107 for any @xmath108 , by the sobolev embedding @xmath109 . the next proposition describes the stream function that we will use for the rotational part of the velocity ; the counterpart of the function @xmath110 in @xcite . while we could use a similar stream function on finite depth , it is more beneficial to work with one that is tailored for finite depth . [ prop : phi_properties ] let @xmath111 and define @xmath112 by @xmath113 then @xmath114 defines a regular distribution , and @xmath115 moreover , the function @xmath116 is harmonic and satisfies @xmath117 where @xmath118 is the newtonian potential introduced in . we will apply ( see ) to prove this result , and thus need a bijective conformal map from the strip @xmath119 to the unit disk @xmath120 , mapping the point @xmath121 to the origin . this is done in three steps : @xmath122 the conformal map for each individual step is well known from elementary complex analysis , see for instance ( * ? ? ? * ii.7 and p. 60 ) . hence @xmath123 defines the desired map from the strip to the unit disk . by the aforementioned theorem , then , @xmath124 solves in @xmath125 . because @xmath126 extends to a meromorphic function on @xmath127 , it is immediate that @xmath114 also solves in @xmath66 ( recall that @xmath111 ) . finally , we have @xmath128 by the final part of the same theorem , which will be important for the asymptotic velocity of the traveling waves that we shall obtain in . note that the stream function @xmath114 introduces a `` mirror vortex '' at @xmath129 , and moreover is @xmath130-periodic in the @xmath96-direction . this is the reason for the limitation on the height of the surface profiles in the set @xmath131 defined in . the next proposition is crucial , because the traces of @xmath114 and its derivatives on the surface enter in the zakharov sulem craig formulation of the problem . having an explicit expression for @xmath114 enables us to prove the proposition in a quite direct way . [ prop : phi_trace ] suppose that @xmath132 , where @xmath133 . then @xmath134 moreover , the dependence on @xmath15 is analytic . we will only treat @xmath114 , as the argument for the derivative is similar . observe that it is sufficient to consider the function defined by @xmath135 for @xmath136 such that @xmath137 , where @xmath138 . since @xmath139 it follows by ( * ? ? ? * theorem 4 of 5.5.3 ) , @xmath140 and @xmath107 being an algebra that the function in lies in @xmath107 and that the dependence on @xmath15 is analytic . another application of the result in @xcite then yields the desired result . as we have seen , because of the reliance on the stream function and the operators @xmath82 and @xmath86 , a central problem is the solution of the laplace equation , @xmath141 on the fluid domain @xmath66 , given @xmath15 and @xmath90 . we have the following theorem , which is adapted from corollary 2.44 in @xcite , and which establishes both existence and uniqueness to in suitable sobolev spaces . functions on the surface will be identified with functions on the real line as in . [ eq : laplace_well_posed ] suppose that @xmath132 for some @xmath142 , and that @xmath143 . then has a unique solution in @xmath144 . while the natural setting for the velocity potential or the stream function on infinite depth is that of homogeneous sobolev spaces , used in both @xcite and @xcite , this is not the case for the stream function on finite depth . because we require @xmath69 to be equal to a constant at the bottom , it must necessarily be the case that @xmath69 tends to the same constant at infinity . otherwise , because of the finite depth , @xmath145 would not decay at infinity ( in the sense that @xmath146 ) , and therefore not describe a localized wave . enables us to define the harmonic extension operator described in as an operator @xmath147 , and using this , defining the dirichlet - to - neumann operator . we refer the reader to @xcite , which is a rich source of results for these operators also in a more general setting . the results there are proved for the velocity potential , but should be adaptable for the stream function . the below theorem is an amalgamation of parts from corollary 2.40 and theorems 3.15 and a.11 in @xcite . see also @xcite . [ thm : boundary_operators ] let @xmath142 and suppose that @xmath132 . then the harmonic extension operator @xmath82 and the dirichlet - to - neumann operator @xmath86 are members of @xmath148 and @xmath149 , respectively . the norms of these operators are uniformly bounded on subsets of @xmath150 that are bounded in the norm on @xmath107 . moreover , the map @xmath151 is analytic for fixed @xmath152 . in the same setting as in , the curvature of the surface is well defined : [ prop : curvature ] the curvature operator @xmath20 is well defined as an operator @xmath153 for any @xmath142 . moreover , the map is analytic . observe that the function @xmath154 defined by @xmath155 is smooth and satisfies @xmath156 . as @xmath157 , the result ( * ? ? ? * theorem 4 of 5.5.3 ) ensures that @xmath158 . since @xmath126 is also analytic , @xmath20 is analytic by the same result . there is one thing we have not yet looked at , namely the vorticity equation . recalling , we will consider velocity fields of the form @xmath159 from we know that the part of the stream function that is generated by the point vortex at @xmath160 is given by the newtonian potential @xmath161 whence reduces to @xmath162(0,-(1-\theta)h ) + \varepsilon\left(\frac{1}{4h}\cot(\pi\theta),0\right),\ ] ] by in . in particular , this means that any solution necessarily must satisfy @xmath163_x(0,-(1-\theta)h ) = 0.\ ] ] for simplicity , we choose to look for @xmath164 in appropriately chosen subspaces of @xmath107 , such that this condition is automatically satisfied . specifically , define @xmath165 which is closed in @xmath107 , and therefore a hilbert space in the inherited norm . we mention that it is still an open question whether asymmetric traveling waves exist . however , it is known that , in many situations , certain properties imply symmetry ( see for instance @xcite ) . furthermore , under suitable assumptions , all symmetric waves are traveling waves @xcite . assume now that @xmath166 , with @xmath142 , and that @xmath167 . then it must necessarily be the case that @xmath168 is even in @xmath47 . hence @xmath169_x$ ] vanishes along the @xmath96-axis and so the vorticity equation reduces further to @xmath170_y(0,-(1-\theta)h ) , \quad \text{where $ c_1 \coloneqq \frac{1}{4h}\cot(\pi\theta)$.}\ ] ] observe also that the dirichlet - to - neumann operator @xmath86 is well defined as an operator @xmath171 for @xmath172 and @xmath142 , and that @xmath20 can be viewed as an operator @xmath173 . one has to be careful with claims about the solution set when @xmath174 . of course actually only needs to be satisfied if @xmath175 . this means that if we impose , then we lose the trivial set of solutions @xmath176 for the other equations , except for the point @xmath177 . this should be kept in mind in any claims of uniqueness . for convenience , define now the spaces @xmath178 which accounts for the limitations on @xmath15 . we proceed to introduce three maps that together will form the basis for our argument . for @xmath142 we define @xmath179 by @xmath180\\ + \frac{(\zeta ' + \varepsilon(1,\eta ' ) \cdot \nabla \phi)^2 + ( g(\eta)\zeta + \varepsilon(-\eta',1)\cdot \nabla \phi)^2}{2\jb{\eta'}^2 } + g\eta -\alpha^2\kappa(\eta ) , \end{gathered}\ ] ] the map @xmath181 by @xmath182 and finally the map @xmath183 by @xmath184_y(0,-(1-\theta)h).\ ] ] in all of these definitions , we really mean the traces of @xmath114 and its derivatives on the surface . the pointwise evaluation in the second term of is allowed because @xmath168 is harmonic . it should be clear that all three maps @xmath185 are smooth . we can now define @xmath186 by @xmath187 and our task will then be to find solutions of the equation @xmath188 one may immediately note that @xmath189 , so that the origin is a trivial solution . it will turn out that in a small neighborhood of the origin in @xmath190 , there is a unique curve of nontrivial solutions parametrized by the vortex strength parameter @xmath102 . we can now finally state and prove the following theorem , establishing the existence of small , localized , traveling wave solutions with a point vortex . for this , we will use an implicit function theorem argument on @xmath191 . note that while we do not apply crandall rabinowitz theorem @xcite , the situation is very much in the spirit of that theorem . we bifurcate from the family of trivial waves described in the remark after by introducing the vorticity equation . [ thm : existence_point_vortex ] let @xmath142 and @xmath99 . then there exists an open interval @xmath192 and a @xmath193-curve @xmath194 of solutions to the zakharov craig sulem formulation , , for a point vortex of strength @xmath102 situated at @xmath160 . the solutions fulfil @xmath195 in their respective spaces as @xmath196 , where @xmath197 is defined by @xmath198 and where @xmath199_y(0,-(1-\theta)h ) , \end{aligned}\ ] ] with @xmath200 as in , @xmath114 as in and @xmath201 as in . moreover , there is a neighborhood of the origin in @xmath202 such that this curve describes all solutions to @xmath203 in that neighborhood . as remarked at the end of , the origin is a trivial solution . in order to apply the implicit function theorem , we require the first partial derivatives of @xmath191 at this point . a direct calculation yields @xmath204_y(0,-(1-\theta)h ) & 1 \end{bmatrix},\ ] ] where the subscript @xmath205 denotes the partial derivative with respect to the variable @xmath206 in @xmath207 . now , every operator on the diagonal of @xmath208 is an isomorphism . indeed , the operator @xmath209 \colon h_\text{even}^s ( { \mathbb{r } } ) \to h_\text{even}^{s-2 } ( { \mathbb{r } } ) } \ ] ] corresponds to the fourier multiplier @xmath210 . since @xmath211 , this operator is invertible , with inverse corresponding to the multiplier @xmath212 . the other two operators on the diagonal are identity operators , and therefore trivially invertible . hence @xmath213 is also an isomorphism . thus we can use the implicit function theorem to conclude that there is an open interval @xmath214 containing zero , an open set @xmath215 containing @xmath216 , and a map @xmath217 such that for @xmath218 , we have @xmath219 furthermore , we obtain @xmath220 which yields the first order terms in . the higher - order terms can be obtained by inserting expansions for @xmath221 and @xmath222 into the equation @xmath223 . this concludes the proof of the theorem . because holds for any @xmath142 , we can get arbitrarily high regularity on the solutions , by possibly making the interval @xmath214 smaller . we have not been able to conclude that they are smooth , however , since the interval could possibly shrink to a point as @xmath224 . observe that , because @xmath200 changes sign at @xmath225 , the direction in which the waves obtained in will travel ( for small @xmath102 ) depends on where the point vortex is in relation to the line @xmath226 . this does not come into play for waves on infinite depth . note that if @xmath227 ( when @xmath228 ) then @xmath229 as @xmath230 , which is in agreement with what was found in @xcite for a point vortex situated at @xmath231 on infinite depth . since @xmath200 vanishes when @xmath232 , also the next term in the expansion for @xmath222 is of interest . we gave an expression for @xmath233 in , but have not determined its sign yet . we will treat the sign of @xmath233 after , which establishes some properties of the function @xmath234 . written out , we have @xmath235 for the function @xmath236 defined in . we will have use for the fact that @xmath236 has an elementary antiderivative @xmath237 and a double antiderivative @xmath238 given by @xmath239 respectively . while there in general seems to be no nice closed form of the leading order surface profile @xmath240 obtained in , we can still give some of its properties . an immediate one is that @xmath234 is smooth . in we give a series expansion for @xmath234 in powers of @xmath241 . furthermore , perhaps more surprisingly , we _ can _ find an explicit expression for @xmath234 in terms of elementary functions whenever @xmath242 is a natural number . if @xmath243 , then @xmath244 are integral powers of @xmath245 , which would appear on the right side of if we had written out @xmath246 and @xmath247 . since @xmath248 spans the kernel of @xmath249 , this explains integral values of @xmath250 being special . before we state and , we need a lemma to simplify some expressions . [ lemma : explicit_expression_for_eta_coefficients ] for @xmath251 and @xmath99 , we have @xmath252 which is equal to @xmath253 whenever @xmath254 . furthermore , for @xmath243 , @xmath255 both sides of define meromorphic functions in @xmath250 on @xmath127 with simple poles in the points @xmath256 . moreover , they are both equal to the integral in when @xmath254 , which can be seen by calculating the integral with both the residue theorem ( around a keyhole contour ) and a laurent series expansion of the integrand . since the interval consists of non - isolated points , we have equality on all of @xmath127 . finally , follows from by taking limits . [ prop : eta_point_vortex_explicit ] if the number @xmath250 in satisfies @xmath251 , then the leading order term of the surface profile from is given by @xmath257 , \end{gathered}\ ] ] while if @xmath243 , then @xmath258 . \end{gathered}\ ] ] these series converge uniformly , and , excluding the origin , so do the series for the first derivative . moreover , when @xmath243 , the function @xmath234 is given explicitly in terms of elementary functions by @xmath259 , \end{gathered}\ ] ] where @xmath260 is defined by @xmath261 it follows from @xmath262 and the definition of @xmath234 , that we may write @xmath234 as the convolution @xmath263 where @xmath264 equivalently @xmath265 through integration by parts , where @xmath237 and @xmath238 are the antiderivatives defined in . we first use to obtain an explicit expression for @xmath234 when @xmath243 . by using the substitution @xmath266 , we find that @xmath267 the fraction in the integrand in the definition of @xmath268 has partial fraction decomposition @xmath269 and since @xmath270 this means that @xmath271 where @xmath272 denotes the principal branch of the logarithm . the result now follows by using the identity @xmath273 valid for all @xmath274 . for the series representation of @xmath234 , we use , because this leads to a series that converges more rapidly . we will assume that @xmath275 ; the case for @xmath243 is similar , except that one needs to use instead of . we use the same substitution as before to arrive at @xmath276 where @xmath277 is defined by @xmath278 one may check that one has @xmath279 for @xmath280 and @xmath281 for @xmath282 . it then follows by termwise integration that @xmath283 on @xmath284 $ ] ( the endpoint is abel s theorem ( * ? ? ? * theorem 17.14 ) ) , and that @xmath285-\frac{1}{m}-\log(2)+\log(x)-2m\sum_{k=1}^\infty ( -1)^k \frac{\cos(k\pi\theta)}{k(m - k)}x^{-k}\\-\left(\frac{2}{m}+4m\sum_{k=1}^\infty ( -1)^k\frac{\cos(k\pi\theta)}{m^2-k^2}\right)x^{-m}. \end{multlined } \end{aligned}\ ] ] for @xmath286 . employing , we find that @xmath234 is given by @xmath287 \end{multlined}\ ] ] for all @xmath274 , by using that @xmath234 is even and observing that for @xmath288 we have @xmath289 and @xmath290 $ ] . if we now apply from in order to get a closed - form expression for the coefficient in front of @xmath291 , we arrive at the desired expansion . the only obstacle to convergence of the series given in is the origin ; thanks to the exponential factor @xmath292 , the convergence is rapid away from the origin . it should also be noted that , while seems to suggest that @xmath234 should be expandable in a series of powers of @xmath293 by equating coefficients in the differential equation defining it , this seems to lead to a series that does not converge . we have kept the series expansion for @xmath234 also when @xmath243 , because the expression in terms of elementary functions is unwieldy , and prone to numerical errors even for small values of @xmath250 . the expressions found in have well defined pointwise limits as @xmath294 ( for @xmath295 ) and @xmath296 . in particular , when @xmath297 these are given by @xmath298\\ \lim_{\theta \uparrow 1 } \eta_2(x ) & = \frac{1}{8\pi^2\alpha^2}[1 + e^{\pi x / h}\log{\abs{1-e^{-\pi x / h } } } + e^{-\pi x / h}\log{\abs{1-e^{\pi x / h } } } ] , \end{aligned}\ ] ] which can can be seen as graphs drawn with thicker lines in , together with @xmath234 for various values of the parameter @xmath101 . in @xmath299 , with @xmath300 . the values of @xmath101 shown are @xmath301 , together with the thicker lower and upper limits @xmath296 and @xmath294 . ] we see from that one gets a depression at the origin , which becomes more pronounced the closer the point vortex is situated to the surface . the profile when the point vortex is close to the surface is very similar to the profile for the infinite depth case , found in @xcite . however , a feature which is not seen on infinite depth is that there is a significant difference between the case @xmath302 and the case @xmath303 ( in addition to the changing sign of @xmath200 ) . for @xmath302 there is a single trough at the origin , and @xmath234 is everywhere strictly negative . when @xmath303 one in addition gets crests on either side of the origin as we can see from , the positions of these crests depend on the position of the point vortex . some of what we have just discussed is not limited to the specific choice of constants that are used in , and for which yields an explicit expression for @xmath234 . we will see that @xmath304 plays a special role in the asymptotic behavior of @xmath234 , however . more precisely , we have the following theorem : [ theorem : asymptotic_behavior_of_eta ] the leading order surface term @xmath234 always satisfies @xmath305 and @xmath306 , meaning that the origin is a depression . when @xmath302 , the function @xmath234 is everywhere negative , and strictly increasing on @xmath307 . for @xmath308 , we have two cases , depending on the number @xmath250 defined in : a. if @xmath309 , then @xmath310 is positive for sufficiently large @xmath311 . in particular , @xmath234 has crests on either side of the origin . b. if @xmath312 , then @xmath310 is negative for sufficiently large @xmath311 . furthermore , @xmath234 has the following asymptotic properties for any @xmath99 : a. for @xmath313 @xmath314 b. if @xmath304 , then @xmath315 c. for @xmath316 @xmath317 we first prove that @xmath305 and @xmath318 , which holds for all values of @xmath250 and @xmath101 . by inserting @xmath319 in , and using the evenness of @xmath236 , we find @xmath320 where the second equality follows from integration by parts , and the function @xmath237 was defined in . since @xmath321 , we also have @xmath322 as @xmath236 achieves a global maximum at the origin . suppose now that @xmath302 . like in , we use the fact that @xmath234 may be written as the convolution @xmath323 which shows that @xmath234 is strictly negative , since @xmath236 is strictly positive when @xmath302 . moreover , some manipulations of the above formula shows that we may write the derivative of @xmath234 as @xmath324 , \end{gathered}\ ] ] where we have used the fact that @xmath236 is even . one may check that @xmath325 is strictly negative for @xmath326 when @xmath302 . this shows that @xmath327 is strictly positive for @xmath326 , and so @xmath234 is strictly increasing on @xmath307 by the mean value theorem . before we consider the case @xmath303 , we prove the asymptotic properties for @xmath234 listed in . these follow by multiplying each side in with the appropriate factor and taking limits . for instance , suppose that @xmath313 , meaning that @xmath328 . for the integral in @xmath329 there are two possibilities : if @xmath232 , then it is possible that the integrand is integrable on the entire real line , meaning that the limit as @xmath330 is zero ; otherwise , the integral tends to @xmath331 , and so @xmath332 by lhpital s rule . the other limits can be treated in a similar way , with one exception : the procedure will show that when @xmath316 , we have @xmath333 where the second and third equality follows from the substitution @xmath334 and an integration by parts , respectively . the result now follows since the integral on the final line is equal to the right - hand side of by . finally , we consider the case of @xmath303 , which is harder to describe completely , as the integrand in the convolution in changes sign . observe that the claims on the sign of @xmath310 for sufficiently large @xmath47 follows for @xmath335 from the limits in . an additional argument is needed for the edge case @xmath336 , because the limit in vanishes . it turns out that also holds in the special case @xmath336 , which can be shown with the same method we used to show the other limits . hence @xmath234 is negative for sufficiently large @xmath47 when @xmath336 , which exhausts the values of @xmath250 . it is likely that @xmath234 has similar properties to those for the case @xmath302 when @xmath303 and @xmath312 , but we have not been able to prove this . we are now in a position where we can give the sign of @xmath233 in the expansion in for @xmath302 . [ prop : velocity_sign ] the constant @xmath233 in is negative when @xmath302 . in particular , if @xmath225 and @xmath102 is sufficiently small , the waves obtained in are left - moving when @xmath337 and right - moving when @xmath338 . recall the definition of @xmath339 in . from we know that @xmath234 is negative , and strictly increasing on @xmath307 . furthermore , the factor @xmath340 is positive and strictly decreasing on the same interval . it follows that also @xmath339 is positive and strictly decreasing on @xmath307 . the harmonic function @xmath341 on @xmath342 assumes the value @xmath343 at the bottom of the domain and @xmath344 at the top of the domain . by the maximum principle , it is positive on the entire domain . thus we may use the hopf boundary point lemma ( see ( * ? ? ? * lemma 3.4 ) ) in order to conclude that @xmath345_y(0,-h ) > 0 $ ] . the result will therefore follow if we can show that @xmath345_y$ ] is increasing along the @xmath96-axis . we will do this by looking at @xmath345_x$ ] on @xmath346 . because of its values on the boundary , it is negative in the interior . another application of the hopf boundary point lemma implies that @xmath345_{xx}$ ] is negative on the @xmath96-axis ( except at the point @xmath347 , where it vanishes ) . since @xmath345_{yy } = - [ h(0)\zeta_3]_{xx}$ ] by the harmonicity of @xmath341 , this concludes the proof . we finish our exposition on a single point vortex with a short discussion on the streamlines of waves obtained in . observe that if @xmath348 denotes the position of a fluid particle at time @xmath12 , then @xmath349 before the new variables in . after introducing the steady variables , becomes @xmath350 meaning that if we only keep the first order terms for @xmath29 and @xmath46 from , we obtain ( keeping the same notation for the paths ) @xmath351 0.495 and @xmath337 . the wave corresponding to @xmath352 propagates to the right , while the wave corresponding to @xmath353 propagates to the left . the arrows illustrating the vector field on the right hand side of have been scaled here for visibility , and only their direction is quantitatively accurate.,title="fig : " ] 0.495 and @xmath337 . the wave corresponding to @xmath352 propagates to the right , while the wave corresponding to @xmath353 propagates to the left . the arrows illustrating the vector field on the right hand side of have been scaled here for visibility , and only their direction is quantitatively accurate.,title="fig : " ] we have used this to obtain , which shows streamlines in the steady frame moving with the wave . the portraits corresponding to @xmath101 and @xmath354 can be obtained from each other by a rotation . when @xmath232 , all the streamlines are closed ( not shown ) , so we will focus on the case @xmath355 . the lines @xmath356 and @xmath357 are nullclines for the system in , and the points @xmath6 with @xmath358 are equilibrium points , corresponding to stagnation points . one may check that @xmath359 as @xmath296 , meaning that the distance between the equilibria is very close to linear in @xmath101 for small @xmath101 ( a corresponding statement holds for @xmath354 small ) . they go off to infinity as @xmath360 from either side . the heteroclinic orbit ( which can be expressed explicitly in terms of @xmath361 ) connecting the two equilibrium points described in encloses a critical layer containing closed streamlines . outside this region the particles always move in the same direction with respect to the steady frame . this direction is either to the left or right depending on the sign of @xmath362 and @xmath102 . we also mention that on infinite depth , the streamlines always look like those in . if the point vortex is situated at @xmath363 , the equilibrium points at the surface will be at @xmath364 , and the points on the heteroclinic orbit between these satisfy @xmath365 which is close to half an ellipse centered at @xmath363 with semiaxes @xmath366 and @xmath367 . the equilibrium points in converge to those on infinite depth as @xmath230 if @xmath368 is held fixed . because only the first order terms in @xmath102 have been kept in , we do not make any claim about the accuracy of the phase portraits in for the full system in . that would require further and more thorough analysis , in particular for the case @xmath225 . still , the phase portraits can give some indication as to how these waves look beneath the surface . one feature will remain the same for : because of the singularity of @xmath114 at @xmath160 , the streamlines will always remain closed sufficiently close to the point vortex . we aim to extend the existence result for traveling waves with a single point vortex in to a finite number of point vortices on the @xmath96-axis . as opposed to the single vortex case , where we could choose @xmath101 freely , there will be limitations on the positions that the point vortices can occupy . we will return to this . suppose that @xmath369 and that we wish to establish the existence of a traveling wave with point vortex at the points @xmath370 the situation being otherwise similar to that of a single point vortex . the admissible surface profiles are those in @xmath371 , as the uppermost point vortex is the most restrictive . for @xmath372 and @xmath373 we may define @xmath374 where @xmath375 in @xmath66 . we will seek solutions of the form @xmath376,\ ] ] cf . for a single point vortex . the main difference from the single point vortex case is of course the vorticity equation , , which needs to be imposed for each of the point vortices . for the @xmath377th point vortex , the vorticity equation reduces to @xmath378=\nabla^\perp[h(\eta)\zeta](0,-(1-\theta_i)h)\\+ \frac{1}{4h}\biggl(\gamma^i\cot(\pi\theta_i ) + \sum_{\substack{j=1 \cr j\neq i}}^n \gamma^j\left[\cot\left(\pi\frac{\theta_i + \theta_j}{2}\right)-\cot\left(\pi\frac{\theta_i-\theta_j}{2}\right)\right],0\biggr ) , \end{multlined}\ ] ] which , if we assume that @xmath15 and @xmath90 are even ( see the discussion before ) , can be written more succinctly as @xmath379_y(0,-(1-\theta_i)h))_{i=1}^n + \theta\gamma.\ ] ] here , we have defined @xmath380 and the matrix @xmath381 by @xmath382 for @xmath383 . as opposed to for one vortex , it is now more natural to use the wave velocity @xmath46 as the bifurcation parameter . we will therefore write @xmath102 instead of @xmath46 in order to have notation that is more consistent with the one vortex case . the idea is to use the vortex strengths @xmath384 in order to balance , which is possible when @xmath385 is invertible . it should be emphasized that this is almost always the case ( ) , but that there always are configurations of @xmath386 point vortices that yield singular @xmath385 ( ) . we have already seen such a configuration , albeit a trivial one : for the case @xmath387 one has @xmath388 when @xmath225 . we make the necessary redefinitions @xmath389 and proceed to define , for @xmath142 , the map @xmath390 by @xmath391\\ + \frac{(\zeta ' + ( 1,\eta ' ) \cdot \nabla \phi^\gamma)^2 + ( g(\eta)\zeta + ( -\eta',1)\cdot \nabla \phi^\gamma)^2}{2\jb{\eta'}^2 } + g\eta -\alpha^2\kappa(\eta ) , \end{gathered}\ ] ] the map @xmath392 by @xmath393 and finally the map @xmath394 by @xmath395_y(0,-(1-\theta_i)h))_{i=1}^n.\ ] ] in all of these definitions , the function @xmath396 and its derivatives are evaluated at @xmath397 , which is suppressed for readability . we now define @xmath398 , and seek solutions of the equation @xmath399 which has the origin as a trivial solution . we are led to the following analog of for several point vortices , establishing the existence of a family of small , localized solutions , assuming that @xmath385 is nonsingular . the resulting waves have one critical layer for each point vortex , assuming that no component of @xmath384 vanishes . [ thm : existence_several_point_vortices ] let @xmath142 , and let @xmath400 . suppose that the matrix @xmath385 defined in is invertible . then there exists an open interval @xmath192 and a @xmath193-curve @xmath401 of solutions with velocity @xmath402 to the zakharov craig sulem formulation , , for point vortices of strengths @xmath403 situated at @xmath404 the solutions fulfil @xmath405 in their respective spaces as @xmath196 , where @xmath406 , the function @xmath197 is defined by @xmath407 and where @xmath408_y(0,-(1-\theta_i)h))_{i=1}^n , \end{aligned}\ ] ] with @xmath409 as in and @xmath201 as in . moreover , there is a neighborhood of the origin in @xmath202 such that this curve describes all solutions to @xmath410 in that neighborhood . as for a single point vortex , we wish to apply the implicit function theorem at the origin . we find the derivative @xmath411_y(0,-(1-\theta_i)h))_{i=1}^n & \theta \end{bmatrix},\ ] ] where @xmath412_y(0,-(1-\theta_i)h))_{i=1}^n$ ] means the operator @xmath413 defined by @xmath414_y(0,-(1-\theta_i)h))_{i=1}^n.\ ] ] recalling that @xmath249 and @xmath385 are invertible by the discussion after and by assumption , respectively , @xmath208 is an isomorphism . hence we can use the implicit function theorem to deduce the existence of an open interval @xmath214 around zero , an open set @xmath415 containing the origin , and a map @xmath217 such that for @xmath416 , we have @xmath417 the terms in the expansion in can be obtained as in the proof of . it is worth mentioning that on infinite depth , the matrix @xmath385 is always invertible . a corresponding existence theorem for infinite depth would thus hold for any configuration . an extension of the existence result in to point vortices that are not all on the same vertical line would require a different argument than the one we have used . the main issue is that assuming @xmath15 and @xmath90 to be even is then no longer sufficient to satisfy the vertical component of the vorticity equation , like we did to obtain . one may note that the sign reversal of the wave velocity about the midpoint @xmath232 that we saw with the single point vortex , can be seen also for several point vortices , albeit in a different manner . if the matrix @xmath385 corresponds to @xmath418 , and we reflect the vortices across the line @xmath419 by considering @xmath420 instead ( without reordering them ) , then the new matrix is @xmath421 . this causes a swap of sign on the leading order vortex strengths , @xmath422 . we have pointed out that the matrix @xmath385 is not invertible for all configurations of point vortices , and gave the trivial example of @xmath225 for a single point vortex . this example , together with , also shows that invertibility of @xmath385 is not a necessary condition for the existence of a traveling wave with point vortices in those points . see also . the only case for multiple point vortices on the @xmath96-axis where we can feasibly describe the admissible positions directly is for @xmath423 . in fact , we give a complete description of when @xmath385 is invertible in ; see also , which presents this result graphically . one may observe that the midpoint between the bottom and surface plays a role also here . for the case @xmath423 as a function of @xmath424 . the determinant vanishes along the solid black curve , which is given explicitly as a parametrization in . ( in the figure , the level curve for @xmath425 is computed numerically . ) ] [ prop : two_point_vortices ] for two point vortices , we have the following : a. if @xmath426 , then @xmath385 is invertible for all @xmath427 . b. if @xmath428 , then @xmath385 is invertible for all @xmath427 except for exactly one value , @xmath429 . the graph of @xmath430 is described by the curve @xmath431 where @xmath432 is defined by @xmath433 it is useful to write the determinant of @xmath385 as @xmath434.\ ] ] one immediately observes that the second term inside the parentheses is always strictly positive . if @xmath426 , then one has in addition that the first term is nonnegative for any @xmath435 . this proves the first part of the proposition . for the second part , let us first prove that there is exactly one value of @xmath436 for each @xmath437 that makes @xmath385 singular , and that this value lies in the interval @xmath438 . for fixed @xmath437 the determinant is strictly increasing in @xmath436 , and tends to @xmath439 as @xmath440 , and to @xmath441 as @xmath442 . hence it vanishes at exactly one value of @xmath436 , say @xmath443 . because the determinant is positive when @xmath444 , this value must necessarily lie in the interval @xmath438 . we now move to the parametrization of the graph of the map @xmath445 . one may note from that there is symmetry across the diagonal line @xmath446 which suggests making a change of variables . by letting @xmath447 we can write the determinant in the form @xmath448,\ ] ] which leads us to solve the quadratic equation @xmath449 for @xmath47 , given @xmath450 . doing this yields the parametrization , by using @xmath451 as the parameter ( some care has to be taken to ensure that one picks the right branches of the functions involved ) and going back to the original variables by inverting . [ remark : existence_when_det_theta_vanishes ] by employing the parametrization of the graph of @xmath452 provided by , one can show that the each column of @xmath385 is linearly independent from @xmath453 when @xmath425 . this implies that an argument similar to that of can be performed , by using the vortex strength @xmath454 as the bifurcation parameter , instead of @xmath46 . thus it is possible to show existence for any configuration when @xmath423 . an extension of this argument to @xmath455 is harder , because it requires the rank of @xmath385 to be @xmath456 . while the set of configurations that make @xmath457 vanish is hard to describe in general when @xmath455 , some observations can be made . of course , if @xmath458 , and as long as the derivative of @xmath457 with respect to the variable @xmath459 does not vanish at a point where @xmath460 , the zero set of @xmath457 is locally a smooth manifold of dimension @xmath456 around that point by the implicit function theorem . when @xmath423 , the zero set is actually the graph of a smooth function in @xmath461 by , and numerical evidence suggests that the zero set is the graph of a smooth function in @xmath462 when @xmath463 . actually checking that the derivative does not vanish is hard , but we have the following theorem : [ thm : point_vortex_configurations_measure_zero ] the subset of configurations of point vortices in @xmath464 such that @xmath385 is not invertible has measure zero . each entry in @xmath385 is analytic in each @xmath465 for @xmath466 fixed . it follows that @xmath457 also has this property , when viewed as a function @xmath467 we first verify that @xmath457 does not vanish identically on @xmath468 . to that end , fix @xmath469 and consider @xmath470 for @xmath471 . the purpose of the upper bound of @xmath472 is to make sure that @xmath473 is well defined for all @xmath474 . observe now that if we let @xmath475 , then @xmath476_{i , j } = \begin{cases } 1 & i = j\\ \tan(\pi\theta_i)\left(\cot\left(\pi\frac{\theta_i + \theta_j}{2}\right ) - \cot\left(\pi\frac{\theta_i - \theta_j}{2}\right ) \right ) & i \neq j \end{cases},\ ] ] where @xmath477 for @xmath478 . it follows that @xmath479 has a limit in @xmath480 as @xmath481 , and that this limit is @xmath482 where we have defined @xmath483 by @xmath484 in particular , @xmath485 is skew - symmetric , which implies that @xmath486 is invertible . since the set of invertible operators is open , so is the matrix @xmath479 for sufficiently small @xmath102 , which in turn means that @xmath385 is invertible for such @xmath102 . finally , the set @xmath468 is connected . hence , since we know that @xmath457 is analytic in each variable and does not vanish identically , we infer that the subset of @xmath468 on which @xmath457 vanishes has measure zero . in general we can not do better than , in the sense that for any @xmath487 there will always be a configuration of @xmath386 point vortices that makes @xmath457 vanish . [ prop : always_bad_configurations ] there are always configurations of point vortices in @xmath464 where @xmath385 is singular . the matrix appearing on the right - hand side of in the proof of has a positive determinant . indeed , the matrix @xmath485 defined in is skew - symmetric , so its spectrum is purely imaginary . moreover , since the matrix is real , the eigenvalues are either zero or appear in complex conjugate pairs . say that the first @xmath250 eigenvalues @xmath488 of @xmath485 are zero and that @xmath489 where the @xmath490 are real . then it follows that @xmath491 because the determinant of a matrix is equal to the product of its eigenvalues ( taking algebraic multiplicity into account ) . by we then have @xmath492 for small @xmath337 ( as in the proof of ) by continuity of the determinant . since all the tangents are also positive , this implies that @xmath493 for small @xmath337 . it remains to exhibit a configuration where @xmath494 . to that end , fix @xmath495 and consider @xmath496 for @xmath471 . proceeding as in the proof of we find @xmath497 where we have defined @xmath498 by @xmath499 this matrix is still skew - symmetric like @xmath485 , and so the right - hand side of has a positive determinant , as before . hence holds for small @xmath102 . however , now @xmath500 is negative and the rest of the tangents are positive , meaning that @xmath457 must be negative . in this section we give some explicit expressions for periodic waves with a point vortex on infinite depth , constructed in @xcite . we will adopt the notation and conventions used there . the fluid domain for the trivial surface is @xmath501 and the waves have period @xmath502 . the stream function for the rotational part is denoted by @xmath110 . we wish to find the stream function @xmath504 corresponding to equally spaced point vortices of unit strength at the points @xmath505 , and which is such that this stream function vanishes at the surface , @xmath506 . by symmetry , it must be the case that @xmath507 vanishes on @xmath508 . this leads us to the boundary value problem @xmath509 on @xmath510 . this equation can be dealt with using in . in order to apply we require a conformal map satisfying the requirements in the theorem statement . one may check ( see ( * ? ? ? * sections 7.1 and 7.2 ) ) that @xmath511 defines a bijective conformal map from the half strip @xmath512 onto the slit unit disk @xmath513 , and which is such that we now move to the surface profile . from @xcite we know that @xmath518 where @xmath236 is defined by @xmath519 written out , we have @xmath520 with the elementary antiderivative @xmath521 in particular , this means that @xmath522 so that reduces to @xmath523 in order to find the fourier series for @xmath515 , we require the fourier series of @xmath236 . we may write @xmath524,\ ] ] which , by expanding into geometric series , means that @xmath525 hence , by termwise differentiation , we obtain @xmath526 which , combined with , yields the result . one may note that @xmath527 as @xmath528 , which agrees with the speed of the solitary waves on infinite depth . when @xmath529 is large , the surface profile is very similar to the surface in the localized case , see . at the other extreme , the first terms in the fourier series will dominate . in this appendix , we provide two theorems that are used in order to get exact expressions for the rotational part of the stream function . except for the final part , is a standard result @xcite . is a less well known extension of . [ thm : greens_functions ] suppose that @xmath531 is a simply connected domain and that @xmath532 . furthermore , suppose that @xmath533 is a bijective conformal map onto the open unit disk , extending continuously to a function @xmath534 and satisfying @xmath535 . then the function @xmath536 defined by @xmath537 is in @xmath538 , extends continuously to the boundary of @xmath539 , and satisfies @xmath540 furthermore , the harmonic function @xmath541 defined by @xmath542 satisfies @xmath543 after identifying @xmath544 and @xmath127 via @xmath545 . we first check the boundary values of the function @xmath546 . by assumption , @xmath126 extends continuously to @xmath547 , and every point on @xmath547 must necessarily be mapped to the unit circle . it is thus immediate that @xmath546 also extends continuosly to the boundary , and moreover , vanishes there . identify now @xmath544 and @xmath127 . observe that since @xmath535 , we have @xmath548 for some holomorphic function @xmath549 , where @xmath550 . indeed , we must have @xmath551 because @xmath126 is injective , and the injectivity of @xmath126 also ensures that there can be no other roots . thus @xmath552 where @xmath553 is harmonic by @xmath550 and the cauchy - riemann equations . hence , by , the function @xmath546 is @xmath554 and satisfies @xmath555 [ thm : greens_functions_mixed ] suppose that @xmath531 is a simply connected domain and that @xmath532 . furthermore , assume that @xmath559 , where @xmath560 is @xmath561 and open in @xmath547 . finally , suppose that @xmath562\times \{0\ } ) } $ ] , where @xmath563 , is a bijective conformal map of @xmath539 onto the unit disk with a slit , satisfying @xmath535 and extending continuously to the boundary . this map should send @xmath564 to the unit circle and @xmath560 to the interval @xmath565 \times \{0\}$ ] , and should extend analytically across @xmath560 ( when viewed as a map on @xmath127 ) . then the function @xmath536 defined by @xmath566 is in @xmath538 , extends continuously to the boundary and satisfies @xmath567 where @xmath568 denotes the normal derivative .

in this paper , we study the existence and stability of travelling wave solutions of a kinetic reaction - transport equation . the model describes particles moving according to a velocity - jump process , and proliferating thanks to a reaction term of monostable type . the boundedness of the velocity set appears to be a necessary and sufficient condition for the existence of positive travelling waves . the minimal speed of propagation of waves is obtained from an explicit dispersion relation . we construct the waves using a technique of sub- and super- solutions and prove their stability in a weighted space . in case of an unbounded velocity set , we prove a superlinear spreading and give partial results concerning the rate of spreading associated to particular initial data . it appears that the rate of spreading depends strongly on the decay at infinity of the stationary maxwellian . * key - words : * kinetic equations , travelling waves , dispersion relation , superlinear spreading .

1307.8325

we address the issue of front propagation in a reaction - transport equation of kinetic type , @xmath1 here , the density @xmath2 describes a population of individuals in a continuum setting , and @xmath3 is the macroscopic density . the subset @xmath4 is the set of all possible velocities . individuals move following a velocity - jump process : they run with speed @xmath5 , and change velocity at rate 1 . they instantaneously choose a new velocity following the probability distribution @xmath6 . unless otherwise stated , we assume in this paper that @xmath7 is symmetric and @xmath8 satisfies the following properties : @xmath9 , and @xmath10 in addition , individuals are able to reproduce , with rate @xmath11 . new individuals start with a random velocity chosen with the same probability distribution @xmath6 . we could have chosen a different distribution without changing the main results , but we do not for the sake of clarity of the presentation . finally , we include a quadratic saturation term , which accounts for local competition between individuals , regardless of their speed . the main motivation for this work comes from the study of pulse waves in bacterial colonies of _ escherichia coli _ @xcite . kinetic models have been proposed to describe the run - and - tumble motion of individual bacteria at the mesoscopic scale @xcite . several works have been dedicated to derive macroscopic equations from those kinetic models in the diffusion limit @xcite . recently it has been shown that for some set of experiments , the diffusion approximation is not valid , so one has to stick to the kinetic equation at the mesoscopic scale to carefully compare with data @xcite . there is one major difference between this motivation and model . pulse waves in bacterial colonies of _ e. coli _ are mainly driven by chemotaxis which create macroscopic fluxes . growth of the population can be merely ignored in such models . in model however , growth and dispersion are the main reasons for front propagation , and there is no macroscopic flux due to the velocity - jump process since the distribution @xmath8 satisfies @xmath12 . for the sake of applications , we also refer to the growth and branching of the plant pathogen _ phytophthora _ by mean of a reaction - transport equation similar to @xcite . there is a strong link between and the classical fisher - kpp equation @xcite . in case of a suitable balance between scattering and growth ( more scattering than growth ) , we can perform the parabolic rescaling @xmath13 in , @xmath14 the diffusion limit yields @xmath15 , where @xmath16 is solution to the fisher - kpp equation ( see @xcite for example ) , @xmath17 we recall that for nonincreasing initial data decaying sufficiently fast at @xmath18 , the solution of ( [ kpp ] ) behaves asymptotically as a travelling front moving at the minimal speed @xmath19 @xcite . in addition , this front is stable in some weighted @xmath0 space @xcite . therefore it is natural to address the same questions for . we give below the definition of a travelling wave for equation . [ def : deftw ] we say that a function @xmath2 is a smooth travelling front solution of speed @xmath20 of equation if it can be written @xmath21 , where the profile @xmath22 satisfies @xmath23 in fact the profile @xmath24 is a solution to the stationary equation in a moving frame , @xmath25 the existence of travelling waves in reaction - transport equations has been adressed by schwetlick @xcite for a similar class of equations . first , the set @xmath7 is bounded and @xmath8 is the uniform distribution over @xmath7 . second , the nonlinearity can be chosen more generally ( either monostable as here , or bistable ) , but it depends only on the macroscopic density @xmath26 ( * ? ? ? ( 4 ) ) . for the monostable case , using a quite general method he was able to prove existence of travelling waves of speed @xmath27 for any @xmath28 , a result very similar to the fisher - kpp equation . we emphasize , that although the equations differ between schwetlick s work and ours , they coincide as far as the linearization in the regime of low density @xmath29 is concerned . on the contrary to schwetlick , we do not consider a general nonlinearity and we restrict to the logistic case , but we consider general velocity kernels @xmath30 . more recently , the rescaled equation ( [ eq : kinkpp2 ] ) has been investigated by cuesta , hittmeir and schmeiser @xcite in the parabolic regime @xmath31 . using a micro - macro decomposition , they construct possibly oscillatory travelling waves of speed @xmath32 for @xmath33 small enough ( depending on @xmath34 ) . in addition , when the set of admissible speeds @xmath7 is bounded , and @xmath35 they prove that the travelling wave constructed in this way is indeed nonnegative . lastly , when @xmath8 is the measure @xmath36 for some @xmath37 , equation ( [ eq : kinkpp ] ) is analogous to the reaction - telegraph equation for the macroscopic density @xmath26 ( up to a slight change in the nonlinearity however ) . this equation has been the subject of a large number of studies in the applied mathematics community @xcite . recently , the authors prove the existence of a minimal speed @xmath38 such that travelling waves exist for all speed @xmath39 @xcite . moreover these waves are stable in some @xmath0 weighted space , with a weight which differs from the classical exponential weight arising for the fisher - kpp equation . as the reaction - telegraph equation involves both parabolic and hyperbolic contributions , the smoothness of the wave depends on the balance between these contributions . in fact there is a transition between a parabolic ( smooth waves ) and a hyperbolic regime ( discontinuous waves ) , see remark [ rem : two velocities ] below . the authors also prove the existence of supersonic waves , having speed @xmath40 ( see remark [ rem : supersonic ] ) . the aim of the present paper is to investigate the existence and stability of travelling waves for equation ( [ eq : kinkpp ] ) for arbitrary kernels @xmath8 satisfying ( [ eq : hypm ] ) . for the existence part , we shall use the method of sub- and supersolutions , which do not rely on a perturbation argument . the stability part relies on the derivation of a suitable weight from which we can build a lyapunov functional for the linearized version of . the crucial assumption for the existence of travelling waves is the boundedness of @xmath7 . we prove in fact that in the case @xmath41 there exists no ( positive ) travelling wave . we then investigate the spreading rate for some particular choices of @xmath8 ( gaussian distribution , cauchy s distribution ) . unfortunately we are only able to give partial answer to this last question . in the last stage of writting of this paper , we realized that this issue was already addressed by mndez , campos and gmez - portillo for a slightly different equation admitting the same linearization near the front edge @xcite . our results are in agreement with their predictions . [ existence - tw ] assume that the set @xmath7 is compact , and that @xmath42 satisfies . let @xmath43 . there exists a speed @xmath44 such that there exists a travelling wave @xmath24 solution of of speed @xmath27 for all @xmath45 . the travelling wave is nonincreasing with respect to the space variable : @xmath46 . moreover , if @xmath47 then there exists no travelling wave of speed @xmath48 . the minimal speed @xmath38 is given through the following implicit dispersion relation : for each @xmath49 there is a unique @xmath50 such that @xmath51 then we have the formula @xmath52 [ rem : two velocities ] in the special case of two possible velocities only @xcite , corresponding to @xmath53 , two regimes have to be distinguished , namely @xmath54 and @xmath55 . in the case @xmath56 the travelling wave with minimal speed vanishes on a half - line . there , the speed of the wave is not characterized by the linearized problem for @xmath57 . note that this case is not contained in the statement of theorem [ existence - tw ] since it is assumed that @xmath42 . this makes a clear difference between the case of integrable @xmath8 and the case of a measure with atoms . [ rem : supersonic ] we expect that travelling waves exist for any @xmath39 , although this seems to contradict the finite speed of propagation when @xmath58 . in fact _ supersonic _ waves corresponding to @xmath58 should be driven by growth mainly , as it is the case in a simplified model with only two speeds @xcite . a simple argument to support the existence of such waves consists in eliminating the transport part , and seeking waves driven by growth only , @xmath59 . integrating with respect to @xmath60 yields a logistic equation for @xmath61 , @xmath62 , which as a solution connecting @xmath63 and @xmath64 for any positive @xmath27 . however these waves are purely artificial and we do not address this issue further . we now define @xmath65 and investigate the dependence of the minimal speed with respect to the velocity kernel @xmath66 . in the following proposition , we give some general bounds on the minimal speed . [ prop : bound on c * ] assume that @xmath7 is symmetric and that @xmath67 for all @xmath5 . the minimal speed satisfies the following properties , 1 . _ [ scaling ] _ for @xmath68 , define @xmath69 , then @xmath70 2 . _ [ rearrangement ] _ denote @xmath71 the schwarz decreasing rearrangement of the function @xmath8 ( see @xcite for a definition of this notion ) and @xmath72 the schwarz increasing rearrangement of the density distribution @xmath8 , then @xmath73 3 . _ [ comparison ] _ if @xmath54 then @xmath74 on the other hand , if @xmath55 then @xmath75 4 . _ [ diffusion limit ] _ in the diffusion limit @xmath13 we recover the kpp speed of the wave , @xmath76 in the case of a bounded set of velocities , we prove that for suitable initial data @xmath77 , the front spreads asymptotically with speed @xmath38 , in a weak sense . [ prop : spreadingbounded ] assume that @xmath7 is bounded and that @xmath78 . let @xmath79 such that @xmath80 for all @xmath81 . let @xmath82 be the solution of the cauchy problem . then 1 . if there exists @xmath83 such that @xmath84 for all @xmath85 and @xmath5 , then for all @xmath86 , @xmath87 2 . if there exists @xmath88 and @xmath89 such that @xmath90 for all @xmath91 and @xmath5 , then for all @xmath92 , @xmath93 where @xmath38 is the minimal speed of existence of travelling waves given by theorem [ existence - tw ] we also establish linear and nonlinear stability in suitable weighted @xmath0 spaces . the keypoint is to derive a correct weight which enables to build a lyapunov functional for the linear problem . we construct a semi - explicit weight @xmath94 , but we believe it is not the optimal one in some sense ( see remark [ rk : phi not optimal ] ) . let @xmath24 be a travelling wave of speed @xmath27 , and let @xmath95 the perturbation of @xmath24 in the moving frame . neglecting the nonlinear contributions , we are led to investigate the linear equation @xmath96 [ linstability ] there exists a weight @xmath97 such that the travelling front of speed @xmath98 is linearly stable in the weighted space @xmath99 . more precisely , the following lyapunov identity holds true for any solution @xmath100 of the linear equation , @xmath101 the weight @xmath97 is explicitly given in definition [ eq : def weight phi ] . using a comparison argument , in the spirit of @xcite , together with the explicit formula of the dissipation for the linearized system , we prove a nonlinear stability result . [ nonlinstability ] let @xmath102 $ ] . let @xmath103 be a travelling wave with speed @xmath98 . let @xmath82 be a solution of . suppose that the initial data satisfies @xmath104 then the following lyapunov identity holds true for the perturbation @xmath105 , @xmath106 \left\vert u \right\vert^2 e^{- 2 \phi ( z , v ) } d z dv \leq 0\,,\ ] ] where @xmath97 denotes the same weight as in proposition [ linstability ] . we expect that nonlinear stability holds true for any @xmath107 $ ] . however this would require to redefine the weight @xmath97 , since we believe it is not the optimal one . boundedness of @xmath7 is a crucial hypothesis in order to build the travelling waves . we believe that it is a necessary and sufficient condition . we make a first step to support this conjecture by investigating the case @xmath41 . we first prove infinite speed of spreading of the front under the natural assumption @xmath108 . as a corollary there can not exist travelling wave in the sense of definition [ def : deftw ] . note that there exist travelling waves with less restrictive conditions than definition [ def : deftw ] , at least in the diffusive regime @xcite . these fronts are expected to oscillate as @xmath109 . we expect that such oscillating fronts do exist far from the diffusive regime . in the case where @xmath41 and @xmath8 is gaussian , we plotted the dispersion relation in the complex plane @xmath110 , for an arbitrary given @xmath111 . we observed that it selects two complex conjugate roots , supporting the fact that oscillating fronts should exist ( results not shown ) . [ prop : spreadingunbounded ] assume that @xmath112 for all @xmath113 . let @xmath79 such that @xmath80 for all @xmath81 and there exists @xmath88 and @xmath114 such that @xmath90 for all @xmath91 and @xmath5 . let @xmath82 be the solution of the cauchy problem . then for all @xmath111 , @xmath115 we can immediately deduce from this result the non - existence of travelling waves when @xmath116 , by taking such a travelling wave as an initial datum @xmath117 in order to reach a contradiction . assume that @xmath118 for all @xmath113 . then equation does not admit any travelling wave solution . next we investigate specific choices for the distribution @xmath8 , both numerically and theoretically . in the case of a gaussian distribution , we expect a spreading rate following the power law @xmath119 ( see also @xcite ) . to support this guess , we prove in fact that spreading occurs _ at most _ with this rate . for this purpose we build a supersolution which is spreading with this rate . this issue has been addressed by mndez , campos and gmez - portillo @xcite in a physical paper for a slightly different equation , where the nonlinearity lies in the diffusion kernel instead of the growth rate . they conjectured that , as for the kpp equation , the front speed is determined through the linearization of the equation near the unstable steady state @xmath64 . we believe that the linearization should give the power law of the propagation , but it is not clear to us that it will give the exact location of the transition . however , it turns out that the linearized equations in @xcite and in the present paper are the same . then , performing some fourier - laplace transform of the solution , mendez et al derived heuristically the power law of the propagation . in particular , for a gaussian kernel , they found out that the spreading rate is @xmath119 . [ thm : unbounded ] let @xmath120 for all @xmath113 . let @xmath79 such that @xmath80 for all @xmath81 . assume that there exists @xmath121 such that @xmath122 let @xmath82 be the solution of the cauchy problem . then for all @xmath123 , one has @xmath124 in the case of a cauchy distribution @xmath125 , we obtain faster spreading rate under similar assumptions ( see proposition [ supersolunbounded - cauchy ] ) , namely for all @xmath123 , one has @xmath126 the phenomenon of accelerating fronts have raised a lot of attention in the literature of reaction - diffusion equations . this phenomenon occurs for the fisher - kpp equation when the initial datum decays more slowly than any exponential @xcite ; for a variant of the fisher - kpp equation where the diffusion operator is replaced by a nonlocal dispersal operator with fat tails @xcite , or by a nonlocal fractional diffusion operator @xcite . recently , accelerating fronts have been conjectured to occur in a reaction - diffusion - mutation model which extends the fisher - kpp equation to a population with heterogeneous diffusion coefficient @xcite . there is some subtlety hidden behind this phenomenon of infinite speed of spreading . in fact the diffusion limit of the scattering equation ( namely @xmath127 ) towards the heat equation makes no difference between bounded or unbounded velocity sets ( see @xcite and the references therein ) . however very low densities behave quite differently , which can be measured in the setting of large deviations or wkb limit . this can be observed even in the case of a bounded velocity set . in @xcite the large deviation ( wkb ) limit of the scattering equation is performed . it differs largely from the classical eikonal equation obtained from the heat equation . the case of unbounded velocities is even more complicated @xcite . to conclude , let us emphasize that low densities are the one that drive the front here ( pulled front ) . so the diffusion limit is irrelevant in the case of unbounded velocities , since very low density of particles having very large speed makes a big difference . we first recall some useful results concerning the cauchy problem associated with : well - posedness and a strong maximum principle . these statements extend some results given in @xcite . they do not rely on the boundedness of @xmath7 . [ global existence : theorem 4 in @xcite ] [ cauchypbm ] let @xmath117 a measurable function such that @xmath128 for all @xmath81 . then the cauchy problem has a unique solution @xmath129 in the sense of distributions , satisfying @xmath130 the next result refines the comparison principle of @xcite in order to extend it to sub and supersolutions in the sense of distributions and to state a strong maximum principle . its proof is given in appendix . [ prop - mp ] assume that @xmath131 are respectively a super- and a subsolution of , _ i.e. _ @xmath132 in the sense of distributions . assume in addition that @xmath133 satisfies @xmath134 for all @xmath135 . then @xmath136 for all @xmath137 . assume in addition that @xmath7 is an interval , and that @xmath78 . if there exists @xmath138 such that @xmath139 , then one has @xmath140 for all @xmath141 such that @xmath142 . if @xmath116 , then this statement reads as in the parabolic framework : if @xmath143 and @xmath144 at @xmath145 , then @xmath146 for all @xmath147 . in the case @xmath148 $ ] we have to take into account finite speed of propagation , obviously . we assume throughout this section that @xmath149 is compact . we construct the travelling waves for @xmath150 . the proof is divided into several steps . it is based on a sub and supersolutions method . the aim of this first step is to solve the linearized equation of at @xmath151 , in the regime of low density @xmath57 . such an achievement gives information about the speed and the space decreasing rate of a travelling wave solution of the nonlinear problem , as for the fisher - kpp equation . the linear problem for writes @xmath152 we seek a solution having exponential decay at @xmath153 . more specifically we separate the variables in our ansatz : @xmath154 , with @xmath155 . the next proposition gathers the results concerning the linear problem . [ propdisp ] there exists a minimal speed @xmath38 such that for all @xmath156 , there exists @xmath157 such that @xmath158 is a nonnegative solution of . the profile @xmath159 is explicitely given by @xmath160 the admissible @xmath161 are solutions of the following dispersion relation , @xmath162 moreover , among all possible @xmath163 for a given @xmath27 , the minimal one @xmath164 is well defined and isolated . here appears the crucial assumption on the boundedness of @xmath7 . if this condition is not fulfilled , it is never possible to ensure that the profile @xmath159 is nonnegative since the denominator is linear with respect to @xmath60 . * * plugging the ansatz @xmath165 into yields @xmath166 the profile is given by @xmath167 the dispersion relation reads @xmath168 , or equivalently . moreover , we require the profile @xmath169 to be nonnegative , which gives the condition @xmath170 for all @xmath5 , which implies @xmath171 . from now on , we focus on the existence of solutions @xmath161 of , with @xmath172 and @xmath173 . let us denote @xmath174 so that we look for solutions of @xmath175 . * # step 2 . * technically speaking , for all @xmath172 , the function @xmath176 is analytic over @xmath177 indeed , as @xmath178 is integrable for all @xmath179 , it is clear that @xmath180 is the analytic development of @xmath181 for @xmath182 . next we observe that @xmath183 is decreasing for all @xmath184 , and that @xmath185 is strictly convex . moreover , the function @xmath181 satisfies the following properties : @xmath186 { } 0 \ , .\end{aligned}\ ] ] the last property relies on the lebesgue s dominated convergence theorem since @xmath187 . * # step 3 . * assume first that @xmath188 . then fatou s lemma gives @xmath189 as a consequence , @xmath190 is well defined and finite for all @xmath172 . it follows from the earlier properties that @xmath191 and @xmath192 . moreover , the regularity and monotonicity properties of @xmath181 guarantee that @xmath193 is continuous and decreasing . hence , there exists @xmath38 such that @xmath194 and there exists @xmath195 such that @xmath196 . next , for all @xmath197 , as @xmath183 is decreasing , one has @xmath198 for all @xmath86 . thus , as @xmath199 , there exists @xmath163 such that @xmath200 for all @xmath86 . second , consider a general @xmath42 possibly vanishing at @xmath201 . to recover the first step , we define for @xmath202 a new maxwellian by @xmath203 over @xmath7 ( and @xmath64 outside of @xmath7 ) , where @xmath204 is the measure of @xmath7 . then @xmath205 since @xmath206 , and thus the earlier step yields that there exists a sequence @xmath207 of minimal speeds associated with @xmath208 . we also associate @xmath209 with @xmath210 through ( [ eq - defi ] ) . we define @xmath211 and we now show that it is the minimal speed . * take @xmath92 . then for all @xmath212 and for some arbitrarily large @xmath179 so that @xmath213 , one has @xmath214 because @xmath215 as @xmath216 , we get @xmath217 thus @xmath200 has no solution for @xmath218 if @xmath92 . * assume that @xmath219 . then one has @xmath220 when @xmath179 is large enough and thus for all @xmath179 sufficiently large , there exists @xmath221 such that @xmath222 . up to extraction , one may assume that @xmath223 converges to some @xmath224 $ ] . fatou s lemma yields @xmath225 . hence , there exists a solution @xmath226 $ ] of @xmath200 and obviously @xmath227 since @xmath199 . * lastly , if @xmath228 , we know that for all @xmath229 , there exists @xmath230 $ ] such that @xmath231 . assuming that @xmath232 $ ] as @xmath233 , we get @xmath234 . [ minlambda ] for all @xmath235 , the quantity @xmath236 is well - defined . moreover , for all @xmath237 , if @xmath238 is small enough , then @xmath239 . we know from the definition of @xmath38 that for all @xmath150 , the set @xmath240 is not empty . thus , we can take a minimizing sequence @xmath241 which converges towards the infimum of @xmath242 . as this sequence is bounded , one can assume , up to extraction , that @xmath243 . then lebesgue s dominated convergence theorem gives @xmath244 . hence @xmath245 . next , we have already noticed in the proof of proposition [ propdisp ] that @xmath246 for all @xmath86 . as @xmath247 , the definition of @xmath164 yields @xmath248 . the conclusion follows from the strict convexity of the function @xmath185 . in this step we construct sub and supersolutions for . we fix @xmath250 and we denote @xmath251 without ambiguity . [ lem - super ] let @xmath252 then @xmath253 is a supersolution of , that is , it satisfies in the sense of distributions : @xmath254 [ lem - sub ] there exist @xmath255 and @xmath238 such that if @xmath256 then @xmath257 is a subsolution of , that is satisfies in the sense of distributions : @xmath258 first , @xmath259 and @xmath260 both clearly satisfy since @xmath261 . next , as @xmath253 is continuous , it immediately follows from the jump formula that , as a minimum of two supersolutions , it is a supersolution of ( [ eq - super ] ) in the sense of distributions . the same arguments as in the proof of lemma [ lem - super ] yield that it is enough to prove that ( [ eq - subper ] ) is satisfied by @xmath257 over the open set @xmath262 . as @xmath86 , proposition [ propdisp ] gives @xmath263 small enough such that @xmath264 and @xmath265 . we compute the linear part : @xmath266 to prove the lemma , we now have to choose a relevant @xmath267 such that @xmath268 holds for all @xmath269 . as @xmath270 for all @xmath271 , one has @xmath272 and thus it is enough to choose @xmath267 such that @xmath273 on the other hand for all @xmath271 such that @xmath274 , we have @xmath275 , meaning that @xmath276 plugging this estimate into , it is enough to choose @xmath267 such that @xmath277 this concludes the proof since such a @xmath267 obviously exists . let @xmath279 , where @xmath38 denotes the minimal speed of proposition [ propdisp ] . in order to prove the existence of travelling waves , we will prove that the solution of the following evolution equation , corresponding to equation in the moving frame at speed @xmath27 , converges to a travelling wave as @xmath280 : @xmath281 the well - posedness of equation immediately follows from proposition [ cauchypbm ] . let now derive some properties of the function @xmath82 from proposition [ prop - mp ] . [ lem - sandwich ] for all @xmath282 , one has @xmath283 as @xmath257 is a subsolution of ( [ eq - approx ] ) and @xmath253 is a supersolution of ( [ eq - approx ] ) , with @xmath284 for all @xmath285 , this result is an immediate corollary of proposition [ prop - mp ] . [ lem - monotonicity ] for all @xmath286 , the function @xmath287 is nonincreasing . take @xmath288 and define @xmath289 . then as @xmath253 is nonincreasing in @xmath290 , one has @xmath291 for all @xmath292 . proposition [ prop - mp ] yields that @xmath293 for all @xmath135 . [ lem - timedec ] for all @xmath294 , the function @xmath295 is nonincreasing . take @xmath296 and define @xmath297 . then lemma [ lem - sandwich ] yields that @xmath298 for all @xmath292 . hence , proposition [ prop - mp ] gives @xmath299 for all @xmath137 . [ lem - lipsch ] the family @xmath300 is uniformly continuous with respect to @xmath81 . moreover , for any @xmath301 , the continuity constants does not depend on @xmath302 . we begin with the space regularity . let @xmath303 . the function @xmath304 is such that @xmath305 is lipschitz - continuous with respect to @xmath290 . therefore there exists a constant @xmath306 such that for all @xmath81 , we have @xmath307 as @xmath308 , it is easily checked that @xmath309 is a supersolution of . hence proposition [ prop - mp ] yields that @xmath310 hence the function @xmath311 is lipschitz continuous with respect to @xmath290 . since the function @xmath311 is bounded from above , @xmath312 is also lipschitz continuous with respect to @xmath290 . the lipschitz constant is uniform with respect to @xmath313 and @xmath314 . we now come to the velocity regularity . for the sake of clarity we first consider the case where @xmath8 is @xmath315 on @xmath7 . the function @xmath316 is @xmath315 too . we introduce @xmath317 . it satisfies the following equation @xmath318 multiplying the equation by @xmath319 we obtain @xmath320 the l.h.s . is linear with respect to @xmath321 and satisfies the maximum principle . the r.h.s . is uniformly bounded since @xmath322 and @xmath82 is uniformly lipschitz with respect to @xmath290 . obviously the constant @xmath323 is a supersolution . we deduce that @xmath324 is uniformly bounded over @xmath325 . in the case where @xmath8 is only continuous over the compact set @xmath7 , thus uniformly continuous , we shall use the method of translations again . however we have to be careful since @xmath7 is bounded . let @xmath326 . we introduce @xmath327 . the function @xmath328 satisfies the following equation @xmath329 let @xmath330 . there exists @xmath331 such that for @xmath332 we have @xmath333 and @xmath334 . moreover we have obviously @xmath335 . we conclude using the same argument as in the @xmath315 case . the modulus of uniform continuity is uniform with respect to @xmath313 and @xmath314 . we are now in position to prove the existence of travelling waves of speed @xmath27 , except for the minimal speed @xmath38 . gathering lemmas [ lem - sandwich ] , [ lem - monotonicity ] and [ lem - timedec ] , we know that @xmath336 is well - defined for all @xmath292 , that @xmath337 is nonincreasing in @xmath290 for all @xmath60 and that @xmath338 . let now prove that @xmath24 defines a travelling wave solution of . as @xmath82 satisfies , converges pointwise and is bounded by the locally integrable function @xmath253 , it follows from lebesgue s dominated convergence theorem that @xmath24 satisfies in the sense of distributions . moreover , lemma [ lem - lipsch ] ensures that @xmath24 is continuous with respect to @xmath339 . we next check the limits towards infinity . let @xmath340 . thanks to @xmath341 , the lebesgue dominated convergence theorem gives @xmath342 . on the other hand , we get @xmath343 in the sense of distributions . integrating over the compact set @xmath7 , we deduce that @xmath344 , _ i.e. _ that @xmath345 or @xmath63 . as @xmath24 is nonincreasing and @xmath338 , one necessarily has @xmath346 and @xmath347 . finally , ( [ eq - limit ] ) gives @xmath348 and @xmath349 for all @xmath5 . this gives the appropriate limits . consider a decreasing sequence @xmath350 converging towards @xmath38 . we already know that for all @xmath179 , equation admits a travelling wave solution @xmath351 , with @xmath352 and @xmath353 , and @xmath354 is nonincreasing . up to translation , we can assume that @xmath355 . moreover , lemma [ lem - lipsch ] ensures that the functions @xmath356 are uniformly continuous with respect to @xmath81 since the continuity stated in lemma [ lem - lipsch ] is uniform with respect to @xmath302 for any @xmath301 . thanks to the ascoli theorem and a diagonal extraction process , we can assume that the sequence @xmath356 converges locally uniformly in @xmath81 to a function @xmath24 . clearly @xmath24 satisfies in the sense of distributions . moreover , as @xmath24 is nonincreasing with respect to @xmath290 , one could recover the appropriate limits at infinity with the same arguments as in the proof of the existence of travelling waves with speeds @xmath86 . [ lem : posharnack ] assume that @xmath357 . for all @xmath358 there exists @xmath359 and a nonnegative , arbitrarily small , compactly supported function @xmath360 which is a subsolution of @xmath361 for the sake of clarity we emphasize the dependence of the function @xmath181 upon the growth rate @xmath11 : @xmath362 we denote by @xmath363 the smallest speed such that there exists a solution @xmath49 of @xmath364 ( see proposition [ propdisp ] ) . let @xmath365 . by continuity we can choose @xmath366 so small that @xmath367 . we claim that there exists @xmath368 such that @xmath369 , with @xmath370 and @xmath371 . indeed we know from the proof of proposition [ propdisp ] [ step 3 ] that @xmath372 under the assumption @xmath373 . using a continuity argument we also have the strict inequality @xmath374 , uniformly with respect to @xmath366 . the complex function @xmath375 is analytic in a neighborhood of @xmath376 . hence , the rouch theorem yields that there exists @xmath377 such that the equation @xmath378 has a solution @xmath379 with @xmath380 arbitrarily close to @xmath376 . we denote by @xmath381 the corresponding velocity profile , @xmath382 by continuity we can choose @xmath383 and @xmath380 such that @xmath384 holds for all @xmath5 . by the very definition of @xmath385 , we have @xmath386 . we denote @xmath387 . recall that we have the strict inequality @xmath374 , uniformly with respect to @xmath366 . using a continuity argument we can impose that @xmath388 . now define the real function @xmath389 by @xmath390,\ ] ] one has @xmath391 and @xmath392 for all @xmath5 . thus , there exists an interval @xmath393\subset { \mathbb{r}}$ ] and a bounded domain @xmath394\times v$ ] such that : @xmath395 on the other hand , as @xmath388 , there exists a constant @xmath396 such that @xmath397 hence , one can choose @xmath398 small enough such that @xmath399 for all @xmath400 we deduce from @xmath401 the following identities , @xmath402 taking real part on both sides , we get for @xmath403 , @xmath404 hence @xmath405 is a subsolution of for all @xmath400 on @xmath406 . we deduce that the truncated function @xmath407 is a subsolution of over @xmath408 . assume that @xmath409 is a travelling wave solution of of speed @xmath410 . according to lemma [ lem : posharnack ] , there exists @xmath411 and a nonnegative compactly supported subsolution @xmath412 of with speed @xmath383 . as @xmath24 is positive and continuous , we can decrease @xmath412 so as to obtain @xmath413 . let @xmath414 and @xmath415 . these two functions are respectively a solution and a subsolution of . as @xmath416 for all @xmath81 , proposition [ prop - mp ] implies @xmath417 taking @xmath418 and letting @xmath280 , we get @xmath419 this is a contradiction . let @xmath86 . consider first the initial datum @xmath420 and let @xmath421 the solution of the cauchy problem . denote by @xmath24 a travelling wave of minimal speed @xmath38 . there exists @xmath422 such that @xmath423 for all @xmath81 . it is straightforward to check that @xmath424 is a supersolution of . hence , the comparison principle of proposition [ prop - mp ] implies that @xmath425 for all @xmath141 . in particular we have , @xmath426 as @xmath427 for all @xmath5 and @xmath86 , we get @xmath428 for all @xmath5 . on the other hand , as @xmath429 is nonincreasing with respect to @xmath430 it follows from the comparison principle that @xmath431 is nonincreasing ( see lemma [ lem - monotonicity ] ) . thus @xmath432 for all @xmath433 and the conclusion follows . for a general initial datum @xmath117 satisfying the hypotheses of proposition [ prop : spreadingbounded ] , one has @xmath434 for all @xmath81 and thus @xmath435 for all @xmath141 , from which the conclusion follows . let @xmath92 . the same arguments as in the first step yield that we can assume that @xmath436 let @xmath412 a compactly supported subsolution of associated with a speed @xmath437 . since @xmath412 can be chosen arbitrarily small , up to translation of @xmath412 , we can always assume that @xmath438 . let @xmath133 the solution of the cauchy problem associated with the initial datum @xmath439 . the comparison principle yields @xmath440 for all @xmath441 . let @xmath442 . this function satisfies @xmath443 clearly @xmath412 is a ( stationary ) subsolution of this equation . the comparison principle proposition [ prop - mp ] yields that @xmath444 is nondecreasing for all @xmath81 ( see also lemma [ lem - timedec ] for a similar proof ) . let @xmath445 . this function is clearly a weak solution of @xmath446 moreover , we have @xmath447 and @xmath448 . [ lem : sliding ] we have @xmath449 . * * first we prove that @xmath450 is positive everywhere . take @xmath451 such that @xmath452 . as @xmath453 satisfies , proposition [ prop - mp ] yields @xmath454 for all @xmath141 such that @xmath455 . as @xmath456 , for all @xmath294 one can take @xmath147 large enough so that @xmath457 . therefore @xmath458 . we thus conclude that @xmath450 is positive over @xmath408 . * # step 2 . * next we prove that @xmath459 . let @xmath460 . define @xmath461 , and @xmath462 as @xmath463 is compactly supported and @xmath450 is positive over @xmath408 and continuous , we have @xmath464 when @xmath465 is small enough . therefore @xmath466 . we argue by contradiction . assume that @xmath467 . the definition of @xmath468 yields that @xmath469 and that @xmath470 . as @xmath463 is compactly supported , this infimum is indeed reached : there exists @xmath471 such that @xmath472 . assume that @xmath473 and take @xmath474 such that @xmath475 . we introduce @xmath476 and @xmath477 . as @xmath478 , proposition [ prop - mp ] gives @xmath479 for all @xmath141 such that @xmath480 , that is : @xmath481 as @xmath482 , for all @xmath430 , one can take @xmath147 large enough so that @xmath483 , leading to @xmath484 for all @xmath81 , a contradiction since equality holds at @xmath485 . hence , @xmath486 , which is also a contradiction since @xmath450 is positive while @xmath463 is compactly supported . we conclude that @xmath487 , namely @xmath488 . evaluating this inequality at @xmath489 , one gets @xmath490 for all @xmath491 . as @xmath492 under the assumption @xmath78 , we have proved in fact that @xmath493 * # step 3 . * as @xmath494 , we can define @xmath495 we know from the previous step that this quantity is positive . if @xmath496 , then the same types of arguments as in step 2 lead to a contradiction . hence @xmath497 , meaning that @xmath498 . as @xmath499 , we conclude that @xmath500 . as a consequence of lemma [ lem : sliding ] we obtain @xmath501 this implies in particular that @xmath502 for all @xmath339 by a sandwiching argument . moreover , as @xmath117 is nonincreasing with respect to @xmath290 , @xmath503 is nonincreasing and thus @xmath504 for all @xmath505 , from which the conclusion follows since @xmath506 . recall that the dispersion relation giving the speed @xmath27 as a function of the exponential decay @xmath163 is @xmath507 , where @xmath508 is defined in . let introduce @xmath509 the function associated with the dilated velocity profile @xmath510 . the function @xmath509 clearly satisfies the scaling relation @xmath511 , therefore we get @xmath512 from the very definition of @xmath38 . we use the symmetry of the kernel @xmath513 to write @xmath514 let define @xmath515 it is an increasing function over @xmath516 , thus @xmath517 . the hardy - littlewood inequality ( * ? ? ? 3 ) yields @xmath518 the dispersion relation is nonincreasing with respect to @xmath27 . it follows immediately that @xmath519 we use the symmetry of the kernel @xmath513 . for @xmath49 the dispersion relation writes @xmath520 since the function @xmath521 is convex on its domain of definition , jensen s inequality yields @xmath522 we recognize the dispersion relation associated with the two - speed model @xcite . we deduce @xmath523 this second - order polynomial has a negative value at @xmath524 , therefore @xmath525 is necessarily greater than the vanishing value , @xmath526 minimizing the right - hand - side with respect to @xmath163 , we deduce that @xmath527 implies : @xmath528 on the other hand we clearly obtain from , @xmath529 by comparison of the relations , as in the proof of proposition [ prop : bound on c*](b ) , we deduce that the speed corresponding to a given maxwellian @xmath30 supported on @xmath530 is smaller than the speed corresponding to @xmath531 . this peculiar case is analysed in @xcite . the minimal speed in this case is @xmath532 the dispersion relation for the rescaled equation reads @xmath533 the previous result guarantees that @xmath534 is bounded from above for @xmath535 , @xmath536 for a given @xmath49 , we perform a taylor expansion of up to second order , @xmath537 uniformly for @xmath538 $ ] , since @xmath7 is bounded . therefore , solving the relation dispersion for the minimal speed boils down to solving @xmath539 we deduce @xmath540 therefore the minimal speed verifies @xmath541 . in this subsection , we focus on the linearized problem around some travelling wave moving at speed @xmath542 . we recall that we consider a solution @xmath100 of the equation associated with the linearization around a travelling wave : @xmath543 where the notation @xmath544 always stands in the sequel for a function of the @xmath545 variable . we shall prove stability of the wave in a suitable @xmath0 framework , inspired by @xcite . we search for an ansatz @xmath546 . the function @xmath547 satisfies : @xmath548 from , we shall derive the dissipation inequality stated in proposition . we test against @xmath547 to obtain the kinetic energy estimate : @xmath549 we aim at choosing a weight @xmath97 such that the dissipation is coercive in @xmath0 norm . let define the symmetric kernel @xmath550 as follows @xmath551 we seek a function @xmath97 such that @xmath552 for a suitable positive function @xmath267 , in the sense of kernel operators . for this purpose we focus on the eigenvalues of the kernel operator @xmath553 . [ eq : def weight phi ] we introduce @xmath554 the smallest solution of the following dispersion relation @xmath555 and we define @xmath556 through the differential equation @xmath557 finally we define @xmath558 recall that @xmath559 , so that the weight @xmath97 is well - defined as soon as @xmath560 is well - defined . a small argumentation is required to prove that @xmath561 is well - defined too . for a given @xmath27 and @xmath562 , define @xmath563 the function @xmath564 is continuous , and satisfies the following properties , @xmath565\,,\\ & g(\lambda ) = \int_{v } \frac { ( 1+r ) m(v ) - r f ( z , v ) } { 1 + \lambda ( c - v ) } dv = 1 - \int_{v } \frac{r f ( z , v ) } { 1 + \lambda ( c - v ) } dv \leq 1\,,\end{aligned}\ ] ] where @xmath163 is chosen such that @xmath507 . thus we can define the smallest @xmath566 $ ] such that @xmath567 . [ eigens ] let @xmath267 be defined as @xmath568 and @xmath569 be the operator associated with the symmetric kernel @xmath570 . the operator @xmath569 is nonpositive . we shall prove that @xmath64 is the perron eigenvalue of the operator @xmath569 . for that purpose we shall exhibit a positive eigenvector in the nullset of @xmath569 . the equation @xmath571 reads @xmath572 plugging the formula for @xmath573 into this expression we get , @xmath574 from the definitions - we have , @xmath575 the weight @xmath97 and the function @xmath267 are chosen such that @xmath576 therefore the equation @xmath577 is equivalent to @xmath578 where the macroscopic quantities @xmath579 and @xmath580 are defined as follows , @xmath581 to resolve this eigenvalue problem , we seek proper values for @xmath579 and @xmath580 . from we deduce that they are solution of a @xmath582 closed linear system , namely @xmath583 this system simplifies thanks to the choice of @xmath561 . indeed we have @xmath584 we are reduced to the following eigenvalue problem , @xmath585 clearly , @xmath586 is the unique solution up to multiplication . we obtain eventually that @xmath587 is given ( up to a multiplicative factor ) by @xmath588^{1/2}}{1 + \lambda ( z ) ( c - v ) } > 0 \,.\end{aligned}\ ] ] as a consequence , we have found that the symmetric operator @xmath569 , which is nonnegative out of the diagonal @xmath589 , possesses a positive eigenvector @xmath587 associated with the eigenvalue 0 . therefore it is a nonpositive operator . this ends the proof of the lemma . we can now conclude the proof of proposition [ linstability ] . lemma [ eigens ] claims that for all @xmath590 such that @xmath591 is solution to the linearized equation , we have @xmath592 which proves the proposition . [ rk : phi not optimal ] we believe that the weight @xmath593 proposed in definition [ eq : def weight phi ] is not optimal with respect to the spectral property of the linearized operator . indeed the dissipation factor @xmath594 is equivalent in the diffusion limit ( @xmath595 ) to @xmath596 , although we expect @xmath597 @xcite . the missing factor @xmath598 is responsible for the restriction @xmath599 in our nonlinear stability result , theorem [ nonlinstability ] . let us recall how to derive the spectral properties of the linearized equation in the diffusive limit , namely the linearized fisher - kpp equation , @xmath600 where @xmath601 is the profile of the travelling wave in the frame @xmath602 . applying the same procedure as in the proof of proposition [ linstability ] , we shall derive an equation for the weighted perturbation @xmath603 , and optimize the dissipation with respect to the weight @xmath97 ( see also @xcite ) , as follows @xmath604 the best choice is achieved when @xmath605 is constant and minimizes @xmath606 , _ i.e. @xmath607_. in the case of the minimal speed @xmath608 , we obtain the following dissipation formula for the linearized operator , @xmath609 notice the factor @xmath598 which is apparently missing in the dissipation term . a systematic way to find the correct weight is to derive the eigenvectors of the operator and its dual , then to use the framework of relative entropy ( see @xcite for a general presentation ) . this was done by kirchgssner @xcite who derived the so - called _ eichform _ for . the linearized operator @xmath610 possesses obviously the nonpositive eigenvector @xmath611 , @xmath612 . the dual operator @xmath613 possesses the nonpositive eigenvector @xmath614 , @xmath615 , as can be checked by direct calculation . therefore the relative entropy identity for the convex function @xmath616 writes for the linearized system as follows , @xmath617 which is equivalent to after straightforward computation ( recall @xmath618 ) . a similar strategy could be performed here : the linearized operator @xmath619 possesses the nonpositive eigenvector @xmath620 , @xmath612 ( recall @xmath621 is nonincreasing ) . to derive the corresponding relative entropy identity , we should find an eigenvector @xmath622 in the nullset of the dual operator @xmath623 existence of such an eigenvector would follow from the krein - rutman theorem . however we were not able to find an explicit formulation of @xmath622 , and thus of the dissipation , which is necessary to derive a quantitative nonlinear stability estimate such as proposition [ nonlinstability ] . this is the reason why we stick to the weight proposed in definition [ eq : def weight phi ] although we believe it is not the optimal one . first , the comparison principle of proposition [ prop - mp ] and yield @xmath624 now , we write the nonlinear equation verified by the weighted perturbation @xmath603 , @xmath625 and as for the linear stability problem we test against @xmath547 : @xmath626 using we deduce @xmath627 this last equation is very similar to . following the same steps as in the proof of proposition [ linstability ] , we find that using the same weight @xmath97 and setting @xmath628 we obtain the estimate . in this section , we provide some numerical simulations to illustrate our results , and to introduce the last section about accelerating fronts . we used a very basic explicit numerical scheme for approximating . the transport part is approximated by a upwind scheme . we show in figure [ fig : vborne ] the expected asymptotic behavior when the velocity space is bounded , namely convergence towards a travelling front with minimal speed . next we have investigated the case of @xmath41 . of course , numerical simulations require that the support of @xmath8 is truncated . we opted for the following strategy : we truncated the velocity set to a finite interval @xmath629 $ ] and we renormalized the distribution @xmath8 . for any @xmath255 we observed the asymptotic regime of a travelling front with finite speed , as expected . however , as @xmath630 , we observed no stabilization of the asymptotic spreading speed . in fact , we observed that the envelope of the spreading speed scales approximately as @xmath631 . hence the front is accelerating with the approximate power law @xmath119 . we expect that the scaling strongly depends on the decay of @xmath8 as @xmath632 . [ fig : vnonborne ] we assume in this section that @xmath116 and that @xmath112 for all @xmath633 . we prove superlinear spreading . we deduce as a corollary that there can not exist a travelling wave solution of . we also give some quantitative features about the spreading of the density in two cases : the case where @xmath8 is a gaussian , and the case where it is a cauchy s law . we construct explicit supersolutions from which we estimate the spreading from above . we expect those estimates to be sharp . in fact they are in accordance with numerical simulations . as we are not able to construct suitable subsolutions , we leave it as an open problem to prove the exact spreading rate , at least for these two specific cases ( gaussian and cauchy s law ) . before we go to the proof , let us give some heuristics concerning the superlinear spreading rate . reaction - diffusion fronts with kpp nonlinearity are _ pulled fronts _ : the spreading rate is determined by the dynamics of small populations at the far edge of the front . in the kinetic model with unbounded velocities , individuals with arbitrary large speeds go at the far edge of the front . no matter their low density , they yield exponential growth of the population and pull the accelerating front . of course we expect the acceleration to depend on the specific tail of the distribution @xmath30 . actually we conjecture that the spreading rates for a gaussian and a cauchy s law scales respectively as @xmath634 , and as @xmath635 . last , let us emphasize that the diffusive limit of leads to the classical fisher - kpp equation , under the assumption that @xmath8 has some finite moments which is obviously the case for a gaussian . the fisher - kpp equation exhibits linear spreading whereas the kinetic equation may exhibit superlinear spreading . the point is that small populations does not show the same scaling at the far edge of the front . to summarize we shall say that asymptotics of large deviations are very much different in the two cases : reaction - diffusion as opposed to kinetic transport - reaction . let @xmath636 so that @xmath637 . for all @xmath638 we define the renormalized truncated kernel and the associated growth rate , @xmath639}(v)}{\int_{-a}^a m } m(v ) \quad \hbox { and } \quad r_a = ( 1+r)\int_{-a}^a m(v)dv -1 \in ( 0 , r)\ , .\ ] ] as @xmath640 is compactly supported , we can apply the results proved when @xmath7 is bounded in order to construct appropriate subsolutions . before we proceed with subsolutions we investigate the dispersion relation in the limit @xmath630 . define for all @xmath641 and @xmath642 : @xmath643 and @xmath644 the corresponding minimal speed defined in lemma [ minlambda ] . [ infspeed ] one has @xmath645 . for all @xmath646 , let @xmath647 such that @xmath648 if @xmath644 does not diverge to @xmath153 as @xmath630 , then it is bounded along a sequence @xmath649 and one has @xmath650 simply by comparison @xmath651 . applying fatou s lemma to , one gets @xmath652 a contradiction . let @xmath653 the solution of @xmath654,\smallskip\\ g_a ( 0,x , v ) = g^0 ( x , v ) \quad & \hbox{in } { \mathbb{r}}\times [ -a , a ] , \end{array}\right.\ ] ] and @xmath655 . clearly , @xmath656 for all @xmath5 . hence , multiplying by @xmath657 , we get @xmath658 extending @xmath659 by @xmath64 outside of @xmath660 $ ] , as @xmath661 , we get that @xmath659 is a subsolution of and it follows from the maximum principle stated in proposition [ prop - mp ] that @xmath662 for all @xmath663 . on the other hand , we know from proposition [ prop : spreadingbounded ] that for all @xmath664 : @xmath665 hence , as @xmath666 and @xmath667 for all @xmath668 $ ] , one gets for all @xmath668 $ ] : @xmath669 . therefore we conclude @xmath670 the conclusion follows from the fact that @xmath671 . we construct below supersolutions for when @xmath41 and the distribution @xmath8 is a gaussian . [ supersolunbounded ] let @xmath116 and @xmath120 . for @xmath672 define @xmath673 let @xmath82 be defined by @xmath674 then @xmath675 is a supersolution of , that is : @xmath676 we shall prove that @xmath82 is a supersolution of . indeed , it will follow that @xmath677 is a supersolution since it is the minimum of two supersolutions . from the duhamel formula , we deduce that @xmath678 to prove that @xmath82 is a subsolution we must prove in fact that @xmath679 this is sufficient to prove that the inequality @xmath680 holds true . computing the expression of @xmath681 we obtain @xmath682 we first deal with the estimate of @xmath683 . we claim the following inequality holds true : for all @xmath684 and @xmath685 $ ] , @xmath686 in fact one has @xmath687^{\frac12 } } \exp \left ( - \frac{1}{2 \sigma^2 } \frac{x^2}{(s+a)^{2 } + ( t - s)^2 } \right)\\ & \leq \displaystyle \frac{1}{\sqrt{2\pi } \sigma}\exp \left ( - \frac{1}{2 \sigma^2 } \frac{x^2}{(t+a)^{2 } } \right),\end{aligned}\ ] ] since @xmath688 , \quad ( t+a)^2 \geq ( s+a)^2 + ( t - s)^2 \geq ( s+a)^2.\ ] ] this yields @xmath689 to estimate @xmath267 , we plug in the formula for @xmath690 , @xmath691 we compute the last integral using the formula for the initial condition @xmath117 , @xmath692 thus , for all @xmath693 : @xmath694 \right ) \exp \left ( -(1+r ) t \right ) \leq \exp \left ( -(1+r ) t \right)\ ] ] as long as @xmath695 and @xmath696 , that is @xmath697 . this concludes the proof . let @xmath330 . for all @xmath698 , we define the zone @xmath699 . from the definition of @xmath677 , we deduce that @xmath677 is a supersolution such that @xmath700 . however , for all @xmath147 and @xmath701 , we have @xmath702 it yields that @xmath703 computations are made easier above since the gaussian distribution is stable by convolution . this is also the case for the cauchy distribution . therefore we are able to derive an inequality similar to . let us comment that specific case before we give the spreading estimate . because the distribution @xmath8 has an infinite variance , we learn from @xcite that the correct macroscopic limit leads to a nonlocal fractional laplacian operator . on the other hand , we expect from @xcite an exponentially fast propagation in the diffusion regime . similarly as for our previous results , we can reasonably expect that the spreading rate is faster in the kinetic model than in the macroscopic limit . therefore we expect a spreading rate faster than exponential . in fact the supersolution that we are able to derive confirms this expectation . [ supersolunbounded - cauchy ] let @xmath116 and @xmath704 . for @xmath705 and @xmath706 $ ] , define @xmath707 let @xmath82 be defined by @xmath674 then @xmath675 is a supersolution of , that is : @xmath676 the proof is the same as for proposition [ supersolunbounded ] . we just show the main computations in the case of the cauchy distribution . to prove we use the residue method as follows , @xmath708 the analog computation for proving goes as follows . first we have @xmath709 thanks to the expression of the initial condition , we compute the latest integral : @xmath710 thus , @xmath711 which holds true if @xmath712 and @xmath713 . we give in this section the proofs of propositions [ cauchypbm ] and [ prop - mp ] . well - posedness relies on a fixed point argument which is also used for the comparison principle . we first state two lemmas . [ lem : eulin ] let @xmath714 and @xmath715 . then there exists a unique function @xmath716 such that @xmath717 in the sense of distributions . this solution also satisfy the duhamel formula : @xmath718 moreover , if @xmath719 and @xmath720 , then @xmath721 in @xmath722 . for @xmath723 we define the operator @xmath724 where @xmath725 take @xmath726 and define @xmath727 and @xmath728 . assume that @xmath729 over @xmath730 . for all @xmath731 , one has : @xmath732 hence , there exists @xmath733 such that for all @xmath734 , @xmath735 is a contraction over @xmath736 . if @xmath737 on @xmath730 , then such an estimate can be derived similarly . hence , @xmath735 admits a unique fixed point , which satisfies over @xmath730 . this gives the local existence and uniqueness of the solution of . moreover , as @xmath738 does not depend on the initial datum @xmath117 , the global existence follows . [ lem : posa_t ] assume that @xmath739 is everywhere positive and that @xmath7 is an interval . then if @xmath740 is nonnegative and if there exists @xmath741 such that @xmath742 , letting @xmath82 the unique solution of , one has @xmath743 for all @xmath141 such that @xmath455 . first , assume by contradiction that there exists @xmath693 such that @xmath744 , with @xmath745 . then integrating over @xmath7 , one gets @xmath746 hence , @xmath747 for all @xmath5 and @xmath748 . letting @xmath749 , one gets @xmath750 for all @xmath5 . as @xmath455 and @xmath7 is an interval , one can take @xmath5 such that @xmath751 , leading to @xmath752 . this is a contradiction since , as @xmath82 is continuous , nonegative and @xmath753 , one has @xmath754 . hence @xmath755 for all @xmath756 such that @xmath745 . define @xmath760 . as in the proof of lemma 6 in @xcite , we first remark that this function satisfies @xmath761 with @xmath762 for all @xmath285 . we define @xmath763 and @xmath764 . writing the integral formulation as in the proof of lemma [ lem : eulin ] gives @xmath765 and thus @xmath766 in @xmath730 for some operator @xmath735 which is monotone and contractive when @xmath767 is small enough . it follows that @xmath768 for all @xmath769 . since @xmath735 is contractive the sequence @xmath770 converges to @xmath64 . we conclude that @xmath771 , meaning that @xmath772 . next , assume that @xmath494 , @xmath7 is an interval , and that there exists @xmath138 such that @xmath139 . we can follow the proof of lemma [ lem : posa_t ] , where @xmath739 defined above is positive everywhere . we deduce that @xmath773 as soon as @xmath142 . aronson , h.f . weinberger , _ nonlinear diffusion in population genetics , combustion , and nerve pulse propagation_. partial differential equations and related topics . lecture notes in math . 446 , springer , berlin , 1975 . e. bouin , v. calvez , n. meunier , s. mirrahimi , b. perthame , g. raoul , and r. voituriez . _ invasion fronts with variable motility : phenotype selection , spatial sorting and wave acceleration_. c. r. math . paris , 350(15 - 16):761766 , 2012 . dunbar , h.g . othmer , _ on a nonlinear hyperbolic equation describing transmission lines , cell movement , and branching random walks_. nonlinear oscillations in biology and chemistry . lecture notes in biomath . 66 , springer , berlin , 1986 . kolmogorov , i.g . petrovsky , n.s . piskunov , _ etude de lquation de la diffusion avec croissance de la quantit de matire et son application un problme biologique _ , moskow univ . math . bull . * 1 * ( 1937 ) , 125 .

systematic calculations of fission barriers allowing for triaxial deformation are performed for even - even superheavy nuclei with charge number using three classes of covariant density functional models . the softness of nuclei in the triaxial plane leads to an emergence of several competing fission pathes in the region of the inner fission barrier in some of these nuclei . the outer fission barriers are considerably affected by triaxiality and octupole deformation . general trends of the evolution of the inner and the outer fission barrier heights are discussed as a function of the particle numbers .

1205.2138

the on - going search for new superheavy elements is motivated by the attempts to provide theoretical and experimental explanations for two open questions in nuclear structure . the first question is related to the limits of the existence of atomic nuclei at large values of proton number , while the second one to the location of the island of stability of superheavy nuclei and the next magic numbers ( if any ) beyond @xmath1 and @xmath2 . heavy and superheavy nuclei decay by spontaneous fission and therefore their stability is defined essentially by the size and the shape of the fission barriers . thus , the inner fission barrier is considered a fundamental characteristic of these nuclei which is important in resolving the questions mentioned above . a systematic investigation of the properties of fission barriers is the best way to address the stability of nuclei against spontaneous fission since it eliminates the arbitrariness of conclusions with respect to the choice of a specific nucleus . different models with triaxiality included have been used for extensive theoretical investigations of the properties of inner fission barriers in the actinides . it has been found that the heights of inner fission barriers are reduced when triaxial shapes are taken into account @xcite . however , this reduction has a strong dependence on the particle number in the proton and neutron subsystems as well as on the applied model . these investigations were performed in the following frameworks : microscopic+macroscopic ( mm ) methods @xcite , the extended thomas - fermi plus strutinsky integral ( etfsi ) method @xcite , and non - relativistic energy density functionals ( edf ) based on zero - range skyrme @xcite and finite - range gogny @xcite forces , and recently in covariant density functional theory ( cdft ) @xcite . these methods have also been used for the study of fission barriers in superheavy nuclei ( see table [ tab4 ] below for a review ) . however , only few of them take into account triaxiality . for example , within the framework of covariant density functional theory , it has been considered only in the study of fission barriers in a single nucleus ( @xmath3hs ) in ref . thus , in order to fill this gap in our knowledge a systematic study of fission barriers of even - even @xmath0 superheavy nuclei with triaxiality included is performed in the current manuscript . we use the same method that has been successfully employed in ref . @xcite for the study of inner fission barriers in actinides , where an average deviation from experiment of 0.76 mev has been found . when extrapolating to superheavy nuclei we want to understand the impact of triaxiality on fission barriers and how they evolve with the change of the particle numbers . an additional goal is to see how the fission barriers depend on the choice of the specific cdft model . while the systematic investigation of inner fission barriers of actinides in ref.@xcite has been performed with the nonlinear meson - nucleon coupling model represented by the nl3 * parametrization of the relativistic mean field ( rmf ) lagrangian , in the current study we use in addition to the nonlinear coupling model also density - dependent meson - nucleon and point coupling models . density functional theories are extremely useful for the microscopic description of quantum mechanical many - body systems . they have been applied with great success for many years in coulombic systems @xcite , where they are in principle exact and where the functional can be derived without any phenomenological adjustments directly from the coulomb interaction . in nuclear physics with spin and isospin degrees of freedom , with strong nucleon - nucleon and three - body forces the situation is much more complicated . however , covariant density functionals exploit basic properties of qcd at low energies , in particular symmetries and the separation of scales @xcite . they provide a consistent treatment of the spin degrees of freedom , they include the complicated interplay between the large lorentz scalar and vector self - energies induced on the qcd level by the in - medium changes of the scalar and vector quark condensates @xcite . in addition , these functionals include _ nuclear magnetism _ @xcite , i.e. a consistent description of currents and time - odd mean fields important for odd - mass nuclei @xcite , the excitations with unsaturated spins , magnetic moments @xcite and nuclear rotations @xcite . because of lorentz invariance no new adjustable parameters are required for the time - odd parts of the mean fields . of course , at present , all attempts to derive these functionals directly from the bare forces @xcite do not reach the required accuracy . however , in recent years modern and very successful covariant density functionals have been derived @xcite which are based on density dependent vertices and one additional parameter characterizing the range of the force . they provide an excellent description of ground states and excited states all over the periodic table @xcite with a high predictive power . modern versions of these forces derive the density dependence of the vertices from state - of - the - art ab - initio calculations and use only the remaining few parameters for a fine tuning of experimental masses in finite spherical @xcite and deformed @xcite nuclei . the manuscript is organized as follows . the theoretical framework and the details of the numerical calculations are discussed in sec . [ theor - sect ] . the results of the investigations of the fission barriers , the role of triaxiality and the comparison with the actinide region are presented in sec . [ res+diss ] . finally , sec . [ conc ] summarizes the results of our work . three classes of covariant density functional models are used throughout this manuscript : the nonlinear meson - nucleon coupling model ( nl ) , the density - dependent meson - exchange model ( dd - me ) and a density - dependent point coupling model ( dd - pc ) . the main differences between them lay in the treatment of the range of the interaction , the mesons and density dependence . the interaction in the first two classes has a finite range , while the third class uses zero - range interaction with one additional gradient term in the scalar - isoscalar channel . the mesons are absent in the density - dependent point coupling model . the density dependence is explicit in the last two models , while it shows up via the non - linearity in the @xmath4-meson in the nonlinear meson - nucleon coupling model . each of these classes is represented in the current manuscript by a set of parameters that is considered to be state - of - the - art . in the meson - exchange models @xcite , the nucleus is described as a system of dirac nucleons interacting via the exchange of mesons with finite masses leading to finite - range interactions . the starting point of covariant density functional theory ( cdft ) for these two models is a standard lagrangian density @xcite @xmath5 \psi\nonumber\\ & + \frac{1}{2}(\partial\sigma)^{2}-\frac{1}{2}m_{\sigma}^{2}\sigma^{2}% -\frac{1}{4}\omega_{\mu\nu}\omega^{\mu\nu}+\frac{1}{2}m_{\omega}^{2}\omega ^{2}\label{lagrangian}\\ & -\frac{1}{4}{\vec{r}}_{\mu\nu}{\vec{r}}^{\mu\nu}+\frac{1}{2}m_{\rho}% ^{2}\vec{\rho}^{\,2}-\frac{1}{4}f_{\mu\nu}f^{\mu\nu}\nonumber\end{aligned}\ ] ] which contains nucleons described by the dirac spinors @xmath6 with the mass @xmath7 and several effective mesons characterized by the quantum numbers of spin , parity , and isospin . they create effective fields in a dirac equation , which corresponds to the kohn - sham equation @xcite in the non - relativistic case . the lagrangian ( [ lagrangian ] ) contains as parameters the meson masses @xmath8 , @xmath9 , and @xmath10 and the coupling constants @xmath11 , @xmath12 , and @xmath13 . @xmath14 is the charge of the protons and it vanishes for neutrons . this linear model has first been introduced by walecka @xcite . to treat the density dependence in this model boguta and bodmer @xcite introduced a density dependence via a non - linear meson coupling replacing the term @xmath15 in eq . ( [ lagrangian ] ) by @xmath16 the nonlinear meson - nucleon coupling is represented by the parameter set nl3 * @xcite ( see table [ tab1 ] ) , which is a modern version of the widely used parameter set nl3 @xcite . apart from the fixed values for the masses @xmath7 , @xmath17 and @xmath18 , there are six phenomenological parameters @xmath19 , @xmath20 , @xmath21 , @xmath22 , @xmath23 , and @xmath24 . the density - dependent meson - nucleon coupling model has an explicit density dependence for the meson - nucleon vertices . there are no nonlinear terms in the @xmath4 meson , i.e. @xmath25 . the meson - nucleon vertices are defined as : @xmath26 where the density dependence is given by @xmath27 for @xmath4 and @xmath28 and by @xmath29 for the @xmath30 meson . _ x _ is defined as the ratio between the baryonic density @xmath30 at a specific location and the baryonic density at saturation @xmath31 in symmetric nuclear matter . the eight parameters in eq . ( [ fx ] ) are not independent , but constrained as follows : @xmath32 , @xmath33 , and @xmath34 . these constraints reduce the number of independent parameters for density dependence to three . this model is represented in the present investigations by the parameter set dd - me2 @xcite given in table [ tab1 ] . [ c]|c|c|c|parameter & nl3 * & dd - me2 + @xmath7 & 939 & 939 + @xmath8 & 502.5742 & 550.1238 + @xmath9 & 782.600 & 783.000 + @xmath10 & 763.000 & 763.000 + @xmath11 & 10.0944 & 10.5396 + @xmath12 & 12.8065 & 13.0189 + @xmath13 & 4.5748 & 3.6836 + @xmath35 & -10.8093 & + @xmath36 & -30.1486 & + @xmath37 & & 1.3881 + @xmath38 & & 1.0943 + @xmath39 & & 1.7057 + @xmath40 & & 0.4421 + @xmath41 & & 1.3892 + @xmath42 & & 0.9240 + @xmath43 & & 1.4620 + @xmath44 & & 0.4775 + @xmath45 & & 0.5647 + the lagrangian for the density - dependent point coupling model @xcite is given by @xmath46 it contains the free - nucleon lagrangian , the point coupling interaction terms , and the coupling of the proton to the electromagnetic field . the derivative terms in eq . ( [ lag - pc ] ) account for the leading effects of finite - range interaction which are important in nuclei . in analogy with meson - exchange models , this model contains isoscalar - scalar , isoscalar - vector and isovector - vector interactions . in the present work it is represented by the dd - pc1 parametrization @xcite given in table [ tab2 ] . [ c]|c|c|parameter & dd - pc1 + @xmath7 & 939 + @xmath37 & -10.04616 + @xmath38 & -9.15042 + @xmath39 & -6.42729 + @xmath40 & 1.37235 + @xmath41 & 5.91946 + @xmath42 & 8.86370 + @xmath44 & 0.65835 + @xmath47 & 1.83595 + @xmath48 & 0.64025 + cccccforce & @xmath49 & @xmath50 & @xmath51 & @xmath52 + nl3 * & 10.7 & -10.4 & 7.40 & 18.9 + dd - pc1 & 10.5 & -7.38 & 7.5 & 19.2 + dd - me2 & 11.0 & -10.3 & 7.9 & 17.0 + + the triaxial relativistic mean field ( rmf)+bcs approach @xcite is used here for the description of fission barriers . this approach has been very successfully applied to a systematic description of the fission barriers in the actinides @xcite . the rmf - equations are solved iteratively and at each iteration the bcs occupation probabilities @xmath53 are determined . these quantities are used in the calculation of densities , energies and new fields for the next iteration . we use a monopole pairing force @xcite with the strength parameters @xmath54 for neutrons ( @xmath55 ) and protons ( @xmath56 ) . for each type of particle we start with a pairing strength parameter @xmath57 and solve at each iteration the gap equation @xcite @xmath58 with @xmath59 , where @xmath60 are the eigenvalues of the dirac equation and the chemical potential @xmath61 is determined by the average particle number . then the occupation probabilities @xmath62 and the gap parameters @xmath63 are determined in a self - consistent way . the pairing energy is defined as @xmath64 the sum over @xmath65 in eqs . ( [ gap - equation ] ) , ( [ delta - bcs ] ) and ( [ epair ] ) run over all states in the pairing window @xmath66 . in ref . @xcite empirical pairing gap parameters @xmath67 have been determined by a fit to experimental data on neutron and proton gaps in the normal deformed minimum . pu nucleus with the nl3 * parametrization of the rmf lagrangian . the energy difference between two neighboring equipotential lines is equal to 0.5 mev . the pink dashed line with solid circles shows the lowest - energy solution as a function of @xmath68 . small red stars show the positions of deformation points at which the numerical results were obtained . the figure is based on the results of ref . further details are given in the text.,width=302 ] these empirical gap parameters form the basis for the definition of the strength parameters @xmath54 in the current manuscript . the following expressions @xcite @xmath69 are used in the calculations . first , using empirical gap parameters of eq . ( [ emp - g ] ) , the values @xmath70 and @xmath71 are obtained for the ground states of all even - even nuclei in the @xmath72 , @xmath73 region . then , the parameters @xmath49 , @xmath50 , @xmath51 and @xmath52 are defined by the least square fit to the set of the @xmath70 and @xmath71 . their values depend on the parameter set of the lagrangian and they are given in table [ tab3 ] . in this way we have strength parameters for the effective pairing interaction depending in a smooth way on the neutron and proton numbers and , because of the changing level density , the gap parameters derived from those values show fluctuations as a function of the particle numbers . note that similarly to ref . @xcite , the cutoff energy for the pairing window is set to e@xmath74=120 mev . & pnp & actinide & she & a / t & @xmath75 + 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 + + mller 2009 @xcite & bcs(@xmath57)@xcite & & yes & yes & yes & t & + dobrowolski 2007 @xcite & bcs(@xmath57 ) & & no & yes & 2 & t & + kowal 2010 @xcite & bcs(@xmath57 ) & @xmath76 @xcite & no & yes & yes & t & + + dutta 2000 @xcite & bcs(@xmath77 ) & & no & 5 & 5 & t & + + bonneau 2004 @xcite & bcs(@xmath57)/bcs(@xmath77 ) & @xmath78/@xmath79 & no & yes & no & a / t & 6 mev + brvenich 2004 @xcite & bcs(@xmath77 ) & across nuclear & no & yes & yes & a & @xcite is chosen to include approximately one additional oscillator shell of states above the fermi level . ] + & & chart @xcite & & & & & + samyn 2005 @xcite & hfb(@xmath77 ) & @xmath80 & yes & yes & yes & t & different @xmath75 are included . mev for the bsk6 , bsk7 and bsk8 skyrme forces , @xmath82 mev for bsk2 force , @xmath83 mev for bsk9 force , @xmath84 mev for sly6 force . ] + staszczak 2006 @xcite & bcs(g ) & @xmath85 and ref.@xcite & no & yes & yes & t & lowest z(n ) states + staszczak 2007 @xcite & bcs(@xmath57)/bcs(@xmath77 ) & @xmath85 & no & yes & yes & t & lowest z(n ) states@xmath86 + + warda 2002 @xcite & hfb(gogny ) & no & no & 5 & no & t & no + delaroche 2006 @xcite & hfb(gogny ) & no & no & yes & no & t & no + + bender 1998 @xcite & bcs(@xmath77 ) & across nuclear & no & no & 3 & t & @xcite@xmath87 + & & chart @xcite & & & & & + brvenich 2004 @xcite & bcs(@xmath77 ) & across nuclear & no & yes & yes & a & @xcite@xmath87 + & & chart @xcite & & & & & + karatzikos 2010 @xcite & rhb(gogny ) & no & no & yes & yes & a & no + abusara 2010 @xcite & bcs(@xmath57 ) & @xmath88 & no & yes & no & t & 120 mev + & & @xmath89 & & & & & + + [ tab4 ] the calculations are performed by successive diagonalizations using the broyden method @xcite and the method of quadratic constraints @xcite . we have also implemented the augmented lagrangian method @xcite for constraints in our computer codes , but we did not find any clear advantage of this method over the method of quadratic constraints for the type of potential energy surfaces ( pes ) we are dealing with . starting from the three multipole operators @xmath90 we calculate in the following investigations three types of potential energy surfaces , using three types of constraints . first , we restrict ourselves to axially symmetric configurations with reflection symmetry , abbreviated throughout the paper by ( a ) . here we use the computer code diz @xcite based on an expansion of the dirac spinors in terms of harmonic oscillator wave functions with cylindrical symmetry and we minimize @xmath91 where @xmath92 is the total energy , @xmath93 denotes the expectation values of the mass quadrupole operators , @xmath94 is the constrained value of the multipole moment , and @xmath95 the corresponding stiffness constant @xcite . in the next step abbreviated throughout the paper by ( t ) we use the triaxial computer code dic @xcite with the d2-symmetry based on an expansion of the dirac spinors in terms of a cartesian oscillator basis and by imposing constraints on axial and triaxial mass quadrupole moments we minimize @xmath96 in addition , starting from the axial reflection symmetric computer code diz of ref . @xcite , we have developed an axial reflection asymmetric [ octupole deformed ] code doz . it is used for a study of the impact of axial octupole deformation on the inner and outer fission barriers . the calculations in this axially symmetric octupole code , abbreviated throughout this paper by ( o ) vary the function @xmath97 it was checked that the numerical results obtained for axially symmetric solutions in the a , t and o calculations differ by no more than 50 kev for all deformations of interest . the truncation of the basis is performed in all these calculations in such a way that all states belonging to the shells up to @xmath98 fermionic shells and @xmath99 bosonic shells are taken into account . this truncation scheme has been tested and used in the actinides in ref . @xcite . although the calculations in such a truncation scheme provide sufficient numerical accuracy , they are also very computationally demanding ( see ref.@xcite for details ) . this is the reason why we treat the pairing channel in the present triaxial rmf calculations in the bcs approximation despite the fact that the triaxial cranked relativistic hartree+bogoliubov ( rhb ) approach with the finite range gogny forces in the pairing channel was developed in the late nineties @xcite . rmf+bcs calculations are less time - consuming and more stable ( especially , in the saddle point region ) than the rhb calculations . [ pes-2d ] shows an example of the distribution of deformation points on a potential energy surface ( pes ) at which the numerical results were obtained using these constraints in a triaxial ( t ) calculation . all mean - field calculations for fission barriers with triaxiality included are computationally demanding . this is the reason why simple pairing interactions ( either seniority pairing with constant strength @xmath57 or the zero - range @xmath77-force ) and the bcs framework are used in the majority of the calculations ( see table [ tab4 ] ) . the overview given in this table focuses mainly on the investigations including triaxial deformation and on the most recent and comprehensive studies within the specific theoretical frameworks . in particular , the studies of fission barriers in superheavy elements ( she ) are included irrespective of whether the calculations include triaxial deformation or not . this table shows that the majority of studies neglect particle number projection and use different prescriptions for the size of the pairing window . the impact of these prescriptions on the height of the inner fission barrier has recently been investigated in ref . @xcite . the inclusion of triaxiality has improved the accuracy of the description of the inner fission barriers in actinides in all state - of - the - art models @xcite . [ n - dep ] and [ z - dep ] show the differences between experimental and calculated heights of inner fission barriers obtained in different theoretical models as a function of neutron and proton numbers , respectively . note that this comparison covers only results of systematic triaxial calculations which simultaneously include even - even th , u , pu , cm and cf nuclei . to our knowledge , no such calculations have been published with dft based on skyrme forces . as a result , these figures cover all existing systematic triaxial studies of inner fission barriers in actinides . the @xmath77-values displayed on the panels of figs . [ n - dep ] and [ z - dep ] show the average deviation from experiment for the calculated heights of inner fission barriers . one can see that they are of the same magnitude in the different approaches and minor differences between the approaches in the @xmath77-values are not important considering the considerable uncertainties in the extraction of inner fission barrier heights from experimental data as seen , for example , in the differences of the compilations of refs.@xcite and @xcite . however , the similarity of the average trends of these deviations ( shown by thick dashed lines in figs . [ n - dep ] and [ z - dep ] ) as a function of neutron and proton numbers is more important considering the differences in underlying mean fields and in the treatment of pairing correlations . at present , it is difficult to find a clear explanation for these trends . although differences in the treatment of pairing correlations ( bcs with monopole pairing and of different pairing windows in the cdft @xcite and mm @xcite calculations versus the hartree - fock - bogoliubov framework based on the d1s force in gogny dft @xcite ) can contribute to deviations between theory and experiment @xcite , it is quite unlikely that they are responsible for the observed trends of the deviations . the nucleus @xmath100120 is predicted to be a spherical doubly magic nucleus in cdft @xcite . its potential energy surface in the @xmath101 plane is shown in fig . [ pes-2d-6panel ] . it is interesting to compare it with the pes of the nucleus @xmath102pu shown in fig.[pes-2d ] . these two pes s are representative examples of typical pes s in actinides and superheavy nuclei . the gross structure of these two pes s is defined by the fact that the total energy is generally increasing when moving away from the @xmath103 axis ; so it looks like a canyon . however , there are local structures inside the canyon which define the differences between the two mass regions with respect to the impact of triaxiality on the inner and outer fission barriers . in @xmath102pu , a large hill is located at the axial shape @xmath104 inside a canyon . as a consequence , the fission path from the normal deformed minimum initially proceeds along the @xmath105 axis , then bypasses the axial @xmath68=0.5 hill via a path with @xmath106 , and then proceeds along the bottom of the canyon on an axially symmetric path again . as a result of this bypass , the inner fission barrier heights of the actinides are lowered by @xmath107 mev due to triaxiality @xcite . however , the calculated outer fission barriers of the actinides are not affected by triaxiality @xcite . nucleus obtained with the nl3 * parametrization . the black solid , red dashed and blue dot - dashed lines display the deformation energy curves for the axially symmetric , triaxial and axial octupole deformed solutions . we show the deformation energy curves for the last two solutions only in the range of @xmath68 values where it is lower in energy than the deformation energy curve of the axially symmetric solution . note that the tr - b fission path via saddle at @xmath108 ( see text for details ) is not shown since this saddle is lower than the one of the ax fission path only by 0.2 mev ( see table [ tab5 ] ) ; the results of ref.@xcite suggests that spontaneous fission exploiting this path is less likely than the one along the ax fission path.,width=302 ] the properties of the pes of the nucleus @xmath100120 , defining the fission pathes , are completely opposite to the case of @xmath102pu since there are two triaxial and one axial hills inside the pes canyon of @xmath100120 . two triaxial hills are located at moderate deformations ( @xmath109 ) , while the axial hill is superdeformed ( @xmath110 ) . the fission path ( shown by red dashed line in fig.[pes-2d-6panel ] ) starts at a spherical shape , then proceeds between two triaxial hills ( @xmath11130@xmath112 ) and bypasses the axial hill at @xmath1130.75 via a @xmath114 path . the @xmath115-softness of the pes , which exists between the two triaxial hills , has only a minor effect on the height of the inner fission barrier ; the triaxial solution is lower than the axial one by 100 - 200 kev at @xmath116 deformations ( see fig . [ pot ] ) . however , this figure shows that _ along this fission path _ the height of inner fission barrier is not affected by triaxiality . the second fission path shown by a solid red line starts at spherical shape , and proceeds along the axially symmetric @xmath117 axis , via a saddle point at ( @xmath118 , @xmath119 ) and then along the first fission path after second minimum . the existence of the valley between the walls of the canyon and triaxial hills at ( @xmath120 ) is a necessary condition for the existence of this path . this type of fission path exists in a number of nuclei , so for convenience we will call it as triaxial tr - b fission path . the unusual physical feature of this second path is the fact that initially the nucleus has to be squeezed along the axis of symmetry , thus creating an oblate nucleus with quadrupole deformation @xmath121 . this is contrary to the usual picture of fission where the prolate or near - prolate nucleus is stretched out along the path of increasing quadrupole deformation . note that the saddle of this fission path is only by 0.2 mev lower than the saddle of the first fission path ( see table [ tab5 ] ) and in addition it is much longer . therefore , considering a dynamical calculation taking into account the action integral along the entire fission path @xcite this path will probably not contribute much to the fission probability . contrary to the actinides , the triaxiality has a considerable impact on the shape and the height of outer fission barrier which is lowered by @xmath122 3 mev in @xmath100120 nucleus ( see fig . [ pot ] ) . note , that in this nucleus , the lowering of outer fission barrier due to octupole deformation is substantially smaller than the one due to triaxiality . it is also evident that the landscape of the pes and the existence of saddle points and valleys depends on the proton and neutron numbers . this is clearly seen in fig . [ pes-2d-6panel ] and in table [ tab5 ] . in the @xmath123 isotopes , the increase of neutron number up to @xmath124 and beyond leads to the emergence of an axial hill at @xmath125 ; this is clearly visible in the nucleus @xmath126120 in fig . [ pes-2d-6panel ] . its existence in high-@xmath127 and neutron - rich nuclei leads to the shift of the first fission path in the deformation space . in lighter systems this path proceeds via the axially symmetric saddle as , for instance , in the nucleus @xmath100120 in fig . [ pes-2d-6panel ] . for heavier systems it starts at spherical shape mev relatively to the superdeformed minimum ) which suggests that these states are extremely unstable against fission . thus , we start the discussion from spherical / weakly(normal)-deformed minima which have better survival probability against fission . ] , proceeds along the axially symmetric @xmath117 axis up to @xmath128 and then via a saddle point at ( @xmath129 ) , located between a triaxial ( @xmath130 ) below . ] and an axial @xmath131 hill , to the axially symmetric superdeformed minimum . this type of fission path exists in a number of nuclei . we will label it as triaxial tr - a for convenience . in addition , there is a second triaxial path tr - b via a saddle at @xmath132 , which is similar to the one present in the nucleus @xmath100120 . the structure of the pes of the nucleus @xmath133116 is similar to the one of the nucleus @xmath126120 in fig . [ pes-2d-6panel ] . the lowering of neutron number below @xmath124 in the nuclei under study leads to the situation that the fission path via the axially symmetric inner saddle ( called as ax below ) becomes energetically favored . this is clearly seen in the pes of the @xmath134116 , @xmath135112 and @xmath136112 nuclei in fig . [ pes-2d-6panel ] and in table [ tab5 ] . it is clear that the landscape of the pes of superheavy nuclei in the region of the inner fission barrier is more complicated than in the case of the actinides . the energetically favored fission path in the majority of these nuclei proceeds through an axially symmetric saddle ( see fig . [ pes-2d-6panel ] and table [ tab5 ] ) . only in a few nuclei ( @xmath137120 , @xmath138 and @xmath133116 ) , triaxial fission path tr - a is lower in energy than axial ax fission path . the triaxial fission path tr - b exists in the majority of the nuclei apart from neutron - poor @xmath139 nuclei and @xmath140 nuclei ( see fig . [ pes-2d-6panel ] and table [ tab5 ] ) . the energy of the saddle point of this fission path is lower than the one of the first fission path ( either ax or tr - a ) in a number of nuclei . however , the results of calculations within the mm method @xcite suggest that although triaxiality lowers the static fission barriers , it plays a minor role in spontaneous fission of superheavy nuclei with @xmath141 . this is because a fission path via an oblate shape and triaxial saddles is substantially longer as compared to the axially symmetric path which leads to a significant reduction of the penetration probability . however , this complexity of the pes and the presence of two fission pathes for the inner fission barriers calls for finding the dynamical path along which the fission process takes place in cdft . work in this direction is in progress . figs . [ pot3a ] and [ pot3b ] show deformation energy curves for several isotopes with charge numbers @xmath142 and 116 for three classes of cdft models . experimental estimates of the inner fission barrier heights were obtained for these nuclei in ref . the potential energy structure of these nuclei is similar to the one seen in @xmath100120 ( fig . [ pes-2d-6panel ] ) ; the presence of two triaxial hills at moderate deformations ( @xmath143 ) defines two possible fission pathes : one between the hills ( either ax or ax - tr pathes ) and another between the hill and the walls of the pes canyon ( tr - b path ) . the @xmath115-softness of the pes , which exists between the two triaxial hills , defines whether the fission path passing between these two hills is fully axially symmetric ( ax path ) or has some degree of triaxiality ( ax - tr path ) . the ax - tr path is generally similar to the ax path but differs from it by the fact that moderate triaxiality @xmath144 appears at the saddle and along the shoulder of inner fission barrier . although the heights of the discussed hills in the region of inner fission barrier depends on the parametrization , the observed features of the potential energy surfaces and fission pathes are rather independent of the parametrization ( fig . [ pes-2d - dif - param ] ) . tables [ tab5 ] and [ tab7 ] show that the saddle point along this path has triaxiality only in the case of the two nuclei @xmath145114 . however , even in these cases the ax saddle is only 50 - 150 kev higher than the tr - a saddle . thus , similarly to the @xmath100120 nucleus triaxiality only marginally affects the inner fission barriers . note that the ground states are somewhat deformed in these nuclei ( see figs . [ pot3a ] and [ pot2]a ) . for these nuclei triaxiality has a considerable impact on the shape and height of the outer fission barriers ; the decrease of the heights of the outer fission barriers due to triaxiality is typically in the range of 1.5 - 2.0 mev and it depends on the particle number and on the rmf parametrization . note that the outer fission barriers are affected also by octupole deformation . in the @xmath140 nuclei , the triaxial saddle is lower in energy than the octupole saddle . the situation is reversed in the @xmath146 nuclei . among the different classes of cdft models , the dd - me2 parametrization always gives the highest values for the inner and outer fission barrier heights . they are ( on average ) by 1 and 1.5 mev higher than the ones obtained in the nl3 * and the dd - pc1 parametrizations . the heights and the shapes of the inner fission barriers are very similar for the nl3 * and dd - pc1 parametrizations . the outer fission barriers also come close to each other in these two parametrizations in axial [ a ] , triaxial [ t ] and octupole [ o ] ( for @xmath140 nuclei ) calculations . however , with increasing of @xmath127 from 114 to 116 the difference between the results of octupole calculations increases ; the difference between the energies of the octupole saddles obtained in the nl3 * and dd - pc1 parametrizations reaches 2 mev in the @xmath147 nuclei ( see fig . [ pot3a ] ) . nuclei as a function of neutron number . ( panel ( b ) ) the energies of axial ( large solid symbols , thick lines ) and triaxial ( either ax - tr or tr - a ; shown by small open symbols and thin lines ) saddles with respect to the spherical / weakly(normal)-deformed minima in these nuclei as a function of neutron number . note that the same type of symbols is used for axial and triaxial saddles in a given isotope chain.,width=302 ] there is only one experimental work @xcite where estimates on the heights of inner fission barriers in superheavy nuclei with @xmath148 112 , 114 , and 116 have been obtained . unfortunately , experimentally the fission barriers are accessible only indirectly and a model - dependent analysis is used to obtain these quantities , which causes an ambiguity in the comparison with theoretical results . even in the actinide region where the fission barrier heights were extracted from a number of independent experiments with high statistics ( see , for example , ref . @xcite ) , a typical uncertainty in the experimental values , as suggested by the differences among various compilations , is of the order of @xmath149 mev @xcite . these uncertainties are expected to be higher in superheavy nuclei since the estimates of ref . @xcite are based on experimental data represented by low statistics and on a method which differs from the methods used in the analysis of fission barrier heights in actinides . in addition , there is no independent confirmation of the inner fission barrier height estimates of ref . the interpretation of experimental data based on cross sections in terms of fission barrier heights becomes even more complicated when the fission path has a double hump structure , which according to many calculations may be the case in superheavy nuclei . the widening of the barrier due to the second hump ( or its remnant ) would require the lowering of the inner fission barrier height ; this possibility has not been taken into account in the analysis of ref . based on this discussion it is clear that the level of confidence of fission barrier height estimates for superheavy nuclei is significantly lower than the one for the actinides . according to ref . @xcite , the estimated lower limits for fission barrier heights in even - even @xmath150 and 116 nuclei shown in fig . [ pot3a ] and [ pot3b ] are 5.5 , 6.7 and 6.4 mev , respectively . our results for the heights of inner fission barrier in these nuclei along the ax and ax - tr fission pathes are always smaller than the experimental data by @xmath151 mev . the saddle along the tr - b fission path is somewhat lower in energy ( by 0.4 - 0.7 mev ) than the saddle along the ax - tr path in @xmath145116 nuclei ( see table [ tab7 ] ) . however , as discussed above ( based on the results of ref . @xcite ) it is unlikely that the fission predominantly proceeds along the tr - b path considering especially ( i ) that the gain in saddle point energy as compared with the ax - tr path is small and ( ii ) the tr - b path is substantially longer as compared with the ax - tr path . considering the discussion above , it is not clear at this moment how serious the discrepancy _ between calculations and experimental estimates _ is . however , the results of the calculations suggest one possible way to increase the heights of inner fission barriers . potential energy surfaces in the ground state region of these nuclei are extremely soft ( see figs . [ pot3a ] and [ pot3b ] ) . for such nuclei , the correlations beyond mean field taken into account , for example , by the generator coordinate method can lower the energy of the ground state by a few mev without affecting much the barrier top , thus effectively increasing the height of inner fission barrier . systematic calculations of inner fission barriers have been performed for even - even @xmath0 nuclei with @xmath152 using the nl3 * parametrization . this is the same parametrization which has been used in systematic calculations of fission barriers in actinides @xcite and which provides 0.76 mev average deviation from experiment for the heights of inner fission barriers in these nuclei . for a given proton number @xmath127 a sequence of nine even - even nuclei is selected such that the middle nucleus of this sequence roughly corresponds to an experimentally observed nucleus ( or a nucleus which is expected to be observed in near future ) . this crudely outlines the region which may be experimentally studied within the next one or two decades . the results of these calculations are summarized in table [ tab5 ] . based on the results of ref . @xcite , it is reasonable to neglect the path tr - b since it is substantially longer as compared with other pathes which leads to significant reduction of the penetration probability . in fig . [ pot2]b we consider the evolution of the heights of the inner fission barriers as a function of the neutron number @xmath153 . the @xmath154 and 116 isotope chains show a generally increasing trend for the barrier heights with increasing neutron number for @xmath155 . [ pot2]a suggests that the origin of this trend can be traced back to the deformation of the ground state . for small neutron numbers these nuclei are deformed in the ground state . however , they gradually become spherical when approaching @xmath124 because there is a spherical shell gap at this neutron number ( see , for example , fig . 28 in ref . the negative shell correction energy at the ground state is larger in absolute value in the vicinity of the @xmath124 spherical shell gap than at lower neutron numbers , where the ground state is deformed and characterized by a larger level density in the vicinity of the fermi level . the level density ( and , as a consequence , the shell correction energy ) at the saddle point of the inner fission barrier does not change so drastically as the one at the ground state . as a result , the heights of inner fission barriers , which are defined as the energy differences between the binding energies of the ground state and saddle point , show the observed features . note that in these nuclei the energies of the ax and ax - tr saddles differ by at most 200 kev , the difference between the energies of the ax and tr - a saddles does not exceed 400 kev . the nuclei with @xmath148 118 and 120 have , with the exception of the nucleus @xmath156 , spherical minima due to the presence of the @xmath123 spherical shell gap . the nucleus @xmath157 has the highest value for the fission barrier among all nuclei under investigation . this is connected with its doubly magic nature in cdft . in the isotope chain with @xmath123 , moving away from the @xmath158 shell closure , shell effects connected with spherical shape become less pronounced and this leads to a decrease of the inner fission barrier , because the barrier height in these nuclei is defined with respect to the spherical ground state . apart from the lightest two and the heaviest @xmath159 isotopes , the fission barrier heights of the @xmath159 chain are nearly constant as a function of neutron number and they are close to 4.5 mev . note that in these nuclei the energies of the ax and ax - tr saddles differ only by 100 - 400 kev . however , the difference between the energies of the ax and tr - a saddles can reach 2 mev ( as in the case of the nucleus @xmath160120 , see table [ tab5 ] ) . it is important to mention that the valley between the axial @xmath161 and triaxial @xmath162 hills leading to a tr - a saddle is rather shallow . its depth with respect to the tip of the triaxial hill varies between 100 and 300 kev . the latter value is obtained , for example , in @xmath163120 nucleus , see fig . [ 302-valley ] . & @xmath164 & ( @xmath68,@xmath115 ) & @xmath165 & ( @xmath68,@xmath115 ) + 1 & 2 & 3 & 4 & 5 & 6 & 7 + + @xmath160120 & 5.11 & 0.20 & 3.04 & ( 0.34,26@xmath112 ) & 2.23 & ( 0.48,22@xmath112 ) + @xmath166120 & 5.00 & 0.20 & 3.46 & ( 0.33,26@xmath112 ) & 2.02 & ( 0.50,22@xmath112 ) + @xmath126120 & 5.46 & 0.21 & 4.51 & ( 0.33,26@xmath112 ) & 3.30 & ( 0.51,21@xmath112 ) + @xmath163120 & 5.37 & 0.19 & 5.00 & ( 0.33,25@xmath112 ) & 3.93 & ( 0.50,24@xmath112 ) + @xmath133120 & 5.35 & 0.32 & 4.93 * & ( 0.31,11@xmath112 ) & 4.08 & ( 0.50,26@xmath112 ) + @xmath167120 & 5.35 & 0.33 & 5.02 * & ( 0.33,12@xmath112 ) & 4.27 & ( 0.49,28@xmath112 ) + @xmath168120 & 5.67 & 0.34 & 5.57 * & ( 0.35,10@xmath112 ) & 4.92 & ( 0.50,27@xmath112 ) + @xmath169120 & 5.27 & 0.34 & & & 4.73 & ( 0.50,27@xmath112 ) + @xmath100120 & 5.82 & 0.36 & & & 5.61 & ( 0.50,27@xmath112 ) + + @xmath126118 & 4.97 & 0.21 & 3.92 & ( 0.33,26@xmath112 ) & 3.03 & ( 0.51,26@xmath112 ) + @xmath163118 & 5.07 & 0.23 & 4.57 & ( 0.33,26@xmath112 ) & 3.41 & ( 0.51,25@xmath112 ) + @xmath133118 & 5.18 & 0.28 & 4.94 * & ( 0.27,15@xmath112 ) & 3.37 & ( 0.51,26@xmath112 ) + @xmath167118 & 5.07 & 0.32 & 4.72 * & ( 0.32,12@xmath112 ) & 3.67 & ( 0.50,27@xmath112 ) + @xmath168118 & 5.00 & 0.32 & 4.73 * & ( 0.33,10@xmath112 ) & 3.87 & ( 0.50,26@xmath112 ) + @xmath169118 & 4.95 & 0.32 & 4.82 * & ( 0.33,10@xmath112 ) & 4.12 & ( 0.50,25@xmath112 ) + @xmath100118 & 4.75 & 0.34 & & & 4.35 & ( 0.50,23@xmath112 ) + @xmath136118 & 4.15 & 0.36 & & & 4.08 & ( 0.49,24@xmath112 ) + @xmath170118 & 3.28 & 0.36 & & & & + + @xmath133116 & 4.83 & 0.26 & 4.48 & ( 0.35,25@xmath112 ) & 3.13 & ( 0.52,22@xmath112 ) + @xmath167116 & 4.78 & 0.28 & 4.57 & ( 0.30,15@xmath112 ) & 3.29 & ( 0.51,23@xmath112 ) + @xmath168116 & 4.53 & 0.32 & 4.32 * & ( 0.32,11@xmath112 ) & 3.35 & ( 0.51,24@xmath112 ) + @xmath169116 & 4.37 & 0.32 & 4.22 * & ( 0.33,7@xmath112 ) & 3.40 & ( 0.50,25@xmath112 ) + @xmath100116 & 3.98 & 0.33 & 3.93 * & ( 0.34,7@xmath112 ) & 3.39 & ( 0.48,26@xmath112 ) + @xmath136116 & 3.39 & 0.34 & & & 3.47 & ( 0.48,27@xmath112 ) + @xmath170116 & 3.55 & 0.34 & & & & + @xmath134116 & 3.43 & 0.36 & & & & + @xmath171116 & 3.40 & 0.37 & & & & + + @xmath168114 & 4.65 & 0.28 & 4.43 * & ( 0.31,15@xmath112 ) & 3.42 & ( 0.49,24@xmath112 ) + @xmath169114 & 4.19 & 0.29 & 4.14 * & ( 0.31,10@xmath112 ) & 3.08 & ( 0.49,25@xmath112 ) + @xmath100114 & 3.81 & 0.32 & & & 3.18 & ( 0.48,25@xmath112 ) + @xmath136114 & 3.40 & 0.34 & & & 3.38 & ( 0.48,26@xmath112 ) + @xmath170114 & 3.52 & 0.32 & & & & + @xmath134114 & 3.32 & 0.33 & & & & + @xmath171114 & 3.07 & 0.38 & & & & + @xmath172114 & 3.08 & 0.42 & & & & + @xmath173114 & 3.20 & 0.42 & & & & + & @xmath164 & ( @xmath68,@xmath115 ) & @xmath165 & ( @xmath68,@xmath115 ) + 1 & 2 & 3 & 4 & 5 & 6 & 7 + + @xmath100112 & 3.94 & 0.30 & & & 3.64 & ( 0.48,27@xmath112 ) + @xmath136112 & 3.48 & 0.31 & & & 3.44 & ( 0.47,26@xmath112 ) + @xmath170112 & 3.07 & 0.31 & & & & + @xmath134112 & 2.90 & 0.32 & & & & + @xmath171112 & 2.65 & 0.32 & & & & + @xmath172112 & 2.35 & 0.41 & & & & + @xmath173112 & 2.70 & 0.43 & & & & + @xmath135112 & 3.06 & 0.44 & & & & + @xmath174112 & 3.34 & 0.44 & & & & + & @xmath175 & ( @xmath68,@xmath115 ) & @xmath165 & ( @xmath68,@xmath115 ) + + @xmath169116 & 3.64 & 0.31 & 3.47 * & ( 0.32,10@xmath112 ) & 2.58 & ( 0.48,27@xmath112 ) + @xmath100116 & 3.63 & 0.32 & 3.55 * & ( 0.32,5@xmath112 ) & 2.83 & ( 0.49,26@xmath112 ) + @xmath136114 & 3.42 & 0.34 & & & & + @xmath170114 & 3.38 & 0.32 & & & & + @xmath134112 & 3.02 & 0.31 & & & & + @xmath171112 & 2.75 & 0.31 & & & & + + @xmath169116 & 5.36 & 0.32 & 5.20 * & ( 0.32,8@xmath112 ) & 4.45 & ( 0.48,25@xmath112 ) + @xmath100116 & 4.77 & 0.32 & 4.68 * & ( 0.32,5@xmath112 ) & 4.24 & ( 0.48,27@xmath112 ) + @xmath136114 & 4.33 & 0.34 & & & & + @xmath170114 & 4.18 & 0.32 & & & & + @xmath134112 & 3.62 & 0.31 & & & & + @xmath171112 & 3.12 & 0.31 & & & & + it is interesting to compare the current results with the ones obtained in other models . the results of skyrme dft calculations of ref . @xcite for the @xmath124 isotones show that the impact of triaxiality on the inner fission barrier is small for @xmath140 , but increases with increasing z ( see fig . 4 in ref . the lowering of the inner fission barrier due to triaxiality is around 2 mev for @xmath123 and exceeds 3 mev for @xmath176 . in the etfsi model calculations @xcite the inner fission barriers are lowered due to triaxiality in the nuclei @xmath177 and @xmath178 by 0.5 and 1.1 mev , respectively . in the macroscopic+microscopic calculations of ref . @xcite the largest reduction of the inner barrier height due to triaxiality is about 2 mev and it appears in the region around @xmath179 , @xmath180 ( see fig . 2 in ref . @xcite ) . as discussed in ref . @xcite on the example of actinide nuclei , the reduction of the inner fission barrier height due to triaxiality is caused by the level densities in the vicinity of the fermi level which are lower at triaxial shape as compared with axial one . the different location of the `` magic '' shell gaps in superheavy nuclei in the macroscopic+microscopic model ( at @xmath139 , @xmath124 ) , in skyrme dft ( predominantly at @xmath176 , @xmath124 ) and in cdft ( at @xmath123 , @xmath158 ) results in different deformed single - particle structures at the deformations typical for the saddles of the inner fission barrier . this is one of the sources of the differences in the predictions of different models . we have to keep in mind , however , that we use in this investigation a seniority zero pairing force with a fixed cutoff energy of @xmath181 mev . it has been shown in ref . @xcite that pairing correlations play an important role for the calculation of the fission barriers and that even if the strength of the pairing force is adjusted to experimental gap parameters at the ground state , the range of the pairing force has an influence on the height of the barrier . this means for zero range forces and for the seniority zero forces that the barriers depend on the cutoff energy . it also has been shown in ref . @xcite that the parameter set dd - me2 in connection with the finite range gogny force d1s in the pairing channel produces in axial symmetric calculations in most of the actinide nuclei inner barriers which are too high as compared with experiment ( see fig . 9 of ref . @xcite ) and that it produces for the superheavy elements with @xmath148 112 , 114 , and 116 , where experimental estimates are available @xcite in axially symmetric calculations barriers which are in rather good agreement with those values ( see fig . 10 of ref . @xcite ) . of course , so far , there exist no relativistic triaxial calculations with the finite range gogny force in the pairing channel and full gogny calculations @xcite are hard to compare because of the different spin - orbit force used in this model . we have , however to keep in mind , that triaxiality reduces the barriers in the actinides , but not in the superheavy elements with @xmath148 112 , 114 , and 116 . therefore , at the moment , we can only conclude that for a final comparison with the experimental data we have to wait for full rhb calculations or at least rmf+bcs calculations with the finite range gogny force in the pairing channel . investigations in this direction are in progress . it is well known that reflection - asymmetric ( octupole deformed ) shapes become important at the second fission barrier and beyond ( see refs . @xcite and references therein ) . our calculations indicate that triaxiality can play a similarly important role at the outer fission barriers . usually , the triaxiality of these barriers is not mentioned in the publications . to our knowledge , it is only discussed in ref . @xcite that non - axial degrees of freedom play an important role in the description of outer fission barriers of actinides . [ ofb ] compares the energies of outer barrier saddle points as obtained in axial reflection symmetric ( a ) , triaxial ( t ) and axial reflection asymmetric [ octupole deformed ] ( o ) calculations . one can see that the inclusion of triaxiality or octupole deformation always lowers the outer fission barrier . the underlying shell structure clearly defines which of the saddle points ( triaxial or octupole deformed ) is lower in energy . for example , the lowest saddle point is obtained in triaxial calculations in proton - rich nuclei ( @xmath182 ) . on the contrary , the lowest saddle point is obtained in octupole deformed calculations in neutron - rich nuclei ( @xmath183 ) . note that the decrease of the saddle point energy by octupole deformation or triaxiality reaches 3 mev in some nuclei . thus , one can conclude that due to the structure of the pes in the fission path valley , we observe in the superheavy region an opposite situation as compared to actinide nuclei where the triaxiality has no impact on the outer fission barriers . our results also suggest that in some superheavy nuclei [ mostly in the nuclei where the deformation energy curves of the a and o calculations are similar in energy ] the combination of two deformations ( triaxiality and odd - multipole deformations ) may be important in the definition of the fission path for @xmath184 . the cdft calculations with both deformations included are at present not yet possible , but require further investigations . nuclei relative to spherical / weakly(normal)-deformed minima as a function of neutron number . the results of calculations of type a ( axial ) , t ( triaxial ) and o ( octupole ) obtained with the nl3 * parametrization are presented.,width=302 ] fig . [ oct - second ] shows that some nuclei are superdeformed in the ground state . a summary for such nuclei is given in table [ sd - summary ] . whether these states are stable or metastable should be defined by the height and the width of the outer fission barrier . the current calculations show that in many nuclei this barrier is appreciable in the axial calculations of type a. however , the inclusion of triaxial or octupole deformation decreases this barrier substantially so it is around 2 mev in the majority of the nuclei . this low barrier would translate into a high penetration probability for spontaneous fission , such that most likely these superdeformed states are metastable . calculations of the spontaneous fission half - lives from spherical / weakly - deformed and superdeformed minima are needed in order to decide which of these minima is more stable against fission . table [ sd - summary ] shows that the saddle point obtained in octupole deformed calculations ( o ) is the lowest in energy in most of the cases . only in proton - rich @xmath169120 and @xmath185118 nuclei , the saddle point of triaxial calculations ( t ) is the lowest . a superdeformed minimum exists also in the doubly magic nucleus @xmath100120 at a low excitation energy of approximately 0.6 mev ( see fig.[pot ] ) . this is definitely not an artifact of the model under consideration , since similar minimum exists also in axial relativistic hartree - bogoliubov calculations with the dd - me2 parametrization with gogny d1s forces in the pairing channel ( see fig . 8 in ref.@xcite ) . its calculated excitation energy depends on the actual strength of pairing and varies between 0 and 2 mev . it is interesting to compare our results for the structure of the outer fission barriers with those obtained in the axial rmf+bcs model ( with and without reflection symmetry ) of ref . @xcite which employed the nl3 and nl - z2 parametrizations . similar to our calculations , the superdeformed minima exist in the calculations of ref . @xcite ( see fig . 5 in this reference ) without octupole deformation . however , in the calculations of ref . @xcite with the nl - z2 parametrization , the heights of outer fission barriers with respect to the sd minimum are lower by approximately 2 mev as compared to our calculations . as a consequence , the inclusion of octupole deformation completely eliminates the outer fission barriers leading to results contradicting ours . on the other hand , the results of the calculations of ref . @xcite ( see fig . 6 in this reference ) for a few selected nuclei based on the nl3 parametrization are very similar to ours , namely , the ground state is superdeformed and the outer fission barrier has a height of approximately 2.5 mev . this result is not surprising considering that the nl3 * parametrization is very similar to nl3 @xcite . however , the results of the calculations for a few selected @xmath150 and 116 nuclei shown in figs . [ pot3a ] and [ pot3b ] reveal that the outer fission barriers survive in the presence of octupole and triaxial deformation not only in the nl3 * parametrization but also in the dd - me2 and dd - pc1 parametrizations . on average , the height of outer fission barrier in these nuclei is around 2 mev . on the contrary , the calculations with the nl - z2 parametrization in ref . @xcite show that octupole deformation kills the outer fission barriers in these nuclei ( see fig . 5 in this reference ) . but for the nuclei @xmath167120 and @xmath166120 . note that for simplicity only the axial solution ( a ) is shown at @xmath186.,width=302 ] but for the nucleus @xmath163120 . in order to show the potential energy surface between the axial @xmath187 and triaxial ( @xmath162 ) hills in greater details the energy difference between two neighboring equipotential lines is set to 0.1 mev and only one fourth of deformation plane of fig . [ pes-2d-6panel ] is shown.,width=302 ] & @xmath188 & @xmath189 + 1 & 2 & 3 & 4 & 5 + + @xmath190 & 4.84 & 4.52 & 2.75 & * 1.62 * + @xmath191 & 4.26 & 4.36 & 3.00 & * 1.82 * + @xmath192 & 3.48 & 4.49 & 3.57 & * 2.07 * + @xmath193 & 2.89 & 5.30 & 3.65 & * 2.55 * + @xmath194 & 2.60 & 6.18 & 3.82 & * 2.61 * + @xmath195 & 1.99 & 5.71 & 3.38 & * 2.25 * + @xmath196 & 1.72 & 4.87 & 2.94 & * 2.52 * + @xmath197 & 0.62 & 4.63 & * 2.24 * & 2.71 + + @xmath198 & 2.68 & 2.41 & 2.14 & * 1.31 * + @xmath199 & 2.92 & 3.58 & 2.42 & * 1.42 * + @xmath200 & 1.78 & 5.16 & 3.40 & * 2.20 * + @xmath201 & 1.70 & 5.34 & 3.54 & * 2.34 * + @xmath202 & 1.48 & 5.17 & 3.15 & * 2.26 * + @xmath203 & 1.15 & 4.99 & 2.82 & * 2.35 * + @xmath204 & 0.69 & 4.58 & * 2.28 * & 2.59 + @xmath205 & 0.40 & 4.03 & * 1.61 * & 2.60 + + @xmath206 & 0.92 & 2.76 & 2.60 & * 1.75 * + @xmath207 & 0.83 & 3.98 & 2.95 & * 1.88 * + we have carried out first systematic investigations of fission barriers in even - even superheavy nuclei with @xmath0 within covariant density functional theory including triaxial shapes with d2-symmetry and octupole shapes with axial symmetry . three different classes of models with the state - of - the - art parameterizations nl3 * , dd - me2 and dd - pc1 were used in the calculations . pairing correlations are taken into account in the bcs approximation using seniority pairing forces adjusted to empirical values of the gap parameters . the following conclusions have been obtained : * the low-@xmath127 and low-@xmath153 nuclei in this region are characterized by axially symmetric inner fission barriers . the increase of the particle numbers leads to a softening of the potential energy surfaces in the triaxial plane . as a result , several competing fission pathes in the region of inner fission barrier emerge in some of the nuclei . their importance in spontaneous fission can be defined only by taking into account the fission dynamics more seriously . however , the results of the calculations within the macroscopic+microscopic method @xcite suggests that the tr - b path ( and even maybe tr - a one ) may not be so important since they are substantially longer as compared with the axially symmetric path because it leads to a significant reduction of the penetration probability . * triaxiality lowers the outer fission barriers by 1.5 - 3 mev in reflection symmetric calculations . in many nuclei the lowering due to triaxiality is even more important than the one due to octupole deformation , which known to be important for the outer fission barriers and beyond from previous calculations ( see ref . @xcite and references therein ) . the underlying shell structure clearly defines that the triaxial [ octupole ] saddle is lower in energy for proton - rich nuclei with @xmath182 [ neutron - rich nuclei with @xmath183 ] . * on average , inner and outer fission barriers obtained for the nl3 * and dd - pc1 parametrizations are similar . on the contrary , the dd - me2 parametrization produces barriers which are by 1 - 1.5 mev higher than the ones obtained with nl3 * and dd - pc1 . * the superdeformed minimum is the lowest in energy in some of these nuclei . in the present calculations , the outer fission barriers with respect to these minima are about 2 mev high . both minima and barriers are present in all three classes of the cdft models . it has to be investigated , however , if these superdeformed minima are stable with respects to more general deformations not taken into account so far . the comparison of our results with those of non - relativistic models clearly shows that cdft predictions for the heights of inner fission barriers in the superheavy region with a seniority zero pairing force still remain on the lower end among nuclear structure models used so far . only axially symmetric calculations with the finite range gogny force d1s in the pairing channel can reproduce the estimates of inner fission barrier heights of ref . @xcite . considering the uncertainties of these estimates , the investigation of other experimental observables that strongly depend on fission barrier heights , especially fission half - lives and @xmath209- or electron capture delayed fission , is needed . the work in this direction is in progress . this work has been supported by the u.s . department of energy under the grant de - fg02 - 07er41459 , by a travel grant to justipen ( japan - us theory institute for physics with exotic nuclei ) under u. s. department of energy grant de - fg02 - 06er41407 and by the dfg cluster of excellence origin and structure of the universe ( www.universe-cluster.de ) .

in this paper we consider the effects of top quark compositeness on the electroweak parameters and and the coupling . we do so by using an effective field theory analysis to identify several promising patterns of mixing between sm - like and vector fermions , and then analyzing simple extensions of the standard model that realize those patterns . these models illustrate four ways in which an extended symmetry , which controls the size of radiative corrections to the observables discussed , may be broken . these models may also be viewed as highly - deconstructed versions of five - dimensional gauge theories dual to various strongly - interacting composite higgs theories . we comment on how our results relate to extra - dimensional models previously considered , and we demonstrate that one pattern of breaking is phenomenologically favored .

1105.5437

new strong interactions are a natural possibility for the dynamics underlying electroweak symmetry breaking . in analogy with the ads / cft correspondence @xcite , such dynamics are expected to have a weakly - coupled dual description in terms of a compactified five - dimensional gauge theory . in the dual five - dimensional description , the ordinary electroweak gauge - bosons are understood as the lightest kaluza - klein " resonances of the five - dimensional gauge theory , whose light masses arise either through the boundary conditions imposed on the five - dimensional gauge fields , as in the case of higgsless models @xcite , or through the vacuum expectation value of a composite scalar higgs @xcite . in both higgsless @xcite and composite higgs models @xcite , the observed mass - eigenstate fermions result from mixing between two kinds of gauge - eigenstate fermions : a set with quantum numbers resembling those of standard model fermions and a new set of vector fermions @xcite . the first set are elementary ( non - composite ) fermion fields that are only weakly coupled to the new strong dynamics , and they correspond to brane - localized " states that are largely confined to the ultraviolet boundary of the compactified space . in contrast , the new vector fermions are kaluza - klein resonances arising from the bulk fields `` of the compactified five - dimensional theory , and in the dual four - dimensional theory they correspond to a tower of composite states arising from the underlying strong dynamics . the fermions observed in experiment correspond to the lighter mass eigenstates resulting from this mixing , and they are mostly composed of elementary fields with a smaller admixture of vector states ; we will refer to these as ' ' ordinary fermions " . the heavier partner mass eigenstates , which have not yet been observed , are predominantly composed of the composite , vectorial states ; we will refer to these as heavy vector fermions " . their mass scale @xmath4 is typically in the @xmath5 gev range . the composition of the lighter mass eigenstates will affect their properties . consider , for instance , an elementary brane - localized fermion @xmath6 , whose left - handed component is a member of the weak doublet @xmath7 , and whose right - handed component is an electroweak singlet . we will denote by @xmath8 and @xmath9 , respectively , the degree of mixing of the weak doublet @xmath7 and weak singlet @xmath10 fields with the new vector fermions @xmath11 bearing the same lorentz and electroweak quantum numbers . then , as suggested by fig . [ fig : mass - fig ] , the mass of the lighter ordinary fermion " mass eigenstate resulting from the mixing will be of order @xmath12 . it is then clear that in order for the light flavors of the ordinary fermions to receive their appropriate masses @xmath8 and/or @xmath9 must be quite small , whereas for the top quark neither @xmath8 nor @xmath9 can be too small if @xmath13 is to have its observed value . hence , from the four - dimensional point of view , the light ordinary fermions will be essentially elementary , while the top - quark must be substantially composite . and composite vector fermions @xmath11 , and a yukawa coupling between @xmath14 and @xmath10 is mediated by scalar @xmath15 . as discussed in the text , the mass of the lighter mass eigenstate will depend on the dirac mass @xmath4 of the vector fermions , and the mixing factors @xmath16.,width=377,height=151 ] in these models , the hierarchy of ordinary fermion masses is then transferred to a hierarchy of the values of the mixing factors @xmath17 . in models with a warped extra - dimension , a possible explanation for such a hierarchy , results from the combination of the exponential factor in the ads@xmath18 metric , along with different fermion profiles in the bulk @xcite . we will assume that the strong electroweak breaking dynamics incorporates either minimal @xcite or next - to - minimal @xcite flavor violation , and therefore that non - sm contributions of these strong dynamics to flavor - changing neutral currents are sufficiently suppressed to avoid conflict with experiment . fermion compositeness can yield significant corrections to low - energy observables ; some are beneficial and others are problematic . in higgsless models , for example , the presence of vector fermions with sm quantum numbers can cancel contributions to the @xmath19 parameter arising from the extended gauge sector @xcite . additional effects are expected to arise predominantly from the top sector , where the mixing factors @xmath20 are largest . two electroweak quantities are particularly sensitive to effects in the top - sector : the @xmath2 coupling , @xmath21 , and the deviation in the ratio of the @xmath22- and @xmath23-boson masses from that predicted in the sm , also known as the electroweak @xmath0 parameter @xcite . potential contributions to both parameters can be understood in terms of the ( approximate ) global symmetry structure of the new strong dynamics . just as in the sm , contributions to @xmath0 can be suppressed if the symmetry breaking sector has an approximate @xmath24 symmetry which , via electroweak symmetry breaking , breaks to a diagonal custodial @xmath25 symmetry @xcite . furthermore , as shown by agashe _ et . al . _ @xcite , corrections to the @xmath2 coupling can be suppressed if custodial symmetry is extended to include a left - right parity symmetry @xmath26 whose action consists in exchanging @xmath27 and @xmath28 charges and making @xmath29 an eigenstate of @xmath26 . in this case , the required overall symmetry structure is @xmath30 . due to the mass splitting between the top- and bottom - quark , neither the conventional @xmath25 nor the extended @xmath31 symmetry can be exact , even in the limit of zero hypercharge . @xmath25 requires that the top and bottom quarks form a doublet , and hence @xmath32 in the symmetry limit . @xmath31 symmetry requires that the top and bottom quark masses both equal that of an additional exotic quark of charge + 5/3 , with which they form a triplet @xcite . in the sm , the dimension - four top - quark yukawa coupling @xmath33 breaks @xmath25 weakly , and is responsible for transmitting electroweak symmetry breaking to the top - quark sector . therefore , the leading sm corrections to both @xmath0 and @xmath21 are proportional to @xmath34 . in models in which the top - quark is composite , _ i.e. _ has a large vector fermion component , the vector fermions can be members of custodial singlets , doublets , or triplets . however , unlike in the sm and simple generalizations thereof @xcite , both positive and negative corrections to @xmath0 and @xmath21 can be generated . this is both dangerous and interesting . it is dangerous because the sm predictions for @xmath0 and @xmath21 are in agreement with experiment . it is interesting because the agreement is not perfect : the sm prediction for @xmath21 is about @xmath35 below the measured value ( with weak dependence on the higgs mass ) @xcite , whereas the sm prediction for @xmath0 is in full agreement with experiment for a light higgs ( @xmath36 gev ) but is @xmath35 below for a heavy higgs ( @xmath37 gev ) @xcite . moreover , the measured left - handed and right - handed @xmath38 couplings have a strong and positive correlation , and the sm predictions are both approximately @xmath35 below their expectation values . in this paper we consider the effect of top compositeness on @xmath0 and @xmath21 , and discuss the patterns of top compositeness that can yield phenomenologically viable models . to begin , we examine the low - energy operators induced @xcite when the heavy vector fermion states are integrated out " at tree - level , and correlate the phenomenological properties of these operators with the custodial quantum numbers of these fermions . in sections iii and iv , we construct simple models that illustrate these effects by extending the sm via the addition of one weak - charged vector fermion multiplet mixing with the left - handed elementary top - bottom doublet , and/or one weak - singlet fector fermion mixing with the elementary right - handed top . these models may be considered as highly deconstructed versions @xcite of full five - dimensional duals @xcite to various underlying strongly - interacting composite higgs theories : deconstructions that de - scope the theory to include only one non - standard kaluza - klein fermion level . taken together , the simple models discussed illustrate the various ways in which the third - generation fermion masses arise in any strongly - interacting composite higgs theory with a weakly coupled five - dimensional dual . finally , in section [ sec : conclusions ] we compare our results to previous calculations in five - dimensional models , note what happens if the right - handed top is mixes with a triplet state , and present our conclusions . in this section we use an effective field theory analysis of top - quark compositeness to understand how integrating out heavy vector fermion states with different quantum numbers correlate with likely phenomenological effects on @xmath0 and @xmath39 . we will consider , in turn , the effects of mixing of new vector fermions with each of the third - generation quark states : the top - bottom doublet , the right - handed top , and the right - handed bottom quark , in situations with and without an overall custodial symmetry . we will catalog ] where @xmath33 is a yukawa coupling and @xmath4 is the mass scale of the heavy vector fermions . the leading effects on both @xmath0 and @xmath39 may then be computed in the low - energy effective field theory @xcite . each of the operators we discuss can be recast in this language for particular choices of the coefficients @xmath40 . ] . the operators that arise from integrating out the heavy vector fermions and identify which are most likely to have significant effects . then , in the following sections of the paper , we will explore those operators more fully by constructing models whose low - energy effective theories give rise to them . to set the stage , let us review a scenario from @xcite where the lagrangian terms that include the heavy vector fermion @xmath11 take the form @xmath41 where @xmath15 is a scalar coupling @xmath14 to @xmath10 . requiring the variation of @xmath42 with respect to @xmath43 to vanish yields the equations of motion @xmath44 which we may solve iteratively in @xmath45 . doing so , we find @xmath46 plugging these expressions into eq . ( [ eq : simplelag ] ) , we obtain the following non - sm operator in the low - energy effective theory @xmath47 where subsequent terms are suppressed by higher powers of @xmath48 . in unitary gauge this term gives rise to an anomalous " coupling of the @xmath23-boson to top - quarks , @xmath49 as sketched in fig . [ fig : ztt - fig ] . since the operator s effects on observables are governed by @xmath50 , they are related to the size of the top - quark mass and therefore are potentially large . mixing with the left - handed top quark can give rise to new effects on the @xmath51 vertex in the low - energy effective theory.,width=302,height=170 ] in fact , we may systematically catalog the operators that can arise from integrating out each of the possible kinds of heavy vector fermions and think about their characteristic phenomenologies when a left - handed state mixing with the left - handed top - bottom doublet has been integrated out , either or both of the following operators may result @xmath52 where @xmath53 . note that @xmath54 transforms as a @xmath55 under the electroweak interactions , like the sm left - handed top - bottom doublet ; the operator ( [ eq : smtl - op ] ) containing this combination of fields arises when the vector fermion @xmath11 state being integrated out carries those sm - like quantum numbers . models in the literature that contain such operators include @xcite . likewise , @xmath56 transforms like an exotic @xmath57 and the associated operator ( [ eq : extl - op ] ) results when the new vector fermion has those exotic quantum numbers ; related models include @xcite . both of these operators will affect @xmath58 at one loop because they alter the @xmath23 but not the @xmath22 propagator ( since the @xmath22 does not couple to @xmath10 ) . neither affects @xmath59 at tree level since they alter the @xmath60 vertex rather than the @xmath61 vertex ; however they could affect @xmath59 at one loop through diagrams with an internal top quark coupled to the decaying @xmath23 boson . both of these operators will be worth exploring further . when a right - handed state mixing with @xmath10 has been integrated out , as in @xcite , one obtains @xmath62 here , @xmath63 has the quantum numbers of the sm @xmath10 quark . this operator will affect both the @xmath64 and @xmath65 couplings ( and may therefore impact single - top production ) . because it does not also affect the @xmath66 coupling , it _ will _ alter @xmath58 at one loop but _ will not _ affect @xmath59 at tree level ; it should generally impact @xmath59 at the one - loop level through diagrams with an internal top quark . this operator also merits further study . in contrast , if a right - handed state mixing with @xmath67 is integrated out , the resulting operator @xmath68 includes @xmath69 which has the quantum numbers of the sm @xmath67 . this operator affects the @xmath70 vertex at tree level and , as such , is very tightly constrained . therefore , this operator does not warrant further study at present . finally , if custodial and flavor symmetry protect the interactions of the sm - like top - bottom doublet , then integrating out the states mixing with @xmath10 and @xmath67 , as in @xcite , yields the custodially - symmetric operator @xmath71 in which @xmath72 and @xmath73 . this operator does not contribute to @xmath58 at all . in the presence of a flavor symmetry , the operator will not alter @xmath59 either @xcite . if one , instead , integrates out heavy vector states mixing with @xmath74 , the resulting operator @xmath75 encodes isospin breaking in the yukawa coupling matrix @xmath76 . this will affect the @xmath60 and @xmath77 couplings with the latter effect suppressed by @xmath78 to a level that is unilkely to be interesting . a @xmath79 coupling can be induced , as occurs in the 3-site model , but this is again suppressed by @xmath80 and is unlikely to yield interesting limits . we will not explore either of these operators further . our discussion has identified three operators of potential phenomenological interest @xmath81 , and shown that they arise from mixing between fundamental top and bottom quarks with particular kinds of ( potentially composite ) vector fermions . in the next sections of the paper , we will introduce specific models that explore the ideas raised by our effective field theory analysis . section [ sec : explicit ] proposes a pair of illustrative models in which the yukawa interactions explicitly break the @xmath25 custodial symmetry of the symmetry breaking sector . in the first model the right - handed top is mixed with a heavy vector singlet , so that this model explores operator ( [ eq : smtr - op ] ) , whereas in the second model the left - handed top - bottom doublet is mixed with a heavy vector doublet so that one instead explores operator ( [ eq : smtl - op ] ) . each of these models include all terms consistent with the gauge symmetries , and each is a renormalizable four - dimensional gauge theory . we will find that neither kind of mixing leads , on its own , to a more viable phenomenology and this implies that models introducing both kinds of mixing at once will be similarly unsuccessful . in section [ sec : soft ] we introduce models incorporating a full @xmath82 symmetry in the yukawa sector ; again , each model includes all terms consistent with the gauge symmetries . electroweak gauge interactions do not respect the @xmath83 global symmetries assumed for the yukawa couplings . however , we find that these effects are negligible so long as the underlying strong dynamics respects @xmath83 . ] we first consider a model in which the left - handed top - bottom doublet is mixed with a heavy vector bi - doublet of @xmath84 ; integrating out the heavy vectors would give rise to the third interesting operator identified in our effective field theory analysis ( [ eq : extl - op ] ) . we find that this particular model has limitations , but also suggests a possible solution . we then instantiate that solution in a particularly economical way by embedding the top - bottom left - handed doublet itself in an @xmath84 bi - doublet @xcite , @xmath85 , while mixing the right - handed top with a heavy vector singlet . for each of the models considered in this section , in the limit of zero gauge couplings , the @xmath3 symmetry is softly broken due to a dirac mass term for the vector fermions . moreover , these two models contain @xmath31 triplet fermions including states of charge @xmath86 and @xmath87 which , after electroweak symmetry breaking , mix with the top - quark . we find that only models with these custodial triplet fermions feature both positive and negative contributions to @xmath0 and @xmath21 , leading to regions of the parameter space in which agreement with experiment is at the @xmath88 level . in this section we examine a pair of scenarios in which the custodial symmetry is explicitly broken by yukawa couplings . we will find that neither of these directions in model building , on their own , provides improved agreement with data on @xmath0 and @xmath21 , compared with the sm . this is because in both models the corrections to both parameters turn out to be uniformly positive and correlated with one another . a vector - like custodial / weak singlet fermion @xmath91 has the same charges as the right - handed top of the sm . adding such a field to the top - sector lagrangian gives @xmath92 where @xmath93 is the elementary left - handed top - bottom doublet , @xmath15 is the @xmath94 higgs doublet , and @xmath53 is the @xmath95 version of @xmath15 . while a yukawa term mixing @xmath96 with @xmath97 is also possible , it can always be removed by redefining @xmath97 and @xmath98 . for our analysis the most relevant feature of this lagrangian is the ( explicit ) hard breaking of custodial symmetry in the yukawa sector ( i.e. the fact that @xmath99 ) . in fact , because the right - handed bottom quark does not contribute significantly to any isospin - violating processes , we will simply set @xmath100 and ignore the right - handed bottom quark altogether . note that in the effective theory resulting from taking the limit @xmath101 in eq . ( [ eq : lsinglet ] ) we generate an operator of the form @xmath102 , that is , an operator of the form of @xmath90 in eq . ( [ eq : smtr - op ] ) . in other words , below the mass of the partner fermion state , the effective theory includes both sm physics and additional effects from @xmath90 . the mass terms in eq . ( [ eq : lsinglet ] ) may be written in matrix notation as @xmath103 where @xmath104 diagonalization shows how to rewrite the gauge eigenstates in terms of the mass eigenstates : @xmath105 where @xmath6 and @xmath106 are , respectively , the ordinary top and its heavy partner , and the mixing factors are : @xmath107^{-1/2 } \ , \quad \sin\theta_r = \left[1+\left(\tan\beta-\frac{\cot\beta\ m_t^2}{m_t^2}\right)^{-2}\right]^{-1/2 } \ .\end{aligned}\ ] ] these rotation angles are conveniently expressed in terms of @xmath108 , which measures the amount of mixing of @xmath97 with the vector fermion @xmath91 : @xmath109 we have also eliminated the parameter @xmath110 in favor of the physical top - quark mass , @xmath13 : @xmath111 in terms of @xmath108 , @xmath112 , and @xmath13 , the mass of the heavy @xmath106 state is : @xmath113 the sm limit of this theory is obtained by sending @xmath114 to infinity , independent of @xmath112 , and hence sending @xmath115 ; in this limit , the left - handed top no longer mixes with new states so the light top eigenstate behaves like the sm top . ( left ) and @xmath39 ( right ) in the model of eq . ( [ eq : lsinglet ] ) . the thin to thick curves are for @xmath116 0.5 , 1 , and 2 tev , respectively , whereas the dashed lines are the sm predictions . in the plot for @xmath0 the lower ( upper ) experimental @xmath88 band is for @xmath36 gev ( @xmath37 gev ) @xcite . the experimental @xmath88 band for @xmath39 has a negligible dependence on the higgs mass @xcite . , title="fig:",width=264,height=188 ] ( left ) and @xmath39 ( right ) in the model of eq . ( [ eq : lsinglet ] ) . the thin to thick curves are for @xmath116 0.5 , 1 , and 2 tev , respectively , whereas the dashed lines are the sm predictions . in the plot for @xmath0 the lower ( upper ) experimental @xmath88 band is for @xmath36 gev ( @xmath37 gev ) @xcite . the experimental @xmath88 band for @xmath39 has a negligible dependence on the higgs mass @xcite . , title="fig:",width=264,height=188 ] with the fermion fields diagonalized it is straightforward to compute the dominant contributions to @xmath0 and @xmath39 . the corrections to @xmath0 arise from the diagrams of fig . [ fig : diagrams](a ) , where @xmath117 and @xmath118 are up - type and down - type fermions , respectively , in the top sector . in the present case @xmath119 and @xmath120 . the dominant corrections to @xmath21 can be computed in the gaugeless limit @xcite , and are given by the diagram of fig . [ fig : diagrams](b ) where @xmath121 and @xmath122 are @xmath123 . the values of @xmath0 and @xmath39 ( defined coupling to left - handed @xmath124 quarks is written as @xmath125 . ] as the new - physics contribution to @xmath21 ) turn out to be positive and correlated ; we find , in agreement with @xcite , the expressions : @xmath126 \ , \label{eq : singlet}\ ] ] where @xmath127 the contribution to the @xmath1 parameter is also positive ( for @xmath128 gev ) , but numerically much smaller than @xmath0 : @xmath129 \ . \label{eq : shatsinglet}\ ] ] in the left - hand pane of fig . [ fig : plotsinglet ] we plot @xmath0 as a function of @xmath130 for @xmath116 0.5 , 1 , and 2 tev ( thin to thick curves , respectively ) , together with the sm predictions ( dashed lines ) . the experimental value of @xmath0 depends on the higgs mass : the lower ( upper ) @xmath88 band is for @xmath36 gev ( @xmath37 gev ) . these bounds are taken for an arbitrary value of @xmath1 : the latter is much smaller than @xmath0 , and within @xmath88 for both light and heavy higgs . we see that new - physics contributions to @xmath0 in this model reduce agreement with the light - higgs limit of the sm and favor a heavier higgs . the right - hand pane of fig . [ fig : plotsinglet ] shows @xmath39 as a function of @xmath130 for the same @xmath112 as before ; here , the experimental value is nearly independent of @xmath131 . the positive correction to @xmath132 in eq . ( [ eq : singlet ] ) does tend to push @xmath21 in the direction required for better agreement with experiment . however , as alluded to above , the values of @xmath130 for which @xmath132 is within @xmath88 of the data disagree with @xmath0 for a light higgs by more than @xmath35 . and while those values of @xmath130 do yield agreement at the @xmath35 level for @xmath0 if a heavy higgs is assumed , overall the model does not surpass the sm in its agreement with the broader electroweak precision data . the chief difficulty lies with @xmath133 . in this kind of model , the large @xmath13 - @xmath134 splitting arises from the fact that @xmath135 ; and we have worked in the limit where @xmath136 is simply set to zero . so in general , this kind of model would predict @xmath137 and in the limit we adopt , @xmath138 . however , as mentioned earlier , the measured left - handed and right - handed @xmath38 couplings have a strong and positive correlation , and the sm predictions are both approximately @xmath35 below their expectation values . therefore , a positive correction to @xmath21 without a corresponding positive correction to @xmath139 can not restore agreement with experiment . a vector - like custodial / weak doublet , @xmath140 , has the same charges as the sm left - handed top - bottom doublet , here denoted by @xmath141 . including @xmath142 in the sm top - sector yields the lagrangian @xmath143 a yukawa term mixing @xmath96 and @xmath97 is also possible , but can always be removed by redefining @xmath96 and @xmath144 . as in the composite singlet model considered above , in this model the breaking of custodial @xmath3 in the yukawa sector is explicit . this time , in the effective theory resulting from taking the limit @xmath145 in eq . ( [ eq : ldoublet ] ) we generate an operator of the form @xmath146 , that is , an operator of the form of @xmath147 in eq . ( [ eq : smtl - op ] ) ; below the mass of the partner fermion state , the effective theory includes both sm physics and additional effects from @xmath147 . ( left ) and @xmath39 ( right ) in the model of eq . ( [ eq : ldoublet ] ) . the thin to thick curves are for @xmath148 0.5 , 1 , and 2 tev , respectively , whereas the dashed lines are the sm predictions . in the plot for @xmath0 the lower ( upper ) experimental @xmath88 band is for @xmath36 gev ( @xmath37 gev ) @xcite . the experimental @xmath88 band for @xmath39 has a negligible dependence on the higgs mass . @xcite , title="fig:",width=264,height=188 ] ( left ) and @xmath39 ( right ) in the model of eq . ( [ eq : ldoublet ] ) . the thin to thick curves are for @xmath148 0.5 , 1 , and 2 tev , respectively , whereas the dashed lines are the sm predictions . in the plot for @xmath0 the lower ( upper ) experimental @xmath88 band is for @xmath36 gev ( @xmath37 gev ) @xcite . the experimental @xmath88 band for @xmath39 has a negligible dependence on the higgs mass . @xcite , title="fig:",width=264,height=188 ] the mass lagrangian is @xmath149 where @xmath150 . diagonalization of the top fields is still expressed by eq . ( [ eq : diagtop ] ) , where now @xmath151 here @xmath152 measures the amount of mixing of @xmath96 with the new vector fermion @xmath142 , @xmath153 in deriving eq . ( [ eq : angles ] ) we have rewritten @xmath110 in terms of the top - quark mass @xmath13 , @xmath154 and in terms of those same variables , the heavy top mass is @xmath155 in this model , because the new vector multiplet includes a partner for the b - quark , @xmath156 ( the gauge eigenstate with sm quantum numbers ) mixes with the new vector fermion @xmath157 . the mass eigenstates are diagonalized by @xmath158 where the mass of the heavy @xmath159 quark is @xmath160 in this case , the sm limit corresponds to @xmath161 with finite @xmath162 ( regardless of the particular value of @xmath162 ) , which implies @xmath163 ; in this limit , the left - handed top no longer mixes with new states and the light eigenstate behaves like the sm top . the one - loop contributions to @xmath0 and @xmath39 are still given by the diagrams of fig . [ fig : diagrams](a ) and [ fig : diagrams](b ) , respectively , with @xmath164 , and @xmath165 . this yields @xmath166 \ , \\ & & \delta g_{lb } = \delta g_{lb}^{\rm sm}\ \log\frac{m_q/\cos\alpha}{m_t } \frac{\cos^4\alpha\ m_t^2}{\sin^2\alpha\ m_q^2}\left[1+{\cal o}\left(\frac{m_t^4}{m_q^4}\right)\right ] \ .\end{aligned}\ ] ] the @xmath1 parameter is , again , numerically much smaller than @xmath0 : @xmath167 \ .\ ] ] in fig . [ fig : plotdoublet ] we plot @xmath0 and @xmath39 , using the same notation as fig . [ fig : plotsinglet ] . the disagreement with data is even worse than in the composite singlet case because @xmath39 is still positive but grows more slowly with decreasing @xmath168 than @xmath0 does . values of @xmath168 for which @xmath39 is within @xmath88 of the data give values of @xmath0 that are several standard deviations larger than the experimental value for any @xmath131 . moreover , the problems with @xmath133 persist ; since @xmath100 the model predicts no shift in @xmath139 from the sm value . in this section , we explore models incorporating a softly - broken extended custodial symmetry . we show that adding a vector fermion bi - doublet to the sm - like fields introduces new contributions to @xmath21 that largely cancel one another , leaving @xmath21 close to the sm prediction . the bi - doublet contribution to @xmath0 is negative : as a consequence , both @xmath21 and @xmath0 lie below the experimental values . this is a different pattern than we found in the models of section [ sec : explicit ] , but is not satisfactory on its own . it is then natural to expect that including an additional vector - like singlet in an @xmath3 symmetric fashion can help to achieve agreement with experiment , since the singlet contributions to @xmath21 and @xmath0 are both positive , as found in section [ sec:2a ] . we will show that this is indeed the case , and that there are regions of the parameter space in which the agreement with experiment is at the @xmath88 level , for either a light or a heavy higgs . let @xmath170 be a heavy vector - like bi - doublet of @xmath84 , @xmath171 note that the field @xmath170 includes both @xmath31 singlet and triplet components , and that the field @xmath172 has electric charge @xmath173 and @xmath174 while the field @xmath175 has electric charge @xmath86 and @xmath87 @xcite . in addition , collect the higgs doublet @xmath15 together with its hypercharge conjugate @xmath176 in another @xmath84 bi - doublet : @xmath177 consider the following top - sector lagrangian that includes the new vector fermion @xmath170 , and the higgs bi - doublet @xmath178 along with the sm - like elementary doublet , @xmath179 and the sm - like elementary singlet @xmath97 : @xmath180 in the effective theory resulting from taking the limit @xmath145 in eq . ( [ eq : lsinglet ] ) we generate an operator of the form @xmath181 , that is , an operator of the form of @xmath169 in eq . ( [ eq : extl - op ] ) . in the limit of zero gauge couplings , and for @xmath182 , this lagrangian features an @xmath183 global symmetry , which is spontaneously broken to custodial @xmath184 . this symmetry is reflected in the identical @xmath33 coefficients for @xmath185 and @xmath186 . the @xmath26 transformations are @xmath187 where @xmath188 switching on @xmath189 introduces a source of soft @xmath3 breaking that does not generate @xmath3-breaking counterterms . in contrast , switching on the gauge interactions leads to a hard breaking of custodial @xmath82 ( for instance , the weak interaction does not respect @xmath26 ) and this generates @xmath3-breaking counterterms . we estimate the impact of this breaking pattern on the quantities of interest to us as follows . if the yukawa interactions are like those of eq . ( [ eq : lbidoublet1 ] ) at some higher - energy `` compositeness scale '' @xmath190 , then at the electroweak scale the coefficients of @xmath185 and @xmath186 will differ by a quantity @xmath191 which can be estimated to be of order @xmath192 for example , taking @xmath19310 tev gives a very small correction , @xmath194 . therefore , we can ignore the running of @xmath191 , and take custodial @xmath3 to be exact in the yukawa sector at the electroweak scale is not generated at loop level , because of a @xmath195 symmetry under which @xmath196 and @xmath197 . the latter is only broken softly by the @xmath189 term . ] . the mass lagrangian may be summarized in matrix form as @xmath198 where @xmath110 is still given by eq . ( [ eq : mthat ] ) . as usual , we can re - express @xmath110 in terms of the mass of the top quark . this gives @xmath199 where @xmath152 is still defined by eq . ( [ eq : alpha ] ) . this shows also that the inequality @xmath200 must be satisfied in order for the top quark mass to attain its observed value . here the sm limit is achieved by taking both @xmath161 and @xmath201 , with @xmath202 finite @xcite . the bottom quark fields are diagonalized as in eq . ( [ eq : bottom ] ) , with the heavy bottom mass still given by eq . ( [ eq : bmass ] ) . the top quark fields are diagonalized by @xmath203 matrices : @xmath204 the perturbative expressions ( in powers of @xmath205 ) we obtained for the heavy top masses and the rotation matrices @xmath206 and @xmath207 are straightforward to calculate , but lengthy and not informative to look at . ( left ) and @xmath39 ( right ) in the model of eq . ( [ eq : lbidoublet1 ] ) . the thin to thick curves are for @xmath148 0.5 , 1 , and 2 tev , respectively , whereas the dashed lines are the sm predictions . in the plot for @xmath0 the experimental @xmath88 band shown is for @xmath36 gev ; the band for larger @xmath131 lies at higher positive valuesof @xmath0 @xcite . the experimental @xmath88 band for @xmath39 has a negligible dependence on the higgs mass @xcite . the theoretical curves cut off at low values of @xmath168 , as required by the bound from eq . ( [ eq : boundalpha ] ) . , title="fig:",width=264,height=188 ] ( left ) and @xmath39 ( right ) in the model of eq . ( [ eq : lbidoublet1 ] ) . the thin to thick curves are for @xmath148 0.5 , 1 , and 2 tev , respectively , whereas the dashed lines are the sm predictions . in the plot for @xmath0 the experimental @xmath88 band shown is for @xmath36 gev ; the band for larger @xmath131 lies at higher positive valuesof @xmath0 @xcite . the experimental @xmath88 band for @xmath39 has a negligible dependence on the higgs mass @xcite . the theoretical curves cut off at low values of @xmath168 , as required by the bound from eq . ( [ eq : boundalpha ] ) . , title="fig:",width=264,height=188 ] the one - loop contribution to @xmath0 is given by the diagrams of fig . [ fig : diagrams](a ) , where in gauge eigenstate basis @xmath208 and @xmath209 . the correction to @xmath21 is given by the diagrams of fig.[fig : diagrams](b ) , where , in mass eigenstate basis , @xmath210 . a straightforward calculation yields @xmath211\frac{m_t^2}{m_q^2/\cos^2\alpha } + { \cal o}\left(\frac{m_t^4}{m_q^4}\right ) \label{eq : thatbidoublet1 } \\ \delta g_{lb}&=&\delta g_{lb}^{\rm sm}\bigg [ \left(\frac{5}{8\sin^2\alpha}-\cos^2\alpha-1\right)\log\frac{m_q/\cos\alpha}{m_t } -\frac{5}{8\sin^2\alpha}\log\frac{m_q}{m_t}+\frac{3}{8\sin^2\alpha}\log\frac{1}{\cos\alpha}\bigg]\frac{m_t^2}{m_q^2 } + { \cal o}\left(\frac{m_t^4}{m_q^4}\right ) \label{eq : glbbidoublet1}\end{aligned}\ ] ] the contribution to @xmath1 is @xmath212\frac{m_t^2}{m_q^2/\cos^2\alpha } \label{eq : shatbidoublet}\end{aligned}\ ] ] once again , the leading order contributions to @xmath1 and @xmath0 are both proportional to @xmath213 , and @xmath1 is numerically much smaller than @xmath0 . in this case the largest contributions to @xmath39 cancel , because of the custodial @xmath82 symmetry ; the remainder has a small and generally negative value . the dominant log contribution to @xmath0 is clearly _ negative _ , with a rather large numerical coefficient . as illustrated in fig . [ fig : plotbidoublet ] , a model based strictly on the present lagrangian , eq . ( [ eq : lbidoublet1 ] ) , does not agree with the data on its own ; some other physics yielding positive corrections to both @xmath0 and @xmath21 would be needed to restore agreement with data at the 1-sigma level . in principle , the situation could be improved by adding a heavy vector singlet to the present model ; this would not introduce sources of o(4 ) hard breaking and , as demonstrated in fig . [ fig : plotsinglet ] , it could provide the needed positive corrections . however , there is also a more economical alternative , which we explore in the next section of the paper . we have just seen that a heavy vector - like bi - doublet , mixed with the sm - like left - handed top - bottom weak doublet , gives a negative contribution to @xmath0 and a very small contribution to @xmath21 : this means that the predictions for @xmath0 and @xmath21 are both below the @xmath214 experimental bands . on the other hand , in section [ sec:2a ] we observed that adding a vector singlet gives positive ( and correlated ) corrections to both @xmath0 and @xmath21 . therefore the most obvious thing to do , in order to restore agreement with experiment , is to build a model with a vector bi - doublet @xmath170 ( mixed with the sm - like left - handed top - bottom doublet @xmath96 ) , a vector singlet @xmath91 ( mixed with the sm - like right - handed top singlet @xmath97 ) , and an @xmath3-symmetric yukawa interaction . in order to simplify our analysis we will let the mixing mass between @xmath96 and @xmath215 become infinite : this effectively removes @xmath96 and @xmath215 from the lagrangian , and promotes @xmath144 to the role of sm - like left - handed top - bottom doublet . in eq . ( [ eq : lbidoublet1 ] ) this is achieved by letting @xmath189 become infinite , whence @xmath163 . we would like to stress , however , that this is only assumed for the sake of simplicity : allowing for a finite @xmath189 does not qualitatively modify our results . in that context , let us consider a model in which a sm - like left - handed top - bottom doublet is part of a bi - doublet @xmath216 : @xmath217 unlike in the previously analyzed models , here the field playing the role of the sm - like fundamental top - bottom doublet is denoted @xmath144 and not @xmath96 , as consistent with the factors discussed above . as in the model of section [ sec:2a ] , the sm - like right - handed singlet , @xmath97 , is allowed to mix with a vector singlet @xmath91 . the structure of this model corresponds to the doublet extended " standard model of @xcite , augmented by the additional @xmath91 weak / custodial singlet field . imposing an unbroken @xmath3 symmetry in the yukawa sector leads to the top - sector lagrangian @xmath218 custodial @xmath3 is now softly broken by @xmath162 and , as in the model of eq . ( [ eq : lbidoublet1 ] ) , we can safely ignore the small @xmath3 breaking due to gauge interactions . here , the sm limit corresponds to taking @xmath219 and @xmath201 ; as in the previous section , the operator arising from integrating out the heavy states will have the form of @xmath169 in eq . ( [ eq : extl - op ] ) . the mass lagrangian is @xmath220 where @xmath110 is still given by eq . ( [ eq : mthat ] ) . as usual , we can rewrite @xmath110 in terms of the mass of the top quark . this gives @xmath221 whence @xmath222 the top quark fields are diagonalized by @xmath203 matrices : @xmath223 as in the heavy bi - doublet scenario of the last section , the perturbative expressions for masses and rotation matrices are straightforward but quite lengthy , and will not be shown in here . , at the @xmath88 level , in the @xmath224 plane , for @xmath225 . the top row ( bottom row ) graphs are for @xmath226 gev . only the white regions are allowed ; see text for details about the various excluded ( shaded , striped , hatched ) zones.,title="fig:",width=188,height=170 ] , at the @xmath88 level , in the @xmath224 plane , for @xmath225 . the top row ( bottom row ) graphs are for @xmath226 gev . only the white regions are allowed ; see text for details about the various excluded ( shaded , striped , hatched ) zones.,title="fig:",width=188,height=170 ] , at the @xmath88 level , in the @xmath224 plane , for @xmath225 . the top row ( bottom row ) graphs are for @xmath226 gev . only the white regions are allowed ; see text for details about the various excluded ( shaded , striped , hatched ) zones.,title="fig:",width=188,height=170 ] + , at the @xmath88 level , in the @xmath224 plane , for @xmath225 . the top row ( bottom row ) graphs are for @xmath226 gev . only the white regions are allowed ; see text for details about the various excluded ( shaded , striped , hatched ) zones.,title="fig:",width=188,height=170 ] , at the @xmath88 level , in the @xmath224 plane , for @xmath225 . the top row ( bottom row ) graphs are for @xmath226 gev . only the white regions are allowed ; see text for details about the various excluded ( shaded , striped , hatched ) zones.,title="fig:",width=188,height=170 ] , at the @xmath88 level , in the @xmath224 plane , for @xmath225 . the top row ( bottom row ) graphs are for @xmath226 gev . only the white regions are allowed ; see text for details about the various excluded ( shaded , striped , hatched ) zones.,title="fig:",width=188,height=170 ] the one - loop contribution to @xmath0 is given by the diagrams of fig . [ fig : diagrams](a ) , where in the gauge eigenstate basis @xmath227 and @xmath228 . the perturbative expressions are quite complicated , so we will only show the results for the limiting cases @xmath201 and @xmath229 , respectively : @xmath230 \ , \nonumber \\ \lim_{m_t\to\infty}\hat{t}&=&\hat{t}^{\rm sm}\left(-8\log\frac{m_q}{m_t}+\frac{22}{3}\right)\frac{m_t^2}{m_q^2 } \left[1+{\cal o}\left(\frac{m_t^4}{m_q^4}\right)\right ] \ .\end{aligned}\ ] ] the @xmath201 limit is positive , and agrees with eq . ( [ eq : singlet ] ) . the @xmath229 limit is negative , and agrees with eq . ( [ eq : thatbidoublet1 ] ) for @xmath163 . for the @xmath1 parameter we obtain @xmath231 \ , \nonumber \\ \lim_{m_t\to\infty}\hat{s}&=&-\frac{1}{96\pi^2}\left(8\log\frac{m_q}{m_t}-15\right)\frac{m_t^2}{m_q^2 } \left[1+{\cal o}\left(\frac{m_t^4}{m_q^4}\right)\right]\ , \end{aligned}\ ] ] respectively in agreement with eq . ( [ eq : shatsinglet ] ) and eq . ( [ eq : shatbidoublet ] ) with @xmath163 . once again , @xmath232 . the correction to @xmath21 is given by the diagrams of fig.[fig : diagrams](b ) , where , in the mass eigenstate basis , @xmath233 . since the perturbative expressions are complicated , we only show the @xmath201 and @xmath229 limits , respectively : @xmath234 \ , \nonumber \\ \lim_{m_t\to\infty}\delta g_{lb}&=&\delta g_{lb}^{\rm sm}\left(-\log\frac{m_q}{m_t}\right)\frac{m_t^2}{m_q^2 } \left[1+{\cal o}\left(\frac{m_t^4}{m_q^4}\right)\right ] \ .\end{aligned}\ ] ] the @xmath201 limit is positive , and agrees with eq . ( [ eq : singlet ] ) . the @xmath229 limit is small and negative , and agrees with eq . ( [ eq : glbbidoublet1 ] ) for @xmath163 . since there are now two heavy fermion mass scales ( @xmath162 associated with the bi - doublet and @xmath112 associated with the heavy singlet ) , we find convenient to plot the regions allowed by data on @xmath0 and @xmath21 in the plane defined by these mass scales . in fig . [ fig : constraints ] we show the allowed and excluded regions , at the @xmath88 level , in the @xmath224 plane , for @xmath225 . the top row graphs are for @xmath36 gev , whereas the bottom row graphs are for @xmath37 . the l - shaped striped region covering the lower left - corner ( thinner yellow - red stripes ) is excluded by the need to reproduce the measured top mass , as required by eq . ( [ eq : boundbeta ] ) . in the adjacent striped region ( thicker , blue - brown stripes ) the top yukawa coupling is larger than @xmath235 : this is mainly possible for small values of @xmath130 , as shown by eq . ( [ eq : mthatmt ] ) and eq . ( [ eq : mthat ] ) . the purple monochrome shaded region is excluded by constraints on @xmath0 : in the upper left region @xmath0 is large and positive , in the lower right @xmath0 is large and negative , whereas the curve centered between the boundaries of the purple monochrome shaded regions traces the locus of the expectation value of @xmath0 . the yellow hatched region is excluded by constraints on @xmath39 : in the left region @xmath39 is large and positive , in the right region @xmath39 is very small , and the curve centered between the boundaries of the yellow hatched regions traces the locus of the expectation value of @xmath39 . finally , the white regions are consistent with all constraints . we see that , for low values of @xmath130 , heavy vector fermions with masses of order 2 tev or more are allowed . note also that if we , alternatively , assume that the model includes no new corrections to @xmath139 beyond those in the sm , then regions of fig . 5 where @xmath236 would , instead , be acceptable at the 99% cl ( these regions lie within the yellow hatched spaces to the right of the white regions in each pane of fig . again , the heavy vector fermions would have masses of at least 2 tev . in this paper we analyzed the effects of top - quark compositeness resulting from mixing between fundamental fields with sm - like quantum numbers and new composite vector fermions . after reviewing the operators that are generated when heavy mostly - vector states are integrated out and identifying which would have the most significant impacts on low - energy observables , we constructed explicit models whose low - energy effective lagrangians include those operators and studied the phenomenology in more detail . our analysis focused on the observables @xmath0 and @xmath21 , which are sensitive to operators that break the extended custodial @xmath3 symmetry , while also commenting on the contributions of the new physics to @xmath1 . we considered two scenarios of @xmath3 breaking : explicit breaking in the yukawa interactions and soft breaking in mixing mass terms . our models can be seen as highly deconstructed versions of extra - dimensional dual models featuring the same global symmetries . in that language , we have studied the effects of the composite fermion sector only , and neglected the composite gauge sector . this is internally consistent , since the contribution from gauge kaluza - klein modes to @xmath0 and @xmath21 is suppressed in models with custodial symmetry @xcite . we found that models with explicit @xmath3 breaking in the yukawa sector are disfavored by experiment , as the contributions to @xmath0 become too large in regions of the parameter space in which @xmath21 is within the @xmath88 bounds . these models correspond to two of the scenarios that our effective field theory analysis had initially identified as potentially interesting . when the right - handed top is composite and , in the limit @xmath101 in eq . ( [ eq : lsinglet ] ) the low - energy effective lagrangian includes an operator of the @xmath90 form @xmath102 . on the other hand , when the left - handed top - bottom doublet is composite and , in the limit @xmath237 ( with @xmath202 fixed ) in eq . ( [ eq : ldoublet ] ) one generates an operator of the @xmath147 form @xmath238 . unfortunately , the presence of either of these operators leads to correlated positive contributions to @xmath0 and @xmath39 that are inconsistent with the data if no other new physics is present . in models with soft @xmath3 breaking , the left - handed top - bottom doublet is mixed with a vectorial quark bi - doublet of @xmath84 . the two @xmath27 doublets embedded in the bi - doublet give canceling contributions to @xmath21 , and an overall negative contribution to @xmath0 : this agrees with the findings of refs . @xcite , where bulk fermion bi - doublets mix with the boundary - localized sm left - handed weak doublets . adding an extra vectorial singlet provides positive and correlated corrections to @xmath0 and @xmath21 that can improve the agreement between model and data : analogous results are found in extra - dimensional models with bulk singlets mixing with the brane - localized sm right - handed weak singlets @xcite . therefore , models of top - compositeness with soft @xmath3 breaking feature a combination of positive and negative contributions to @xmath0 and @xmath21 , and agreement with experiment , at the @xmath88 level , is possible for both light- and heavy - higgs regimes . although our results agree with refs . @xcite , our analysis further includes independent fermion mass scales , in addition to the arbitrary compositeness parameters @xmath168 and @xmath130 . in the language of effective field theory @xcite , we see that the patterns of deviation in @xmath0 and @xmath39 that arise in the various models are strongly correlated with the quantum numbers of the heavy vector fermions with which the top - quark mixes . in particular , the interesting and potentially _ negative _ contributions arise from the existence of the exotic @xmath239 state in the bidoublet field(s ) introduced in models with extended @xmath82 custodial symmetry @xcite . it is integrating out these states that produces the crucial low - energy operator @xmath240 of the @xmath169 form . while introducing only these extra bi - doublet states does not give rise to a phenomenologically acceptable model @xcite , allowing the top quark to mix both with an exotic @xmath239 state , as well as an ordinary @xmath241 singlet or @xmath242 doublet state ( in section [ sec:3b ] ) can yield a phenomenologically viable model . such scenarios also necessarily include ditop " quarks ( @xmath172 ) with charge @xmath173 which would decay into two like - sign @xmath22 s and a bottom quark and could provide a potentially interesting pair - production signature at the lhc @xcite . finally , we note that there is another possible implementation of a a softly - broken extended @xmath82 symmetry , in which the @xmath10 field ( or the heavy field with which it mixes ) is embedded in a @xmath243 _ triplet _ field @xcite . as shown in appendix [ sec : appx ] , such models generically include dangerous tree - level corrections to @xmath39 . in the @xmath82 limit where these tree - level corrections are absent , integrating out the heavy vector fermions gives rise to a low - energy operator of _ precisely _ the same form as @xmath90 . the low - energy phenomenology of this model , therefore , is similar to those discussed above . rsc and ehs were supported , in part , by the us national science foundation under grant phy-0854889 . they acknowledge the hospitality of the institute for advanced study ( princeton ) and the aspen center for physics , where part of this work was completed . the work of rf is supported by iisn and the belgian science policy ( iap - vi-11 ) . in the models studied in this paper , we have embeded the @xmath10 field ( or the heavy fields with which it mixes ) in a @xmath244 custodial singlet field . alternatively @xcite , one can choose to embed @xmath10 as a member of a @xmath243 _ triplet _ field . here , we construct the simplest realization of this possibility _ i.e. _ we construct a model with a softly - broken extended custodial symmetry in which the @xmath10 is embedded as a triplet . while models including triplets generically yield tree - level contributions to the @xmath245 coupling , we show that these can be avoided in the @xmath82-symmetric limit . finally , we discuss the form of the low - energy operators that arise once soft symmetry breaking is introduced , and show that the operator is the familiar @xmath90 , whose phenomenology is studied elsewhere in this paper . let us consider an analog of the doublet extended symmetry breaking " model ( desm ) of @xcite . the yukawa sector of such a model contains the following terms , @xmath246 where @xmath247 is a bidoublet fermion field ( which transforms as a @xmath248 under @xmath249 ) , @xmath250 is the usual bidoublet higgs field with @xmath251 , @xmath252 and @xmath253 are @xmath254 and @xmath255 triplet fermion fields , and the @xmath256 are the usual pauli matrices . the @xmath26 symmetry requires that @xmath257 . we gauge the usual @xmath258 subgroup of @xmath259 , and obtaining the correct fermion electric charges requires the @xmath260 charges of @xmath14 , @xmath261 , and @xmath262 to be @xmath86 . we may break the @xmath263 symmetry of the top sector softly in the usual way : we will include mass terms for all the extra fermions , just as in @xcite . hence , we will include chiral partners and fermion mass terms for the @xmath264 and @xmath262 fields . the @xmath261 field includes two exotic fermions and , as its @xmath265 member , the @xmath10 . following @xcite , we will add chiral partners and fermions masses for the @xmath266 fields but leave @xmath267 massless and without a chiral partner . the model now has precisely the same properties as the desb model the top yukawa sector has an extended @xmath263 symmetry which is softly broken by the mass terms for the extra fermions , and it reduces to the standard model in the limit where the extra fermion masses become large . we now consider the form of the low - energy operators induced after integrating out the heavy triplet fermion states . we will do so in two steps : first , let us integrate out complete sets of fermion triplets @xmath252 and @xmath253 , with masses @xmath268 and @xmath269 . as we noted above , however , we should not integrate out the state @xmath267 so we will add back in the contribution of this state and see what kind of operator results . let us begin by integrating out all six states in the complete right - handed multiplet . doing so results in an operator of the following form @xmath270 + \frac{\lambda^2_f}{m^2_f } \left[{\rm tr}(\bar{\psi}_l \sigma^a \phi ) i\fsl{d}(\phi^\dagger \sigma^a \psi_l)\right]~. \label{eq : optf}\end{aligned}\ ] ] we proceed by using the fierz identity are the conventionally normalized @xmath271 generators . ] @xmath272 the first term in the operator then becomes @xmath273~. \label{eq : opt}\ ] ] in the low energy lagrangian , we make the substitution @xmath274 since the heavy exotic @xmath275 fields are decoupled . therefore @xmath276 and the second term in the operator yields precisely the familiar @xmath277 of eq . ( [ eq : smtr - op ] ) . in particular , this combination does not affect the @xmath278 vertex at tree - level . using the leibniz product rule , the first term in eq . ( [ eq : opt ] ) can be manipulated as follows @xmath279 = { \rm tr}\left[\bar{\psi}_l \phi \phi^\dagger ( i\fsl{d } \psi_l)\right ] + { \rm tr } \left[\bar{\psi}_l ( \phi i \fsl{d } \phi^\dagger ) \psi_l \right]~,\ ] ] after electroweak symmetry breaking , @xmath280 , the first contribution yields a renormalization of the usual @xmath74 kinetic energy term and produces no observable effects . the second contribution , however , renormalizes the @xmath23 and @xmath22 couplings of @xmath74 . using @xmath281 we see that @xmath282 defining @xmath39 in terms of the @xmath283 coupling @xmath284 we therefore find that this second contribution from @xmath261 exchange yields @xmath285 the analysis of the @xmath253 exchange contributions in eq . ( [ eq : optf ] ) proceeds in a similar fashion , and we find @xmath286 we therefore find that the tree - level shift in the @xmath245 coupling that arises from integrating out a complete right - handed custodial triplet multiplet ( as summarized in the operator in eq . ( [ eq : optf ] ) ) is @xmath287 we explicitly see that , in the @xmath82 limit where @xmath257 and @xmath288 , there is no tree - level correction to the @xmath245 vertex . the calculation above integrated out the entire heavy @xmath252 multiplet , including the @xmath289 fermion . in our softly - broken symmetry model , however , @xmath267 remains massless and has no chiral partner . hence , to obtain the net effect on the @xmath290 vertex in our model , we must isolate the specific contribution from the @xmath291 state to the @xmath290 vertex . the @xmath291 state gives rise to the following term in eq . ( [ eq : optf ] ) @xmath292~. \label{eq : op3}\ ] ] using the low - energy restriction of eq . ( [ eq : sub ] ) , we see that @xmath293 and hence the operator in eq . ( [ eq : op3 ] ) just yields @xmath90 , an operator we have already studied , and which does not affect the @xmath245 vertex at tree level . in a model with a softly broken custodial symmetry with right - handed triplets , generically @xmath294 and there can be dangerous tree - level corrections to @xmath39 . when @xmath295 , however , the effective field theory analysis reduces to the analysis of operator @xmath296 that is discussed in the main body of the paper . m. s. carena , a. delgado , e. ponton , t. m. p. tait and c. e. m. wagner , phys . d * 68 * , 035010 ( 2003 ) [ arxiv : hep - ph/0305188 ] . m. s. carena , a. delgado , e. ponton , t. m. p. tait and c. e. m. wagner , phys . d * 71 * , 015010 ( 2005 ) [ arxiv : hep - ph/0410344 ] . m. s. carena , e. ponton , j. santiago and c. e. m. wagner , nucl . b * 759 * , 202 ( 2006 ) [ arxiv : hep - ph/0607106 ] . m. s. carena , e. ponton , j. santiago and c. e. m. wagner , phys . d * 76 * , 035006 ( 2007 ) [ arxiv : hep - ph/0701055 ] .

we present a spectral line survey of the c - rich envelope in the 2 mm and 1.3 mm bands carried out with the arizona radio observatory ( aro ) 12 m telescope and the heinrich hertz submillimeter telescope ( smt ) . the observations cover the frequency ranges of 131160ghz , 219244ghz , and 252268ghz with typical sensitivity limit ofmk . a total of 74 individual emission features are detected , of which 69 are identified to arise from 21 molecular species and isotopologues , with 5 faint lines remaining unidentified . two new molecules ( ch and chcn ) and seven new isotopologues ( co ,sic ,sio ,sio ,cs , cs , and cs ) are detected in this object for the first time . the column densities , excitation temperatures , and fractional abundances of the detected molecules are determined using rotation diagram analysis . comparison of the spectra of to that of suggests that the spectral properties of are generally consistent with those of . for most of the molecular species , the intensity ratios of the lines detected in the two objects are in good agreement with each other . nevertheless , there is evidence suggesting enhanced emission from cn and hcn and depleted emission from hcn , sis , and ch in . based on their far - ir spectra , we find that probably has a lower dust - to - molecular gas ratio than . to investigate the chemical evolution of evolved stars , we compare the molecular abundances in the agb envelopes and and those in the bright proto - planetary nebula . the implication on the circumstellar chemistry is discussed .

0808.3226

the late stages of stellar evolution from the asymptotic giant branch ( agb ) to planetary nebulae ( pn ) are now recognized as an active period of chemical synthesis of molecules . the detection and analysis of millimeter wave molecular emission lines are fundamental to the understanding of the physical conditions and chemical processes leading to chemical synthesis . due to the rapid evolution of the star , the changing physical conditions , including dust , stellar winds , shock waves , uv emission and x - rays from the central star , etc . play different roles in circumstellar chemistry . this leads to corresponding different circumstellar chemical compositions in different evolutionary stages . the envelopes around c - rich stars , with their enhanced carbon abundance , provide a perfect cradle for molecule formation . hitherto , more than 60 molecular species have been detected in c - star envelopes @xcite , most of which were discovered through their rotational lines at millimeter wavelengths . recent improvement in telescope design and receiver performance enable us to detect new molecular emission with a higher sensitivity , and thus with the possibility of shedding new light on circumstellar chemistry . the most frequently investigated c - star envelope is , which is one of the richest molecular sources in the sky . several molecular line surveys have been presented for this object ( see * ? ? ? * ; * ? ? ? * and the references therein ) , which was found to harbor extremely abundant carbon chain and metal - containing molecules . has been frequently used as a standard reference for the chemical compositions of late - type stars . this inevitably invites the issue whether is a chemically unique late - type star . to settle this question , we require systematic surveys of molecular line emission from other c - star envelopes . in the present study , we report a spectral - line survey of the c - star envelope at millimeter wavelengths . this allows us to compare the similarity and difference in the chemical compositions between the two c - star envelopes . ( ) was first discovered during the caltech 2-@xmath11 m sky survey and was among the 14 very red infrared - bright optical - faint sources found @xcite . is characterized by its very low color temperature , implying that the star is surrounded by a very thick dust envelope and has been identified as a long - period variable with a period of about 628 days @xcite . from the period - luminosity relation , @xcite estimated the distance of to be @xmath12pc , which is slightly more distant than , which has a distance of between @xmath13pc @xcite and @xmath14pc @xcite . is believed to be more evolved and has a lower mass loss rate compared to ( see * ? ? ? * e.g. ) . the large polarization found in the visible and infrared wavelengths implies that the distribution of circumstellar material around the star is asymmetric @xcite . multi - wavelength imaging observations have been performed to study the structure of the nebula around . the optical images obtained by the hubble space telescope ( hst ) and the near - ir images of obtained by the keck - i telescope have revealed a bipolar dust envelope and an elongated component with time - variable asymmetry @xcite . several scattering arcs were revealed by the hst - nicmos imaging polarimetry @xcite . these arcs are nearly concentric and extend to large stellar radii . mid - ir images of were obtained by @xcite using the eso 3.6-m telescope . a cometary - like feature was revealed in their 9.7@xmath15 m image . there have been several observations of molecular lines in at millimeter wavelength . @xcite reported observations of a few molecular lines in and between 18 and 150ghz . they found that relative abundances of observed molecules in the two sources have no significant differences . using the nobeyama 45 m radio telescope , @xcite observed a few transitions in the frequency ranges between 3947ghz and 8591ghz in a sample of evolved stars including and found that the abundance of hnc increases with the evolutionary stage of the stars . @xcite presented observations of 10 molecular transitions in c - rich and o - rich circumstellar envelopes including with the iram 30 m radio telescope at 1.3 mm , 2 mm , and 3 mm windows . a recent molecular line survey was presented by @xcite using the sest 15 m and onsala 20 m telescopes . they found that stands out from the other c - rich envelopes due to its high cn / hcn ratio and low hnc / hcn ratio . to date , the molecular species positively detected in at millimeter wavelengths include co , @xmath8co , cn , @xmath8cn , cs , sio , sis , @xmath5si@xmath16s , c@xmath17h , sic@xmath17 , hcn , h@xmath8cn , hnc , c@xmath18n , hc@xmath18n , hc@xmath8ccn , hcc@xmath8cn , and hc@xmath19n . in this paper , we present the first systematical line survey of at the 2 mm and 1.3 mm windows , using the arizona radio observatory ( aro ) 12 m telescope and the heinrich hertz submillimeter telescope ( smt ) . the observations are described in sec . 2 . in sect . 3 we present the identifications and abundance calculations of the detected molecular species . in sect . 4 we discuss the implication of our findings on circumstellar chemistry . the conclusions are given in sect . the observations were carried out in beam switching mode with an azimuth beam throw of 2@xmath20 during the period from 2005 april to 2006 september . pointing was checked by observations of a planet every 2 hours . the 131160ghz ( 2 mm window ) spectra were obtained with the aro 12 m telescope at kitt peak , and the 219244ghz and 252268ghz ( 1.3 mm window ) spectra with the smt 10 m telescopes on mount graham , arizona . the 2- and 1.3-mm dual - channel sis receivers were employed , operated in single sideband dual polarization mode with a typical image rejection ratio of @xmath21db . at the aro 12 m , the spectrometer back - ends were two 256-channel filter banks ( fbs ) with a channel width of 1mhz and a millimeter autocorrelator ( mac ) with 3072 channels and 195khz per channel . the spectrometers utilized at smt were a 2048-channel acousto - optical spectrometer ( aos ) with a channel width of 500khz and 1024-channel forbes filterbanks ( ffbs ) with a channel width of 1mhz . the system noise temperatures were typically 150400k at 2 mm and 400700k at 1.3 mm . the temperature scales at the aro 12 m and the smt , @xmath22 and @xmath23 , were obtained using standard vane calibration . a @xmath24 calibration error was obtained from a few strong lines detected in different spectrometers and different epochs . the main beam brightness temperatures were derived through @xmath25 and @xmath26 , for the 12 m and the smt data , respectively , where @xmath27 is the corrected beam efficiency ( @xmath28 ; see the aro 12 m manual for its definition ) and @xmath29 the beam efficiency ( @xmath30 ) . both telescopes have a large beam size , covering the whole emission region . the class software package in gildas was used to reduce the spectra . after discarding the bad scans which are seriously affected by bandpass irregularities , we co - added the spectra from individual scans . the baseline was fitted with a low - order polynomial . in order to improve the signal - to - noise ratios , the spectra were smoothed and rebinned by a factor of 3 , yielding a typical @xmath31 noise temperature of @xmath32mk in main beam brightness temperature units and a spectral resolution of @xmath33km / s . since each line has been observed by two different spectrometers , we can reduce the uncertainties caused by ripples and bad channels . a total of 74 individual features were detected in our survey , including some less than certain detections . line identification is mainly based on the jpl catalog @xcite and the cologne database for molecular spectroscopy ( cdms , * ? ? ? * ; * ? ? ? * ) , which give the molecular line frequencies based on theoretical calculations . we also used the nist recommended rest frequencies for observed interstellar molecular microwave transitions , which provides the molecular line frequencies derived from observations of various sources . moreover , we utilized the recent observations of and by @xcite , @xcite , and @xcite for reference . the complete spectra obtained with the aro 12 m telescope and the smt 10 m telescope are plotted in figs . [ spe_cit6_12 m ] and [ spe_cit6_smt ] along with identified lines marked . [ pro_12 m ] and [ pro_smt ] give the detected molecular line profiles . note that the two features at 221.459ghz and 241.458ghz clearly seen in fig . [ spe_cit6_smt ] are the strong co lines from the image sideband . among the detected features , 69 are identified to arise from 21 molecular species and their isotopic variants . table [ line ] lists their assignments , peak and integrated intensities and line widths species - by species . the half maximum line widths ( @xmath34 ) were obtained by fitting gaussian line profiles . if a line has an unresolved hyperfine structure , the peak intensity of the strongest transition is given , and the integrated intensity and the @xmath34 are those of the combined feature . five faint features remain unidentified and they are listed in table [ uline ] . comparing with the unidentified line listing in the nist frequency table , we find that the 148444 , 255940 , and 265936 mhz lines have no previous recorded detections elsewhere , and the 257035 and 262255 mhz may have corresponding u lines at 257015 and 262273 mhz detected in sgr b2 ( n ) @xcite . a detailed description of the molecules detection is given below . the @xmath361 transitions of co , @xmath8co , and c@xmath4o were detected at the 1.3 mm window . [ pro_smt ] clearly shows that the shape of the co line differs from that of the @xmath8co line . the co line shows a parabolic profile , while the @xmath8co line has a double - peaked profile , suggesting the former is optically thick , while the latter is optically thin . @xcite observed the co transitions from @xmath370 to @xmath385 in . the shape of the co ( 21 ) line obtained by our observations is very similar to their results ( see their fig . the co ( 21 ) and @xmath8co ( 21 ) lines have also been observed by @xcite using the iram telescope in 1991 . they obtained the integrated intensities of the co ( 21 ) and @xmath8co ( 21 ) lines to be 230 and 16.5kkms@xmath39 respectively , a factor of @xmath40 larger than our results . as the beam size of the iram telescope is about half of that of the smt , the discrepancy of the co intensity can be explained with different beam dilution effects . hence , we can conclude that the variation of co intensity is relatively small ( within 10@xmath41 ) during the past 15 years . the extremely faint c@xmath4o emission should be a new detection for this object . sic@xmath6 has been detected by @xcite in the spectra of through the sic@xmath6 ( 5@xmath434@xmath44 ) ) transition at 115.382ghz . in our line survey , a number of sic@xmath6 transitions were prominently detected at the 2 mm and 1.3 mm windows . these lines have a similar profile with an average linewidth of @xmath45kms@xmath39 . we also detected two faint @xmath5sic@xmath6 transition although their intensities are only at a 23@xmath46 noise level . we can not find previous papers reporting on the detections of these sic@xmath6 and @xmath5sic@xmath6 transitions tabulated in table [ line ] . @xmath47 . is characterized by bright cn emission . the frequencies of three cn ( 21 ) fine - structure groups lie in the region of the smt spectrum . all of them were clearly seen in fig . [ spe_cit6_smt ] . the three cn ( 21 ) fine - structure groups consist of 18 hyperfine structure components . @xcite carried out a survey of cn ( 21 ) and ( 10 ) emission in a sample of evolved stars . they detected the strongest two cn ( 21 ) groups in . their observations suggests the intensity ratio of high- and low - frequency fine - structure groups to be 1.5 , in excellent agreement with our result of 1.6 . @xmath48 . strong cs ( 32 ) emission was detected in the 2 mm window . this line had previously been discovered by @xcite and @xcite . we also clearly detected its isotopic transitions , @xmath8cs ( 32 , 54 ) and c@xmath10s ( 32 , 54 ) . the faint c@xmath9s ( 54 ) emission was only marginally discovered . the cs ( 54 ) transition at 244.9ghz does not lie in the frequency region observed in the work because of a gap between 244.5252.5ghz in the survey . [ pro_12 m ] and [ pro_smt ] show some evidences that the line profiles of cs and its isotopologue emission are slightly asymmetric in shape with a brighter red wing . the @xmath385 and @xmath504 transitions of sis in have previously been observed by many researchers @xcite . in this work , three sis transitions with higher @xmath51 ( 87 , 1312 , and 1413 ) were clearly detected with an average linewidth of @xmath52kms@xmath39 . sio emission in this carbon star has been extensively studied . the transitions observed by previous studies include @xmath361 , 32 , 54 , and 87 @xcite . in this work , we report the detection of the sio ( 65 ) transition and its @xmath5s and @xmath7s isotope lines as well . the sio ( 65 ) line detected in this survey has a similar width to other sio transitions detected by previous researchers . the @xmath550 transition of c@xmath6h in this object has been detected by @xcite . the @xmath562 transition lies in our surveyed frequency range . the rotation transition is split in six hyperfine - structure lines grouped in three fine - structure groups , all of which were detected at the 2 mm window . the two main components ( 3@xmath572@xmath58 and 3@xmath582@xmath59 ) are quite strong and have a well - defined profile , which is similar to that of the cs line with higher flux in the red wing . the 3@xmath572@xmath58 transition is blended with a weak sic@xmath6 line . the 3@xmath582@xmath58 transition is too faint to obtain a reliable intensity . the @xmath61 - 2 transitions of hcn and h@xmath8cn were clearly detected . the hcn ( 32 ) transition is the second brightest line after the co ( 32 ) transition in our survey . this line was first observed by @xcite using the smt , who obtained an integrated intensity of 43.1kkms@xmath39 , in excellent agreement with our measurement . given its line shape , the line should be optically thick . there are five favorable vibrationally excited lines of hcn present in the surveyed frequency range . three of them were clearly detected , as illustrated in table [ line ] . the weak @xmath62 @xmath612 transitions are below the 3@xmath46 noise level . in , hcn is the only species with detected vibrationally excited transitions . [ pro_smt ] shows that these vibrationally excited lines are probably narrower than the hcn ( 32 ) line . discarding the uncertain detection of the @xmath62 @xmath612 transitions and the @xmath63 @xmath612 transition which is partially blended with the hcn ( 32 ) line , we obtain the average width of the vibrationally excited lines of hcn to be 14.2kkms@xmath39 . the narrow linewidth of the vibrationally excited lines suggests that they might arise from the hot inner region with low expansion velocity . we do not detect the vibrationally excited lines of h@xmath8cn . @xcite and @xcite observed the c@xmath3n ( 1110 a , b ) transitions in . both lines are relatively weak . due to fine - structure interactions , every rotational transition of c@xmath3n is split into doublets of similar intensity . there are 12 c@xmath3n transitions present in our survey range with six at the 2 mm window and six at the 1.3 mm window . we detected the strongest four c@xmath3n transitions ( 1413 a , b and 1514 a , b ) at the 2 mm window with @xmath65mk . the 1615 a , b transitions at around 158ghz have a comparable intensity with the other four transitions at the 2 mm window . however , they fall within a spectral region with a high noise level , and thus are not listed in table . [ line ] . @xmath66 . to our knowledge , c@xmath2h has not been detected in before this work . @xcite and @xcite estimated the intensity upper - limits of the c@xmath2h ( 109 a , b ) transitions , which indicate that c@xmath2h emission in this object is relatively weak . analogous to that of c@xmath3n , every rotational transition of c@xmath2h is split into two components with a similar intensity . there are 12 favorable lines of c@xmath2h present in the frequency range surveyed here . eight of them were detected with @xmath6735mk . the c@xmath2h ( 1514 a , b ) and ( 1615 a , b ) transitions were clearly detected with well - defined profiles at the 2 mm window . the c@xmath2h ( 1413 a , b ) lines fall within a spectral region with a high noise level , and thus are not listed in table [ line ] . the four c@xmath2h lines ( 2423 a , b and 2827 a , b ) at the 1.3 mm window are extremely faint , and are only marginally detected . the c@xmath2h ( 2524 a , b ) transitions at around 238ghz are overwhelmed by noise . seven hc@xmath3n transitions from @xmath6914 to @xmath7028 are in the frequency range of our survey . all of them were prominently detected with @xmath71mk . previously only lower @xmath51 transitions of hc@xmath3n have been reported @xcite and to the best of our knowledge all these detections are new . because of a gap in the 1.3 mm spectrum , the hc@xmath3n ( 2726 ) at 245.606ghz was not detected here . as shown in table [ line ] , there is a trend that the widths of hc@xmath3n lines decrease with increasing @xmath51-values , suggesting that the high-@xmath51 transitions might originate in hot inner regions which have a lower velocity compared to the regions from where the low-@xmath51 transitions arise . all hc@xmath3n lines detected in this survey show asymmetric profiles ( see fig . [ pro_12 m ] and [ pro_smt ] ) . opposite to cs and c@xmath6h lines , hc@xmath3n lines have a brighter blue wing relative to the red one . this indicates that the molecular envelope is asymmetric along the line of sight , i.e. , the chemical compositions and/or physical conditions in the red and blue sides are different . there are 14 favorable ch@xmath3cn transitions in the survey region . we detected the blended features of the ( 12@xmath7311@xmath73 ) and the ( 12@xmath7411@xmath74 ) transitions with @xmath75mk , which are the strongest two among the 14 ch@xmath3cn transitions . @xcite failed to detect ch@xmath3cn in this object and only estimated the intensity upper limit of the ( 6@xmath735@xmath73 ) transition . this is therefore the first detection of ch@xmath3cn in . all the lines discovered in this survey have been detected in @xcite . compared to the spectrum of in the same frequency range , the non - detected species in the spectrum of include c@xmath3h , c@xmath3h@xmath6 , c@xmath6s , c@xmath3s , h@xmath6co , sic , sin , pn , and metal containing molecules , all of which have only weak emission in the spectra of , and thus are below our detection limit . a number of vibrationally excited species have been detected in ( see * ? ? ? * and the references therein ) . for , however , no vibrationally excited lines except those of hcn are strong enough to be detected . moreover , we find no evidence for the presence of ionic species in . the standard rotation - diagram " method was applied to calculate the excitation temperatures ( @xmath76 ) and column densities ( @xmath77 ) of the molecules observed in our spectra . from the equation of radiative transfer and assuming that the lines are optically thin , the level populations are in local thermal equilibrium ( lte ) , and @xmath78 , where @xmath79 is the cosmic background radiation temperature ( 2.7k ) , we have the well - known relation , @xmath80 @xmath81 , @xmath82 , and @xmath83 are the population , degeneracy , and excitation energy of the upper level , @xmath84 is the integration of the source brightness temperature over the velocity , @xmath85 is the line strength , @xmath11 is the dipole moment , @xmath86 is the line frequency , and @xmath87 is the rotational partition function . if several transitions arising from levels covering a wide energy range are observed , @xmath76 and @xmath77 can be deduced using a straight - line fit to @xmath88 versus @xmath89 . the rotation - diagram provides important tools for studies of excitation conditions . departure from the linear relation can be caused by different excitation mechanisms or misidentification . for sic@xmath6 , sis , hc@xmath3n , and c@xmath2h , there are adequate numbers of detected transitions covering a wide range of excitation energy and their rotation diagrams are given in fig . [ dia_cit6 ] . good linear correlations were obtained for these species . for the calculations , we have corrected the effect of beam dilution through @xmath90 , where @xmath91 is the antenna full beam at half - power ( @xmath92 and 30 for the aro 12 m and the smt respectively ) and @xmath93 is the source diameter . @xmath93 may be different for different species . we followed the assumption by @xcite and took an common @xmath93 of @xmath94 for all the species . since the source size is likely to vary from species to species @xcite , this assumption will introduce a @xmath95 uncertainty in the derived column densities . the derived excitation temperatures and column densities are given in table [ col_cit6 ] . for these species for which only one line was detected or observed transitions arise from a narrow range of energy levels , the method of rotation diagram method can not be employed and a constant @xmath76 of 40k was assumed for the calculations of their column densities . assuming that the molecular envelope is a spherical shell , the emission is optically thin , @xmath76 is uniform throughout the envelope , mass loss rate and expansion velocity are constant during the formation of the envelope , and the molecular density follows an @xmath96 law , we determined the fractional abundances of the observed species with respect to h@xmath6 through the formula proposed by @xcite , @xmath97 where @xmath98 is given in kkms@xmath39 , the full half power beam width @xmath91 is in arcsec , the expansion velocity @xmath99 is in kms@xmath39 , @xmath100 is the distance in pc , @xmath101 is the mass loss rate in m@xmath102 , @xmath103 the line frequency in ghz , @xmath82 is the statistical weight of the upper level , @xmath104 is the einstein coefficient for the transition , @xmath105 is the energy of the lower level , and @xmath106 with @xmath107 and @xmath108 the inner radius and outer radius of the shell . for the calculations , we adopted @xmath109pc @xcite . for the determination of molecular abundances using eq . 2 , we first derive the expansion velocity @xmath99=18 km s@xmath39 from the profile of the co ( 21 ) line . the mass loss rate of @xmath110m@xmath111 is obtained by applying eq . 2 of @xcite to the co ( 21 ) line assuming @xmath112 . these values of @xmath99 and @xmath101 are in excellent agreement with those given by @xcite . these parameters suggest that has a similar expansion velocity and about one order of magnitude lower mass loss rate in comparison with . the resultant abundances are given in table . [ col_cit6 ] . combined with the uncertainties introduced by calibration , noise , baseline , and source size , we estimate that the errors of the column densities and abundances amount to a factor of @xmath113 . one should bear in mind that when the emission is optically thick the @xmath77 and @xmath114 listed in table . [ col_cit6 ] represent only lower limits . a comparison of our results with those derived by @xcite and @xcite are also given in table . [ col_cit6 ] and no significant discrepancies are found . similar to , the molecular envelope of cit 6 is also characterized by a lack of oxygen - bearing compounds and abundance of carbon - bearing compounds . in , three o - bearing molecules are observed , co , sio and hco@xmath115 @xcite whereas in , only two o - bearing molecules ( co and sio ) are detected . we find that the sio ( 65)/@xmath8co ( 21 ) integrated intensity ratio in is 1.05 , in excellent agreement with the value of 1.13 in @xcite . in our discussion , @xmath8co is taken as a reference molecule because the @xmath8co ( 21 ) line is likely optically thin and the @xmath116c/@xmath8c isotopic ratios are similar in and ( see sect . sio has been commonly detected in c - rich envelopes @xcite , implying the presence of icy comets surrounding the stars @xcite and/or non - equilibrium chemical processes @xcite . hco@xmath115 has a relatively low abundance in @xcite , and is presumably below the current detection limit even if it might be present in . we observed abundant carbon chains and radicals in , including co , sic@xmath6 , cn , hcn , cs , c@xmath6h , c@xmath3n , c@xmath2h , hc@xmath3n , and ch@xmath3cn , all of which are linear . this characteristic feature is similar to those of and ( see * ? ? ? * ) although these lines are much fainter in . the most intriguing characteristic of is the strong cn emission . the integrated intensity ratio of the cn ( 21 ) group and the @xmath8co ( 21 ) transition is 4.6 , a factor of 2.2 larger than the value in @xcite . cn is mainly formed through the photodissociation of hcn , @xmath117 according to @xcite , the h@xmath8cn ( 32)/@xmath8co ( 21 ) integrated intensity ratio in is 4.5 , a factor of 3.2 larger than that in . therefore , our observations provide strong evidence that reaction ( [ hcn ] ) dominates the chemistry of cn and hcn in agb stars and the photodissociation is more efficient in the more evolved c - rich envelope . the above discussion also suggests that about 30@xmath41 cn formed from hcn has been destroyed . on the other hand , cn can be reprocessed into hc@xmath3n through the reaction @xmath118 we do find that the hc@xmath3n line intensities relative to the @xmath8co ( 21 ) transition in are a factor of @xmath40 larger than those in , indicating efficient formation of hc@xmath3n in . we did not find evidence for the enhancement of the c@xmath3n radical , suggesting that photodissociation of hc@xmath3n into c@xmath3n is insignificant in this object . shows strong c@xmath6h emission . the c@xmath6h radical is dominantly produced through the photodissociation reaction @xmath119 our observations show that the c@xmath6h line intensities relative to the @xmath8co ( 21 ) transition in are almost the same at those in , suggesting that there is no significant c@xmath6h enhancement in compared to . this probably has an implication that c@xmath6h is dominantly processed in early agb stages . since c@xmath6h and hc@xmath3n have the same chemical precursor , the @xmath120(c@xmath6h)/@xmath120(hc@xmath3n ) abundance ratio can be used to test the chemical formation path @xcite . our results yield a ratio of 4.2 for @xmath120(c@xmath6h)/@xmath120(hc@xmath3n ) , which is in good agreement with those found in interstellar clouds by @xcite and is consistent with the prediction of gas phase chemistry . on the other hand , as shown in figs . [ pro_12 m ] and [ pro_smt ] , hc@xmath3n lines show profiles that differ from those of the c@xmath6h lines . this suggests that the @xmath120(c@xmath6h)/@xmath120(hc@xmath3n ) ratio is not a constant in the envelope . the chemistry structures in the red and blue sides of the star are not identical . this is probably a consequence of an inhomogeneous density distribution or physical environment . a complete understanding of the molecular environment of calls for a comprehensive 3d photochemistry model and high - resolution mapping observations . c@xmath2h is positively detected in . the c@xmath2h radical can be formed via @xmath121 @xcite suggested that ion - molecule reactions also play an important role for the production of c@xmath2h , unlike those for hc@xmath3n . based on millimeter interferometer observation , @xcite found that the photochemical model underestimates the c@xmath2h abundance in by a factor of 5 . our observations suggest that has a @xmath120(c@xmath2h)/@xmath120(hc@xmath3n ) abundance ratio of 3.1 , which is lower than that in @xcite by a factor of @xmath40 . on the other hand , c@xmath2h can be photodissociated into c@xmath6h . however , as no enhancement of c@xmath6h is found in , destruction of c@xmath2h should be insignificant in this c - rich envelope . therefore , the high radical abundance of c@xmath2h in still remains mystery . ch@xmath3cn is a new finding for this c - rich envelope . this molecule is a symmetric top and has been widely used as diagnostics of excitation temperature . we detected only one weak ch@xmath3cn line in , and thus can not use it to derive an excitation temperature . ch@xmath3cn can be produced via @xmath122 followed by @xmath123 the formation of ch@xmath3cn may be very efficient in since strong hcn emission is detected . has a ch@xmath3cn ( 1211)/@xmath8co ( 21 ) integrated intensity ratio of 0.058 , in good agreement with the value of 0.043 in @xcite . strong cs emission has been detected in our survey . according to @xcite , cs can be rapidly destroyed by shocks which might occur after a star leaves the agb stage and ejects material in a very fast wind @xcite . therefore , the high abundance of cs in suggests that shocks are not important for the chemistry in this c - rich envelope . three refractory si - bearing species ( sio , sis , and sic@xmath6 ) were detected in . although other si - bearing species were detected in @xcite , emission from the other si - bearing species is relatively faint and should be below the detection limit of our observations of . we find that the abundances of sio and sic@xmath6 in are similar to those in determined by @xcite . although the situation may be complicated by optical depth effects , the hcn / sio line intensity ratio has the potential to provide a useful tool to discriminate between c - rich and o - rich envelopes and is a good tracer of mass loss rate for m and s stars ( see e.g. * ? ? ? the hcn ( 32)/sio ( 65 ) intensity ratio in is 8.4 , in good agreement the value of 9.7 in @xcite . there is no evidence showing that the hcn / sio line intensity ratio has dependance on the mass loss rate of c - rich stars . @xcite and @xcite found a correlation between the mass loss rate and the sio abundance for agb stars . this is described as freeze - out of sio molecules onto dust grains . the similarity of the sio abundances in and suggests that the depletion of sio onto dust grains might be insignificant for the two c - rich envelopes . our observations show that the sis abundance in is lower than that in . the sis ( 1413)/sio ( 65 ) intensity ratio in is 1.1 , about half of that in found by @xcite . @xcite did not find a strong correlation between the mass loss rate and the sis abundance , suggesting that sis molecules are less likely to be depleted onto dust grains than sio molecules . therefore , freeze - out should not be the reason for the depletion of sis in since no depletion of sio is found for this object . on the other hand , with a high efficiency , sis can be photodissociated into si@xmath115 , which initiates circumstellar sic@xmath124 chemistry @xcite . hence , we infer that efficient photodissociation in has destroyed sis , and silicon chemistry has been ongoing in this evolved c - star envelope . interferometric observations of these si - bearing molecules are obviously needed to verify the conjecture . isotopic ratios of various elements provide substantial tests for nucleosynthesis of low- and intermediate - mass stars ( lims ) . when a lims evolves into the agb stage , the nucleosynthesized products synthesized through the cno cycle inside the star are dredged up to the surface and then are ejected into the circumstellar envelope . consequently , the isotopic composition in the circumstellar shell can be markedly changed . based on the fractional abundances proposed in table [ col_cit6 ] , we deduce the isotopic ratios ( or their lower limits ) of carbon , oxygen , silicon , and sulfur in . the results are listed in table [ isoto_cit6 ] . the errors estimated from the measurement and calibration are given . for comparison , we also list the isotopic ratios for @xcite and the sun @xcite . the @xmath116c/@xmath8c abundance ratio is the most studied isotopic abundance in lims . standard stellar models predict that the @xmath116c/@xmath8c abundance ratio can be significantly increased during the nucleosynthesis and dredge - up processes in the agb stage . however , extensive observations have shown that the @xmath116c/@xmath8c abundance ratios in limss are considerably lower than those expected by standard stellar models ( e.g. * ? ? ? @xcite proposed an extra mixing process to account for the low @xmath116c/@xmath8c . in low - mass agb stars , the nonstandard mixing called cool bottom processing may decrease the @xmath116c/@xmath8c ratio to @xmath125 @xcite . for agb stars more massive than @xmath126 , the hot bottom burning may take place and induce @xmath116c/@xmath8c to further decrease to its equilibrium value of @xmath127 @xcite . current observations of the co isotopologues in pns @xcite suggest that the @xmath116c/@xmath8c ratio is in the range of 2.240 , supporting the theory including nonstandard mixing processes . three @xmath8c - bearing species have been detected in this survey , including @xmath8co , @xmath8cs , and h@xmath8cn . however , their main lines are likely optically thick . therefore , the abundance ratios of co , cs , and hcn and their isotopologues only provide lower limits of the @xmath116c/@xmath8c ratio . our results are in good agreement with those presented by @xcite who derived the abundance ratios @xmath116co/@xmath8co@xmath128 and h@xmath116cn / h@xmath8cn@xmath129 for . we have detected the rare isotopes , @xmath116c@xmath10s and @xmath8c@xmath16s . if the @xmath16s/@xmath10s abundance ratio were known , we could obtain the @xmath116c/@xmath8c ratio using the two optically thin species . @xcite found that the sulfur isotopic ratios in are close to solar . therefore , we reasonably assume that the @xmath16s/@xmath10s ratio in is the solar value . it follows that we obtained the @xmath116c/@xmath8c ratio of @xmath130 in , which is in perfect agreement with that found in and is significantly lower than the solar value . the @xmath116c/@xmath8c ratios found in these c - rich envelopes are also lower than the value of @xmath131 in the orion bar proposed by @xcite using the c@xmath132o/@xmath8c@xmath132o abundance ratio . this is consistent with the hypothesis that nonstandard mixing processes have decreased the @xmath116c/@xmath8c ratios in the envelopes around agb stars . as shown in table [ isoto_cit6 ] , different @xmath116c/@xmath8c values are obtained for if different species are used for the calculations . if completely ascribing the @xmath116c/@xmath8c discrepancies found for to the opacity effects of the main lines , we can estimate the optical depths of the co ( 21 ) , cs ( 32 ) , and hcn ( 32 ) lines , which are 1.3 , 1.2 , and 2.0 , respectively . if the isotopic lines are also optically thick , the optical depths obtained here should be considered as lower limits . the nucleosynthesis and dredge - up processes in the agb stage can lead to strong enrichment of @xmath4o relative to @xmath133o and @xmath132o ( see * ? ? ? * for a recent review ) . @xcite found that the @xmath4o/@xmath132o ratios in c - rich envelopes are markedly higher than the terrestrial and interstellar values , but the @xmath133o/@xmath132o ratios are comparable to the solar value . we have detected c@xmath4o , allowing us to derive the @xmath133o/@xmath4o ratio in . the c@xmath133o / c@xmath4o abundance ratio give a lower limit of @xmath134 . we may also use the optically thin species @xmath8c@xmath133o and @xmath116c@xmath4o to derive the @xmath133o/@xmath4o ratio . assuming @xmath116c/@xmath8c@xmath135 ( see above ) , we obtain a @xmath133o/@xmath4o ratio of @xmath136 . @xcite did not obtain the oxygen isotopic ratios . @xcite calculated the @xmath133o/@xmath4o ratio for a sample of c - rich envelopes . they found that @xmath133o/@xmath4o@xmath137 and @xmath138 for and , respectively . these values are in good agreement with our result , agree with each other , and are lower than the solar value by a factor of about three , which is consistent with predictions of stellar models . no @xmath132o - bearing species was detected . the non - detection of the c@xmath132o ( 21 ) transition at 219560mhz seems to suggest that @xmath4o/@xmath132o@xmath139 in . however , this result should be taken with some caution since the c@xmath132o line is very close to the edge of the spectrum . the nucleosynthesis in lims is expected to hardly affect the elements in the 3rd row of the periodic table , such as si and s. @xcite indeed found that the si and s isotopic ratios in are compatible with the solar values . the silicon isotopes , @xmath5si and @xmath7si , have been detected through faint emission from @xmath5sio , @xmath7sio , and @xmath5sic@xmath6 . since the sio and sic@xmath6 lines are probably optically thick , the lower limits of the @xmath140si/@xmath7si and @xmath140si/@xmath5si ratios were derived . given the low abundances of @xmath5si and @xmath7si , the @xmath5si/@xmath7si ratio is not affected by opacity effects . we obtained @xmath5si/@xmath7si@xmath141 , comparable with the values in and in the sun . the c@xmath16s / c@xmath10s ratio gives a lower limit of @xmath16s/@xmath10s in . c@xmath9s was only marginally detected . the two optically thin species , c@xmath9s and c@xmath10s , give a @xmath9s/@xmath10s ratio of @xmath142 . consequently , we did not find a significant deviation of the s isotopic ratios in from those in and in the sun . given its brightness and abundant molecular emission , is the most surveyed object and has frequently served as a standard reference for studying circumstellar chemistry . a commonly asked question is whether can truly represent c - rich agb stars . to investigate the problem , we systematically compare the spectra of with those of obtained in the same survey program @xcite . we find that all lines detected in this work have also been observed in , and all the strong lines detected in are seen in . in fig . [ linecomp ] , we compare the intensity ratios of the lines detected in both objects . the results show that the line intensity ratios in the two objects are in agreement within one order of magnitude with an average value of 0.08 with a standard deviation of 0.05 . the intensity ratios , however , might be affected by beam dilution . is a more compact object and thus suffers from a larger beam dilution effect . if we correct for the beam dilution effect , the @xmath143(cit6)/@xmath143(irc+10216 ) ratios should increase by a factor of @xmath144 , where @xmath145 assuming @xmath146 and @xmath147 @xcite and using @xmath148 and 30@xmath149 for the aro 12 m and the smt respectively , we obtain the @xmath144 values of 1.8 and 1.6 for the 12 m and smt data , respectively . while recognizing that eq . 9 assumes uniform brightness temperature which is probably not realistic , the result does suggest that there is no large correction factor difference between the @xmath0 2 mm and 3 mm bands . our conclusion that for most of the species the @xmath143(cit6)/@xmath143(irc+10216 ) line ratios are in good agreement therefore stands . consequently , we conclude that is indeed likely representative of c - rich envelopes although comparisons with a larger sample of objects will be needed to make a stronger statement . the richness of molecular lines in the spectra of is mainly due to its relatively nearby distance , and not due to any special circumstances . [ linecomp ] also shows that there are a few molecular species for which the @xmath143(cit6)/@xmath143(irc+10216 ) ratios depart from the average value . for the cn and hc@xmath3n lines , the ratios are higher , whereas for hcn , sis , and c@xmath2h , the ratios are lower . as discussed in sect . [ carbon ] , this partly reflects the chemical evolution in the circumstellar envelope around the more evolved agb star . however , the cause of the abnormally strong c@xmath2h emission in remains unknown . far - ir spectra of and have been obtained by the iso long wavelength spectrometer ( lws ) @xcite . the rotational transitions revealed by the iso spectra can trace the inner regions of the circumstellar envelopes @xcite . in fig [ iso_con ] , we compare the far - ir spectra of the two objects . the spectra were retrieved from the iso archive . inspection of the figure shows that their far - ir spectra are dominated by thermal continuum emission from the dust with some superimposed molecular lines . @xcite fitted the dust continuum of irc+10216 with a single blackbody of 510 k and the continuum of cit 6 with two blackbodies of 1000 k and 510 k. in the long wavelength region , the dust temperatures of the two objects are therefore almost identical . [ isocomp ] gives the continuum - subtracted iso lws spectra of the two c - rich envelopes . a number of lines from co , hcn , h@xmath8cn , and vibrationally excited hcn have been identified by @xcite in the spectrum of . [ isocomp ] shows that the relative flux ratios of the lines detected in the two object are in good agreement although the detection of some weak lines is difficult in as it is fainter . the far - ir line fluxes detected in are a factor of about 10 higher than those in , consistent with that found for the molecular lines at millimeter wavelengths . however , we find that the far - ir dust continuum emission in is a factor of about 20 times higher than that of , as shown in fig [ iso_con ] . this probably suggests that has a larger dust - to - molecular gas ratio . since has been proposed to be a highly evolved agb star on the verge to become a proto - pn ( or ppn ) @xcite , it would be useful to compare the molecular abundances of with the corresponding molecular abundances in the evolved agb star and a ppn . the archetypical ppn has abundant molecular emission . it is one of the brightest molecular sources in the sky and is an ideal target for investigating circumstellar chemistry ( see e.g. * ? ? ? all three objects are carbon rich , and are commonly assumed to belong to a common evolutionary sequence . any systematic difference in molecular abundances in these three objects can be used to infer on the processes of chemical synthesis and destruction . [ compare ] gives the fractional abundances of the species in the three objects as a function of their photodissociation rates taken from the umist database for a temperature of 300k . the abundances in and are taken from @xcite and @xcite , respectively . to facilitate the comparison , we follow @xcite and give the molecular abundances relative to hc@xmath3n . according to this figure , except for cn and hcn , the molecular abundances in are obviously closer to those in than to . has lower abundances of sio and cs , suggesting that destruction by shocks and depletion onto dust grains may play an important role for the chemistry in this object . compared to the other two objects , , having a b0 central star , is exposed in a stronger uv radiation field . however , we do not find a correlation between the abundance differences of these objects and the molecular photodissociation rates , suggesting that the destruction by photodissociation is not a factor affecting the chemistry in these objects during the agb - ppn transition . however , we do find that some molecules have been efficiently reprocessed during this transition . since the dynamical timescale of ppn is @xmath15010@xmath151 years , our results therefore suggest a rapid change of the chemical compositions after the star evolves into the ppn stage . we should note that this study is limited to simple molecules in the gas phase . there is strong evidence from infrared spectroscopy that the solid - state phase chemistry is very active in the agb - ppn evolutionary transition , with many aromatic and aliphatic compounds being formed in the circumstellar envelope @xcite . even some simple gas - phase molecules ( e.g. , acetylene and benzene ) are difficult to detect via rotational transitions in the mm / submm region , and complex organics even more so . consequently , the technique of rotational mm / submm spectroscopy is unable to serve as a complete diagnostic tool for circumstellar chemistry . it is more useful as a probe of photo- or shock - chemistry in the pn phase , or during the formative stage of early agb evolution . the presence of rich molecular species around evolved stars provides an opportunity to study the evolution of chemistry in circumstellar envelopes , which have been widely suggested as one of the main sources of organic compounds in space . as part of our project of investigating circumstellar chemistry , this paper reports a spectral line survey of the carbon - rich envelope , covering the frequency range between 131160 , 219244 , and 252268ghz with a high sensitivity . a total of 74 lines are reported in the survey . we identify 69 lines belonging to 21 different molecular species and isotopologues , most of which are carbon - bearing species . the new species include two carbon - chain molecules , c@xmath2h and ch@xmath3cn , and seven c , o , s , and si isotopologues . several new transitions from known species have been detected for the first time in this object . the species with the largest number of detected emission lines in our survey is sic@xmath6 , which has 19 lines . it is followed by hc@xmath3n , with 7 lines . we find that the line profiles for some molecular species have different shapes , suggesting that the chemical structure is asymmetric in the envelope . a comprehensive 3d photochemistry model is required to account for the line intensities and profiles in . the excitation temperatures , column densities and abundances of the detected molecules are determined through rotation - diagram analysis . the spectra of are characterized by a large cn / hcn abundance ratio . our results suggest that there is evidence for the photodissociation of hcn and sis and the formation of cn and hc@xmath3n in the evolved agb envelope . the strong sio and cs emission may suggest that depletion onto dust grains and destruction by shocks are insignificant in this object . an abundance comparison with the ppn implies to a rapid chemical evolution after a star leaves the agb stage . in order to investigate whether the molecular environment of is intrinsically unique , we systematically compare its spectra with those of . according to the comparison , we find that the molecular species can be classified into three groups , a ) for most of the species , the intensity ratios of individual lines in the two objects are in good agreement with each other ; b ) the emission from hc@xmath3n and cn may be enhanced in ; c ) the emission from sis , hcn , and c@xmath2h may be depleted in . the differences of the line - intensity ratios in the two objects are probably a consequence of chemical evolution with the exception of c@xmath2h , for which a high abundance in can not be explained by photochemical models . the iso lws spectra show that has a lower continuum - to - line ratio than , suggesting that the latter might have a larger dust - to - molecular gas ratio . using the same telescope settings , we also obtained the spectra of the agb stars and , the ppn , and the young pn . a detailed study of the chemical compositions in different evolutionary stages will be published in a separate paper . we are grateful to the aro staff for their help during the observing run . otto peng and yu - chin huang assisted in the observations . we also thank jun - ichi nakashima and jin - hua he for useful discussions . we acknowledge an anonymous referee for comments which helped strengthen the paper . the work was supported by a grant from the research grants council of the hong kong special administrative region , china ( project no . hku 7028/07p ) . lllccccc 230538.0 & co & j=21 & 4.2 & 2.555 & 67.590 & 23.7 + 220398.7 & @xmath8co & j=21 & 5.6 & 0.163 & 4.996 & 29.0 + 224714.4 & c@xmath4o & j=21 & 7.1 & 0.022 & 0.258 & 25.6 + 137180.8 & sic@xmath6 & 6@xmath1525@xmath43 & 6.8 & 0.040 & 1.040 & 24.2 + 140920.1 & sic@xmath6 & 6@xmath1535@xmath154 & 4.8 & 0.062 & 1.459 & 22.6 + 141751.5 & sic@xmath6 & 6@xmath1555@xmath156 & 6.1 & 0.073 & 1.671 & 23.4 & * + 141755.4 & sic@xmath6 & 6@xmath1575@xmath158 & & & & & * + 145325.8 & sic@xmath6 & 6@xmath1545@xmath159 & 6.5 & 0.088 & 1.867 & 21.1 + 158499.2 & sic@xmath6 & 7@xmath1606@xmath152 & 9.5 & 0.086 & 2.548 & 25.3 + 220773.7 & sic@xmath6 & 10@xmath1619@xmath162 & 5.6 & 0.087 & 2.282 & 22.8 + 222009.4 & sic@xmath6 & 9@xmath1638@xmath164 & 6.4 & 0.069 & 1.887 & 24.8 + 232534.1 & sic@xmath6 & 10@xmath1659@xmath166 & 4.0 & 0.072 & 1.937 & 25.2 + 234534.0 & sic@xmath6 & 10@xmath1679@xmath168 & 4.8 & 0.030 & 0.452 & 16.1 : & * + 234534.0 & sic@xmath6 & 10@xmath1699@xmath167 & & & & & * + 235713.0 & sic@xmath6 & 10@xmath1709@xmath171 & 5.2 & 0.045 & 1.101 & 25.2 & * + 235713.1 & sic@xmath6 & 10@xmath1719@xmath172 & & & & & * + 237150.0 & sic@xmath6 & 10@xmath1739@xmath174 & 5.8 & 0.038 & 0.859 & 20.7 + 237331.3 & sic@xmath6 & 10@xmath1749@xmath175 & 5.8 & 0.036 & 0.771 & 23.5 + 241367.7 & sic@xmath6 & 11@xmath17610@xmath161 & 4.2 & 0.074 & 1.770 & 22.7 + 254981.5 & sic@xmath6 & 11@xmath17710@xmath165 & 3.9 & 0.094 & 2.448 & 23.6 + 258065.0 & sic@xmath6 & 11@xmath16910@xmath167 & 9.3 & 0.039 & 0.471 & 10.5 : + 258065.0 & sic@xmath6 & 11@xmath17810@xmath169 & & & & & * + 259433.3 & sic@xmath6 & 11@xmath17910@xmath170 & 7.8 & 0.054 & 1.301 & 27.5 & * + 259433.3 & sic@xmath6 & 11@xmath17010@xmath171 & & & & & * + 261150.7 & sic@xmath6 & 11@xmath18010@xmath173 & 4.1 & 0.044 & 1.049 & 21.5 + 261509.3 & sic@xmath6 & 11@xmath17310@xmath174 & 4.1 & 0.051 & 1.270 & 26.5 + 261990.7 & sic@xmath6 & 12@xmath18111@xmath176 & 7.1 & 0.387 & 9.945 & 26.1 & $ + 238347.0 & @xmath5sic@xmath6 & 11@xmath17610@xmath161 & 5.3 & 0.013 & 0.366 & 35.3 + 255631.6 & @xmath5sic@xmath6 & 11@xmath17910@xmath170 & 4.2 & 0.010 & 0.127 & 32.9 : & * + 255631.8 & @xmath5sic@xmath6 & 11@xmath17010@xmath171 & & & & & * + 226287.4 & cn & n(j , f)=2(3/2,1/2)1(3/2,1/2)&4.9 & 0.034 & 1.278 & 53.0 : & * + 226298.9 & cn & n(j , f)=2(3/2,1/2)1(3/2,3/2 ) & & & & & * + 226303.0 & cn & n(j , f)=2(3/2,3/2)1(3/2,1/2 ) & & & & & * + 226314.5 & cn & n(j , f)=2(3/2,3/2)1(3/2,3/2 ) & & & & & * + 226332.5 & cn & n(j , f)=2(3/2,3/2)1(3/2,5/2 ) & 4.9 & 0.052 & 1.819 & 25.6 & * + 226341.9 & cn & n(j , f)=2(3/2,5/2)1(3/2,3/2 ) & & & & & * + 226359.9 & cn & n(j , f)=2(3/2,5/2)1(3/2,5/2 ) & & & & & * + 226616.6 & cn & n(j , f)=2(3/2,1/2)1(1/2,3/2 ) & 4.9 & 0.053 & 1.593 & 26.8 & * + 226632.2 & cn & n(j , f)=2(3/2,3/2)1(1/2,3/2 ) & & & & & * + 226659.6 & cn & n(j , f)=2(3/2,5/2)1(1/2,3/2 ) & 4.9 & 0.178 & 6.092 & 29.3 & * + 226663.7 & cn & n(j , f)=2(3/2,1/2)1(1/2,1/2 ) & & & & & * + 226679.3 & cn & n(j , f)=2(3/2,3/2)1(1/2,1/2 ) & & & & & * + 226874.2 & cn & n(j , f)=2(5/2,5/2)1(3/2,3/2 ) & 4.9 & 0.394 & 12.176 & 24.6 & * + 226874.8 & cn & n(j , f)=2(5/2,7/2)1(3/2,5/2 ) & & & & & * + 226875.9 & cn & n(j , f)=2(5/2,3/2)1(3/2,1/2 ) & & & & & * + 226887.4 & cn & n(j , f)=2(5/2,3/2)1(3/2,3/2 ) & & & & & * + 226892.1 & cn & n(j , f)=2(5/2,5/2)1(3/2,5/2 ) & & & & & * + 226905.4 & cn & n(j , f)=2(5/2,3/2)1(3/2,5/2 ) & & & & & * + 146969.0 & cs & j=32 & 4.8 & 0.303 & 7.058 & 21.6 + 138739.3 & @xmath8cs & j=32 & 4.6 & 0.023 & 0.495 & 21.0 + 231221.0 & @xmath8cs & j=54 & 5.0 & 0.027 & 0.846 & 27.6 + 242913.6 & c@xmath9s & j=54 & 7.4 & 0.032 & 0.563 : & 12.4 : + 144617.1 & c@xmath10s & j=32 & 4.9 & 0.040 & 1.039 & 26.1 + 241016.1 & c@xmath10s & j=54 & 4.6 & 0.062 & 1.315 & 23.1 + 145227.0 & sis & j=87 & 6.5 & 0.067 & 1.745 & 22.1 + 235961.1 & sis & j=1312 & 5.2 & 0.119 & 2.598 & 20.6 + 254102.9 & sis & j=1413 & 3.6 & 0.213 & 5.791 & 25.0 + 260518.0 & sio & j=65 & 4.8 & 0.233 & 5.233 & 20.3 + 254216.7 & @xmath7sio & j=65 & 3.6 & 0.022 & 0.581 & 21.5 + 257255.2 & @xmath5sio & j=65 & 6.9 & 0.021 & 0.583 & 30.3 + 262005.3 & c@xmath6h & n@xmath182=3@xmath572@xmath58 & 7.1 & 0.387 & 9.945 & 26.1 & $ + 262066.1 & c@xmath6h & n@xmath182=3@xmath582@xmath59 & 7.1 & 0.244 & 6.596 & 25.4 + 262208.4 & c@xmath6h & n@xmath182=3@xmath582@xmath58 & 6.0 & 0.024 & 0.539 : & 22.9 : + 265886.4 & hcn & j=32 & 3.7 & 1.959 & 43.942 & 21.2 + 265852.8 & hcn & @xmath183=1@xmath184 j=32 & 3.7 & 0.067 & 1.392 & 20.4 + 267109.1 & hcn & @xmath183=2@xmath185 j=32 & 4.2 & 0.011 & 0.098 : & 22.5 : & * + 267120.1 & hcn & @xmath183=2@xmath186 j=32 & & & & & * + 267199.3 & hcn & @xmath183=1@xmath187 j=32 & 4.2 & 0.050 & 0.876 & 14.4 + 267243.2 & hcn & @xmath183=2@xmath188 j=32 & 4.2 & 0.020 & 0.318 & 14.0 + 259011.8 & h@xmath8cn & j=32 & 9.6 & 0.296 & 7.120 & 20.8 + 138515.7 & c@xmath3n & n=1413 a & 7.3 & 0.039 & 0.747 & 21.5 + 138534.5 & c@xmath3n & n=1413 b & 7.3 & 0.031 & 0.674 & 25.5 + 148409.1 & c@xmath3n & n=1514 a & 4.2 & 0.021 & 0.537 & 29.9 + 148427.8 & c@xmath3n & n=1514 b & 4.2 & 0.021 & 0.521 & 35.6 + 142728.8 & c@xmath2h & n=1514 a & 6.0 & 0.022 & 0.567 & 33.7 + 142767.3 & c@xmath2h & n=1514 b & 6.0 & 0.021 & 0.450 & 23.1 + 152243.6 & c@xmath2h & n=1615 a & 5.7 & 0.018 & 0.433 & 23.6 + 152282.1 & c@xmath2h & n=1615 b & 5.7 & 0.020 & 0.505 & 27.2 + 228348.6 & c@xmath2h & n=2423 a & 4.1 & 0.015 & 0.224 : & 19.7 : + 228387.0 & c@xmath2h & n=2423 b & 4.1 & 0.035 & 0.256 : & 13.1 : + 266389.9 & c@xmath2h & n=2827 a & 3.9 & 0.008 & 0.140:&29.9 : + 266428.2 & c@xmath2h & n=2827 b & 3.9 & 0.009 & 0.175 : & 19.1 : + 136464.4 & hc@xmath3n & j=1514 & 6.6 & 0.160 & 4.131 & 24.0 + 145560.9 & hc@xmath3n & j=1615 & 6.5 & 0.139 & 3.772 & 24.1 + 154657.3 & hc@xmath3n & j=1716 & 7.0 & 0.131 & 3.165 & 24.4 + 227418.9 & hc@xmath3n & j=2524 & 6.3 & 0.064 & 1.404 & 22.0 + 236512.8 & hc@xmath3n & j=2625 & 6.2 & 0.053 & 1.078 & 21.7 + 254699.5 & hc@xmath3n & j=2827 & 3.9 & 0.068 & 1.160 & 21.7 + 263792.3 & hc@xmath3n & j=2928 & 7.9 & 0.048 & 0.990 & 19.6 + 220743.0 & ch@xmath3cn & 12(1)11(1 ) & 5.6 & 0.023 & 0.290 & 17.6 : & * + 220747.2 & ch@xmath3cn & 12(0)11(0 ) & & & & & * + lccrccr sic@xmath6 & 57.6 & 1.53(14 ) & & & 2.4(-6 ) & 3.1(-6 ) + @xmath5sic@xmath6 & & 2.26(13 ) & & & 3.6(-7 ) & + sis & 29.6 & 2.96(14 ) & & & 3.4(-6 ) & 3.1(-6 ) + c@xmath2h & 53.9 & 3.04(14 ) & @xmath189(13 ) & & 4.0(-6 ) & @xmath190(-6 ) + hc@xmath3n & 41.0 & 7.69(13 ) & @xmath191(14 ) & & 1.3(-6 ) & 2.4(-6 ) + co & & 1.63(17 ) & & & & + @xmath8co & & 1.35(16 ) & & & & + c@xmath4o & & 6.89(14 ) & & & & + cs & & 1.72(14 ) & & & 2.0(-6 ) & 2.5(-6 ) + @xmath8cs & & 1.28(13 ) & & & 1.5(-7 ) & + c@xmath9s & & 6.18(12 ) : & & & 7.2(-8 ) : & + c@xmath10s & & 2.58(13 ) & & & 3.0(-7 ) & + sio & & 2.15(13 ) & & & 7.0(-7 ) & 1.0(-6 ) + @xmath7sio & & 2.45(12 ) & & & 8.0(-8 ) & + @xmath5sio & & 2.43(12 ) & & & 7.9(-8 ) & + hcn & & 1.10(14 ) & 5.13e15 & & 1.2(-6 ) & 1.1(-5 ) + h@xmath8cn & & 1.83(13 ) & & & 2.0(-7 ) & 3.0(-7 ) + cn & & 1.22(15 ) & & & 2.6(-5 ) & 2.0(-5 ) + c@xmath6h & & 7.43(14 ) & @xmath192(14 ) & & 5.4(-6 ) & 6.9(-6 ) + c@xmath3n & & 2.16(13 ) & & & 1.9(-6 ) & 2.6(-6 ) + ch@xmath3cn & & 7.88(12 ) & & & 1.1(-7 ) & @xmath193(-7 ) + lllll @xmath116c/@xmath8c & @xmath116c@xmath10s/@xmath8c@xmath16s & 45.4@xmath194 & 45@xmath195 & 89 + & @xmath116co/@xmath8co & 12.1@xmath196@xmath197 & & + & @xmath116cs/@xmath8cs & 13.4@xmath198@xmath197 & & + & h@xmath116cn / h@xmath8cn & 6.0@xmath199@xmath197 & & + @xmath133o/@xmath4o & @xmath8c@xmath133o/@xmath116c@xmath4o & 890@xmath200 & & 2680 + & c@xmath133o / c@xmath4o & 237 @xmath201 & & + @xmath5si/@xmath7si & @xmath5sio/@xmath7sio & 1.0@xmath202 & 1.45@xmath203 & 1.52 + @xmath140si/@xmath7si & @xmath140sio/@xmath7sio & 8.8@xmath204@xmath197 & 20.3@xmath205 & 29.9 + @xmath140si/@xmath5si & @xmath140sio/@xmath5sio & 8.9@xmath206 & 15.4@xmath205 & 19.6 + & @xmath140sic@xmath6/@xmath5sic@xmath6 & 6.7@xmath207 & & + @xmath16s/@xmath10s & c@xmath16s / c@xmath10s & 6.7@xmath208@xmath197 & 21.8@xmath209 & 22.5 + @xmath9s/@xmath10s & c@xmath9s / c@xmath10s & 0.2@xmath210 & 0.18@xmath211 & 0.18 +

a class of surfaces - graphs in a riemannian 3-space with a prescribed projection of one field of principal directions onto a surface is considered . a problem of determination of such surfaces when both principal curvatures are given over a line in is formulated and studied . the geometric problem is reduced to the cauchy problem for quasilinear pde s which , under certain conditions for data , are hyperbolic and admit a unique solution . it is shown that the parallel curved ( pc ) surfaces in space forms provide a special class of global solutions to the geometrical problem with weaker regularity assumptions . such solutions may be found by an iteration function sequence . riemannian space , surface , principal curvature / direction , hyperbolic pde s * ams subject classification : * 53b20 , 53b25 , 53c99 , 53a05

1003.2184

[ s-01 ] the surfaces possessing nontrivial deformations which preserve principal curvatures and directions ( or , equivalently , the shape operator ) were investigated by several authors , see @xcite , @xcite and review with bibliography in @xcite . it is known that surfaces with one family of principal curves being geodesic ( as for parallel curved ( pc ) surfaces recently studied in @xcite@xcite ) represent degenerate case in studying immersions of simply connected surfaces with a prescribed shape operator . recent studying of reconstruction of surfaces by their partially given principal curvatures and directions may be useful for applications of differential geometry to computer graphics , the wavefront analysis in applied optics , etc . in what follows , @xmath1 denotes a @xmath2-regular riemannian 3-space with coordinates @xmath3 , @xmath4 ( @xmath5 ) for some @xmath6 , @xmath7 a @xmath2-regular surface , @xmath8 the coordinate curve , and @xmath9 the _ curvilinear projection_. in the paper we consider surfaces - graphs @xmath10 with _ prescribed curvilinear projection ( onto @xmath0 ) of one field of principal directions_. we show ( see theorem [ t - main0 ] ) that such surfaces depend on two arbitrary functions of one variable , namely , the principal curvatures over @xmath11 which are assumed close enough to corresponding values for @xmath0 . more precisely , we study the following . [ prob-1]given @xmath1 , a vector field @xmath12 transversal to @xmath11 on @xmath0 , functions @xmath13 and @xmath14 of class @xmath15 , * find * a function @xmath16 of class @xmath17 , whose graph @xmath18 in @xmath19 satisfies the conditions : \(i ) _ the projection ( @xmath20 ) onto @xmath0 of the field @xmath21 of principal directions _ _ corresponding to the principal curvature @xmath22 ( of @xmath23 ) coincides with @xmath12 _ , \(ii ) _ the principal curvatures @xmath24 ( of @xmath23 ) over @xmath11 coincide with @xmath25 : @xmath26 _ , \(iii ) _ the values of @xmath16 and @xmath27 at the point @xmath28 of @xmath11 are given_. our approach is based on reducing the * problem [ prob-1 ] * to the cauchy problem for a quasilinear system of pde s which , under certain conditions for data , is hyperbolic and admits a unique local smooth solution . the pc surfaces in @xmath29 represent a special class of solutions when a family of curvature lines projects onto @xmath0 as parallel lines or concentric circles . such surfaces are recovered by an iteration function sequence and using the reconstruction of planar curves by their curvature ( see theorem [ t - rpc2 ] and proposition [ t - rpc1 ] ) . notice that the space of pc surfaces free of umbilics and having the same shape operator depends on arbitrary function of one variable , see @xcite . the structure of the work is the following . * section [ sec : main - res ] * represents main results ( theorems [ t - main0][t - rpc2 ] and corollary [ t - main3 ] for @xmath30 ) . * section [ sec : proofs ] * contains proofs . * section [ sec : appendix ] * contains necessary facts on pc surfaces and examples . we shall use the following notation : @xmath31 ; @xmath32 , @xmath33 the linear space of bounded continuous functions @xmath34 ( @xmath35 is a domain in @xmath36 ) ; @xmath37 the norm in @xmath33 ; @xmath38 is the set of functions @xmath34 , having in @xmath35 continuous partial derivatives of order @xmath39 . for short , we omit @xmath40 from the above notations . for simplicity , we assume in what follows that @xmath0 is a totally umbilical surface with the normal curvature @xmath41 ( if @xmath42 then @xmath0 is totally geodesic ) . the main result of the paper is the following . [ t - main0 ] let @xmath43 be functions of class @xmath44 , @xmath12 a vector field of class @xmath45 , that is transversal but not orthogonal to @xmath11 . if @xmath46 are small enough , then * problem [ prob-1 ] * admits in @xmath1 a unique local solution . namely , there are @xmath47 such that if @xmath48 @xmath49 , then for some @xmath50 $ ] there a function @xmath16 of class @xmath51 on @xmath52 with the properties : the principal curvatures @xmath24 of @xmath53 satisfy @xmath54 , @xmath12 is tangent to the @xmath20-projection onto @xmath55 of @xmath22-curvature lines , and @xmath56 . moreover , there is @xmath57 $ ] such that the solution @xmath16 is in the class of @xmath2-regular functions satisfying @xmath58 . one may apply theorem [ t - main0 ] to surfaces in 3-space forms ( see also section [ sec : warped2 ] ) . we illustrate this for @xmath29 with cartesian coordinates . [ t - main3 ] let @xmath59 be a function of class @xmath60 on a rectangle @xmath61 in @xmath29 ( with cartesian coordinates ) , and @xmath43 functions of class @xmath62 on the segment @xmath63 . if @xmath64 are small enough , then * problem [ prob-1 ] * admits a unique local solution . namely , there are @xmath47 such that if @xmath65 , then for some @xmath50 $ ] there a function @xmath16 of class @xmath51 on @xmath66 with the properties : the principal curvatures @xmath24 of the graph @xmath67 in @xmath29 satisfy @xmath68 , the vector field @xmath69 is tangent to the projection onto @xmath55 of @xmath22-curvature lines , and @xmath56 . moreover , there is @xmath70 such that the solution @xmath16 is in the class of @xmath2-regular functions satisfying @xmath71 . [ r-01 ] ( a ) the condition that @xmath0 is totally umbilical can be dropped . in this case , @xmath25 should be close enough to corresponding principal curvatures of @xmath0 along @xmath11 , and @xmath12 close enough to one of principal directions on @xmath0 . the values of @xmath72 and @xmath73 can be taken small enough ( see proposition [ p - twoodes ] ) , in the present text for simplicity we assume them zero . if @xmath74 in corollary [ t - main3 ] , then @xmath23 is a pc surface , i.e. , the planes @xmath75 intersect @xmath23 by curvature lines , see section [ sec : ex-003 ] . pc surfaces in spherical coordinates , see also section [ sec : ex-003 ] , illustrate theorem [ t - main0 ] . \(b ) a _ normal geodesic graph in @xmath1 of a function _ @xmath76 ( @xmath77 is the focal radius of @xmath0 ) is a surface @xmath78 , where @xmath79 is a unit speed geodesic normal to @xmath0 . the _ semi - geodesic coordinates _ @xmath80 on @xmath0 with the base curve @xmath81 ; have the metric is @xmath82 , where @xmath83 , @xmath84 ( see @xcite ) , and @xmath85 is the gaussian curvature of @xmath0 . let @xmath19 has the coordinates @xmath86 ( @xmath5 ) such that * * @xmath80 are semi - geodesic coordinates on a surface @xmath7 , * * the curve @xmath8 is a simple geodesic in @xmath0 , and * * @xmath4 is the signed distance to @xmath0 ( hence @xmath87 and @xmath88 ) . one may obtain corollary of theorem [ t - main0 ] , where @xmath89 means the nearest point " in @xmath0 ( for @xmath29 we again have corollary [ t - main3 ] ) . we will formulate for @xmath19 ( and study in the paper for @xmath29 ) the problem that plays essential role in solving * problem [ prob-1 ] * for pc surfaces . [ prob-2]given a vector field @xmath90 on @xmath91 , an integral curve @xmath92 of @xmath12 through @xmath93 , a curve @xmath94 transversal to @xmath95 through @xmath96 and functions @xmath97 on @xmath98 ( @xmath99 ) , * find * a function @xmath16 of class @xmath17 , whose graph @xmath18 in @xmath19 satisfies the conditions : ( i ) _ the projection ( @xmath20 ) onto @xmath0 of the field @xmath21 of principal directions _ _ corresponding to the principal curvature @xmath22 of @xmath23 , coincides with @xmath12 _ , \(ii ) _ the principal curvatures @xmath100 of @xmath23 coincide with @xmath25 over @xmath98 : @xmath101 _ , \(iii ) _ the values of @xmath16 and @xmath27 at the point @xmath96 are given_. define a rectangle @xmath102 in the @xmath103-plane of @xmath29 , where @xmath104 are positive . the pc surfaces represent a special class of global solutions ( i.e. , on the domains @xmath105 with an arbitrary @xmath106 ) to * problems [ prob-1 ] * and * [ prob-2 ] * with weaker regularity assumptions . [ t - rpc2 ] -0.8pt let @xmath107 be real , @xmath108)$ ] and @xmath109)$ ] . then for any @xmath110 one can choose @xmath111 such that if @xmath112}<\delta$ ] @xmath49 , then there exists a unique pc surface @xmath113 , where @xmath114 , with the properties -2 mm the @xmath22-principal direction @xmath21 on @xmath23 is parallel to the plane @xmath115 , -2 mm the principal curvatures of @xmath23 satisfy @xmath116 . theorem [ t - rpc2 ] is based on the following result concerning existence of a solution to * problem [ prob-2 ] * in the class of pc surfaces . [ t - rpc1 ] given @xmath107 , let @xmath117 and @xmath118 be line segments in @xmath103-plane of @xmath29 . let @xmath119 and @xmath120 . then for any @xmath110 there is @xmath111 such that if @xmath121 then there exists a unique pc surface @xmath67 in @xmath29 , where @xmath114 , with the properties -3 mm the principal direction @xmath21 is orthogonal to the vector @xmath122 , -3 mm @xmath123 is the principal curvature of @xmath23 over @xmath95 corresponding to @xmath21 , and -3 mm @xmath14 is the principal curvature of @xmath23 over @xmath124 corresponding to the second principal direction @xmath125 . the solution can be found using reconstruction of two planar curves by their curvature functions . in section [ sec : results ] we prove theorem [ t - main0 ] and its corollary [ t - main3 ] . based on the euler formula for the principal curvatures we deduce a system of pde s ( proposition [ t - compr ] ) . using compatibility conditions , we transform above equations to equivalent quasi - linear system ( proposition [ t-01 ] ) , for which we formulate the cauchy problem . the initial values are analyzed in lemma [ l-04 ] , where @xmath23 is recovered over @xmath11 . section [ sec : th2 - 3 ] shows that the pc surfaces in space forms ( i.e. , a family of curvature lines projects onto parallel lines or concentric circles ) represent a special class of global solutions to the geometrical problem with weaker regularity assumptions ( @xmath126),\bar k_{2}\in c^{0}([-a_1 , a_1])$ ] ) . we approximate such solutions by the iterated function sequence , and apply the procedure of reconstruction of two plane curves by their curvature functions ( theorem [ t - rpc2 ] and proposition [ t - rpc2-b ] ) . section [ sec : axresults ] contains auxiliary lemmas and necessary facts on hyperbolic pde s . a _ surface - graph _ @xmath23 in @xmath19 is defined by equation @xmath127 . let @xmath128 be coordinate vector fields on @xmath19 , @xmath129 their restrictions on @xmath23 , and @xmath130 the coordinate vector fields ( lifts under @xmath20 of @xmath131 from @xmath0 ) on @xmath23 . for simplicity assume in what follows that @xmath132 ( see also remark [ r-01 ] ) . we have @xmath133 and @xmath134 , where @xmath135 and @xmath136 . the metric on @xmath23 is given by @xmath137 . the coefficients of the first and the second fundamental forms of @xmath23 are denoted by @xmath138 suppose that @xmath139 are the principal directions on @xmath23 . we compute the product @xmath140 . hence @xmath141 . denote by @xmath142 . from above and lemma [ l - efg-1 ] ( section [ sec : axresults ] ) it follows @xmath143 define the functions @xmath144 , where @xmath145 @xmath146 , see ( [ e - alpha-3])@xmath147 , and @xmath148 are given in ( [ e - l1-m1-n1 ] ) of section [ sec : axresults ] . for @xmath29 we get @xmath149 , where @xmath150 , @xmath151 and @xmath152 . [ ex - pi]let @xmath153 , hence @xmath154 . by lemma [ l - efg-1 ] ( section [ sec : axresults ] ) , the coefficients of the 1-st and the 2-nd fundamental forms are @xmath155 the functions ( [ e - hijkr ] ) on @xmath0 have the following form : @xmath156 where @xmath157 . the function @xmath158 ( of the principal direction of @xmath0 ) satisfies @xmath159 the principal curvatures @xmath160 of @xmath0 are solutions to the quadratic equation @xmath161 since @xmath0 is totally umbilical , @xmath162 are the roots of ( [ e - tumb2 ] ) , hence @xmath163 in this case , ( [ e - tumb1 ] ) is satisfied by any @xmath158 . first , we will prove propositions [ t - compr ] , [ t-01 ] and lemma [ l-04 ] . [ t - compr ] let @xmath10 be the graph of @xmath164 . then @xmath165 are the principal curvatures and @xmath166 ( with @xmath167 ) is the projection onto @xmath0 of @xmath22-principal direction of @xmath23 if and only if @xmath168 if @xmath169 , @xmath170 and @xmath171 , then the compatibility conditions for ( [ e-02-r],b ) are reduced to pde s @xmath172 where @xmath173 are known functions ( see the proof ) . the characteristics of ( [ e-3.2-r ] ) are the projections onto @xmath0 of curvature lines of @xmath23 . -3 mm * proof*. let @xmath174 be a vector on @xmath23 . the functions @xmath175 are the principal curvatures of @xmath23 , and @xmath176 is the projection onto @xmath0 of @xmath22-principal direction if and only if they satisfy the euler formula @xmath177/i(v)$ ] , with the normal curvature @xmath178 . hence @xmath179k_1+[\,g(v,\partial_2)^2/i(\partial_2)]k_2.\ ] ] notice that if @xmath180 is the principal direction on @xmath23 then @xmath181 is the second principal direction . we have @xmath182 since @xmath183 is arbitrary , ( [ e - euler ] ) is equivalent to the system @xmath184 which , in view of @xmath185 and @xmath186 ( see lemma [ l - efg-1 ] in section [ sec : axresults ] ) , yields ( [ e-02-r],b ) . if @xmath169 and @xmath170 , then the compatibility conditions for ( [ e-02-r],b ) , i.e. , @xmath187 take a form @xmath188 substituting ( [ e-02-r],b ) into ( [ e-18-r ] ) , we obtain pde s for @xmath189 @xmath190 where the matrices @xmath191 , @xmath192 are @xmath193-regular , and the components of the vector @xmath194 are @xmath195 notice that @xmath196 when @xmath197 . the direct computation shows that @xmath198 , where @xmath199 and @xmath200 . hence ( [ e-3.1-r ] ) is equivalent to the system ( [ e-3.2-r ] ) with @xmath201.@xmath202 [ r - petcod]let @xmath203 be a @xmath2-regular surface without umbilical points parameterized by coordinates @xmath204 of the curvature lines , @xmath205 . hence @xmath206 . let @xmath207 be the principal curvatures of @xmath23 . in this case , if @xmath208 , then @xmath209 , @xmath210 and other @xmath211 are zero . hence @xmath212 , @xmath213 and other @xmath214 are zero . hence , ( [ e-3.1-r ] ) is reduced to the system @xmath215 that is equivalent to peterson - codazzi formulae ( see @xcite ) @xmath216 . one may compare the last formulae with our compatibility equations ( [ e-18-r ] ) . [ t-01 ] let the functions @xmath217 of class @xmath193 on @xmath0 ( in particular , on @xmath55 ) satisfy ( [ e-04-r ] ) . \(i ) if ( [ e-02-r ] ) holds in @xmath0 , then @xmath169 and ( [ e-18-r ] ) holds in @xmath0 . \(ii ) if ( [ e-02-r ] ) holds for @xmath218 and ( [ e-18-r ] ) holds in @xmath0 , then ( [ e-02-r ] ) holds in @xmath0 . * proof*. ( i ) if ( [ e-02-r ] ) is satisfied in @xmath0 , then , since @xmath219 are of class @xmath193 , equations ( [ e-02-r],b ) imply that @xmath220 are of class @xmath221 . hence @xmath169 and , by commutativity of partial derivatives , ( [ e-18-r ] ) holds in @xmath0 . \(ii ) denote for short @xmath222 . by ( [ e-02-r ] ) with @xmath223 and ( [ e-04-r ] ) , we conclude that @xmath220 and @xmath16 satisfy in @xmath0 the integral equations @xmath224 ( for short we omit the variables @xmath22 and @xmath225 in @xmath219 ) . by conditions imposed on @xmath226 , @xmath227 , @xmath16 and @xmath24 , one may differentiate by @xmath228 the first integrand in ( [ e-31b ] ) . using ( [ e-04-r ] ) and @xmath229 , we get @xmath230\/d\eta\\ \hskip-10 mm = h_{11}(u,0,f(u,0),p(u,0),q(u,0))\\ \hskip-10 mm + \int\nolimits_{0}^v\frac{d}{d\eta}h_{11}(u,\eta , f(u,\eta),p(u,\eta),q(u,\eta))\,d\eta\\ \hskip-3 mm + \int\nolimits_{0}^v\big[h_{12,p}(p_{u}{-}h_{11 } ) { + } h_{12,q}(q_{u}{-}h_{12}){+}h_{12,f}(f_{u}-p){+}h_{12,u}{+}h_{12,p}h_{11}\\ \hskip-3 mm + h_{12,q}h_{12}+h_{12,f}p - h_{11,v}{-}h_{11,p}h_{12 } -h_{11,q}h_{22 } -h_{11,f}q\big]\,d\eta . \end{array } \end{aligned}\ ] ] define the functions @xmath231 , and @xmath232 by ( [ e-18-r])@xmath233 , from ( [ e-32 ] ) it follows @xmath234\,d\eta . \end{array}\ ] ] similarly , differentiating ( [ e-31c ] ) by @xmath228 , and using ( [ e-04-r ] ) and ( [ e-18-r])@xmath235 , we obtain @xmath236\,d\eta . \end{array}\end{aligned}\ ] ] differentiating ( [ e-31d ] ) by @xmath228 and using the first equation in ( [ e-04-r ] ) , we obtain @xmath237\,d\eta . \end{array}\ ] ] hence the following equation is satisfied : @xmath238 for each @xmath239 $ ] the system of integral equations ( [ e-34]-c ) is equivalent to cauchy problem for linear homogeneous ode s with initial conditions @xmath240 . hence @xmath241 , and ( [ e-02-r ] ) are satisfied . @xmath202 1.0 mm first , we will recover the graph @xmath23 of @xmath242 $ ] infinitesimally along @xmath11 , i.e. , to solve ( [ e - pqfa1 ] ) for @xmath243 and @xmath244 . define the quantity @xmath245 [ l-04 ] let @xmath246 . then for any @xmath57 $ ] there is @xmath247 $ ] such that for @xmath248 satisfying @xmath249 , the cauchy problem @xmath250 has on @xmath11 a unique @xmath62-regular solution @xmath251 satisfying @xmath252 . * proof*. denote @xmath253 for @xmath99 . substituting @xmath254 into ( first two equations of ) ( [ e - pqfa1 ] ) and using ( [ e - tumb4 ] ) gives us along @xmath11 @xmath255 with @xmath256 , and @xmath257 . due to example [ ex - pi ] , @xmath258 on @xmath0 , see ( [ e - tumb4 ] ) . hence @xmath259 is the solution to ( [ e - tumb3b ] ) with @xmath260 . let us take @xmath57 $ ] . we claim that if @xmath261 are close enough to @xmath262 , then the cauchy problem ( [ e - tumb3b ] ) has on @xmath11 a unique smooth solution satisfying @xmath263 . in aim to apply proposition [ p - twoodes ] to ( [ e - tumb3b ] ) and to the same system with @xmath264 ( @xmath265 in the last case ) , denote by @xmath266 since the functions @xmath267 and @xmath268 have continuous partial derivatives w.r . to @xmath269 , they satisfy in @xmath270 the lipschitz condition ( for @xmath269 ) with some @xmath271 . by ( [ e - alpha-3])@xmath233 , @xmath272 , and by ( [ e - hijkr ] ) , there is @xmath273 such that @xmath274 on @xmath275 . hence , @xmath276 where @xmath277 is defined by ( [ dfnbrhoa ] ) . assume that @xmath278 by the theory of ode s , there is a maximal interval @xmath279)$ ] , in which ( [ e - tumb3b ] ) admits a unique solution with the property @xmath280 on the other hand , in view of ( [ estqmp ] ) , ( [ e - rhoa - ineq ] ) and proposition [ p - twoodes ] , this solution satisfies in @xmath281 $ ] strong inequalities @xmath282 , @xmath283 and @xmath284 . if @xmath285 or @xmath286 , due to the theory of ode s the solution can be extended on a larger interval with the property ( [ prop ] ) . hence , @xmath287 . @xmath202 1.0 mm * proof of theorem [ t - main0]*. since @xmath12 is transversal to @xmath11 , one may assume @xmath288 for some function @xmath289 of class @xmath60 in a neighborhood of @xmath11 in @xmath0 . since @xmath12 is not orthogonal to @xmath11 , we have @xmath290 along @xmath11 . there is @xmath291 $ ] such that @xmath292 over @xmath11 for @xmath293 . define the functions @xmath294 and @xmath295 by ( [ e - efg],c ) in what follows . in view of @xmath296 , see ( [ e - alpha-3])@xmath235 , we get @xmath171 on the set @xmath297 for some @xmath298 $ ] . restricting ( [ e-02-r ] ) on @xmath11 and denoting @xmath299 , @xmath300 and @xmath301 , yields the system ( [ e - pqfa1 ] ) for the functions @xmath302 and @xmath303 , @xmath304 . by lemma [ l-04 ] , there exist @xmath247 $ ] such that if @xmath305 , the cauchy problem ( [ e - pqfa1 ] ) has on @xmath306 $ ] a unique solution @xmath251 of class @xmath193 , satisfying @xmath307 . by propositions [ t - compr ] and [ t-01 ] , the * problem [ prob-1 ] * is reduced to the cauchy problem , see ( [ e-04-r ] ) and ( [ e-3.2-r ] ) , @xmath308 with the initial conditions for the functions @xmath309 and @xmath207 , @xmath310 notice that there exists @xmath311 $ ] such that @xmath171 if @xmath312 , and ( [ e - system-5pdes ] ) is non - singular . since @xmath313 , from the definition of @xmath314 ( see the proof of proposition [ t - compr ] ) it follows that the functions @xmath315 , where @xmath316 the normal curvature @xmath41 of a @xmath2-surface @xmath0 is @xmath193-regular . from above it follows that functions in right hand side of ( [ e - system-5pdes ] ) belong to class @xmath317 . the system ( [ e - system-5pdes ] ) for the functions @xmath318 is hyperbolic , and it has a diagonal form in its main part ( containing the derivatives of unknown functions ) . first three families of characteristics of ( [ e - system-5pdes ] ) are lines @xmath319 and the last two families of characteristics are integral curves of ode s @xmath320 and @xmath321 . denote @xmath322 , where @xmath323 is the projection of @xmath324 onto the space of variables @xmath309 . by theorem a ( and remark after it , with @xmath325 for @xmath24 ) there is @xmath326 $ ] such that the cauchy problem ( [ e - system-5pdes ] ) , ( [ e - bc0-bb2],b ) admits a unique solution @xmath327 with @xmath328 . by proposition [ t-01 ] , equations ( [ e-02-r],b ) are valid in @xmath55 and @xmath329 . furthermore , by proposition [ t - compr ] , the functions @xmath330 are the principal curvatures of the graph @xmath53 , and , in view of ( [ initbrk ] ) , the conditions @xmath331 are satisfied . thus , the surface @xmath23 represents a solution of * problem [ prob-1]*. suppose that a @xmath2-regular surface - graph @xmath53 over @xmath55 is a solution of * problem [ prob-1 ] * ( with principal curvatures @xmath332 ) satisfying the condition @xmath333 , where @xmath85 , @xmath334 and @xmath335 have been chosen above . by propositions [ t - compr ] and [ t-01 ] , @xmath336 is a solution to ( [ e - system-5pdes ] ) and ( [ e - bc0-bb2],b ) , in which @xmath251 is a solution to ( [ e - pqfa1 ] ) . since solutions of these cauchy problems are unique , the solution of * problem [ prob-1 ] * is also unique in the class of functions under consideration . @xmath202 1.0 mm * proof of corollary [ t - main3]*. let @xmath23 in @xmath29 ( with cartesian coordinates ) be a surface - graph of @xmath337 defined on @xmath61 . by lemma [ l - efg-1 ] or directly , we find that @xmath338 $ ] is the unit normal to @xmath23 , and the 1-st and the 2-nd fundamental forms of @xmath23 are @xmath339 , and @xmath340 . from proposition [ t - compr ] it follows that @xmath165 are the principal curvatures and the vector @xmath341 ( where @xmath342 ) is the projection onto @xmath0 of @xmath22-principal direction of @xmath23 if and only if @xmath343 for @xmath344 , the system ( [ e-04-r])([e-3.2-r ] ) with @xmath345 has a form ( see also ( [ e - hijkr - r3 ] ) ) @xmath346 where @xmath347 , @xmath348 , @xmath349 . if @xmath169 and @xmath350 , then ( [ e - system-4pdes])@xmath351 are compatibility conditions for ( [ e-02],b ) , i.e. , @xmath352 , @xmath353 . similarly to the proof of theorem [ t - main0 ] , one may show that there are @xmath354 and @xmath355 such that if @xmath65 , then for some @xmath356 the cauchy problem ( [ e - system-4pdes ] ) with @xmath357 is non - singular and admits a unique solution @xmath358 of class @xmath359 with @xmath360 . moreover , there is a unique function @xmath16 of class @xmath361 such that @xmath362 and @xmath56 . as in the proof of theorem [ t - main0 ] , one may show that @xmath329 and the surface - graph @xmath363 represents a unique solution of * problem [ prob-1 ] * in the class of functions from the formulation of the corollary . @xmath202 1.0 mm a surface @xmath364 in @xmath365 is called _ parallel curved _ ( pc ) if there is a totally umbilical ( or totally geodesic ) surface @xmath366 such that at each point @xmath367 there is a principal direction tangent to @xmath368 ( a parallel surface to @xmath369 on the distance @xmath370 ) . surfaces of revolution and cylinders in euclidean space @xmath29 provide examples of pc surfaces ( see @xcite and @xcite ) . we study pc surfaces in space forms @xmath365 ( see also section [ sec : warped2 ] ) in relation to * problems [ prob-1 ] * and * [ prob-2]*. a pc surface @xmath23 in @xmath365 ( a riemannian 3-space of constant curvature @xmath39 ) can be recovered locally by exact procedure . we will prove the claim for pc surfaces in @xmath29 . in proposition [ t - rpc1 ] we solve * problem [ prob-2 ] * using twice the reconstruction a plane curve by its curvature . 1.0 mm * proof of proposition [ t - rpc1 ] * will be divided into three steps . 1 . we apply geometrical construction of proposition [ l - rpc-1a ] when @xmath371 is a plane with the normal @xmath372 . in this case , @xmath176 is a constant vector field ( its trajectories are parallel lines ) . the planes parallel to @xmath369 intersect @xmath23 transversally by curvature lines , fig . [ f-00y](a ) . let @xmath373 be the orthogonal projection onto the plane @xmath374 . let @xmath375 be a @xmath22-curvature line on @xmath23 through the origin @xmath376 . the normals to @xmath377 ( lines in @xmath378 ) and parallel curves ( of constant distance ) to @xmath377 form a _ semi - geodesic net _ on @xmath378 on a neighborhood of @xmath377 . let us parameterize @xmath377 @xmath379 in @xmath380-coordinates of @xmath378 . notice that @xmath381 ( @xmath382 is a unit normal to @xmath378 ) is the orthonormal frame of @xmath29 . the curvature of @xmath377 is @xmath383 from ( [ e - v0-gamma0 ] ) and ( [ e - ku ] ) it follows the inequality @xmath384^{3/2}}\,ds $ ] . by lemma [ appl ] , if @xmath385 , then @xmath386 ( apriori estimate ) for @xmath387 . if we take @xmath388 , then a unique solution @xmath389 to ( [ e - v0-gamma0 ] ) , ( [ e - ku ] ) , defined for @xmath390 , satisfies the inequalities @xmath391 since @xmath392 is @xmath193-regular , from ( [ e - ku ] ) it follows that @xmath389 is @xmath2-regular . hence , the curve @xmath393^{1/2}},\ v_0(u)+\frac{\rho}{[1+{v_0'}^{2}(u)]^{1/2}}]$ ] , ( on the distance @xmath394 to @xmath377 in the plane @xmath378 ) is @xmath395-regular for small enough @xmath394 , its curvature is @xmath396 , see @xcite . assume now that @xmath397 ( the function of @xmath398 ) and translate @xmath399 on the height @xmath398 in the normal direction to @xmath378 . then , the obtain @xmath400-parametrization of @xmath23 near @xmath377 with @xmath378-level curves @xmath401 , @xmath402 using the equation for principal curvatures , @xmath403 , since @xmath206 , we find @xmath404 notice that @xmath405 is the curvature of @xmath401 multiplied by @xmath406 , see ( [ e - k12])@xmath233 , where @xmath407 is the angle between @xmath378 and the tangent plane to @xmath23 through the intersection point . the 2-nd principal curvature , @xmath408 ( does not depend on @xmath228 ) is simply the curvature of the curve @xmath409 $ ] in the vertical plane @xmath410 with the @xmath394-axis @xmath411 . one may recover a @xmath221-regular function @xmath412 for small enough @xmath398 from ( [ e - k12])@xmath235 , solving the bvp @xmath413 if @xmath414 , then by lemma [ appl ] , a unique solution @xmath412 to cauchy problem ( [ e - rhok2 ] ) exists for @xmath415 , and satisfies the inequalities @xmath416 one may recover @xmath392 from ( [ e - k12])@xmath233 for @xmath417 with known @xmath418 . finally , @xmath389 is a unique @xmath221-regular solution to ( [ e - ku ] ) with initial values ( [ e - v0-gamma0 ] ) . it was shown that _ if @xmath419 , then the parametrization ( [ e - mpar ] ) defines a regular surface over the rectangle of parameters @xmath400 _ @xmath420 equations ( [ e - mpar ] ) of @xmath23 in cartesian coordinates @xmath421 take a form @xmath422 in view of ( [ e - mpar ] ) , ( [ e - mpar2 ] ) , @xmath2-regularity of @xmath389 and @xmath221-regularity of @xmath412 , the surface @xmath23 is @xmath221-regular . \3 . along @xmath95 ( i.e. , @xmath417 ) we get @xmath423_{|(u,0)}= [ \frac{1}{\sqrt{\alpha^2 + 1}},\ \frac{-\alpha}{\sqrt{\alpha^2 + 1}},\ 0]$ ] , and @xmath424}_{|(u,0)}= [ \frac{\alpha}{\sqrt{\alpha^2 + 1}},\ \frac{1 } { \sqrt{\alpha^2 + 1}},\ v_0'(u)].$ ] hence @xmath425 @xmath426 , and @xmath23 regularly projects onto a neighborhood of @xmath95 in @xmath103-plane . it is not difficult to see that @xmath23 is @xmath221-regular . let us show that for any @xmath110 there is @xmath111 such that if @xmath427 , the surface @xmath23 regularly projects onto @xmath105 . based on ( [ e - mpar])@xmath428 , we write ( [ e - mpar2])@xmath429 in the equivalent form @xmath430 . consider the mapping @xmath431 $ ] associated with above system . we will show that for any @xmath432 the system @xmath433 admits a unique solution @xmath400 in @xmath434 . notice @xmath435 is the image of @xmath105 under the linear mapping @xmath436 , @xmath437 . assume that the metric in @xmath434 is induced by the norm @xmath438 . in order to apply the banach fixed point theorem , we will show that for a small enough @xmath439 the mapping @xmath440 maps @xmath434 into itself , and that @xmath440 is a contraction . if @xmath441 , then using ( [ estrh1])@xmath233 , we obtain for @xmath432 and @xmath442 that @xmath443 and @xmath444 under above condition for @xmath439 , @xmath440 maps @xmath434 into itself . to show that @xmath445 is a contraction , we find the differential of @xmath440 at @xmath446 : @xmath447 using ( [ estv0 ] ) , ( [ estrh1 ] ) , we obtain the following estimates : @xmath448 hence , for a small enough @xmath439 the norm @xmath449 is less than @xmath450 for any @xmath446 , that is the mapping @xmath440 is a contraction in @xmath434 . thus , for small enough @xmath439 , the projection @xmath451 ( the orthogonal projection onto the @xmath103-plane ) realizes a bijection between the surface @xmath452 and @xmath105 . since the linear operator @xmath453 is invertible for any @xmath446 , by the implicit function theorem this projection is regular.@xmath202 1.5 mm * proof of theorem [ t - rpc2]*. the condition @xmath454 determines the intersection curve of @xmath23 ( see ( [ e - mpar2 ] ) ) with @xmath455-plane of the form @xmath456 , where @xmath457 . now , @xmath458 and @xmath389 are functions to be found using the boundary values of the principal curvatures . denote @xmath459 . consider the system @xmath460 where @xmath461 , @xmath462 , @xmath463 , @xmath464 is a continuous in @xmath465 $ ] function , and @xmath466 , @xmath467 in the integrand . due to the proof of proposition [ t - rpc1 ] , we are looking for the parametric form ( [ e - mpar2 ] ) of @xmath23 , where @xmath468 ( the functions of @xmath469 ) are given in ( [ e - mpar ] ) , and @xmath389 satisfies ( [ e - ku ] ) with unknown function @xmath392 . by proposition [ t - rpc1 ] , the principal curvature @xmath225 of @xmath23 satisfies @xmath470 . it is sufficient to show that for small enough @xmath25 there exists a unique pair @xmath471 ( and hence @xmath23 ) such that @xmath472 for the principal curvature @xmath22 of @xmath23 . from ( [ e - mpar2 ] ) with @xmath473 we conclude that the pair @xmath474 satisfies the equation @xmath475 for @xmath476 $ ] @xmath477 . in order to satisfy @xmath472 , we rewrite ( [ e - k12])@xmath233 as @xmath478 substituting @xmath392 of ( [ e - ku ] ) into ( [ e - d - u ] ) , we get the integral equation @xmath479 this leads to the system ( [ eqhu1 ] ) for the functions @xmath458 and @xmath480 . we claim that for small enough @xmath13 and @xmath14 , ( [ eqhu1 ] ) admits a unique solution in the product space @xmath481 , where @xmath482 is the closed ball of radius @xmath70 in @xmath483 centered at the function @xmath484 . first we will investigate ( [ eqhu1])@xmath233 . set @xmath485 . by the proof of proposition [ t - rpc1 ] , if @xmath486 , the cauchy problem ( [ e - rhok2 ] ) admits a unique solution @xmath412 defined for @xmath487 and the estimates ( [ estrh1 ] ) are valid . denote @xmath488 from ( [ estrh1])@xmath429 we get that if @xmath489 . hence @xmath490 for any @xmath491\times i\times{{\mathbb r}}$ ] , that is @xmath492\to[-a_1,a_1],\\ \label{dfdx } |\partial_x f|\le\frac{\sqrt{\alpha^2 + 1}}{\alpha}|\tilde\rho'(x)|\le \frac{\sqrt{2}\,a_1}{\alpha\,\sqrt{\alpha^2 + 1}}\,\bar k\le\frac{1}{\sqrt{2}}.\end{aligned}\ ] ] hence @xmath493 for any @xmath494\times i\times{{\mathbb r}}$ ] . we conclude that ( [ eqhu1])@xmath233 admits a unique solution @xmath495 $ ] for any @xmath496 . since the function @xmath497 is @xmath193-regular in @xmath306\times i\times{{\mathbb r}}$ ] , by the implicit function theorem , the function @xmath498 is @xmath193-regular in @xmath499 . let us substitute @xmath500 into ( [ eqhu1])@xmath235 , which is equivalent to cauchy problem @xmath501 we will show that ( [ eq - dwdu ] ) admits a unique solution @xmath502 in @xmath483 . by estimates ( [ estrh1])@xmath429 and definition of function @xmath503 , we have @xmath504 for @xmath505 . since @xmath506 , @xmath507 and @xmath508 are @xmath193-regular in @xmath306 $ ] , the ode ( [ eq - dwdu ] ) satisfies the conditions required for local existence and uniqueness of a solution to cauchy problem . in order to show that the solution to ( [ eq - dwdu ] ) does not blow up in @xmath509 , we need an apriori estimate of a solution to ( [ eqhu1])@xmath235 with @xmath500 . let @xmath510 be a continuous solution to this equation in @xmath511\subseteq i$ ] . from ( [ eqhu1])@xmath235 and ( [ estblw ] ) it follows @xmath512).\ ] ] by lemma [ appl ] , the estimate @xmath513 is valid in @xmath511 $ ] , if @xmath514 . since the bound for @xmath515 in @xmath511 $ ] does not depend on @xmath516 , the solution @xmath502 exists and is continuous on @xmath509 . so , we have proved the claim ( @xmath517 ) in the product space @xmath518 , moreover , by above apriori estimate , the solution belongs to the smaller space @xmath519 . the desired @xmath221-regular surface @xmath23 is given by ( [ e - mpar2 ] ) , ( [ e - mpar ] ) , where @xmath520 . using ( [ e - d - u])@xmath235 and proposition [ t - rpc1 ] , we obtain that for any @xmath521 there exists @xmath111 such that if @xmath427 , then @xmath23 projects regularly onto @xmath105 and @xmath23 is a unique surface with the properties indicated in the theorem.@xmath202 [ t - rpc2-b ] the solution in theorem [ t - rpc2 ] can be represented in the form ( [ e - mpar ] ) , ( [ e - mpar2 ] ) , where @xmath520 and w(u ) is the second component of the solution @xmath522 to ( [ eqhu1 ] ) . moreover , @xmath522 is the limit of the iterated function sequence for the operator @xmath523 in @xmath524 @xmath525\ ] ] with the starting point @xmath526 . * proof*. indeed , a fixed point of @xmath523 is a solution to ( [ eqhu1 ] ) . let us prove that for a small enough @xmath439 the solution @xmath527 to ( [ eqhu1 ] ) ( that determines the surface @xmath23 ) can be found by an iterative process . first we will show that for a small enough @xmath439 the operator @xmath523 maps @xmath528 into itself . denote @xmath529 . by ( [ estrh1])@xmath428 and ( [ estblw ] ) , for @xmath530 we have @xmath531 by this and ( [ propf ] ) , for a small enough @xmath439 the operator @xmath523 maps @xmath532 into itself . let us show that @xmath533 is a contraction w. r. to some metric on @xmath532 for a small enough @xmath439 . from ( [ estrh1])@xmath428 we obtain @xmath534 , if @xmath491\times i\times{{\mathbb r}}$ ] . this estimate and ( [ dfdx ] ) mean that @xmath535 and @xmath536 becomes arbitrary small for a small enough @xmath439 , hence the first component of @xmath537 satisfies w.r . to @xmath538 the lipschitz condition with the lipschitz constant @xmath539 for a small enough @xmath439 . let us compute the differential of the second component @xmath540 ( of @xmath523 ) at a point @xmath541 : @xmath542 where @xmath543 and @xmath544 . using @xmath545 and ( [ estrh1 ] ) , ( [ estblw ] ) , yields that the function @xmath546 satisfies the lipschitz condition w.r . to @xmath538 in @xmath306\times[-1,1]$ ] with a lipschitz constant @xmath547 for a small enough @xmath439 . notice that @xmath548 is not arbitrary small for a small enough @xmath439 , because the expression for @xmath549 contains the derivative @xmath550 that is not assumed to be small . in aim to show that @xmath523 is a contraction , let us define the following metric in the second component @xmath551 of the product space @xmath532 : @xmath552 ( see @xcite ) , where @xmath553 will be chosen in the sequel . clearly , this metric is equivalent to the original @xmath554-metric @xmath555 in @xmath551 . the metric of @xmath532 is @xmath556 let us estimate for @xmath557 , @xmath558 , @xmath559 : @xmath560 a similar estimate is valid for @xmath561 . thus , @xmath562 above arguments imply that if @xmath563 and @xmath439 is small enough , then @xmath523 is a contraction in @xmath532 w.r . to the metric defined above . by the banach fixed point theorem , for any @xmath564 in @xmath532 the iterated function sequence @xmath565 converges uniformly on @xmath509 to the unique fixed point @xmath566 of @xmath523 in @xmath532 . since @xmath567 , this point can be chosen as starting one in the iterative process . we consider a first order quasilinear system of pde s , @xmath568 equations in @xmath568 unknown functions @xmath569 and two variables @xmath570 , @xmath571 where @xmath572 is an @xmath573 matrix , @xmath574 is an @xmath568-vector . the _ cauchy problem _ for ( [ e - pde-1 ] ) is the problem of finding @xmath228 such that ( [ e - pde-1 ] ) and @xmath575 are satisfied , where @xmath576 is given . when the coefficient matrix @xmath577 and the vector @xmath578 are functions of @xmath579 and @xmath580 only , the system is _ linear_. when @xmath577 and @xmath578 are functions of @xmath579 , @xmath580 and @xmath228 , the system is _ quasilinear_. the system ( [ e - pde-1 ] ) is called _ hyperbolic _ in the @xmath580-direction at @xmath581 ( in an appropriate domain of the arguments of @xmath577 ) if the ( right ) eigenvectors of @xmath577 are real and span @xmath36 . in this case , let @xmath582 $ ] be the matrix of the ( right ) eigenvectors @xmath583 of @xmath577 . for a solution @xmath584 to ( [ e - pde-1 ] ) , the corresponding eigenvalues @xmath585 are called the _ characteristic speeds _ , the vector field @xmath586 is the @xmath587-_characteristic field _ , and its integral curves are @xmath587-_characteristics_. -2 mm * theorem a * ( see @xcite[th - eu1 ] ) _ let the system of pde s ( [ e - pde-1 ] ) be such that _ -3 mm ( i ) it is hyperbolic in the @xmath580-direction in @xmath588 for some @xmath589 ; -3 mm ( ii ) the matrices @xmath590 and the vector @xmath578 are @xmath62-regular in @xmath591 ; -3 mm ( iii ) it is satisfied an initial condition @xmath592 for which @xmath593}<r$ ] and @xmath576 is @xmath62-regular in @xmath465 $ ] . then there is @xmath594 $ ] such that ( [ e - pde-1 ] ) and ( [ e - pdes-2 ] ) admit a unique @xmath62-regular solution @xmath584 in the trapeze @xmath595 , where @xmath596 . in theorem a , one may use the norm @xmath597 for @xmath598 . to show this one should replace unknown functions @xmath599 ( hence @xmath600 ) to reduce to original theorem a for @xmath601 . we use theorem a in the proof of theorem [ t - main0 ] for diagonal matrices @xmath602 . the next proposition is known . for convenience of a reader we prove it . [ p - twoodes ] let vector functions @xmath267 and @xmath268 satisfy the lipschitz condition @xmath603 ( with the same @xmath604 ) for @xmath605 $ ] and @xmath606 ( @xmath591 a domain ) . let @xmath607 and @xmath608 are solutions to ode s @xmath609 , and @xmath610 , resp . , where @xmath605 $ ] and @xmath611 . then @xmath612}\le ( mh + \|y_0 - z_0\|_{\infty})\,e^{\bar l\,h}$ ] , where @xmath613\times\omega}$ ] . * proof*. we present bvp equivalently in the integral form @xmath614\,dx . \end{array}\ ] ] hence @xmath615 . from the gronwall - bellmann integral inequality @xmath616 with @xmath617 and @xmath618 it follows the claim.@xmath202 [ appl ] let a function @xmath619 of class @xmath620)$ ] obeys the inequality @xmath621 with @xmath622 . then @xmath623 for @xmath624 $ ] . * proof*. denote @xmath625 . then @xmath626 is the solution to the ode @xmath627 with the initial condition @xmath628 . since @xmath622 , @xmath629 is continuous in @xmath630 $ ] . clearly , @xmath631 satisfies in @xmath632 the integral equation @xmath633 denote by @xmath634 and @xmath635 . hence @xmath636 . since @xmath637 is non - negative and continuous in @xmath638 , the function @xmath639 is also non - negative and continuous in @xmath630 $ ] . furthermore , by conditions of the lemma and ( [ intequ-3 ] ) , @xmath640 satisfies in @xmath630 $ ] the integral inequality @xmath641 , which can be written as the equation @xmath642 , where @xmath643)$ ] is non - negative . the above linear integral equation of volterra type can be solved by the iterative method @xmath644 . since all the functions of the sequence are non - negative in @xmath630 $ ] , their limit @xmath640 is also non - negative , that is the desired estimate is valid.@xmath202 the _ covariant derivative _ of a @xmath645-tensor @xmath646 in @xmath1 is defined by @xmath647 where @xmath648 are christoffel symbols . [ l - efg-1 ] let @xmath649 be the graph of a function @xmath164 in @xmath1 . then the unit normal @xmath650 to @xmath23 , the coefficients of the 1-st and the 2-nd fundamental forms of @xmath23 are @xmath651 where @xmath652 , @xmath146 , and @xmath653 the proof of lemma [ l - efg-1 ] is based on the following * proposition a * ( see @xcite ) . _ the equations @xmath654 define a regular surface @xmath23 in @xmath1 if and only if @xmath655 are regular ( of class @xmath221 ) , and the rank of @xmath656 is equal to 2 . the first @xmath657 and the second @xmath658 fundamental forms of @xmath23 with the unit normal @xmath659 are given by @xmath660 where @xmath661 are christoffel symbols of the 2-nd kind on @xmath19 . _ * proof of lemma [ l - efg-1]*. from the definition @xmath662 , @xmath663 it follows ( [ e - efg ] ) . let @xmath664 be a unit normal to @xmath23 . we find @xmath665 from @xmath666 here @xmath667 , see ( [ e - alpha-3])@xmath147 . the expressions for @xmath668 and @xmath669 of ( [ e - n123 ] ) follow from the linear system @xmath670 from ( [ e-12aminov])@xmath235 , and @xmath671 , @xmath672 , we have @xmath673 hence , the coefficients @xmath674 and @xmath675 of ii are given by @xmath676 where @xmath677 are the covariant derivatives , see ( [ e - covder ] ) . @xmath202 we survey basic properties of pc surfaces in aim to illustrate that the pc surfaces provide a special class of solutions to the geometrical problem . \(a ) for @xmath678 , a solution to ( [ e - bc0-bb2],b ) , ( [ e - system-4pdes ] ) is a pc surface in @xmath29 . if @xmath679 , we get a cylinder @xmath680 of radius @xmath681 with the axis @xmath682 . we will build a pc surface with @xmath683 , see corollary [ t - main3 ] . let @xmath684 be a surface of revolution in @xmath685 , where @xmath686 is an increasing @xmath193-regular function . revolving about @xmath687-axis , @xmath688 and replacing the function @xmath689 , we obtain @xmath690 a parallel @xmath691 of above @xmath23 lies in the plane @xmath692 , and projects onto @xmath103-plane as a line segment . ( b ) consider spherical coordinates @xmath693 in the domain @xmath694 of @xmath29 , where @xmath695 and @xmath696 and @xmath697 . the _ curvilinear projection _ onto @xmath698 ( with @xmath699 ) is given by @xmath700 . .5 mm ( b)@xmath233 denote @xmath701 the line segment in @xmath0 . let @xmath43 be the functions of class @xmath702)$ ] and @xmath12 a vector field of class @xmath45 that is transversal but not orthogonal to @xmath11 . by theorem [ t - main0 ] , if @xmath261 are small enough in the @xmath554-norm , then * problem [ prob-1 ] * admits a unique smooth solution @xmath703 on @xmath704 . .5 mm ( b)@xmath235 assume that @xmath23 is a pc surface relative to the sphere @xmath705 . take @xmath706 for some @xmath707 . the @xmath22-curvature lines of @xmath23 project onto concentric circles @xmath708 on @xmath0 , hence @xmath709 is transversal but not orthogonal to @xmath11 ( theorem [ t - main0 ] is applicable ) . now let @xmath108)$ ] and @xmath109)$ ] are small enough in the @xmath554-norm . follow the proof of theorem [ t - rpc2 ] ( section [ sec : th2 - 3 ] ) , one may show that @xmath703 can be recovered over a curvilinear rectangle @xmath710 . the geometric construction of a pc surface @xmath23 is as follows . the spheres @xmath711 intersect @xmath23 transversally by @xmath22-curvature lines . let @xmath712 be the _ radial projection _ onto the unit sphere , i.e. , @xmath713 . let @xmath714 be a @xmath22-curvature line on @xmath23 , and @xmath714 belongs to @xmath715 for some @xmath516 . the curve @xmath716 is homothetic to @xmath11 ( the coefficient of homothety is @xmath717 ) . the great circles on @xmath718 orthogonal to @xmath377 and the curves of constant distance to @xmath377 form a _ semi - geodesic net _ on @xmath718 near @xmath377 , see lemma [ l-4nets ] . 1.0 mm a 1-parameter family of geodesics and their orthogonal curves on @xmath369 is called a _ semi - geodesic net _ ( it is uniquely determined by the base curve @xmath377 ) . -3 mm ( a ) , @xmath720 : @xmath377 is a line for @xmath721 , ( horocycle for @xmath722 ) , -3 mm ( b ) , @xmath723 : @xmath377 is a circle , -3 mm ( c ) , @xmath724 : normals to @xmath377 are tangent to a curve @xmath725 , ( @xmath377 is _ evolvent _ of @xmath725 ) . the cartesian and polar nets correspond to cylindrical surfaces ( @xmath726 ) and surfaces of revolution ( @xmath727 along @xmath728-curves , the axis is orthogonal to @xmath369 ) , resp . ( case ( c ) appears on pc surfaces illustrated in fig . [ f-00y](b ) ) . * proof*. it is known that the normals to a regular curve in @xmath729 ( locally ) form one of four families : ( super-)parallel lines , lines through a point and enveloping a smooth curve @xmath95.@xmath202 * proof*. the curves of @xmath732 belong to totally umbilical surfaces @xmath368 ( on the distance @xmath370 to @xmath369 ) , and the curves are parallel on @xmath734 . hence , @xmath733 ( that is orthogonal to @xmath732 ) consists of geodesics of @xmath23 . let @xmath735 be unit vector fields tangent to @xmath736 , @xmath568 a unit normal to @xmath23 , @xmath737 a unit normal to @xmath369 . then @xmath738 is orthogonal to @xmath568 and @xmath737 . by ( 2 ) of lemma [ l - chen1 ] ( see section [ sec : warped2 ] ) and rodrigues theorem ( see @xcite ) , @xmath739 where @xmath740 is the covariant derivative . hence , the angle between surfaces @xmath23 and @xmath368 along the curvature lines @xmath732 ( the intersection ) is constant . the projections of @xmath732 onto @xmath369 are parallel curves @xmath741 , hence their orthogonal trajectories @xmath742 are geodesics on @xmath369 . thus @xmath743 is a semi - geodesic net on @xmath369 . in coordinates of curvature lines we have @xmath744 , see remark [ r - petcod ] . since @xmath745 ( @xmath733-curves are unit speed geodesics ) , we obtain @xmath746 , hence @xmath747 along @xmath732-curves . one may show that ( as in @xmath29 , see @xcite , @xcite and ( [ e - k12])@xmath235 in what follows ) the curves of @xmath733 are congruent in @xmath748 each to another , and lie in planes through geodesics @xmath749 and orthogonal to @xmath369.@xmath202 1.0 mm let @xmath750 be a riemannian 2-space of constant curvature @xmath39 , and @xmath751 . the _ riemannian warped product _ 3-space is @xmath752 where @xmath753 . @xmath754 contains no open subsets of constant curvature if and only if @xmath755 on @xmath509 , @xcite . a surface @xmath756 is a _ slice _ of @xmath754 . the mean curvature vector of @xmath757 is defined by @xmath758 , where @xmath398 the second fundamental form of @xmath23 . a surface @xmath23 is we conclude that @xmath759 when @xmath11 is a geodesic in @xmath523 . on the other hand , by ( 3 ) and ( 4 ) of lemma [ l - chen1 ] , @xmath776 if and only if @xmath775 . hence , if @xmath759 then @xmath776 , that is @xmath11 is a geodesic in @xmath523 . @xmath202 an @xmath761-surface @xmath766 is totally umbilical with @xmath777 if and only if @xmath769 is totally geodesic , see @xcite . any such @xmath769 over an @xmath523-geodesic @xmath11 will be named @xmath761-_plane_. by lemma [ l - chen1 ] , the gaussian curvature of @xmath769 is @xmath778 a surface @xmath779 is called _ parallel curved _ ( pc ) relative to @xmath523 if it does not belong to a slice , and at each point @xmath367 at least one principal direction is tangent to @xmath780 passing through @xmath579 . a pc surface is _ regular _ if such principal directions form a 1-dimensional foliation ( @xmath732 ) . proposition [ l - rpc-1a ] can be extended as follows [ l - rpc-1 ] let @xmath779 be a regular pc surface - graph over domain in @xmath523 . then the 2-nd family of curvature lines ( @xmath733 ) consists of geodesics on @xmath23 which lie in @xmath761-planes . two families ( @xmath732 and @xmath733 ) of curves project onto @xmath523 as a semi - geodesic net . ferapontov e.v . surfaces in 3-space possessing nontrivial deformations which preserve the shape operator , in differential geometry and integrable systems " ( tokyo , 2000 ) , 145159 , contemp . math . , 308 , ams

the recently discovered dwarf galaxy perseus i appears to be associated with the dominant plane of non - satellite galaxies in the local group ( lg ) . we predict its velocity dispersion and those of the other isolated dsphs cetus and tucana to be 6.5 , 8.2 , and , respectively . the ngc 3109 association , including the recently discovered dwarf galaxy leo p , aligns with the dwarf galaxy structures in the lg such that all known nearby non - satellite galaxies in the northern galactic hemisphere lie in a common thin plane ( rms height 53 kpc ; diameter 1.2 mpc ) . this plane has an orientation similar to the preferred orbital plane of the milky way ( mw ) satellites in the vast polar structure . five of seven of these northern galaxies were identified as possible backsplash objects , even though only about one is expected from cosmological simulations . this may pose a problem , or instead the search for local backsplash galaxies might be identifying ancient tidal dwarf galaxies expelled in a past major galaxy encounter . the ngc 3109 association supports the notion that material preferentially falls towards the mw from the galactic south and recedes towards the north , as if the mw were moving through a stream of dwarf galaxies . [ firstpage ] galaxies : dwarf galaxies : individual : perseus i galaxies : kinematics and dynamics local group dark matter

1402.4130

the milky way ( mw ) is surrounded by a vast polar structure ( vpos ) of satellite objects including the satellite galaxies , young halo globular clusters and several stellar and gaseous streams @xcite . the proper motions of the 11 classical satellite galaxies reveal that these almost exclusively co - orbit in this vpos , which allowed us to predict the proper motions of the remaining satellite galaxies @xcite . have first identified this planar alignment as being inconsistent with cosmological simulations based on the cold dark matter paradigm with a cosmological constant , @xmath1cdm . this finding subsequently triggered an ongoing debate on whether such structures can be reconciled with cosmological expectations ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) or rather points at a different origin such as the formation of phase - space correlated tidal dwarf galaxies . @xcite and @xcite have recently discovered a similar great plane of andromeda ( gpoa ) , a co - orbiting alignment consisting of about half of the satellite galaxies of the andromeda galaxy ( m31 ) , the other major galaxy in the local group ( lg ) . motivated by this discovery that satellite galaxies appear to preferentially live in phase - space correlated structures , @xcite set out to search for similar structures on a lg scale . they have discovered that all but one of the 15 lg dwarf galaxies more distant than 300kpc from the mw and m31 are confined to two narrow ( short - to - long axis ratios of @xmath2 ) and highly symmetric planes , termed lgp1 and lgp2 . lgp1 is the dominant plane both by number of objects ( about nine ) , and alignment with additional features , such as the magellanic stream which traces the positions and line - of - sight ( los ) velocities of the lgp1 plane members in the southern galactic hemisphere . given that the number of known dwarf galaxies in the lg more distant than 300kpc from both major galaxies is still low , each additional detection poses a chance to test the existence of the planar lg structures and to potentially refine our understanding of these structures . such an opportunity is now provided by the recent discovery of the dwarf spheroidal ( dsph ) galaxy perseus i at a distance of 374kpc from m31 @xcite . in the following , we test whether it can be considered to be associated to either lgp1 or lgp2 . in addition , we predict the velocity dispersion of perseus i and two other non - satellite dsphs as expected in modified newtonian dynamics ( mond , * ? ? ? * ; * ? ? ? similar predictions have been made for other dwarf galaxies in m31 s vicinity @xcite and these have successfully passed the test of observations @xcite . unfortunately , no similar predictions are possible in the @xmath1cdm framework . another recently discovered nearby dwarf galaxy , leo p @xcite , has lead @xcite to re - investigate the ngc 3109 association , a group of dwarf galaxies at a distance of about 1.31.4mpc from the mw that consists of ngc 3109 , antlia , sextans a and sextans b @xcite . they realised that leo p aligns with the four other members of the association in a very narrow , linear structure . as the ngc 3109 association is very close to the lg and has a linear extent of 1.2mpc , similar to its distance from the mw , we will investigate its orientation in the context of the lg planes of non - satellite dwarf galaxies . this reveals an intriguing alignment with the other three nearby non - satellite galaxies in the northern hemisphere of the mw and leads us to discuss suggested origins for the ngc 3109 association in light of the geometry of the lg . the paper is structured as follows . in sect . [ sect : perseus ] we determine whether perseus i is associated with one of the dwarf galaxy planes in the lg . in sect . [ sect : mond ] we predict the velocity dispersion of the distant dsphs in the lg , perseus i , cetus and tucana . in sect . [ sect : ngc3109association ] we determine the orientation of the ngc 3109 association in the same coordinate system used in @xcite , discuss possible origins for the alignment and conclude that the association is likely part of the lg dwarf galaxy structures . in sect . [ sect : backsplash ] we discuss how the search for cosmological backsplash galaxies in the lg might give rise to two additional small - scale problems of cosmology and how it could falsely identify tdgs as backsplash objects . in sect . [ sect : scheme ] we present a sketch of the lg dwarf galaxy structures and their preferred direction of motion and discuss open questions and limitations in sect . [ sect : openquestions ] . finally , we summarize our results in sect . [ sect : conclusion ] . ( yellow dots ) and lgp2 ( green squares ) in the lg ( looking along @xmath3 = [ 308.9^\circ , 16.8^\circ]$ ] ) . the best - fit planes are plotted as solid lines , the dashed lines denote their rms heights . the black ellipses indicate the positions and orientations of the mw and m31 and the blue plus signs ( red and black crosses ) indicate mw ( m31 ) satellites . grey lines mark the 1@xmath4 distance uncertainties . the newly discovered dwarf galaxy perseus i ( yellow star ) is aligned with lgp1@xmath5 . compare to the similar plot shown as fig . 9 in @xcite . in addition , the plot shows the positions of the dwarf galaxies associated to the linear ngc 3109 association ( grey diamonds ) which is situated behind the lg in this view . the ngc 3109 association is almost parallel to the dominant non - satellite plane in the local group , but offset by 0.30.5mpc . _ bottom panel _ : edge - on view of lgp1@xmath5 and the plane fitted to the five members of the ngc 3109 association plus the lgp1@xmath5 members leo t , leo a and ugc 4879 ( looking along @xmath3 = [ 348.3^\circ , -54.9^\circ]$ ] ; the view of the upper panel would originate from approximately the lower left of this plot ) . the ngc 3109 association might be related to the dominant lg plane ( lgp1@xmath5 ) . the planes are inclined to it by only @xmath6 and the intersection of the two planes lies close to the mw . , title="fig:",width=321 ] ( yellow dots ) and lgp2 ( green squares ) in the lg ( looking along @xmath3 = [ 308.9^\circ , 16.8^\circ]$ ] ) . the best - fit planes are plotted as solid lines , the dashed lines denote their rms heights . the black ellipses indicate the positions and orientations of the mw and m31 and the blue plus signs ( red and black crosses ) indicate mw ( m31 ) satellites . grey lines mark the 1@xmath4 distance uncertainties . the newly discovered dwarf galaxy perseus i ( yellow star ) is aligned with lgp1@xmath5 . compare to the similar plot shown as fig . 9 in @xcite . in addition , the plot shows the positions of the dwarf galaxies associated to the linear ngc 3109 association ( grey diamonds ) which is situated behind the lg in this view . the ngc 3109 association is almost parallel to the dominant non - satellite plane in the local group , but offset by 0.30.5mpc . _ bottom panel _ : edge - on view of lgp1@xmath5 and the plane fitted to the five members of the ngc 3109 association plus the lgp1@xmath5 members leo t , leo a and ugc 4879 ( looking along @xmath3 = [ 348.3^\circ , -54.9^\circ]$ ] ; the view of the upper panel would originate from approximately the lower left of this plot ) . the ngc 3109 association might be related to the dominant lg plane ( lgp1@xmath5 ) . the planes are inclined to it by only @xmath6 and the intersection of the two planes lies close to the mw . , title="fig:",width=321 ] .parameters of the fitted planes [ cols="<,^,^",options="header " , ] parameters of the planes fitted to the lg dwarf galaxies , as discussed in sects . [ sect : perseus ] and [ sect : ngc3109leoplane ] . these are : @xmath7 : @xmath8- , @xmath9- and @xmath10-position of the centroid of the plane in the coordinate system introduced in @xcite . @xmath11 : the direction of the normal vector ( minor axis ) of the best - fit plane in galactic longitude @xmath12 and latitude @xmath13 . @xmath14 : uncertainty in the normal direction . this and all other uncertainties were determined by varying the galaxy positions within their uncertainties and then determining the standard deviation in the resulting plane parameters . @xmath15 and @xmath16 : offset of the planes from the mw and m31 position . @xmath17 : rms height of the galaxies from the best - fit plane . @xmath18 and @xmath19 : ratios of the short and intermediate axis to the long axis , determined from the rms heights in the directions of the three axes . @xmath20 : number of galaxies associated with the planes used for the fitting . in particular lgp1 and lgp2 might have additional satellite galaxies as members , but these were not included in the plane fits compiled here . recently , @xcite reported the discovery of a dsph galaxy in the vicinity of andromeda , persus i. at a distance of 374kpc from m31 it is a non - satellite galaxy according to the categorisation of @xcite , which considers only galaxies closer than 300 kpc , i.e. within the typically assumed virial radii of the mw and m31 , to be satellites . all but one of the 15 previously known non - satellite galaxies in the local group were found to be close to one of two thin , highly symmetric lg planes ( lgp1 and lgp2 ) . is perseus i a member of one of the two non - satellite galaxy planes in the lg ? to determine whether perseus i lies close to one of the two lg planes as reported by @xcite ( see for example their table 3 ) , we adopt the position and distance modulus of perseus i from table 1 of @xcite . in the cartesian coordinate system of @xcite , this places persues i at @xmath21kpc , with a position - uncertainty along the line connecting perseus i and the sun of 65kpc . perseus i indeed lies in the vicinity of one of the planes , lgp1 ( offset by @xmath22kpc , compared to the plane s root - mean - square [ rms ] height of @xmath23kpc ) , as defined by the dwarf galaxies ugc 4879 , leo a , leo t , phoenix , tucana , wlm , cetus , ic 1613 and andromeda xvi . all other dwarf galaxy planes , in particular lgp2 , the second non - satellite galaxy plane in the lg , are more distant than @xmath24kpc . while andromeda xvi is considered a member of lgp1 for formal reasons in @xcite , this galaxy is perfectly aligned with the gpoa ( offset of only @xmath25kpc ) , its line - of - sight velocity shows that the galaxy follows the co - orbiting trend of the other gpoa members and it is at a distance of only 323kpc from m31 . it is therefore more likely that andromeda xvi belongs to the gpoa rather than the lgp1 . removing it from the plane - fit results in a rms height of @xmath26kpc ; short - to - long axis ratio of @xmath27 and intermediate - to - long axis ratio of @xmath28 ; offset from mw of @xmath29kpc and from m31 of @xmath30kpc . the normal to the best - fit plane points to @xmath31 . of the galaxies within 300kpc of m31 , triangulum / m33 and its potential satellite andromeda xxii are both very close to the best - fit plane ( @xmath32 and @xmath33kpc , respectively ) . perseus i is at a considerably smaller offset ( @xmath34kpc ) from this plane fit than from the one including andromeda xvi . this warrants inclusion of the galaxy in the modified lgp1 sample ( lgp1@xmath5 ) , which now consists of ugc 4879 , leo a , leo t , phoenix , tucana , wlm , cetus , ic 1613 and perseus i. the resulting parameters for lgp1@xmath5 are compiled in tab . [ tab : allplanes ] . they are similar to the fit without perseus i : rms height of @xmath35kpc ; short - to - long axis ratio of @xmath36 and intermediate - to - long axis ratio of @xmath37 ; offset from mw of @xmath38kpc and from m31 of @xmath39kpc . the normal to the best - fit plane points to @xmath40 which is only 5 degree inclined relative to the normal of the original lgp1 ( @xmath41 = [ 220^\circ , -22^\circ]$ ] ) . m33 and andromeda xxii both remain close to the best - fit plane ( @xmath42 and @xmath43kpc , respectively ) . as can be seen in fig . [ fig : lgplanes ] , which shows an edge - on view of the lg planes lgp1@xmath5 and lgp2 , perseus i is clearly aligned with this thin plane . the newly discovered galaxy therefore confirms the finding by @xcite that essentially all non - satellite galaxies of the lg are confined to two very thin planes . perseus i is offset by only @xmath44kpc from the best - fit plane , which is to be compared with the largest extent of lgp1@xmath5 of 2.2 mpc between tucana and ugc 4879 . the offset might in fact be a weak indication of a bending of the plane . as seen in fig . [ fig : lgplanes ] , perseus i and other nearby galaxies are offset to the right from the line indicating the edge - on view of the best - fit plane in fig . [ fig : lgplanes ] , while those at the top and bottom are offset to the left . the significance of a satellite galaxy plane can be tested by comparing the observed distribution with an expected one , which in most cases is assumed to be isotropic . due to the existence of a preferred axis in the lg ( the mw m31 line ) , this comparison can not be easily adopted for planes spanning the whole lg . a proper determination of the plane significance therefore requires an expected model for the distribution of the non - satellite galaxies in the lg , which is not available . furthermore , such a test has to take observational biases like the sky coverage of surveys searching for lg dwarf galaxies into account . due to the very inhomogeneous nature of the galaxy data , this is currently not feasible . we nevertheless try to get a rough estimate for how likely it is to find two similarly thin planes of non - satellite galaxies in the lg . assuming that andromeda xvi is part of the gpoa , we use the following 15 lg dwarf galaxies for this test : andromeda xviii , andromeda xxviii , aquarius , cetus , ic 1613 , leo a , leo t , ngc 6822 , pegasus dirr , perseus i , phoenix , sagittarius dirr , tucana , ugc 4879 and wlm . instead of assuming a model for the expected galaxy distribution , we generate 1000 randomised realisations by rotating each of the dwarf galaxy positions by individual random angles around the mw m31 axis . this preserves their distances from both the mw and from m31 . for each realisation , we then proceed as follows . ignoring one of the lg galaxies ( analogously to ignoring the pegasus dirr galaxy which is neither part of lgp1@xmath5 consisting of 9 dwarf galaxies nor of lgp2 consisting of 5 dwarf galaxies ) , we split the remaining galaxies into two samples , constructing all possible combinations of 9:5 objects . planes are then fitted to both samples for each of these combinations . both planes rms heights @xmath17 and axis ratios @xmath18 are recorded for all 30030 possible combinations for each of the 1000 realisations . due to the large number of possible combinations we refrain from varying the observed galaxy distances within their uncertainties and only use the most - likely values . we then test how many of the randomised galaxy distributions contain planes which are similarly thin ( @xmath45kpc ) or have similar axis ratios ( @xmath46 ) as the observed lgp1@xmath5 . the mean of the minimum rms height for the 9-galaxy combinations over all realisations is 83 kpc , with a standard deviation of 22 kpc , while the mean minimum axis ratio is 0.140 , with a standard deviation of 0.037 . thus , on average a plane of galaxies as narrow as the lgp1@xmath5 is not expected . of the 1000 realisations , 29 ( 2.9 per cent ) contain a group of 9 galaxies which can be fitted by a plane with a rms height at least as small as that of lgp1@xmath5 . only 10 ( 1.0 per cent ) can be fitted by a plane with a sufficiently small axis ratio . if we simultaneously require that five out of the remaining six galaxies can be fitted by a plane having a similar rms height or axis ratio as lgp2 ( @xmath47kpc , @xmath48 ; @xcite ) , a situation similar to the one observed in the lg occurs in only 13 ( 1.3 per cent , rms height criterion ) or three ( 0.3 per cent , axis - ratio criterion ) of the 1000 random realisations . this test demonstrates that the lg planes are unexpected . we have not tested for the symmetry of the planes and their alignments , and varying the galaxy positions by only rotating them around the mw m31 axis has a large chance of preserving information on the possible lg planes because they are parallel to the mw m31 line . our results should therefore be considered as upper limits . in addition to the structures that dwarf galaxies trace in the lg , their internal kinematics are also of interest . these objects are generally inferred to be dark matter dominated , though there was no reason to anticipate this a priori . lcdm models can be constructed to match this observation , but do not provide the ability to predict the velocity dispersion of any particular dwarf . in contrast , it is possible to predict a dwarf s velocity dispersion given its photometric properties using mond . we employ here the method described by @xcite . being quite remote from m31 ( 374kpc ) , perseus i is well into the isolated deep mond regime for which the characteristic velocity dispersion follows directly from the stellar mass ( @xmath49 ) . this makes it one of the best test cases among the dwarfs of andromeda . most ( though not all ) of the other dwarfs are in the regime dominated by the external field effect so that the predicted velocity dispersions are less certain as they depend on the properties of m31 as well as those of the dwarfs themselves ( see * ? ? ? * ) . given the luminosity reported by @xcite , we predict that perseus i should have a velocity dispersion of @xmath50 . for consistency with @xcite we assume a stellar mass - to - light ratio of @xmath51 , the first uncertainty reflects a factor of two variation in mass - to - light ratio while the second propagates the stated observational uncertainties . predictions of this type have proven largely successful so far @xcite ; perseus i provides another opportunity to test this a priori prediction of mond . there exist two other dwarf spheroidals in the local group that are far removed from both the milky way and andromeda : cetus and tucana . being far from major perturbers , they should also be in the isolated mond regime , and provide correspondingly good tests . however , they tend to be overlooked since they are not grouped together with the dwarfs that are obvious satellites . applying the same procedure described above given the luminosities and most - likely distances tabulated by @xcite , we predict for cetus @xmath52 and for tucana @xmath53 . as before , the first uncertainty represents the range of mass - to - light ratios @xmath54 . the second uncertainty represents the effect of the stated uncertainty in luminosity on the velocity dispersion for the nominal assumed mass - to - light ratio of @xmath55 . any systematic error in distance will have a strong effect , since @xmath56 . the predicted velocity dispersions of both cetus and tucana compare poorly with observed values . @xcite observe @xmath57 for cetus and measure @xmath58 for tucana . the observed values are a factor of @xmath59 and 3 higher than predicted , respectively . in terms of formal significance , the observed velocity dispersions are @xmath60 ( cetus ) and @xmath61 ( tucana ) above the predicted range . the velocity distributions of cetus and tucana are not particularly well described as gaussians , so it is not obvious how to interpret their fitted velocity dispersions and uncertainties . the errors on the velocities of individual stars are typically @xmath62 , so the velocity dispersions predicted here would not be resolved : improved observations are warranted . nevertheless , in the absence of systematic errors , either in the overestimation of the velocity dispersion or the underestimation of the luminosity , these two dwarfs are problematic for mond . after the above text was written but shortly before we submitted this paper , @xcite reported new observations of cetus . they measure a velocity dispersion of @xmath63 . this agrees well with our prediction of @xmath64 . the dwarf galaxy leo p has recently been discovered in the vicinity of the lg by the alfalfa survey @xcite . it can be considered to be a member of the ngc 3109 association of galaxies ( see e.g. * ? ? ? * ; * ? ? ? * ) , consisting of ngc 3109 , antlia , sextans a and sextans b. @xcite have shown that leo p s position and velocity is consistent with it belonging to the highly elongated , essentially linear association , as was also noted by @xcite . given the existence of correlated satellite galaxy planes and the recent discovery that essentially all lg dwarf galaxies are confined to one of two thin and highly symmetric planes @xcite , it it worthwhile to investigate how the ngc 3109 association relates to these structures . in the following , we use the galaxy positions as compiled by @xcite . however , like @xcite we use the homogeneous set of tip of the red giant branch ( trgb ) distance moduli from @xcite for the members of the ngc 3109 association . for leo p , we adopt the recent trgb distance modulus by @xcite , according to which the galaxy is at a distance of @xmath65mpc from the sun . the galaxies in the ngc 3109 association all have similar galactocentric distances of 1.31.4mpc , and leo p is also consistent with this distance range . the association is therefore oriented approximately perpendicular to our the line - of sight , the angle between the long axis of the association and the line connecting the associations centroid with the position of the mw is @xmath66 . the galaxies also have very similar distances from m31 and from the center of the lg , such that the association is also oriented approximately perpendicular as seen from those points ( 85 and @xmath67 , respectively ) . the ngc 3109 association is therefore almost perpendicular from the line connecting its centroid with the position of m31 . we have determined the associations s orientation in the same cartesian coordinate system used previously by employing the method used in @xcite . this effectively fits an ellipsoid to the points by determining the eigenvectors of the moments of inertia tensor defined by the non - mass - weighted galaxy positions . this gives the orientations of the major , intermediate and minor axes and the root - mean - square ( rms ) heights of the distribution along these axes . the resulting rms axis ratios of the ngc 3109 association are indicative for a very elongated distribution which is extremely flat in one direction ( short - to - long axis ratio @xmath68 ) and a bit more extended along the intermediate axis ( @xmath69 , still narrow compared to the long axis ) , like a ruler . the long axis points to @xmath70 or @xmath71 , with an uncertainty of @xmath72 . the association therefore aligns ( being only @xmath73 inclined ) with the supergalactic plane @xcite . it is also almost parallel to lgp1@xmath5 ( @xmath74 inclined ) , but offset by 0.3 ( leo p ) to 0.5 mpc ( antlia and ngc 3109 ) , as can be seen in the upper panel of fig . [ fig : lgplanes ] . the lg is the closest major galaxy group to the ngc 3109 association and none of the other nearby galaxy groups listed in are well aligned with the line defined by the main axis of the ngc 3109 association . the closest alignment , at an angle of @xmath75 , is found for the m 81 group , but at a distance of @xmath76mpc from the association , two times as distant as the lg , this is most likely a chance alignment . the short axis ( normal direction if it were a plane ) of the ellipsoid describing the ngc 3109 association points to @xmath77 and @xmath78 . this is extremely close to the pole of the supergalactic plane ( @xmath41 = [ 47^\circ.4 , 6^\circ.3]$ ] ) and also similar to the normal to the lgp1@xmath5 pointing to @xmath40 . intriguingly , the normal to the gpoa has a similar orientation as well ( @xmath41 = [ 206^\circ , 8^\circ]$ ] , @xcite ) , even though it is defined by the co - orbiting satellite galaxies of m31 on the opposite side of the mw than the ngc 3109 association . however , the mw is offset by 0.8mpc and m31 by 0.9mpc from the ngc 3109 association along this short axis direction . thus , while some hints for a connection exist it is not immediately obvious whether the ngc 3109 association is related to the dwarf galaxy planes in the lg . in the following we will discuss additional indications why it might be part of the lg and its dwarf galaxy geometry . @xcite mention several types of possible origins of the alignment of the ngc 3109 association . these include a tidal encounter with the mw which has stretched a pre - existing group of dwarf galaxies along its orbital plane , the formation of the galaxies as phase - space correlated tdgs , or the formation of the ngc 3109 association in a thin and cold cosmological filament which is just now starting to fall towards the lg . here we discuss these suggestions in more detail . assuming the ngc 3109 association to be bound and spherically symmetric , @xcite estimate that it would have to have a mass of @xmath79 . this is a significant fraction ( up to one third ) of the total mass of @xmath80 currently estimated for the mw halo @xcite . if the ngc 3109 association were embedded in such a massive dark matter halo , this halo should significantly influence the dynamics of the lg . in addition , a past interaction with the massive dark matter content of the association would have likely lead to it merging with the mw because dynamical friction must be significant in this mass range . a spherical dark matter halo encompassing the whole association would furthermore have to have a radius of at least 600 kpc , half the associations diameter . this is twice as much as the virial radii assumed for the mw and m31 and would almost reach leo a at a distance of about 800kpc from the ngc 3109 association . we therefore deem it unlikely that the ngc 3109 association is a distinct gravitationally bound entity . @xcite suggest a cosmological accretion scenario in which the ngc 3109 association is a `` thin and cold cosmological filament '' , assuming that the galaxies in the association have recently left the hubble flow and are currently falling into the lg for the very first time . however , simulated dark matter filaments around mw - like haloes today have diameters which are larger than the virial radii of the major haloes they feed @xcite . due to its much more narrow configuration , the ngc 3109 association can not be identified with such a major dark matter filament . the dark matter hypothesis does , however , allow the freedom to propose that the ngc 3109 association is embedded in a dark matter `` sub - filament '' , which would have to be oriented almost perpendicular to its closest major mass concentration . while cosmological simulations might be searched for the existence of thin sub - filaments of this kind , as of now such a suggestion unfortunately remains non - testable via observations . however , one wonders how such a dark matter filament in the vicinity of the major dark matter halo potentials of the mw and m31 could have remained essentially a straight line over a length of 1.2mpc without being aligned towards the mass concentration . the interpretation that the association is a filament falling towards the lg is in conflict with the orientation of the ngc 3109 association being almost perpendicular to the direction towards the mw or the lg barycenter . simulated filaments of cold gas , which would be narrow enough to accommodate the thin association , point towards the central galaxy and are considered to be important at high redshifts of @xmath81 only ( * ? ? ? * ; * ? ? ? more importantly , the ngc 3109 association does not fall towards but recedes from the mw with @xmath82 , a velocity similar to those of galaxies at larger distances of 1.6 to 2.2 mpc ( see e.g. fig . 5 of @xcite ) . the association appears to be situated beyond the zero - velocity surface of the lg ( @xmath83 1 to 1.5mpc ) and therefore should follow the hubble flow . if it follows the hubble flow , the galaxies should never have been close to the lg ( except during the big bang ) . however , the members of the ngc 3109 association appear to recede faster than the expected hubble flow velocity at their distance ( see e.g. figure 6 of @xcite ) . if this is the case and if they are on radial orbits ( otherwise their motion is even faster ) , the galaxies must have been close to the mw in the past . assuming that all galaxies in the ngc 3109 association are receding on radial orbits from the mw ( i.e. the tangential velocity is zero ) , and that they have been travelling with their current galactocentric velocity ( no acceleration ) gives a rough estimate for their travel time . if we transform the measured heliocentric velocities to galactocentric ones by assuming a circular velocity of the local standard of rest ( lsr ) of 220 km / s and a peculiar motion of the sun as measured by @xcite , the travel times are @xmath84gyr for ngc 3109 , @xmath85 gyr for sextans b , @xmath86gyr for antlia and @xmath87gyr for sextans a. due to the large distance uncertainty for leo p its travel time is in the range of 7.6 to 10.6 gyr . the stated uncertainties are based on the distance uncertainties and an assumed uncertainty in the line - of - sight velocities of the galaxies of 2 kms@xmath88 . of course , the major inaccuracy comes from the over - simplified assumption of radial orbits without acceleration . another major source of uncertainty is the lsr velocity . if we assume a lsr velocity of 240 kms@xmath88 , instead of the previously used 220 kms@xmath88 , the travel times become 0.75 gyr longer on average . if the galaxies in the ngc 3109 association do not have a significant tangential velocity , this timing estimate demonstrates that they have been close to the mw in the past . more importantly , the galaxies are consistent with all having been nearby at about the same time , @xmath89gyr ago . only ngc 3109 s timing appears to be off by 1.52 gyr , but unaccounted - for tangential motion , gravitational acceleration by the lg galaxies and others and the possible interaction between members of the ngc 3109 association make this estimate unreliable on such time scales . that the ngc 3109 members have been close to the mw at a similar time in the past is also consistent with results by @xcite . in their model of the dynamical history of the lg , the ngc 3109 association members had a close passage ( pericenter @xmath90kpc ) with the mw about 7gyr ago . however , at such a distance dynamical friction , in particular for massive objects such as ngc 3109 or the putative dark matter halo needed if one requires that the association as a whole is gravitationally bound , can not be neglected . as @xcite do not consider dynamical friction in their study , the orbit of the association s members as derived from their study becomes unreliable as soon as they come close to the mw halo . another independent indication that the members of the ngc 3109 association have been close to the lg before is their identification as probable backsplash galaxies , galaxies which have passed through the virial volume of the mw in the past but are now situated outside of it . by comparing the distances and line - of - sight velocities of lg galaxies with sub - haloes in the via lactea ii simulation , @xcite identify all four of the then - known galaxies in the ngc 3109 association as likely backsplash haloes that have interacted with the mw before . it therefore appears likely that the ngc 3109 association had a past encounter with the mw , in conflict with the interpretation as an infalling dark matter filament . a scenario in which a pre - existing group of dwarf galaxies collided with the mw and was tidally stretched along its orbit appears plausible , but the effects of dynamical friction in such a scenario will need to be assessed . it is currently not possible to discriminate such a scenario from one in which the galaxies were formed as tdgs in a past encounter in the lg , but there are additional coincidences which make a tdg origin look more appealing . as shown before , given their current distances and line - of - sight velocities , the galaxies in the ngc 3109 association are consistent with having been in the vicinity of the mw at roughly the same time about 7 to 9 gyr ago . this leads us to the other possible origin of the association discussed by @xcite : the galaxies might have been expelled as tdgs in a galaxy encounter . interestingly , the estimated travel times are in good agreement with independent estimates for the timing of major galaxy encounters suggested as the origin of a population of tdgs in the lg : * based on the ages of the young halo globular clusters , the vpos around the mw was estimated to have formed 912gyr ago @xcite . * a major merger might have formed m31 89gyr ago . this event must have produced tdgs , some of which would constitute the gpoa today while others would have been expelled towards the mw , maybe forming the vpos @xcite . * in mond the observed baryonic masses of the mw and m31 and their relative velocity require a past encounter between the two major galaxies to have happened 711gyr ago . the ngc 3109 association is therefore consistent with having formed in the same major galaxy encounter and thus such an origin can not be ruled out at present . the sfhs of the non - satellite galaxies do not provide conclusive information either , because most non - satellite galaxies in the lg are dirrs which show ongoing star formation , consistent with expectations for gas - rich tdgs if the present high - resolution simulations of tdg formation and early evolution can be extrapolated to many gyr . differing environmental effects will furthermore diversify the evolutionary histories of tdgs born in the same event over time . the apparent dark matter content of some of these galaxies , unexpected for tdgs in a cold dark matter universe , is not a conclusive argument against the possible tdg nature of the ngc 3109 association either , as inflated velocity dispersions are expected in mond as exemplified in sect . [ sect : mond ] . to be consistent with a scenario in which a major fraction of the lg galaxies was formed in one common major encounter , the members of the ngc 3109 association must show signs of being related to the other dwarf galaxy structures known in the lg , in particular the vpos , the gpoa and potentially the lg dwarf galaxy planes , because tdgs form as a phase - space correlated population , i.e. in a highly flattened tidal tail . if the ngc 3109 association shows alignments with other lg dwarf galaxies , this might indicate that the ngc 3109 association is , contrary to current belief , not an isolated entity and might share a common origin with other dwarf galaxy structures . given that the total extent of the ngc 3109 association ( the distance between leo p and antlia ) is 1.2 mpc , while the three lgp1 members ugc 4879 , leo a and leo t are at distances of only 0.81 mpc from their respectively closest member of the ngc 3109 association , these eight galaxies might well constitute one common structure of square - megaparsec size . the difference in the galactocentric line - of - sight velocities between the ngc 3109 association ( receding ) and the other three galaxies ( slowly approaching ) might then be simply due to the stronger gravitational attraction acting on the more nearby galaxies . indeed , as seen from the center of the lg the members of the ngc 3109 association align along a similar band as defined by ugc 4879 , leo a and leo t ( see fig . [ fig : asp ] ) . this indicates that they are in a common plane running through the center of the lg . fitting a plane to the eight galaxy positions confirms that the galaxies are confined to a thin planar structure , which we will refer to as the _ great northern plane_. the parameters of the best - fit plane are compiled in tab . [ tab : allplanes ] . the plane has a normal vector pointing to @xmath91 and rms axis ratios of @xmath92 and @xmath93 . as expected , it runs through the center of the lg ( see lower panel in fig . [ fig : lgplanes ] ) . the galaxy with the largest offset of only 86kpc from the best - fit plane is ugc 4879 . compared to lgp1@xmath5 , with which it shares three members , the plane has a similar offset from the mw ( @xmath94kpc ) and a similar rms height ( @xmath95kpc ) . there is no other known dwarf galaxy in the northern hemisphere of the mw and between galactocentric distances of 0.3 to 1.7 mpc from the mw , such that all currently known nearby non - satellite dwarf galaxies to the north of the mw are confined to this one common plane . to determine how significant this planar alignment is , we follow the same principle as in sect . [ sect : perseus ] ( thus the same caveats apply here ) . we randomise the positions of the eight non - satellite galaxies in the northern hemisphere of the mw and then determine the rms height of the planes fitted to these new positions . we generate @xmath96 realisations for each of two randomisation methods . the first method draws the galaxy positions randomly from an isotropic distribution , but we force the galaxies to be confined to @xmath97 . this make sure they are in the northern galactic hemisphere and in an area not obscured by the mw disc . this method conserves the distances from the mw , but not from m31 . in this case , only five ( 0.005 per cent ) of all realizations have @xmath98kpc . in the second method we instead conserve the distances from both the mw and from m31 , by rotating the galaxy positions by a random angle around the mw m31 axis . again they are required to be at @xmath97 . in this case only seven ( 0.007 per cent ) of the realisations result in sufficiently thin best - fit planes . however , antlia might be interpreted as a satellite of ngc 3109 @xcite , such that these two galaxies might not be treated as two independent systems . if we exclude antlia from this analysis ( thus using only seven instead of eight galaxies ) , the probabilities to find a plane at least as narrow as the observed one increase to 0.079 per cent if the galaxy positions are chosen from an isotropic distribution and to 0.037 per cent if the galaxy positions are individually rotated by a random angle around the mw a plane as narrow as the observed great northern plane is therefore very unlikely to occur by pure chance . the plane is inclined relative to lgp1@xmath5 by only 27@xmath99 . as can be seen in the lower panel of fig . [ fig : lgplanes ] , which shows these two planes edge - on , the line at which they intersect passes close to the mw . interestingly , the plane is also inclined by only 25@xmath99 from the average orbital pole of the mw satellites with known proper motions , around which the individual satellite orbital poles scatter with a spherical standard distance of @xmath100 @xcite . the plane is therefore consistent with being aligned with the preferred orbital plane of the mw satellites in the vpos , but is somewhat less polar than the vpos . relative to the gpoa of m31 satellites it is inclined by @xmath101 . interestingly , even though planes of satellite galaxies can be dynamically stable , depending on the galactic potential in which they are embedded their orientation can change due to precession and inclined planes can evolve into a more polar orientation ( klarmann et al . in prep . ) . it is possible that lgp1@xmath5 bends close to the position of the mw , such that the ngc 3109 association is in fact a part of the same structure of non - satellite galaxies in the lg . for such a causal connection to be feasible , the ngc 3109 members must have been closer to the lg in the past , which appears to be supported by the timing estimate presented before . the ngc 3109 association therefore appears to be connected to the previously - identified dwarf galaxy structures of the lg such that the lg geometry does not rule out the possibility of a tdg origin for its members . of lg dwarf galaxies is plotted against their galactocentric distance @xmath102 . compare to fig . 6 of @xcite . the black circle denotes m31 . dwarf galaxies identified as likely backsplash objects by @xcite are marked in blue , the others in red . leo a has a non - negligible but below 50 per cent likelihood of being a backsplash galaxy . leo p was discovered only after the study of @xcite and should also be interpreted as a backsplash candidate . the grey dots are tidal debris particles from a @xmath103-body model of a galaxy collision at about 9gyr after the first pericenter . two disc galaxies embedded in @xmath104 dark matter haloes merge , with one falling in on a polar , prograde orbit . the model parameters have not been fine - tuned to reproduce the mw or the surrounding lg galaxies , but the distribution of particles illustrates that tidal debris have similar properties as the backsplash galaxy candidates in this plot . this demonstrates that tdgs might be wrongly identified as backsplash galaxy candidates . note that more than one tidal tail can be formed if there are several pericenter passages before the final merger and that a different origin of the tdgs , e.g. in m31 or at the barycenter of the lg in case of a past mw m31 encounter , would result in a more complex distribution of tidal debris . , width=321 ] interestingly , of the seven galaxies in the sample by @xcite which lie to the north of the mw ( @xmath105 ) , five are identified as likely backsplash galaxies and one of the remaining two ( leo a ) has a non - negligible likelihood if sub - haloes close to a m31 analogue are excluded from the analysis . in the whole northern hemisphere of the mw ( the southern hemisphere is more complicated due to the presence of m31 and its satellite galaxy population ) the majority of dwarf galaxies are likely backsplash objects , which is unexpected from cosmological simulations . according to @xcite , overall only 13 per cent of all sub - haloes between 300 and 1500kpc from the mw should be backsplash haloes . this can be expressed in another way , which might highlight a more curious coincidence : all galaxies in the great northern plane except for ugc 4879 are possible backsplash objects , posing the question why they should all be in the same thin , planar structure . even more puzzling is that most of the remaining five likely backsplash galaxies of @xcite situated in the southern hemisphere of the mw are also close to the great northern plane defined by the northern non - satellite galaxies only : ngc 185 at a distance of 45kpc from the best - fit plane , phoenix at 87kpc , whereas more recent measurements yield @xmath106kms@xmath88 @xcite or even @xmath107kms@xmath88 @xcite . these result in phoenix approaching the mw much faster than assumed in the analysis by @xcite , such that identification as a backsplash candidate will be less likely for this galaxy . ] , tucana at 113kpc and maybe cetus which is offset by 204kpc . only ngc 6822 has a much larger offset of 460kpc . the distance uncertainties in the galaxy positions result in uncertainties in the offsets from the best - fit plane of 8 to 11kpc . that all distant dwarf galaxies in the northern hemisphere of the mw are confined to one single plane , which in addition is inclined by less than @xmath108 from both the average orbital plane of the mw satellites co - orbiting in the vpos and from the lgp1 might be more than a lucky coincidence . the study by @xcite , according to which the distance and the line - of - sight velocity makes five out of seven of these galaxies more consistent with being backsplash galaxies than infalling ones , whereas the latter type should be seven times more abundant , might indicate another small - scale problem for the current @xmath1cdm cosmology . to find at least 5 out of 7 galaxies to be backsplash objects if only 13 per cent of all galaxies in this distance range should be backsplash objects has a likelihood of only 0.12 per cent . even if half of the candidates are false positives , i.e. we would expect to find that 26 per cent of all galaxies are identified as candidates , the chance to draw 5 out of 7 candidates is still only 3 per cent . if leo p is included as a backsplash object , these likeliehoods drop even more . unless cosmological simulations can show that there is a large excess of backsplash haloes in one hemisphere from a central galaxy , the @xmath1cdm model will face an `` overabundant backsplash problem '' in the lg . already now we have to wonder how , in a whole hemisphere , there can be more galaxies which have been close to the mw before and escaped to large distances than there are galaxies which are falling in , unless at least some of them have formed in and been expelled from the lg as tdgs . in addition , if the identification as backsplash galaxies by @xcite is confirmed by other studies , we might even face a `` planar backsplash problem '' as almost all of the current backsplash candidates lie within about 100 kpc of the same plane , which is defined by only the northern non - satellite galaxies and measures more than 1mpc in diameter . again , this might be well understood in a tdg scenario where new galaxies form in a tidal tail as a phase - space correlated population , but there is no reason to anticipate a planar configuration among backsplash galaxies . as illustrated in fig . [ fig : backsplashdebris ] , tdgs expelled from a major galaxy encounter will have similar orbital properties as backsplash - sub - haloes . they are receding from the main galaxy with high velocities on almost radial orbits because they originate from material in the galactic disc . their formation in a common event furthermore implies that the galaxies have similar travel times , in contrast to backsplash galaxies which might have been individually accreted . this could be a promising route to observationally discriminate between the two origins . such a discrimination will be important because the search for backsplash galaxies in the lg might in fact be good at identifying ancient tdgs expelled in a past major galaxy encounter . the approximate alignment of the mw satellites co - orbiting in the vpos @xcite , the m31 satellites co - orbiting in the gpoa @xcite , the dominant non - satellite galaxy plane lgp1@xmath5 in the lg and its connection to the ngc 3109 association ( this work ) motivate our attempt to schematically sketch the geometry of the lg in fig . [ fig : lgsketch ] . it shows the before - mentioned structures approximately face - on . starting on the left , we see the orientation of m31 ( solid ellipse ) and the gpoa consisting of more than half its satellite galaxies ( dashed circle ) . the arrow heads along the circles indicate the sense of rotation . m31 is moving towards the mw on an almost radial orbit @xcite . towards the bottom of the sketch we see a number of dwarf galaxies belonging to lgp1@xmath5 which appear to form a kind of bridge between the galaxies ( small circles ) . as seen from the mw these are also approaching , except for tucana ( which , as one of the only two distant dsphs in the lg , is also morphologically different from most of the shown dwarf galaxies which are predominantly gas - rich dirrs ) . these five bridge-galaxies have similar positions on the sky and similar line - of - sight velocities like the magellanic stream @xcite , which falls in towards the mw with the lmc and smc in the southern galactic hemisphere . the mw itself is oriented almost edge - on in this view ( flatter black ellipse ) , but its polar structure of satellite galaxies is seen face - on ( dashed circle ) . we can see that both satellite galaxy structures orbit in the same sense . in the north of the mw an over - density of hypervelocity stars ( hvs ) recede in the direction of the constellation of leo @xcite and opposite to the magellanic stream @xcite , which is relevant in this context because the tidal disruption of a dwarf galaxy close to the galactic center has been suggested as a possible formation mechanism for grouped hvss @xcite . whether this over - density continues to lower galactic latitudes is currently unknown . the two galaxies at intermediate distance in the same constellation , leo t and leo a both have small approaching galactocentric line - of - sight velocities . they might just have turned around , falling back towards the mw after having passed and receded from it some time ago . this is supported by their low velocities . @xcite predict that to be on its first infall towards the mw , leo a would have to have a large tangential velocity of two times its radial velocity . furthermore , phoenix on the other side of the mw is at essentially the same distance as leo t ( 415 compared to 422kpc ) , but it is falling towards the mw with a twice as large galactocentric line - of - sight velocity ( -103 compared to -58kms@xmath88 ) . finally , the members of the ngc 3109 association , starting with leo p in the same constellation mentioned before and continuing to lower galactic latitudes , are all receding with high velocities which indicate that they have been close to the mw in the past . not shown in the sketch is leo i , the most distant of the classical mw satellite galaxies . it appears not to co - orbit in the vpos and in contrast to all other classical satellites has a larger radial than tangential velocity @xcite . it moves away from the mw and might even be unbound to the mw , which would be unexpected if leo i traces a dark matter sub - halo @xcite . leo i s 3d velocity vector @xcite points into the direction of leo t : as seen from leo i , leo t s position is at @xmath109 and leo i s most - likely velocity vector points to @xmath110 . in the future , the galaxy might therefore become a part of the great northern plane . ugc 4879 , the topmost galaxy in the sketch , does not fit in . it has a very low line - of - sight velocity and its free - fall time from its present position onto the lg is larger than a hubble time @xcite . it is therefore thought to have never interacted with a major lg galaxy @xcite . in addition , the galaxy has the largest offset from the great northern plane ( see sect . [ sect : ngc3109leoplane ] ) and according to @xcite almost all its stars appear to have formed more than 10gyr ago . finally , of all galaxies in the great northern plane , ugc 4879 has the lowest gas to star mass ratio @xcite . ugc 4879 therefore seems to be the most unlikely galaxy to be associated to the dwarf galaxy structure in the lg . overall , there appears to be a trend that a variety of objects preferentially fall in from the galactic south while those in the galactic north recede . this is confirmed by fig . [ fig : m31anglevel ] , which plots the cosine of the angle between the position on the sky of a non - satellite lg galaxy and the position of m31 , @xmath111 , against the galaxy s galactocentric radial velocity @xmath112 . for comparison , the coloured lines illustrate the average radial velocities of sub - haloes in the elvis suite of cosmological simulations modelling lg - like galaxy pairs @xcite . they are measured in 10 bins in @xmath113 , where @xmath114 is the angle between a sub - halo position , as seen from one major dark matter halo ( representing the mw ) , and the position of the other major dark matter halo ( representing m31 ) . all sub - haloes within 0.3 and 1.8mpc from the mw equivalent and outside of 300kpc from the m31 equivalent are included . the dashed lines indicate the scatter in sub - halo velocities , 95 per cent of all sub - halo velocities are below the upper dashed lines and similarly 95 per cent are above the lower . two sets of lines are shown for each pair of galaxies , each assuming a different main halo of each halo pair to be the mw equivalent . for clarity only the four simulations with the largest volume uncontaminated by low - resolution particles are included , but the remaining 8 simulations in the elvis suite cover the same range in @xmath112 . [ fig : m31anglevel ] shows that the lg galaxies tend to approach from the direction of m31 and recede in the opposite direction . if the lg were simply collapsing towards the lg barycenter , the galaxies in the direction of m31 are expected to approach but those on the opposite side are expected fall towards the lg barycenter together with the mw , resulting in radial velocities closer to zero . this is confirmed by the cosmological simulations , which at @xmath115 scatter by @xmath116 around @xmath117 . the observed velocities , however , continue to larger receding values of 150 to 190kms@xmath88 for the direction opposite to m31 . such high radial velocities are not found for dark matter sub - haloes in the simulations . while part of the velocity gradient will thus be due to the mw falling towards m31 , the continuation of this trend in the direction opposite to m31 indicates that we might witness a larger - scale flow of dwarf galaxies within a flattened structure . the linear trend of increasing velocities with decreasing @xmath111 can be interpreted as due to the mw moving through the lg dwarf galaxy population , such that its gravitational potential is deforming the structure ( in addition to external tides , ) and the mw might have accreted or is still accreting dwarf galaxies from it ( such as the lmc if it is on its first infall ) , while others are leaving on the opposite side after having passed close to the mw ( such as the potentially unbound leo i ) . while this schematic representation currently is speculative ( as we do not know the tangential motions ) and does not prove any scenario right , the geometry presented in fig . [ fig : lgsketch ] might help to constrain attempts to model the lg . , against the galactocentric radial velocity , @xmath112 , of galaxies in the lg ( within 1.8mpc from the mw ) . black dots represent galaxies more distant than 600kpc from the mw while grey diamonds represent galaxies at a distance between 300 and 600kpc , which might be more affected by the gravitational pull of the galactic potential . galaxies closer than 300kpc to the mw or m31 , except for m31 itself , are not shown . the coloured lines represent sub - halo velocities measured from one major dark matter halo in the elvis suite of cosmological simulations of lg equivalents @xcite , where @xmath114 is measured from the position of the other major halo in the simulated group . the solid line is a fit to the observed galaxies at more than 600kpc distance from the mw resulting in @xmath118 . , width=321 ] despite the intriguing alignments discovered among the currently known lg galaxies , it is to be expected that additional nearby galaxies will be discovered in the future . as our current knowledge is based on a large number of different sources , it is impossible to accurately consider selection effects and the possibility , however unlikely , that additional dwarf galaxies will preferentially be found outside of the structures does exist . we essentially do not know what we have not seen yet . nevertheless , for searches for nearby dwarf galaxies , the northern galactic hemisphere is generally considered the best - studied direction because the sdss sky coverage initially focussed on this region . this might give some confidence in the significance of the great northern plane . this is further supported by the present non - detection of new bright dwarf galaxy candidates in single - epoch data of the pan - starrs1 survey . while based on an approximately 1 magnitude shallower photometric depth than the sdss , it covers a significantly larger fraction of the sky ( away from the vpos ) and therefore supports the satellite galaxy anisotropy revealed by the presently known dwarf galaxies @xcite . however , despite being covered by the sdss , leo p was only found by the less extended alfalfa hi survey , indicating that additional faint galaxies might hide in the lg . only once proper motions of the lg dwarf galaxies are known will we be able to directly determine whether the planar structures are dynamically stable . this is so far only possible for the 11 classical mw satellites in the vpos , of which nine are found to be consistent with moving within the structure @xcite , while one ( leo i ) is moving towards the great northern plane ( as discussed in sect . [ sect : scheme ] ) . promising developments in this regard , such as the measurement of the proper motions of the m31 satellites m33 and ic 10 via water masers or the optical proper motion measurement of the mw s most - distant classical satellite leo i @xcite and of m31 @xcite , indicate that acquisition of three - dimensional velocity data for a larger sample of lg galaxies might be feasible in the near future . in addition to the tentativeness of the dynamical stability of the dwarf galaxy structures , all of the different scenarios of their formation have yet to address serious issues . most importantly , if the observed dwarf galaxies are identified with dark matter sub - haloes such as those produced in cosmological simulations , these do not naturally result in the observed planar and co - orbiting structures . most claims of the contrary @xcite have subsequently been addressed in the literature and shown to lack important aspects of the observed situation or to be inconsistent with additional observations ( * ? ? ? * ; * ? ? ? * ibata et al . in prep , pawlowski et al . in prep . ) furthermore , the inability of present simulations to reproduce the observed structures comes in addition to the other known small - scale problems of the @xmath1cdm cosmology . in the alternative tdg scenario phase - space correlated dwarf galaxies occur naturally , but there are two crucial questions that yet lack decisive answers : why do the observed lg dwarf galaxies have large velocity dispersions that are classically interpreted as a strong indication for dark matter ? and how could tdgs , born out of pre - processed material stripped from much larger galaxies , end up on the mass - metallicity relation ? modified gravity models and non - equilibrium dynamics ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * yang et al . in prep . ) might provide answers to the first , and the early formation of tdgs at a redshift of @xmath119 and out of less - metal rich material from the rim of interacting galaxies might indicate a possible approach to the second . however , none of these issues have yet been satisfactorily investigated in a full and self - consistent model forming a lg - like group of galaxies . we have investigated how the recently discovered dsph galaxy perseus i and the ngc 3109 association extended with the recently discovered dwarf galaxy leo p are related to , and might fit in with , the dwarf galaxy structures present in the lg . our work has shown that the ngc 3109 association can not necessarily be interpreted as an independent group of galaxies , but might be related to the lg dwarf galaxy population and as such might provide important constraints on attempts to model the whole lg . the main results of our analysis are : 1 . perseus i is consistent with being part of the lgp1@xmath5 , the dominant plane of non - satellite galaxies in the lg , at least if andromeda xvi is associated with the gpoa . 2 . in the context of mond , we have predicted perseus i s velocity dispersion to be @xmath120 . the corresponding prediction for cetus ( @xmath121 ) is in much better agreement with the more recent observational value ( @xmath122 : * ? ? ? * ) than with the previous measurement ( @xmath123 : * ? ? ? the prediction for tucana ( @xmath124 ) is in conflict with the available measurement . we note that the observations of lack the spectral resolution to resolve the predicted velocity dispersion . unfortunately , no similar predictions are possible in a @xmath1cdm context . the orientation of the ngc 3109 association consisting of the dwarf galaxies ngc 3109 , antlia , sextans a , sextans b and leo p has been determined in the same coordinate system used to review the planes of co - orbiting satellite galaxies around the mw and m31 and the symmetric larger - scale dwarf galaxy structure in the lg @xcite . the association aligns with the supergalactic plane , is almost perpendicular to our line - of - sight and parallel but offset by 300 - 500kpc to lgp1@xmath5 . 4 . the members of the ngc 3109 association have large receding velocities which indicate that they have been close to the mw in the past , possibly at the same time about 79gyr ago . this is consistent with their orbits passing within @xmath90kpc of the mw suggested by @xcite and the identification as likely backsplash galaxies by @xcite . together with the association s extremely narrow extent and perpendicular orientation this argues against the association tracing a thin and cold cosmological filament . the timing is consistent with independent timing estimates for several suggested major galaxy encounter scenarios in the lg , during which phase - space correlated populations of tdgs could have formed that would today give rise to the observed dwarf galaxy structures . the association aligns with the other three distant ( @xmath125kpc ) lg galaxies in the northern hemisphere of the mw in a narrow plane ( rms height of @xmath126kpc and axis ratios of @xmath127 and @xmath128 ) . this `` great northern plane '' passes through the center of the lg , is inclined to lgp1@xmath5 by only @xmath6 and to the gpoa by @xmath101 and is consistent with being aligned with the preferred orbital plane of the mw satellites in the vpos . five out of seven ( 6 of 8 if the later discovered leo p would be included ) of the galaxies in the great northern plane have been identified as likely backsplash objects by @xcite , and most of the remaining five backsplash candidates in the southern galactic hemisphere are also situated close to the same plane . as only a small fraction of sub - haloes in simulations are identified as backsplash objects the finding of a majority of such galaxies in one hemisphere is extremely unlikely ( @xmath2 per cent ) and might constitute an `` overabundant backsplash problem '' for @xmath1cdm . it would mean that more galaxies are receding in one direction from the mw than are being accreted onto the mw from that direction . a natural explanation for this would be the local formation of galaxies as tdgs , which , if expelled to large distances , have very similar orbital properties like cosmological backsplash galaxies . that the backsplash candidates preferentially lie in a common plane is also consistent with an interpretation as phase - space correlated tdgs . lg galaxies are found to be preferentially infalling in the galactic south and receding in the galactic north , which possibly indicates that the mw is moving through a stream of dwarf galaxies . this would be in qualitative agreement with the m31-merger scenario by @xcite , in which our galaxy passes through the tidal debris expelled in a past merger forming m31 ( see also * ? ? ? * ; * ? ? ? . however , the tdg origin would be in m31 , such that the galaxies move away from m31 sufficiently fast which would imply that they have a tangential velocity component relative to the mw , rendering the timing estimate of sect . [ sect : timing ] somewhat useless . the scenario should however be testable on lg scales if proper motion measurements of the distant dwarf galaxies could be obtained . we thank heather morrison and paul harding for very useful discussions and shea garrison - kimmel for making the elvis simulations publicly available . we also thank the referee , rodrigo ibata , for very helpful comments .

an arbitrarily small concentration of impurities can affect the spin hall conductivity in a two - dimensional semiconductor system . we develop a boltzmann - like equation that can be used for impurity scattering with arbitrary angular dependence , and for arbitrary angular dependence of the spin - orbit field around the fermi surface . for a model applicable to a 2d hole system in gaas , if the impurity scattering is not isotropic , we find that the spin hall conductivity depends on the derivative of with respect to the energy and on deviations from a parabolic band structure , as well as on the angular dependence of the scattering . in principle , the resulting spin hall conductivity can be larger or smaller than the `` intrinsic value '' , and can have opposite sign . in the limit of small angle scattering , in a model appropriate for small hole concentrations , where the band is parabolic and , the spin hall conductivity has opposite sign from the intrinsic value , and has larger magnitude . our analysis assumes that the spin - orbit splitting and the transport scattering rate are both small compared to the fermi energy , but the method is valid for for arbitrary value of .

cond-mat0509702

an electric current passing through a semiconductor can induce spin polarization near lateral edges of the sample , with opposite sign at opposite edges . such an effect , commonly referred to as the _ spin hall effect _ , had been predicted more than 30 years ago @xcite and was observed recently.@xcite this effect results from the coupling between spin and momentum of an itinerant electron . two specific mechanisms have been considered for the spin - hall effect . the `` extrinsic '' mechanism couples spin to momentum during events of impurity scattering in the mott skew scattering channel @xcite . as a result , scattered electrons of different spin polarizations propagate towards opposite boundaries of the system . as opposed to the extrinsic , the `` intrinsic '' effect originates from a spin - split band structure @xcite rather than from impurity scattering . experiments have been performed on both electron - doped @xcite and hole - doped @xcite gaas - based semiconductor structures of different dimensionalities . a spin hall effect in three - dimensional @xmath5-doped gaas films @xcite has been explained in terms of extrinsic skew and side - jump impurity scattering @xcite while the intrinsic mechanism was estimated @xcite to yield spin accumulation smaller by an order of magnitude than actually observed in the experiment . the two - dimensional electron gas on a ( 110 ) surface of gaas , studied in [ ] , was also in the dirty limit , and the spin hall effect observed there was attributed to extrinsic effects . on the other hand , @xmath6-doped two - dimensional ( 2d ) samples used in the experiments of ref . [ ] were estimated to contain very few impurities and it was suggested that the observations reflect a spin - hall effect of the intrinsic type . while experiments directly measure spin accumulation along the sample edges , theoretical contributions to the field of spin - hall effect are mostly concerned with a different quantity which is easier to calculate , namely , the spin current . the latter is conventionally defined as the expectation value of the operator @xmath7 , where @xmath8 and @xmath9 are , respectively , the operators of the carrier spin and velocity ( which is , in general , also spin - dependent ) , and the normalization constant @xmath10 has been chosen differently by different authors . ( in the present paper , we define @xmath11 to have eigenvalues @xmath12 , and we choose @xmath13 , so that the spin current is equal to @xmath14 times the difference in the particle currents for carriers with opposite spins . ) apart from the normalization , the definition of spin current is largely a matter of convenience since it _ is not _ rigorously related to spin polarization ( accumulation ) by a conservation law , as opposed to _ e.g. _ the electric charge conservation , namely , @xmath15 . non - conservation of spin - current ( due to spin precession ) makes the problem of its conversion into polarization highly non - trivial . moreover , since spin current is even with respect to time inversion , it is not necessarily absent even in equilibrium @xcite . in the present paper we concentrate on the 2d electron and hole systems . in this case the components of spin current @xmath16 polarized along the direction ( @xmath17 ) perpendicular to the 2d plane appear due to non - equilibrium conditions only . the paper [ ] predicted a universal value for the intrinsic spin - hall current in a 2d electron gas , with spin - orbit coupling of the rashba form , which , using the normalization of the present paper is given by @xmath18 independent of the strength of spin - orbit interaction . ( the prediction of ref . [ ] for _ three - dimensional _ hole systems , though dependent on the electron density , was also independent of the strength of so coupling ) . these results immediately posed a question of how impurities would affect the spin - hall conductivity . the derivation of formula ( [ sin ] ) neglected scattering , but also assumed a steady state . the latter , however , is impossible in an infinite system if momentum relaxation is absent . soon a number of papers appeared about the role of impurity scattering , which , initially , reached differing conclusions . some studies encountered decreasing of the magnitude of the effect with increasing scattering @xcite , while others found a universal value ( [ sin ] ) even in the presence of scattering . however , it is now generally agreed that in the simplest model of a 2deg , with a spin - orbit coupling that is linear in the wavevector @xmath19 , the dc spin hall conductivity will vanish , even in the case of arbitrarily small impurity concentration . inoue et al.@xcite first reached this conclusion in the clean limit , @xmath20 , where the spin orbit splitting @xmath1 exceeds the transport scattering rate @xmath3 , while entertaining the possibility of a finite effect for finite values of @xmath21 . the present authors @xcite developed a theory based on a quantum kinetic equation to describe spin - polarized transport in the two - dimensional electron systems for _ arbitrary _ ratio of spin - orbit splitting and scattering rate , as long as they are both small compared to the fermi energy ( @xmath22 ) . applying this theory to the spin - hall effect in a 2deg with ( @xmath23-linear ) rashba coupling and isotropic impurity scattering , we established the vanishing of spin hall conductivity for arbitrary value of the product @xmath21 . however , the cancellation is complete only in the bulk . near the contacts which inject unpolarized electrons , spin - currents are non - zero and , hence , a spin - polarization normal to plane can accumulate near the corners of a sample . the models used in refs . [ ] and [ ] assumed short - range impurity potentials , with isotropic scattering . the result of vanishing spin hall conductivity in a 2deg with rashba coupling was subsequently confirmed by several studies @xcite , and were shown to hold for arbitrary angular dependence of the impurity scattering . early numerical calculations @xcite seemed to contradict the theoretical work but later ones supported the vanishing of spin - hall current @xcite . results for a 2deg with rashba coupling can also be applied directly to a model for a 2deg on an ( 001 ) surface with pure @xmath23-linear dresselhaus coupling , as the two problems are related by a rotation in spin space . as was pointed out by dimitrova@xcite , the vanishing of _ dc _ spin currents @xmath16 in an _ infinite _ 2d system with @xmath19-linear spin - orbit coupling has a simple explanation in terms of the following operator argument . consider a system with the spin - independent disorder and the `` rashba''-type spin - orbit coupling @xcite @xmath24 , where @xmath25 represents the set of three pauli spin matrices , and @xmath19 is the canonical momentum . the equation of motion for the operator of spin polarization yields ( @xmath26 ) : @xmath27/\hbar=-4\alpha m j^z_l/\hbar^3 $ ] , where @xmath28 is the electron effective mass . it follows from this identity that a nonzero expectation value of the spin current @xmath29 would result in a time - dependent spin polarization along the corresponding direction @xmath30 , which is impossible in a steady state . this operator argument applies also to a model with a combination of rashba and @xmath19-linear dresselhaus coupling . in agreement with this argument , analytical calculations confirm the vanishing of weak - localization ( of the first order in @xmath31 corrections to the spin hall conductivity @xcite . it should be emphasized that this vanishing of the spin hall conductivity is characteristic of the 2d electron system with so coupling _ linear _ in momentum . cubic dresselhaus terms in the spin - orbit hamiltonian , which are @xmath32 for a 2d electron system on a ( 001 ) surface , should result in a finite spin hall conductivity @xcite . for a 2d _ hole _ system in an asymmetric confining well , the rashba effect gives rise to a spin orbit field @xmath33 whose direction winds three times around a circle in the x - y plane as @xmath19 moves once around the fermi circle . it turns out , in this case , that for short - range impurities there is no cancellation in the clean limit , and one then obtains the full value of the intrinsic spin hall effect , at least in a simplified model which neglects cubic anisotropy.@xcite curiously , in contrast to the 2deg , it is the vertex corrections that vanish in this case , analogous to the absence of the vertex corrections to the electrical conductivity for short - range impurities @xcite . vertex corrections are also absent in an @xmath5-doped 3d system with dresselhaus interaction , @xmath34 , and short - range impurity scattering , in this case due to the cubic crystal symmetry . in the absence of vertex corrections , the impurities lead to a smooth relaxational suppression of the spin - hall conductivity due to the broadening of the spectral function of carriers . in principle , one should obtain the full intrinsic value of the spin - hall conductivity in the clean limit , but 3d systems are always in the dirty limit , where the intrinsic spin hall conductivity is reduced by a factor @xmath35 . as mentioned above , the spin hall effect observed in 3d @xmath5-type gaas is most probably due to extrinsic scattering effects , omitted from the model discussed here.@xcite it is natural to ask whether the absence of vertex corrections for the 2d hole system is special to the assumption of short range impurities , or whether it applies equally well to a model with arbitrary angular dependence of the impurity scattering . in the present paper , we develop equations to calculate the spin hall conductivity for arbitrary form of the impurity potential , and for arbitrary form of the spin - orbit field @xmath33 in a 2d system . solution of these equations is simple in models which have an overall circular symmetry , and we obtain analytic results in the limit of small angle scattering , resulting from smooth disorder . this type of disorder is a good assumption for many 2d systems in which the impurity centers ( dopants ) reside relatively far from the heterostructure interface . the result of our analysis is that the spin - hall conductivity , in general , does depend on the form of the impurity scattering , and except for the case of isotropic scattering , depends also on the energy - dependence of the spin - orbit coupling and on the deviation from parabolicity of the energy dispersion for the carriers . in our circularly symmetric model , the magnitude of the spin - orbit field @xmath33 is independent of the direction of @xmath19 , while its direction winds @xmath36 times around the unit circle in the @xmath37-@xmath38 plane , at a uniform rate , as @xmath19 moves around the fermi circle . ( @xmath36 must be an odd integer . ) the energy - dependence of @xmath1 and deviations of the energy dispersion @xmath39 from a parabolic form near the fermi energy , are then characterized by two parameters @xmath40 and @xmath41 , defined in eq . ( [ b235 ] ) below , and for a given form of the impurity scattering , the spin hall conductivity is found to depend on @xmath36 , @xmath40 and @xmath41 . in section ii , below , we define precisely the models we are considering and the relevant parameters . in section iii , the transport quantities are expressed in terms of the wigner distribution function , for which boltzmann - like equations of motion are presented in section iv . in section v , these equations are solved in the case of our circularly symmetric model . in section vi , we obtain explicit formulas , eqs . ( [ c8 ] ) and ( [ c9 ] ) , for the two limiting cases of isotropic scattering and small angle scattering by the impurities . the relation of our results to previous theoretical results and to recent experiments is discussed in section vii . some details of the derivations are presented in appendix a , and an alternate derivation of the kinetic equations of motion , using green s function methods , is presented in appendix b. we consider a two - dimensional system of non - interacting electrons or holes , described by a one - body hamiltonian of the form : @xmath42 where @xmath43 is the potential due to impurities , and @xmath19 is the kinetic momentum operator . in the presence of a vector potential @xmath44 , we have @xmath45 , where @xmath46 . throughout this paper , @xmath47 is the charge of the carriers , and we set @xmath48 except where explicitly noted . we may include a uniform electric field @xmath49 by letting @xmath50 depend on time . by time reversal symmetry , the spin - orbit field @xmath51 must satisfy @xmath52 . note that vectors such as @xmath19 and @xmath53 , which refer to spatial coordinates , have two cartesian components , while vectors in spin space , such as @xmath51 have three components . we have assumed for simplicity that @xmath39 , the energy dispersion in the absence of spin orbit coupling , is isotropic and a monotonic function of @xmath54 . the formulas derived below can be extended in a straightforward manner to more complicated band structures , but the formulas would then be much less transparent , and the resulting integral equations are more difficult to solve , as a further simplification , we shall later specialize to models where @xmath51 has a simple rotational symmetry ; i.e. , we assume @xmath55 where @xmath56 is a complex number , whose magnitude may vary with the magnitude of @xmath54 , but whose phase is independent of @xmath19 . this will allow us to obtain analytic solutions to the equations . the final results will , of course , depend crucially on the winding number @xmath36 . we shall see that results can also be affected by the energy dependence of @xmath1 and by deviations of the energy dispersion @xmath39 from a parabolic form near the fermi energy . we characterize these dependences by constants @xmath40 and @xmath41 , with @xmath57 for @xmath54 equal to the fermi momentum . in other words , we assume that near the fermi energy , @xmath58 , and @xmath59 for a 2d electron system on a ( 100 ) surface of a iii - v semiconductor , the case @xmath60 corresponds to pure rashba coupling , while @xmath61 corresponds to a @xmath23-linear dresselhaus coupling . the case @xmath62 arises in a circularly symmetric model of a 2d hole system , which ignores warping due to the tetragonal symmetry . in the limit of small hole doping , the band structure is parabolic and @xmath63 ; hence @xmath64 and @xmath65 . at larger values of the doping , when the fermi - energy becomes comparable to the splitting between the light and heavy hole bands , the values of @xmath40 is reduced , and the value of @xmath41 will be negative . at still higher doping , both light and heavy hole bands become occupied , and the situation is more complicated . the hamiltonian ( [ b1 ] ) assumes that there is no direct spin - orbit coupling associated with the impurity potential or with applied electric field . for example , it omits terms of the form @xmath66 , which are generally present in systems with spin orbit coupling . this term leads to skew - scattering by the impurities , giving an extrinsic contribution to the spin hall conductivity . since there is no direct general relation between the coupling constant @xmath67 and the magnitude of the intrinsic spin - orbit splitting @xmath1 , it is at least logically consistent to consider a model with @xmath68 . moreover , the skew scattering cross - section is higher order in the impurity potential @xmath43 than the ordinary transport scattering cross - section , so it is normally dominated by places where the carrier enters a region of strong potential gradients , close to the impurity . in a remotely doped 2d hole system one might expect that the skew scattering contribution to the spin hall conductivity should be relatively small , at least if one can ignore scattering by residual impurities close to the 2d system . we concentrate on systems which are large compared to the mean free path . the `` mesoscopic '' spin - hall effect @xcite in ballistic heterostructures is , therefore , beyond the scope of the present paper . we begin by defining a wigner distribution function for the carriers , which is a @xmath69 matrix in spin space , given by @xmath70 in thermal equilibrium , if spin orbit coupling is absent ( @xmath71 ) and impurities are neglected , we have the standard result @xmath72 where @xmath73 is the fermi function . for @xmath74 , there are two different wave vectors at a given energy @xmath75 and a given direction @xmath76 , corresponding to spin states parallel or antiparallel to @xmath51 . introducing an index @xmath77 to distinguish these two states , one finds in equilibrium , @xmath78 with @xmath79 impurity scattering broadens the @xmath80 functions in these expressions , but does not shift their centers or alter their amplitudes , provided that the scattering rate is small compared to the fermi energy @xmath81 . we define the distribution as a function of energy and direction in @xmath19-space by @xmath82 with @xmath83 . ( we suppress here the indices @xmath53 and @xmath84 , and we shall omit other indices as well when the meaning is clear . ) substituting ( [ b4 ] ) into ( [ b5 ] ) , we find , in equilibrium , to first order in @xmath85 , @xmath86 where @xmath87 , and @xmath88 is the value of @xmath54 such that @xmath39 is equal to the given value of @xmath75 . for @xmath75 at the fermi energy , @xmath89 is the fermi momentum in the absence of spin - orbit coupling , and @xmath90 is the corresponding fermi velocity . the last term in ( [ b6 ] ) may be understood as arising from the difference in the densities of states for the two spin states @xmath77 , which may be written as @xmath91 , where @xmath92 and @xmath93 . the shifts in momentum and velocity have the opposite sign for the two values of @xmath94 , and the final term in eq . ( [ b6 ] ) is obtained by expanding to first order in these shifts . finally , we introduce functions @xmath95 and @xmath96 , which describe the excess number and spin densities in momentum direction @xmath76 and energy @xmath75 , defined by @xmath97 if @xmath5 and @xmath98 are known , we can readily compute the particle current and spin current , as well as the particle density and spin density at a spatial point @xmath53 . we define the spin density and spin current by @xmath99 @xmath100 where @xmath101 is the velocity operator , @xmath102 at any point @xmath53 , we then have for densities and currents @xmath103 @xmath104 @xmath105 @xmath106 \ .\ ] ] here @xmath107 is the `` bare '' velocity , and @xmath108 is the spin - orbit contribution . in the equation for @xmath109 , we have dropped terms of order @xmath110 . this is justified because we assume @xmath111 , and in the situations we will consider , values of @xmath112 are smaller than the characteristic values of @xmath113 by a factor of @xmath114 or more . we will be interested in the linear response of the system to an infinitesimal applied electric field @xmath49 , which might , in general , depend on space as well as time . in developing the equations of motion , then , we need only consider terms which are of first order in @xmath98 and @xmath5 , or first order in @xmath115 ; we may ignore terms of order @xmath116 or @xmath117 . in the absence of spin - orbit coupling , the linearized equation of motion for @xmath118 may be written as @xmath119 where @xmath120 is the standard advection term , @xmath121 describes acceleration of particles by the external field , and @xmath122 , \ ] ] is the collision integral . the scattering kernel @xmath123 is given , in the born approximation , by @xmath124 here @xmath125 is the fourier component of the scattering potential at wavevector @xmath126 , and we have assumed that its mean - square value is independent of the direction in space . in ( [ b10 ] ) , we have assumed that the electric field @xmath49 is oriented along the @xmath37-axis . the equations assume that spatial variations are slow on the scale of the fermi wavelength , and they may need to be supplemented by boundary conditions derived from microscopic equations , in the case of a sharp interface or sample edge . the equations given above for @xmath127 remain valid in the presence of spin - orbit coupling , to first order in @xmath128 . the equations of motion for @xmath98 , however , are affected in an essential way by a nonzero @xmath1 . we find here @xmath129 @xmath130 @xmath131 @xmath132 @xmath133\nonumber\\[5pt ] & + & \int_0^{2\pi } \ ; d\theta'\ ; [ { \bf m}(\theta,\theta ' ) n ( \theta')-{\bf m}(\theta',\theta ) n ( \theta ) ] \ . \nonumber \\[5pt]\end{aligned}\ ] ] spin - orbit contribution to the scattering kernel can be divided into two different parts : @xmath134 where @xmath135 , \ ] ] originates from spin - orbit correction to the density of states , and @xmath136 } { 4k_\epsilon v_\epsilon } \tan\left(\frac{\theta-\theta ' } { 2}\right)\frac{\partial k(\theta-\theta ' ) } { \partial\theta}\ ] ] is due to the spin - dependence of the momentum transfer entering the scattering matrix element . in the case of small angle scattering , ( [ b18 ] ) and ( [ b19 ] ) , which are correct to first order in @xmath51 , require not only that @xmath137 , but also @xmath138 , where @xmath139 is the typical scattering momentum . a more detailed derivation of these equations will be given in appendices a and b. we give here just a brief indication of the origin of the various terms . we then shall explore some consequences of these equations in simple cases , which will shed additional light on their meaning and will provide some non - trivial consistency checks . the contribution given by ( [ b15 ] ) may be understood as arising from precession of the spin , induced by the spin - orbit field @xmath51 . the second term on the right hand side of this equation is zero in any case where the direction of @xmath51 is determined by the direction of @xmath19 , independent of the magnitude of @xmath54 , and we shall omit it in the following . the electric field term ( [ b15e ] ) reflects the action of @xmath49 on the equilibrium distribution ( [ b4 ] ) , as explained in appendix a. the scattering kernel @xmath140 entering ( [ b16 ] ) gives a contribution to @xmath141 proportional to @xmath5 , which is present even when @xmath142 . the term @xmath143 , given by ( [ b18 ] ) , arises because there are two different fermi radii at a given energy @xmath75 and a given direction @xmath76 , corresponding to spin states parallel or antiparallel to @xmath51 , and the densities of final states are generally different for particles scattered into these two spin states . the second term @xmath144 , given by ( [ b19 ] ) arises when the scattering matrix element depends on @xmath145 , because the momentum transfer actually depends not only on @xmath146 , but also on the initial and final fermi radii , and hence on the component of the spin in the directions of @xmath147 and @xmath148 . first consider a situation where @xmath149 , and @xmath95 is independent of @xmath76 . the equations of motion for @xmath98 then have a time - independent solution where @xmath150 a change in electron density independent of @xmath76 is just what one would find if one makes a small change in the chemical potential , by an amount @xmath151 such that @xmath152 the change in the equilibrium distribution produced by a shift @xmath153 can be obtained from ( [ b6 ] ) by replacing @xmath154 by @xmath155 . we see that this would change the polarization by an amount precisely equivalent to ( [ d1 ] ) , as is required . as another example , we may consider a situation where there is no impurity scattering as well as no electric field . we may then construct spatially - varying time - independent solutions of the equations of motion of the form @xmath156 , \ ] ] where @xmath157 is a unit vector parallel to @xmath158 , @xmath159 is a unit vector perpendicular to @xmath157 , @xmath160 and @xmath161 are arbitrary constants , which may depend on @xmath76 and @xmath75 , and @xmath162 this steady state corresponds to a situation where we have populated eigenstates of the hamiltonian with energy @xmath75 and wavevector - direction @xmath76 , whose spin at @xmath163 is polarized in direction @xmath164 . the spin polarization precesses with a spatial - frequency @xmath165 as one moves away from the origin . finally , we may consider a situation where impurities exist only in a small region about the origin , whose radius is small compared to @xmath166 but large on the scale of the fermi wavelength . we can then construct a scattering wave solution of the hamiltonian , for which there is an incident plane wave of definite energy @xmath75 and wavevector @xmath19 , with polarization parallel or antiparallel to @xmath33 . the scattered wave radiating from the origin will have spin parallel to the spin of the incident wave for @xmath53 close to the origin , but will have a spin direction that depends on @xmath53 away from the origin . if one moves out along a constant direction @xmath167 , the spin direction will precess about the direction @xmath168 , where @xmath169 is a vector of magnitude @xmath89 and direction @xmath167 . the polarization @xmath170 produced by such a scattering wave solution will give a time - independent solution of the equations motion derived above , at least within the born approximation . we now specialize to the case where @xmath171 has the simple dependence on @xmath76 given by ( [ b22 ] ) . we wish to find the linear response to a uniform electric field , in an infinite homogeneous system . when the angular dependence of @xmath51 is given by ( [ b22 ] ) , the equations can be solved by fourier transform in @xmath76 , i.e. , by expanding @xmath172 in circular harmonics : @xmath173 \phi^+&\equiv&\phi_x+i\phi_y=\sum_m\phi_m e^{im\theta},\nonumber\\[5pt ] k(\theta-\theta')&=&\sum_m k_me^{im(\theta-\theta')}.\end{aligned}\ ] ] since @xmath174 is real , we have @xmath175 . also , since @xmath123 is real and an even function of @xmath176 , we see that @xmath177 is real and @xmath178 . equations ( [ b15 ] ) and ( [ b15e ] ) can then be written as @xmath179 \left.\frac{d\phi_m } { dt}\right |_b&= & ib_0\phi_{m - n}^z,\nonumber\\[5pt ] \left.\frac{d\phi_m^z } { dt}\right |_e&= & 0,\nonumber\\[5pt ] \left.\frac{d\phi_m } { dt}\right |_e&=&- \frac{ee\eta_0m}{2 } \left [ \delta_{m,(n+1)}-\delta_{m,(n-1)}\right ] \frac{\partial f_0}{\partial\epsilon},~~\end{aligned}\ ] ] @xmath180 it is seen by inspection that the solution of these equations has all components zero except for @xmath191 and @xmath192 , which are therefore determined by a set of three coupled linear equations . for a dc applied field , where @xmath193 , these equations may be written in the form @xmath194 @xmath195 @xmath196 where @xmath197,% \;\;\ ; \nonumber\\ \alpha^*_{(- ) } \equiv \left [ -\gamma_{(-)}^ * + \eta_0^ { * } ( n-1 ) \right].~\end{aligned}\ ] ] note that @xmath198 are dimensionless constants that depend on the shape of the angular dependence of the impurity scattering , but not on its overall strength . solution of these equations gives @xmath199 @xmath200 .\ ] ] note that the right - hand side of ( [ c5 ] ) is real , so that @xmath201 is pure imaginary . integrating ( [ c5 ] ) over the energy @xmath75 , we find the spin currents @xmath202 and @xmath203 , with a spin hall conductivity given by @xmath204 in the clean limit , @xmath205 , one can replace the denominator @xmath206 in eq . ( [ c7 ] ) by @xmath207 . since @xmath208 , we see that @xmath209 is independent of @xmath182 and independent of the magnitude of @xmath210 in this limit . by contrast , in the dirty limit , when @xmath211 , we see that @xmath212 note that spin - hall conductivity cancels at @xmath213 for an arbitrary scattering kernel @xmath214 . indeed , according to eq . ( [ b32 ] ) , @xmath215 is identically zero for @xmath213 , while @xmath216 and @xmath217 given by eqs . ( [ b32 ] ) and ( [ b33 ] ) are equal . thus , both @xmath218 , and @xmath219 are zero , therefore the first term in the eq . ( [ c7 ] ) vanish . also , @xmath220 by definition ( see eq . ( 53 ) ) , and the spin - hall conductivity is identically zero . this result is in agreement with a general argument given in ref . for the case of isotropic impurity scattering , we have @xmath221 , and @xmath222 for @xmath223 . we also have @xmath224 , @xmath225 , and @xmath226 . then , for @xmath227 , we find in the clean limit , @xmath228 this is just the `` universal intrinsic value '' of the spin hall conductivity for this model , as has been found previously by other authors , written for our present normalization . the result will be reduced by a factor of @xmath229 if @xmath230 is not small . in the limit of small angle scattering , we have @xmath231 k_0-k_m&=&\frac{m^2}{2\pi\tau } , \nonumber\\[5pt ] \widetilde{k}(\theta)&=&\frac{\theta}{2 } \frac{\partial}{\partial\theta } k(\theta ) \ .%\nonumber\\[5pt]\end{aligned}\ ] ] to compute @xmath232 , one has to integrate by parts and expand the integrand to the second order in @xmath76 . the result is : @xmath233 we then find @xmath234 \gamma_0^w&=&3\eta_0,\nonumber\\ \lambda^w&=&3n\eta_0 , \end{aligned}\ ] ] using eqs . ( [ b31 ] ) , ( [ c4 ] ) , ( [ c6 ] ) and ( [ c7 ] ) , we thus obtain , in the limit of small angle scattering and @xmath235 , @xmath236 as expected , this result gives @xmath237 for @xmath238 . for @xmath227 , the result for small angle scattering can be larger or smaller than ( [ c8 ] ) , and can even have opposite sign , depending on the values of @xmath40 and @xmath41 . for a 2d hole system in gaas , at low doping , with @xmath239 and @xmath64 , we obtain , for small angle scattering , @xmath240 which has different sign and is approximately twice as large as the result ( [ c8 ] ) for isotropic scattering . in a remotely - doped 2d electron or hole system , the charged impurities that compensate the carriers are located far from the 2d system , so they contribute only long - wavelength fourier components to the disorder potential , and produce only small angle scattering . as the set back distance is made large , their contribution to the transport scattering rate decreases , so that eventually the latter may be dominated by a small concentration of residual impurities close to the layer , whose scattering amplitudes may be nearly isotropic . thus , it is natural to consider a model where both types of impurities are important , so the scattering kernel @xmath241 contains both a narrow peak at @xmath242 , and an isotropic @xmath76-independent part . if the two contributions are weighted so that a fraction @xmath6 of the transport scattering rate arises from small angle scatterers and a fraction @xmath243 arises from large angle scatterers , then doing the fourier transform , one finds that the coefficients @xmath198 entering ( [ c7 ] ) are given by @xmath244 , for @xmath223 . [ from ( [ b21 ] ) and ( [ c4 ] ) , it may be seen that for arbitrary @xmath241 , the transport scattering rate is determined by @xmath245 , so that one always has @xmath246 . ] the constants @xmath247 and @xmath248 given by eqs . ( [ b32])-([b34 ] ) have no contribution from the isotropic part , and are therefore each reduced from the values given in ( [ b37 ] ) by a factor of @xmath6 . the quantity @xmath249 is independent of @xmath6 . substituting these values in the eq . ( [ c6 ] ) and ( [ c7 ] ) , one finds , in the clean limit , for @xmath250 : @xmath251 \right\ } \ . \nonumber \end{aligned}\ ] ] eq . ( [ sh - p ] ) reproduces the `` universal intrinsic value '' ( [ c8 ] ) for @xmath252 , and the small - angle scattering limit ( [ c9 ] ) for @xmath253 . note , however , that it does not simply interpolate between the two values : eq . ( [ sh - p ] ) is not necessarily a monotonic function of @xmath6 . in general , the parameters @xmath254 and @xmath41 are related to each other . however , they enter eq . ( [ sh - p ] ) only via the combination @xmath255 . in order to illustrate the behavior of our model for different values of @xmath256 , we plotted @xmath257 at different values of @xmath41 , keeping @xmath40 fixed ( see fig . [ fig1 ] ) . note that spin - hall conductivity can even change its sign for @xmath258 . also , @xmath259 can exceed the `` universal intrinsic value '' @xmath260 . the model with a combination of isotropic and small angle scattering can also be considered as an approximation to a situation where there is a single type of impurity with a scattering kernel that is neither isotropic nor strongly peaked at small angles . in agreement with previous theoretical work , we have found that there is no spin hall conductivity in the case @xmath261 , applicable to a 2d electron system with pure rashba or @xmath19-linear dresselhaus coupling , for arbitrary form of the impurity scattering , while the spin hall conductivity has the full intrinsic value for the case @xmath62 , which models a 2d hole system , when the impurity scattering is isotropic . for angle - dependent scattering , with @xmath62 , we find that the spin hall conductivity is altered , so that its precise value , and even its sign , can depend on such details as the ratio of small angle to large angle scattering , and the energy dependences of the spin - orbit coupling and of the hole velocity . for the 2d hole systems that actually occur in gaas , an accurate computation should also take into account warping of the fermi surface due to tetragonal anisotropy , which has been omitted from our model . our results imply that spin - hall conductivity for @xmath262 is not universal . quantization of @xmath263 in units of @xmath264 is broken by small - angle scattering processes even in the clean limit ( @xmath265 ) , when spin - orbit field is strong . in contrast , charge hall conductivity in the strong field limit is insensitive to the details of disorder scattering . this difference probably reflects the non - topological origin of spin - hall conductivity . this conclusion is indirectly supported by the result of numerical laughlin gauge flux experiment performed in ref . [ ] for @xmath213 . in a recent article , liu and lei @xcite have performed numerical calculations of the spin hall conductivity for a two - dimensional hole model , with scattering from impurities set back 50 nm from the layer . their model has a quadratic energy spectrum , and @xmath266 spin - orbit coupling , corresponding to our model with @xmath267 , and has densities varying from @xmath268 to @xmath269 . ( the densities in the published version of ref . were misstated by a factor of @xmath270 ) . their numerical results for @xmath209 range from slightly smaller than the intrinsic value , at their lowest densities , to roughly twice the intrinsic value at high densities , in all cases quite different from the prediction of our eq . ( [ c10 ] ) . for high densities , it appears that their model is in the regime where the spin - orbit splitting @xmath271 is larger than the momentum transfer in a scattering event , so that our formulas would not apply . ( the numerical results actually coincide with a prediction for this regime in a new work by khaetskii @xcite ) however , we do not have a complete understanding of the numerical results at lower densities . although we do not know the precise value to be expected for the spin hall conductivity in the 2d hole samples studied by wunderlich et al.,@xcite it seems plausible that the magnitude should be similar to the universal intrinsic value for @xmath62 . in recent work , nomura et al.@xcite have analyzed the geometry of the experiment in ref . [ ] and conclude that the amount of polarized light obtained is consistent in magnitude with what might be expected to arise from the predicted intrinsic spin hall conductivity . nevertheless , more work seems necessary before one can be confident that one has a complete understanding of the origin of the the observed polarization in these experiments . as has been noted by various authors , spin polarization near a boundary depends on the boundary conditions , and may not be simply related to spin hall currents away from the boundary . as an example , for the 2d _ electron _ system with pure rashba coupling , where @xmath237 , one may still find spin polarization in the @xmath17-direction near the lateral boundaries of a sample if spin - flip processes are strong at the boundary,@xcite or if electrons can cross the boundary into a contact or a region with different spin - orbit coupling@xcite . in any case , it is expected that the polarization near a boundary will be insensitive to processes that occur further from the boundary than a few times the length scale for spin relaxation , and this length scale is quite short for holes in gaas . it seems unlikely that one will be able to obtain a _ quantitative _ measure of the bulk spin hall conductivity in such samples based on observations of spin accumulation near a boundary . in an earlier theoretical work by two of the present authors,@xcite a set of boltzmann - like kinetic equations was derived for a matrix distribution , denoted @xmath272 , which contained spin information but was integrated over the energy @xmath75 . the model was restricted to the case of @xmath19-linear spin - orbit coupling , and impurities were not included . the focus of ref . [ ] was on the time evolution of distributions that are non - uniform in space , and the linear response to a time - dependent perturbation in the absence of impurity scattering . the formalism developed there can not be directly applied to spin - dependent transport in the presence of impurities , which is the focus of the present work . this work was supported in part by nsf grants phy 01 - 17795 and dmr 02 - 33773 . research of avs is supported by u.s . doe under contract deac 02 - 98 ch 10886 . we are very thankful to a. khaetskii , who derived independently spin - hall conductivity for small angle scattering@xcite , and discussions with whom helped to correct errors in the previous versions of our papers@xcite . the authors are also grateful to s. y. liu for calling their attention to the existence of ref . and for discussions of the model used in that paper . discussions with e. i. rashba are also gratefully acknowledged . we give here a derivation of the two terms @xmath143and @xmath144 in the scattering kernel ( [ b17 ] ) , using an approach based on fermi s golden rule . the term @xmath143 arises from the spin - orbit corrections to the density of final states . consider an unpolarized electron of energy @xmath75 , traveling in a direction @xmath273 , incident on an impurity located at a point @xmath274 . ( the incident electron may be considered to be an incoherent 50 - 50 mixture of spin up and spin down states along any convenient axis of quantization . ) suppose the electron is scattered into a direction @xmath76 . at the angle @xmath76 , there are two possible momenta @xmath275 , with @xmath276 , where @xmath77 denotes spin parallel or antiparallel to @xmath277 . as in the discussion following ( [ b6 ] ) , the corresponding densities of states , per unit @xmath278 , are given by @xmath279 , and the difference between @xmath77 is given by @xmath280 . if we ignore possible differences in matrix elements for the two spin states , the difference in transition rates is given by @xmath281 , where @xmath282 is defined in ( [ b12 ] ) . this leads to a net polarization of the scattered spin , giving a contribution to @xmath283 equal to @xmath284 , where @xmath143 is given by ( [ b18 ] ) . since spin is conserved during the scattering event , we will also have a contribution to @xmath285 equal to @xmath286 . the term @xmath144 arises from spin - dependent momentum shifts in the matrix elements . the momentum transferred during the scattering process depends on the spin orientation of both the initial and final state . letting @xmath287 distinguish between initial spin parallel or antiparallel to @xmath288 , we see that the actual momentum transfer is @xmath289 , where @xmath145 is given by eq . ( [ b12 ] ) and @xmath290 ( \partial q / \partial \epsilon ) \ ] ] is due to spin - orbit interaction . as we are assuming that @xmath291 , and we only need the scattering kernel correct to linear order in @xmath1 , we may consider the effects of @xmath292 and @xmath293 separately , and add the results at the end . suppose that @xmath294 . then @xmath295 ( \partial q / \partial \epsilon ) $ ] . the change in the squared matrix element for the scattering process is given by @xmath296 , which leads to different scattering rates for @xmath297 . the out - scattering process thus creates a net polarization of the electrons remaining at angle @xmath273 , and gives a contribution to @xmath298 equal to @xmath299 since spin is conserved in the scattering process , we must have a contribution to @xmath300 which is the negative of ( [ contrib1 ] ) . next consider the case where @xmath301 . now @xmath302 depends on @xmath94 , so the scattering rate depends on which of the two final states is involved . the difference in matrix elements leads to a contribution to the polarization at @xmath76 given by @xmath303 again , since spin is conserved in the scattering process , we must have a contribution to @xmath304 which is the negative of ( [ contrib2 ] ) . the sum of these four contributions give the contributions proportional to @xmath305 in eq . ( [ b16 ] ) . although we have used fermi s golden rule and the born approximation in deriving these results , we expect that eqs . ( [ b18 ] ) and ( [ b19 ] ) should hold more generally , within the model given by the hamiltonian ( [ b1 ] ) , as long as the impurities are sufficiently dilute . the direct effect of a uniform electric field term in the hamiltonian , in an infinitesimal time interval @xmath306 , is to displace the momentum distribution by an amount @xmath307 , while keeping fixed the orientations of the spins . to see the effect of this , it is convenient to integrate over energy and the magnitude of the momentum , and define a spin density at an angle @xmath76 by @xmath308 , \ ] ] with @xmath19 oriented in the direction @xmath76 . in thermal equilibrium , @xmath309 is given by @xmath310 under the assumption that @xmath1 and the temperature @xmath311 are small compared to @xmath312 . the electric field term conserves the total spin , but gives rise to a flow of @xmath309 around the unit circle at a velocity @xmath313 , for @xmath49 along the @xmath37-axis . then , to lowest order in the electric field , we have @xmath314 by definition , @xmath315 = \int d \epsilon \bphi ( \theta , \epsilon ) $ ] . moreover , it is clear that @xmath142 for energies far from the fermi energy , and for any fixed value of @xmath76 , we must have @xmath316 . equation ( [ b15e ] ) then follows from ( [ dmdt ] ) the kinetic equation ( [ b13])-([b19 ] ) contains non - diagonal elements of the spin density matrix and is therefore a quantum kinetic equation , rather than a conventional boltzmann equation , which contains only occupation numbers . in this section we show that this equation can be derived from the microscopic hamiltonian ( [ b1 ] ) by green s function methods . in our derivation , we shall mainly follow the route outlined in ref . [ ] . as usual , we introduce one - particle retarded and advanced green s function @xmath317 and @xmath318 , and keldysh function @xmath319 describing non - equilibrium state of the system . ( note that all green s functions are operators in spin space . ) treating the disorder potential as a perturbation , one may write dyson equation for these function in a matrix notation : @xmath320 here the lower bar denotes matrix in keldysh space , and @xmath321 . in this work , we neglect localization effects , and consider only diagrams with non - crossing impurity lines . this gives the standard approximation for the self - energy part @xmath322 where @xmath323 is the disorder correlation function . for a time - independent disorder and in the absence of electron - electron and electron phonon interactions , the functions @xmath324 and @xmath318 are independent of keldysh function @xmath319 . in fact , they are hermitian conjugate to each other : @xmath325 and it is sufficient to find only one of them . since the functions @xmath324 and @xmath318 are independent of the non - equilibrium state , they depend only on @xmath326 in a translationally invariant system . going to the fourier representation , one finds for @xmath324 : @xmath327 where @xmath39 is the hole dispersion without spin - orbit , and @xmath328 where @xmath329 is the fourier transform of the disorder correlation function @xmath330 . in general , the self - energy part ( [ selfenergy ] ) is an operator in spin space . its hermitian part describing disorder - induced renormalization of electron spectrum and spin - orbit coupling will be ignored in the following . the anti - hermitian part of @xmath331 which we denote @xmath332 describes the decay of one - particle state because of elastic scattering . it is convenient to project the function @xmath333 onto spin eigenstates : @xmath334 where @xmath335 are corresponding eigenvalues , and @xmath336 are lifetimes . since the imaginary part of @xmath337 is peaked near @xmath338 , one can replace it by a delta - function to get the self - energy in the leading order in impurity concentration . thus , we find @xmath339 in this work , we restrict ourselves by the first order corrections in spin - orbit coupling . also , we will be primarily interested in the decay time on the `` mass shell '' : @xmath340 expanding eq . ( [ tau - so ] ) to the first order in @xmath341 , one finds , on the mass shell , @xmath342 \ . \nonumber\end{aligned}\ ] ] thus , in general , scattering rates are different for spin majority and spin minority bands . having found the retarded function @xmath343 , one can now rewrite the remaining equation for @xmath319 to make it uniform in @xmath319 ( see ref . [ ] for details ) : @xmath344 where @xmath345 and operators @xmath346 and @xmath347 act on the left and the right argument of @xmath319 respectively . we now apply wigner transform : @xmath348 and the left hand side of eq . ( [ gk-1 ] ) becomes @xmath349 + \frac{1}{2 } \left\ { \frac{1}{\hat{\tau}_{\bf k } } , \hat{g}_{{\bf k}e}\right\ } \ , \nonumber\end{aligned}\ ] ] where the scattering rate operator @xmath350 is given by eq . ( [ tau - so - final ] ) . the right hand side of eq . ( [ gk-1 ] ) takes the form @xmath351 \nonumber\end{aligned}\ ] ] ( here we assumed that @xmath352- and @xmath84- dependence of @xmath353 is smooth on the scale of fermi wavelength ) . calculating transport quantities , we are normally interested in keldysh green s function at coinciding temporal points , @xmath354 , which in wigner representation corresponds to energy - integrated functions . however , eqs . ( [ gk-1])-([gk-3 ] ) contains green s function for arbitrary energies . to express all quantities in terms of functions at @xmath355 , we use the following ansatz for @xmath353 : @xmath356 where the function @xmath357 depends only on momentum . integrating eq . ( [ h - def ] ) over @xmath115 , one finds : @xmath358 therefore , the function @xmath359 can be used to compute transport quantities . in fact , the function @xmath360 can be interpreted as a distribution function , with diagonal elements of @xmath361 being the occupation numbers . now we can derive the equation for the function @xmath362 by integrating left- and right - hand sides of quantum kinetic equation ( [ gk-2 ] ) and ( [ gk-3 ] ) . since the left - hand side depend on @xmath115 only via @xmath363 , the integration is trivial . the terms on the right hand side ( eq . ( [ gk-3 ] ) ) contain @xmath115 explicitly , and should be carefully integrated . substituting the definition ( [ h - def ] ) into the eq . ( [ gk-3 ] ) , one gets four different terms . two of them contain the product of two retarded ( or two advanced ) green s functions , and vanish after integration over @xmath115 . the two remaining terms are anti - hermitian conjugate to each other . one of them has the form @xmath364 projecting green s function onto eigenstates with the help of eq . ( [ g - projection ] ) and integrating over @xmath115 , one finds : @xmath365 where the spin - majority and spin - minority energy bands are given by the eq . ( [ eigenvals ] ) . assuming the scattering rate @xmath366 in the denominator to be small , we now rewrite this term using sokhotsky formula : @xmath367 where @xmath368 denotes principal value . subtracting hermitian conjugate and expanding to the first order in spin - orbit field @xmath369 , one finds , for the right hand side of quantum kinetic equation : @xmath370 } { ( \epsilon_{{\bf k } } - \epsilon_{{\bf k}'})^2 } \nonumber \ .\end{aligned}\ ] ] the first term in the eq . ( [ terms ] ) is a conventional scattering term , while the second is the spin - orbit correction to the scattering rate . the third term is due to renormalization of one - particle spectrum by the disorder , and should be ignored in the following , since we already neglected similar terms in the real part of self - energy . now we can use the eq . ( [ terms ] ) and ( [ gk-2 ] ) to write the equation for @xmath371 : @xmath372 \nonumber \\ & = & 2\pi \int \frac{d^2{\bf k}'}{(2\pi)^2 } w({\bf k } - { \bf k } ' ) \left\ { ( \hat{h}_{{\bf k } ' } - \hat{h}_{{\bf k } } ) \delta ( \epsilon_{\bf k } ' - \epsilon_{{\bf k } ' } ) \right . \nonumber \\ & - & \left . \frac{1}{2 } \left\{\hat{h}_{{\bf k } ' } - \hat{h}_{{\bf k } } , { \bm \sigma } \cdot ( { \bf b } ( { \bf k } ) - { \bf b}({\bf k}'))\right\ } \delta ' ( \epsilon_{{\bf k } } - \epsilon_{{\bf k } ' } ) \right\ } \nonumber\end{aligned}\ ] ] ( we grouped together terms proportional to @xmath373 ) . to this point , we did not take into account the effects of external electric field . to incorporate the electric field into kinetic equation , consider a very smooth external scalar potential @xmath374 . strictly speaking , this potential breaks translational invariance , and the calculations we have done so far become invalid . however , for small field the potential @xmath374 is smooth on the spin - orbit precession length @xmath375 and does not lead to transitions between spin subbands . therefore , @xmath374 can be considered as a local shift of electron energy . the term @xmath376 now contains @xmath377 , which becomes , after wigner transform , @xmath378 . thus , the electric field term in the kinetic equation takes the standard form . rewriting kinetic equations in terms of function @xmath360 , one finds : @xmath379 + e{\bf e } \frac{\partial \hat{f}_{{\bf k}}}{\partial { \bf k } } \nonumber \\ & = & 2\pi \int \frac{d^2{\bf k}'}{(2\pi)^2 } w({\bf k } - { \bf k } ' ) \bigl ( % \left\ { ( \hat{f}_{{\bf k } ' } - \hat{f}_{{\bf k } } ) \delta ( \epsilon_{\bf k } ' - \epsilon_{{\bf k } ' } ) % \right . \nonumber \\ & - & % \left . \frac{1}{2 } \left\{\hat{f}_{{\bf k } ' } - \hat{f}_{{\bf k } } , { \bm \sigma } \cdot ( { \bf b } ( { \bf k } ) - { \bf b}({\bf k}'))\right\ } \delta ' ( \epsilon_{{\bf k } } - \epsilon_{{\bf k } ' } ) % \right\ } \bigr ) \nonumber \ .\end{aligned}\ ] ] the kinetic equation ( [ ke - momentum - final ] ) is sufficient to compute spin transport in a 2d hole gas . note , however , that in general , spin - orbit terms on the right hand side mix states with different values of @xmath380 because of the presence of the derivative of delta - function . thus , in general , one has to consider all electron states with different energies simultaneously . this mixing occurs because the quantity @xmath380 is not conserved in the scattering process . instead , the true electron energy @xmath381 is conserved . we are now going to use this integral of motion to separate contribution of electrons with different energies , and make energy conservation explicit . we shall make the change of variables , expressing the electron momentum in terms of energy @xmath75 and momentum direction @xmath76 . note , however , that the electron energy @xmath75 contains spin - dependent part , and therefore replacing momentum length by @xmath75 is not a simple algebraic transformation : one has to ensure the proper ordering of spin operators in the resulting expression . to figure out this operator order , consider the effect of spin - dependent gauge transformation @xmath382 on the one - particle density matrix @xmath383 . in the first order in @xmath384 , one finds : @xmath385 \nonumber \\ & - & \frac{i { \bf q } \cdot ( { \bf r } - { \bf r}')}{2 } \left\{\hat{\rho}_{{\bf r},{\bf r } ' } , { \bm \sigma}\cdot{\bf b}\right\}\end{aligned}\ ] ] by doing wigner transform with respect to @xmath386 and @xmath387 , one arrives at : @xmath388 \nonumber \\ & + & \left\ { ( { \bf q}\cdot \nabla_{\bf p } ) \hat{\rho}({\bf r } , { \bf p } ) , ( { \bm \sigma}\cdot{\bf b})\right\ } \ .\end{aligned}\ ] ] the second term in the eq . ( [ rho - transform ] ) describes spin rotation due to transformation @xmath389 , and the third term describes spin - dependent momentum shift . note that spin operator in the last term acts symmetrically on @xmath390 . this symmetry is related to the antisymmetric action of momentum operator on @xmath391 and @xmath392 in @xmath393 . one can undo the spin rotation described by the second term in the eq . ( [ rho - transform ] ) , and use the third term as a prescription of how the spin - dependent momentum shift should be performed : @xmath394 now we can replace momentum in the eq . ( [ ke - momentum - final ] ) by @xmath395 where @xmath396 is the unit vector in the momentum direction , and @xmath89 and @xmath397 are defined in sec . [ boltzmann - formulation ] . to rewrite the gradient term in this representation , one can simply replace the momentum according to the eq . ( [ momentum - transform ] ) : @xmath398 the spin - orbit rotation term already contains spin - orbit field and should not be modified at all : @xmath399 \ .\ ] ] to find the electric field contribution , note that for small field it should be computed only for an equilibrium distribution . in the equilibrium , the occupation numbers @xmath400 depend only on the total energy . thus , @xmath401 \ . \nonumber\end{aligned}\ ] ] ( the second term here takes into account the momentum shift given by eq . ( [ momentum - transform ] ) ) . in the scattering term , the delta - function and its derivative combine into @xmath402 , due to energy conservation . in zeroth order , the scattering contribution is @xmath403 where both @xmath23 and @xmath404 are taken on the surface @xmath405 . spin - orbit corrections to the eq . ( [ scattering-0 ] ) arise from either volume element @xmath406 or the matrix element @xmath407 . the matrix element contribution is due to spin - orbit induced changes of @xmath23 and @xmath404 : @xmath408 transforming this expression as explained in appendix a , one recovers the scattering kernel @xmath409 . the volume element is computed as follows : @xmath410 the derivative @xmath411 should be computed to the first order in spin - orbit field , as explained in comments to eq . ( [ b6 ] ) , and the volume element becomes @xmath412 ( as usual , its matrix part should be applied symmetrically ) . this gives additional contribution to the collision integral : @xmath413 the evolution of the density matrix @xmath414 in @xmath415 representation is then given by @xmath416 where all terms are given by eqs . ( [ term - g ] ) , ( [ term - b ] ) , ( [ term - e ] ) , ( [ term - w ] ) and ( [ term - d ] ) . to simplify calculation of physical quantities , one can remove the spin dependence from the volume element . to do that , we include the volume element in the distribution function via the following transformation : @xmath417 one can see that in the first order in spin - orbit @xmath418 the function @xmath419 can be now identified with the distribution used throughout the paper . substituting eqs . ( [ n - subst ] ) and ( [ b7 ] ) into eqs . ( [ term - e ] ) , ( [ term - w ] ) and ( [ term - d ] ) , one can reproduce eqs . ( [ b15e ] ) , ( [ b19 ] ) , and ( [ b18 ] ) respectively . indeed , the spin - orbit correction to the electric field term is given by the sum of contributions from eq . ( [ term - e ] ) and eq . ( [ n - subst ] ) : @xmath420 collecting the terms proportional to @xmath421 , one finds : @xmath422 thus reproducing eq . ( [ b15e ] ) . the matrix element correction to the collision integral ( [ term - w ] ) reproduces eq . ( [ b19 ] ) . the density of states correction ( [ term - d ] ) , however , is modified by a similar contribution that arises from transformation of the collision integral ( [ scattering-0 ] ) . in the latter , one has to express @xmath423 in terms of @xmath419 and _ then _ transform the resulting contribution according to eq . ( [ n - subst ] ) to get @xmath424 . as a result , four spin - orbit terms arise . then , the term in eq . ( [ term - d ] ) containing @xmath425 is cancelled by the corresponding contribution to the eq . ( [ scattering-0 ] ) . similar cancellation occur for @xmath426 term . as a result , one recovers eq . ( [ b18 ] ) . this concludes the derivation of kinetic equation ( [ b13 ] ) by green s function method .

driven elastic manifolds in random media exhibit a depinning transition to a state with non - vanishing velocity at a critical driving force . we study the depinning of stiff directed lines , which are governed by a bending rigidity rather than line tension . their equation of motion is the ( quenched ) herring - mullins equation , which also describes surface growth governed by surface diffusion . stiff directed lines are particularly interesting as there is a localization transition in the static problem at a finite temperature and the commonly exploited time ordering of states by means of middleton s theorems ( a. middleton , phys . rev.lett . * 68 * , 670 ( 1992 ) ) is not applicable . we employ analytical arguments and numerical simulations to determine the critical exponents and compare our findings with previous works and functional renormalization group results , which we extend to the different line elasticity . we see evidence for two distinct correlation length exponents .

1406.4270

elastic manifolds in random media are one of the most important model systems in the statistical physics of disordered systems , which exhibit disorder - dominated pinned phases with many features common to glassy systems @xcite . likewise , the depinning of an elastic manifold from a disorder potential under the action of a driving force is a paradigm for the non - equilibrium dynamical behavior of disordered systems capturing the avalanche dynamics of many complex systems if they are driven through a complex energy landscape @xcite . in particular , the problem of a directed line ( dl ) or directed polymer ( i.e. , an elastic manifold in @xmath0 dimensions ) in a random potential and driven by a force has been subject of extensive study @xcite . at zero temperature , there is a threshold force , at which the manifold changes from a localized state with vanishing mean velocity to a moving state with a non - zero mean velocity . the depinning transition has been treated within the framework of classical critical phenomena by functional renormalization group techniques starting from the more general problem of depinning of @xmath1-dimensional elastic interfaces ( with @xmath2 corresponding to lines ) . in @xmath3 dimensions , `` critical '' exponents at depinning can be calculated by functional renormalization using dimensional regularization in an @xmath4-expansion @xcite . at finite temperature , there is experimental evidence for a creep motion at any non - vanishing driving forces which can be understood qualitatively as thermally activated crossing of energy barriers which result from an interplay of both elastic energies of the line and the disorder potential . the energy of dls such as flux lines , domain walls , wetting fronts is proportional to their length ; therefore , the elastic properties of directed lines are governed by their line tension , which favors the straight configuration of shortest length . here , we concentrate on stiff directed lines ( sdls ) , whose elastic energy is given by the curvature of the line and , thus , represents a bending energy . this gives rise to configurations which are locally curvature - free and straight but , in contrast to the dl , the straight segments of sdls can assume any orientation even if this increases the total length of the line . there are a number of applications and interesting general theoretical issues concerning sdls in random media . the overdamped equation of motion of a sdl is the ( fourth - order ) herring - mullins linear diffusion equation @xcite , which also describes surface growth governed by surface diffusion . the depinning dynamics of the herring - mullins equation in quenched disorder has been subject of a number of prior studies @xcite , whose findings ( e.g. , an unphysically small roughness exponent ) differ in part significantly from ours as we will point out below ( see sec . [ sec : prev ] ) . sdls also describe semiflexible polymers with contour lengths smaller than their persistence length for bending fluctuations , such that the assumption of a directed line is not violated @xcite . our results can be applied to the depinning dynamics of semiflexible polymers such as dna or cytoskeletal filaments like f - actin in a random environment , such as a porous medium , as long as the correlation length of the depinning transition is smaller than the persistence length . as in other semiflexible polymer phase transitions ( such as adsorption ) non - universal quantities such as the value of the depinning threshold itself will be governed by the bending elasticity . at the depinning transition , where the correlation length diverges , semiflexible polymers will exhibit a crossover to critical properties of effectively flexible lines ( dls ) with a segment length set by the persistence length . moreover , static sdls in a random potential feature a disorder - driven localization transition at finite temperatures already in @xmath0 dimensions @xcite . due to an interesting dimensional shift in the problem , an analogous transition occurs for dls only in higher dimensions . in principle , this offers an opportunity to observe new phenomena arising from an interplay of depinning and delocalization for sdls ( the disorder used in refs . @xcite does not feature such a transition in the static problem ) . the localization transition at a finite temperature also offers the opportunity to test the usage of static quantities in the treatment of creep motion , because the sdl is not pinned by disorder above the critical temperature . a lot of the progress for the depinning theory of dls has been based on two basic theorems due to middleton @xcite , which essentially state that a forward moving dl can only move forward and will stop in a localized configuration , if such exists . this allows for an unambiguous time - ordering of a sequence of states . we will show that these theorems do not hold for the sdl , which can be seen as a consequence of the next - to - neighbor terms that are introduced by the bending elasticity . the paper is organized as follows . we present the model and the relevant equations of motion in sec . [ sec : model ] , where we also comment on some equilibrium properties and previous work on the sdl . in sec.[sec : ana ] we present analytical results based on scaling arguments and functional renormalization group calculations . in sec . [ sec : num ] we present numerical methods and results and , in sec.[sec : middle ] , we comment on the non - applicability of middleton s theorems for sdls . we conclude with a summary of our results in sec.[sec : concl ] . a general approach to driven elastic manifolds starts from @xmath5 dimensional manifolds with an elastic energy @xmath6 as an external pulling force @xmath7 intrinsically couples to only one transversal displacement component @xmath8 , there is no substantial gain in treating more than one transverse dimension , and we will restrict our analysis to this case . we assume that the manifold s internal coordinates @xmath9 are bounded to the @xmath1-dimensional hypercube @xmath10^d$ ] and call @xmath11 the system size ( or length in @xmath2 ) . we are mostly interested in the simplest case of lines with @xmath2 . the order of the derivative @xmath12 distinguishes different kind of elasticities , @xmath13 is the directed line ( dl ) with tensional elasticity , and @xmath14 is the stiff directed line @xcite ( sdl ) with bending elasticity . the equilibrium statistics of the sdl and dl model are related as there is a mapping of the sdl to the dl model in higher transverse dimensions in problems with short - ranged potentials @xcite . in an earlier work we extended this mapping between sdl and dl in short - ranged random potentials @xcite . in the context of polymers , the sdl model is often used as a weak - bending approximation to the so - called worm - like chain or kratky - porod model @xcite , which is the basic model for inextensible semiflexible polymers , such as dna , cytoskeletal filaments like f - actin , or polyelectrolytes . the weak - bending approximation is only applicable on length scales below the persistence length @xmath15 , where tangent fluctuations @xmath16 remain small . the persistence length contains thermal and disorder contributions as discussed in refs . additionally , there is a relation of the sdl model to surface growth models , on which we will comment below . in suitable units , the overdamped equation of an elastic line with an elastic energy @xmath17 as in can be written as @xmath18 at zero temperature . the first term on the right hand side represents the elastic forces as obtained from variation of the elastic energy . the force @xmath7 denotes a static , uniform pulling force , which tends to depin the line from the disordered medium , and @xmath19 is a quenched force due to the disordered medium . for this quenched force we distinguish two cases in the following , random - field and random - potential disorder . for random - field ( rf ) disorder , @xmath19 is a random variable with zero mean and short - ranged correlations @xmath20 whereas for random - potential or random - bond ( rb ) disorder the force @xmath21 stems from a random potential @xmath22 , which features zero mean and short - ranged correlations . hence , after a fourier transformation in the transverse dimension , we can write @xmath23 within this work we will mostly focus on random - potential ( rb ) disorder ; in particular , all our numerical results in sec . [ sec : num ] are for rb disorder . the analytical arguments in sec . [ sec : ana ] , i.e. , scaling relations and functional renormalization group results will be applied to both types of disorder . if thermal fluctuations at temperature @xmath24 are included , an additional time - dependent white noise @xmath25 with zero mean and correlations @xmath26 is added on the right hand side of eq . . we use @xmath27 to denote averages over time , @xmath28 $ ] for spatial averages at a given time , @xmath29 for the average over realizations of the quenched disorder and @xmath30 for the average over realizations of the thermal noise . a subscript @xmath31 at an average denotes a cumulant . at zero temperature , there is a finite threshold value @xmath32 at which the velocity @xmath33}\end{aligned}\ ] ] of a driven elastic manifold in a pinning potential becomes nonzero in the limit of very large times @xmath34 @xmath35 the shape of the driven line and its dynamics are usually characterized by the roughness exponent @xmath36 and the dynamical exponent @xmath37 , which describe how the roughness or width of the line , @xmath38 ^ 2 ] } = \overline { [ u(x , t)^2 ] _ c } , \label{eq : defw}\end{aligned}\ ] ] scales with the system size @xmath11 and the time @xmath34 : @xmath39 with the typical time scale @xmath40 sdls with @xmath14 are closely related to surface growth models for molecular beam epitaxy ( mbe ) @xcite . in the presence of surface diffusion , mbe has been described by a ( quenched ) herring - mullins linear diffusion equation @xcite for a surface described by a height profile @xmath41 , @xmath42 in the herring - mullins limit , it is assumed that effects from a surface tension can be neglected as compared to surface diffusion effects . surface tension would give rise to additional @xmath43-terms . in this context , the quantity @xmath7 describes the constant flux of particles onto the surface and @xmath19 random fluctuations in the deposition process . in suitable units , which reduces to the given form by rescaling @xmath44 , @xmath45 and @xmath46 with @xmath47 given by the discretization , @xmath48 and @xmath49 . ] , the herring - mullins equation ( [ eq : mulher ] ) is equivalent to the overdamped equation of motion of the sdl . without external forces ( @xmath50 , @xmath51 ) the exponent @xmath36 and @xmath37 take their thermal values @xmath52 and @xmath53 @xcite . the observation of the `` super - rough '' @xmath54 in tumor cells @xcite hints towards further experimental relevance of the herring - mullins equation . for the dl with @xmath13 , the equation of motion is the quenched edwards - wilkinson equation @xcite @xmath55 and the thermal exponents are @xmath56 and @xmath57 . generally , the thermal exponents are given by @xmath58 and @xmath59 . the equilibrium ( @xmath51 ) problem of a sdl in a @xmath0-dimensional medium with rb disorder features a localization transition at a finite temperature @xmath60 as we pointed out in refs . @xcite . the roughness exponent is @xmath61 in the disorder - dominated phase for @xmath62 and assumes the thermal value @xmath63 for @xmath64 . in contrast to the sdl , the dl with one transverse dimension is localized for all temperatures with a roughness exponent @xmath65 @xcite . this implies that the sdl in rb disorder offers the opportunity to study the dynamics of an unlocalized elastic manifold in disorder for @xmath64 and the interplay of the delocalization transition at @xmath66 and a depinning transition at @xmath67 . for rf disorder , functional renormalization group approaches @xcite using an expansion in @xmath68 around the upper critical dimension @xmath69 give a static roughness exponent @xmath70 to at least two ( and possibly all ) orders in an expansion in @xmath71 and in good agreement with numerical results both for the dl @xcite and the sdl @xcite . in a discrete model that directly implemented surface diffusion and was proposed to correspond to the undriven quenched herring - mullins equation a differing exponent @xmath72 was found for the sdl @xcite . the result @xmath70 is the simple scaling or `` flory '' result , that follows from balancing the typical elastic energy of a line with displacement @xmath8 , which scales as @xmath73 , with the typical disorder energy @xmath74 as the disorder energy is picked up at @xmath75 independent sites and its correlator decreases linearly in @xmath8 for large enough @xmath8 @xcite . similar arguments fail to reproduce the non - trivial rb roughness exponent , but can provide bounds to it as discussed in refs . @xcite . there has been some previous work on the depinning of sdls with rf disorder . from renormalization group analysis it is expected that the critical exponents of the depinning transition are universal for all disorders with shorter ranged correlations than rf ( including rb ) , although a different scenario is possible in principle @xcite . for the dl the exponents do coincide for rf and rb disorder @xcite . the roughness exponents previously found at the depinning of a sdl in rf disorder are @xmath76 and @xmath77 and a dynamical exponent @xmath78 @xcite . furthermore , in a discrete model @xcite based on the quenched herring - mullins equation @xmath79 and @xmath80 have been found at depinning . one obvious problem with these values for the roughness exponent @xmath36 is that they are _ smaller _ than the thermal value @xmath52 , i.e. , that disorder decreases the roughness of the line . we will comment below in more detail on similarities and differences in the findings of these studies to ours . in order to describe the depinning of driven elastic lines in a random medium within the framework of classic critical phenomena @xcite , the roughness exponent @xmath36 and dynamical exponent @xmath37 introduced in eq . ( [ eq : defzetaz ] ) are not sufficient but one additional exponent related to the control parameter , the driving force @xmath7 , is needed . in the vicinity of the depinning threshold @xmath32 we can introduce two exponents describing the `` order parameter '' , which is the velocity @xmath81 of the center of mass , and the correlation length @xmath82 : @xmath83 the correlation length @xmath82 gives the typical length of segments that rearrange during the avalanche - like motion close to the threshold ; the typical time scale for this segment motion is @xmath84 . we can use one of these exponents , e.g. the correlation length exponent @xmath85 , to obtain from the equilibrium scaling relation @xmath86 with a scaling function @xmath87 ( with @xmath88 and @xmath89 for @xmath90 ) , which underlies eq . ( [ eq : defzetaz ] ) , a corresponding scaling relation close to depinning , @xmath91 with scaling functions @xmath92 ( for forces above ( + ) and below ( - ) the threshold and with @xmath93 and @xmath94 for @xmath95 ) . there are two scaling laws relating the exponents @xmath96 and @xmath85 to the roughness exponent @xmath36 and the dynamical exponent @xmath37 at the depinning transition . the first scaling law simply establishes a relation between @xmath96 and @xmath85 using that @xmath97 ( for @xmath98 ) , which results in @xmath99 this relation is valid independently of the form of the elastic energy , i.e. , independent of @xmath12 . as all exponents should be positive this also implies @xmath100 . the other relation comes from an additional tilt symmetry of the equation of motion @xcite , which leads to @xmath101 for the sdl ( @xmath14 ) or , for general elasticity , to @xmath102 . the relations ( [ eq : betascalrel ] ) and ( [ eq : nuscalrel ] ) should hold both for rf and rb disorder at depinning . for the analysis of simulation data it is convenient to infer exponents from the short time scaling properties of the velocity , which follows from the scaling ( [ eq : scalingwtf ] ) and @xmath103 : @xmath104 where we introduce two auxiliary exponents @xmath105 for convenient data analysis [ using the scaling relation ( [ eq : betascalrel ] ) in eq . ( [ eq : delta ] ) ] . the exponent values obtained previously in ref . @xcite are @xmath106 and @xmath107 at the sdl depinning transition ( for rf disorder ) . these values are problematic as they violate the scaling relation ( [ eq : nuscalrel ] ) . one reason for this problem might be that the exponent @xmath36 has been determined by direct measurement of the roughness @xmath108 and its scaling for different system sizes @xmath11 . however , such an approach is strongly influenced by the choice of the transverse system size ( which should be @xmath109 ) because the value for the critical force @xmath32 depends also on the transverse system size . as ref . @xcite contains two other independently measured exponents , namely @xmath110 and @xmath96 in our nomenclature , and scaling relations ( [ eq : betascalrel ] ) and ( [ eq : nuscalrel ] ) imply @xmath111 . we can give a resulting `` scaling '' roughness exponent @xmath112 , which strongly differs . there is a another exponent that is often referred to as @xmath85 or @xmath113 describing the scaling of the sample - to - sample fluctuations of the threshold force @xmath114 in a system of finite size @xmath11 . in general , @xmath85 and @xmath115 do not have to coincide . for the dl , @xmath116 has been confirmed @xcite , whereas for the charge density wave problem ( periodic potential ) , @xmath85 and @xmath115 are distinct @xcite . this might affect the scaling relations ( [ eq : betascalrel ] ) and the auxiliary exponents @xmath117 and @xmath110 , see eqs . ( [ eq : gamma ] ) and ( [ eq : delta ] ) , which could read @xmath118 , @xmath119 , and @xmath120 . this happens if threshold force fluctuations by sample - to - sample disorder fluctuations on a scale @xmath11 , @xmath121 , are larger than the excess to the threshold force necessary to depin a segment of length @xmath11 , @xmath122 , see eq . ( [ eq : defnu ] ) . therefore , we expect @xmath123 if @xmath85 and @xmath115 are distinct . fluctuations in the depinning force @xmath32 origin from the fluctuations in the disorder . a finite manifold of size @xmath11 and width @xmath124 occupying a volume @xmath125 should at least pick up the same free energy fluctuations as a summation of i.i.d . random numbers which are @xmath126 . this results in a general lower limit for the depinning force fluctuations @xcite @xmath127 thus giving @xmath128 it has been argued that @xmath129 for an elastic line as long as the line continuously `` explores '' new regions of the disorder @xcite . in this interpretation , distinct correlation length exponents @xmath85 and @xmath115 for the charge density wave are a manifestation of the fact that the line `` knows '' the total potential at each point due to its periodicity . if @xmath129 holds , the bound ( [ eq : nufsbound ] ) is equivalent to a lower bound to the roughness exponent @xmath36 at depinning , @xmath130 which is valid for all elastic energies of the form ( [ eq : ham ] ) , i.e. , for all @xmath12 . the contraposition is equally important : if the roughness is less than @xmath131 , this implies that @xmath85 and @xmath115 are distinct . an upper bound to the roughness exponent comes from studying the line in the larkin approximation with a constant ( @xmath8-independent ) random force acting on every segment of the line @xcite . the resulting larkin roughness exponent is @xmath132 . as the line can gather unbound energy via large undulations in accordance with the force this represents an upper bound to the problem with finite potential range and , therefore , @xmath133 which holds for the roughness exponents @xmath36 below , at , and above depinning . . the length scale @xmath134 is determined via @xmath135 . the two - loop solution corresponds to a roughness exponent given by eq . . the small deviations between first and second order give rise to distinct roughness exponents ( [ eq : zeta12 ] ) . ] it has originally been suggested that the roughness exponent @xmath36 at the threshold force @xmath67 is independent of the type of disorder ( rb or rf ) and coincides with the static roughness of the line in a medium with rf disorder @xmath70 to all orders of @xmath71 @xcite . the discrepancy to numerical simulations @xcite has been solved by means of the two - loop functional renormalization group ( frg ) @xcite , which gives for the roughness exponent at depinning both for rb and rf disorder @xmath136 with the euler - mascheroni constant @xmath117 and a constant @xmath137 that depends on the form of the elastic energy , especially @xmath138 and @xmath139 . the frg approach is based on a flow equation for the disorder force correlator @xmath140 defined by @xmath141 at two - loop order the frg flow converges to the same fixed point disorder correlator ( shown in fig . [ fig : frg ] ) both for rb and rf disorder . therefore , two - loop frg predicts identical roughness exponents at depinning . as @xmath142 is rather large for the sdl , the two - loop contribution is important . this can be seen in the significant deviation of the two naive results @xmath143 using a direct evaluation of ( pad approximants give values @xmath144 to @xmath145 ) . thus the sdl roughness at depinning is expected to be above the static value for zero external force @xcite @xmath146 for rb disorder , but below the static result @xmath147 for rf disorder . we note that any extrapolation of the results in eq.([eq : frg ] ) necessarily violates the bound ( [ eq : zetafsbound ] ) , which holds if @xmath129 , as the negative two - loop contribution leads to @xmath148 for some finite @xmath71 . this is an indication for two distinct exponents @xmath85 and @xmath115 . in fig . [ fig : frg ] we show the numeric solution of the frg fixed point for the disorder correlator @xmath140 verifying that eq . ( [ eq : frg ] ) does indeed correspond to the unique non - negative faster than algebraically decaying ( convex in double logarithmic plot ) solution . we followed the numerical procedure outlined in refs . the new second order contribution @xmath149 giving the two - loop contribution in the ansatz @xmath150 can be approximated by the taylor series @xmath151 the frg calculation presented in ref.@xcite is in principle also capable of determining the dynamical exponent @xmath37 and , thus , all exponents . however , to two loops this involves the evaluation ( to leading order in @xmath71 ) of the `` correction to friction '' which remains an open task . it is possible ( details are given in appendix [ app : bounds ] ) to find bounds for the value of the dynamical exponent @xmath37 in two - loop order @xmath152 the exponents at one - loop order for the sdl are given in table [ tab : exp ] together with the previous numerical findings . l|ccccc exponent & @xmath36 & @xmath37 & @xmath85 & @xmath96 & @xmath110 + frg & @xmath153 & @xmath154 & @xmath155 & @xmath156 & @xmath157 + simulation ref . @xcite & @xmath158 & @xmath159 & @xmath160 & @xmath161 & @xmath162 + simulation this paper & @xmath163 & @xmath164 & @xmath165 & @xmath166 & @xmath167 + an interesting question in the frg analysis is the stability of the fixed point solution . it has been argued @xcite that the two previously defined correlation length exponents coincide , that is @xmath129 , if the fixed point solution for the disorder correlator is stable . this is in agreement with the results for charge density waves ( fixed point unstable @xcite , @xmath168 @xcite ) and the dl ( fixed point presumably stable @xcite , @xmath169 @xcite ) . we did not try to perform a full stability analysis , but we note that the simple argument of ref . @xcite for the instability of the fixed point for charge density waves might also hold for sdls : after integrating the frg flow equation from @xmath170 to @xmath171 it reads @xmath172 the second contribution on the right hand side is negative because ( to two loops ) @xmath173 and @xmath174 ( with the shorter notation @xmath175 ) . more importantly , this implies that @xmath176 and , thus , the frg fixed point of @xmath177 is unstable . the instability of the fixed point of eq . leads to a flow of the form @xmath178 with @xmath179 . thus an additional constant ( @xmath8-independent ) contribution to the fixed point @xmath180 is generated , that grows as @xmath181 goes to zero if @xmath176 . this means that to two - loop order a random force of the larkin - type is generated . this random force generates the larkin - like roughness @xmath182 which for the sdl with @xmath183 implies a separate correlation length exponent @xmath184 according to the tilt - symmetry scaling relation ( [ eq : nuscalrel ] ) . for sufficiently large external forces we can generalize the perturbative arguments for dls from refs . @xcite to general elastic manifolds in @xmath5 dimensions with an elastic energy of the form ( [ eq : ham ] ) . then , to second order in perturbation theory , the velocity of the center of mass of the line is @xmath185 here we assumed a short - ranged random potential that is completely uncorrelated along the internal dimensions . for the problem at hand ( @xmath2 , @xmath14 ) this implies that the first correction at large forces should scale as @xmath186 this is in agreement with our numerical results ( see sec . [ sec : feigel ] ) for not too large forces . the asymptotic behavior for very large forces can be understood with the same perturbative reasoning that led to eq.([eq : feigel ] ) , but neglecting the elastic forces and considering the effective single - particle ( sp ) equation @xmath187 . this leads to @xmath188 the crossover should happen , when the length scale @xmath189 on which the elastic adjustments to the forced induced motion are relevant becomes significantly smaller than the lattice spacing @xmath190 . this is the length scale that corresponds to the time scale @xmath191 for a moving line to get to the next `` disorder site '' for the free dynamic exponent @xmath192 . at finite temperatures @xmath193 there is a thermally activated motion , @xmath194 , for any driving force @xmath7 . for the dl this dynamical phenomenon has successfully been described via the thermal activation over barriers that are determined from a static consideration as the motion is expected to be very slow for low temperatures and forces @xcite . for forces @xmath195 below depinning this involves activation over large energy barriers ( diverging in the limit @xmath196 ) and results in so - called creep motion . as a result of thermal activation , the sharp depinning transition at @xmath67 is rounded . for forces @xmath197 above depinning the line moves with finite velocity and additional thermal activation has only little effect . for the sdl , there is an additional complication because of the disorder - induced localization transition at a finite temperature @xmath60 @xcite . for temperatures @xmath62 , we expect the sdl to behave qualitatively similar to a dl , i.e. , to exhibit creep for @xmath195 , thermal rounding of the depinning transition at @xmath67 , and only minor modifications of the flow behavior for @xmath197 . in order to derive the sdl creep law via a scaling argument , we consider the static equilibrium energy fluctuations , which scale as @xmath198 with the ( equilibrium ) energy fluctuation exponent @xmath199 . in a static framework , a depinning force @xmath7 simply tilts the energy landscape @xmath200 balancing these two contributions to optimize the total barrier energy @xmath201 , one gets the energy of the effective barriers scaling as @xmath202 with the barrier exponent @xmath203 for lines with @xmath2 , this gives @xmath204 for the dl and @xmath205 for the sdl . the velocity follows from the arrhenius law to be @xmath206 } \text{. } \label{eq : arrhenius}\end{aligned}\ ] ] for the dl this has been confirmed experimentally @xcite . for temperatures @xmath64 , on the other hand , the scenario is less clear . in the static problem , the sdl then depins already by thermal fluctuations . the roughness in the static problem is only larger than the thermal roughness , @xmath207 , for temperatures @xmath62 below the critical temperature @xcite . for @xmath64 , the static sdl is thermally rough @xmath208 , and there are no macroscopic energy fluctuations . assuming that the static equilibrium physics is indeed relevant for low driving forces ( as in the derivation of the creep law ) , the conclusion could be that there are only finite energy barriers of characteristic size @xmath209 , and the velocity is given by the so called thermally assisted flux flow ( taff ) @xcite @xmath210 } \text{.}\end{aligned}\ ] ] the treatment within the frg @xcite suggests that ( at least to one - loop order ) the force - force correlator @xmath211 is only affected by a finite temperature within the `` thermal boundary layer '' of width @xmath212 ( especially , there is only a `` cusp '' for @xmath213 ) . within this layer the line assumes the static roughness @xmath214 , whereas on larger scales the dynamic roughness @xmath215 becomes apparent ( for finite @xmath216 ) . thus , the frg seems to be in line with our previous reasoning , that is @xmath217 for @xmath64 and a taff - like velocity - force curve . however , this comes with the substantial caveat that , to our knowledge , the frg theory in its present form is not apt to describe the full temperature dependence and , in particular , the existence of a transition to thermal roughness ( which should manifest itself in the emergence of a fixed point solution with roughness @xmath218 ) at finite temperature . one basic difficulty is that a disorder - induced localization transition at a finite temperature @xmath60 does only occur for low dimensions @xmath219 , whereas the frg uses an expansion around an upper critical dimension @xmath220 . in section [ sec : rounding ] , we will present numerical evidence that the thermal rounding of the depinning transition at the threshold force @xmath32 is very similar for dls and sdls . this surprising result suggests that the thermal depinning transition of the sdl in the absence of a driving force does _ not _ change the depinning by a driving force qualitatively . the numerical determination of the barrier exponent @xmath221 is an unsolved problem even with algorithms specifically designed to capture the creep dynamics @xcite . the thermal rounding of the transition leads to a temperature dependent velocity at the ( zero - temperature ) threshold force , which , for the dl , has been found to follow a power - law @xmath222 it has been suggested that @xmath223 @xcite ( the perturbative argument in ref . @xcite is equally valid for the sdl ) . numerically and experimentally a value @xmath224 ( @xmath225 @xcite ) has been found for the dl with rb disorder @xcite . we make use of a recently presented implementation @xcite for graphics processing units ( gpus ) . the high number of parallelly executed computations becomes very advantageous for large lengths , with an effective speedup of two orders of magnitude for the dl @xcite . as the different elastic force generates only little additional branching ( the determination of the next - nearest neighbors with periodic boundaries ) , the gpu implementation is also favorable for sdls . additionally , we implemented an equivalent simulation for cpus . an euler integration scheme is used for the benefit of computational simplicity . in the numerical simulations we focus on random potential ( rb ) disorder ( as opposed to ref . the random potential is implemented by drawing random numbers from a normal distribution on a @xmath226 lattice where @xmath227 is the transverse size of the system @xcite . between the lattice points the potential is interpolated by periodic , cubic splines in @xmath8-direction . disorder averages were performed over 1000 samples . our system is periodic in @xmath9- and @xmath8-direction and the simulation starts with a flat line @xmath228 . we set the lattice spacing in both directions equal to one , @xmath229 , and approximate the fourth derivative by the central finite difference @xmath230 , which is of second order in space . a von - neumann stability analysis shows that ( without external forces ) the euler integration scheme becomes unstable for @xmath231 . throughout this work we used time steps @xmath232 , unless stated differently . as the width of the line can be influenced by various effects on different length scales , it can be difficult and error - prone to infer @xmath36 directly from the line width @xmath233 . therefore , it is helpful to study the structure factor @xmath234 where @xmath235 is the fourier transformation of @xmath41 . this assumes self - affinity of the line . for a sdl with @xmath236 ( @xmath237 , @xmath238 ) . the initial conditions ( flat line ) persist on long length scales and cross over to a regime with larkin roughness ( @xmath239 , solid line ) on short length scales . after adjustment to the disorder till about @xmath240 the conformation of the line does not change . ] for sufficiently large @xmath227 the system is ergodic only above the threshold , when the line moves . thus , for forces smaller than the threshold force ( cp . [ fig : fltfc ] ) the line does not show the static roughness , but adjusts itself to the potential on short length scales ( large wavenumbers ) leading to larkin roughness @xmath239 , whereas the conformation on longer length scales ( small wavenumbers ) depends on the initial conditions . here and in the following we plot the structure factor as a function of @xmath241 to correct for lattice artifacts . ( top ) and roughness @xmath242 ( bottom ) in a system with @xmath237 , @xmath238 for various forces and short times . both quantities exhibit power - law regime consistent with @xmath243 at @xmath244 . for clarity we plotted the data for these data as full ( @xmath245 ) and hollow ( @xmath246 ) circles and used dashed ( @xmath247 ) and dotted ( @xmath248 ) lines for the rest ( the key given in the upper figure holds for both ) . we note that the time - averaged roughness @xmath249 becomes maximal around @xmath250 . ] alike previous studies @xcite we employ short - time dynamics scaling to determine the critical exponents of the sdl at depinning . from eqs . ( [ eq : scalingwtf ] ) and ( [ eq : vf ] ) we know that , at @xmath67 , velocity and line width scale as @xmath251 these two observables contain the same information regarding the critical exponents . the behavior for forces near the threshold can be used to extract the exponent @xmath252 from rescaling the velocity according to eq . ( [ eq : vf ] ) , @xmath253 however , with actual data it turns out to be difficult to extract precise and unambiguous exponents from this finite - size scaling like procedure . additionally , the roughness exponent @xmath36 becomes apparent in the structure factor @xmath254 for wavenumbers below some @xmath255 , see eq . ( [ eq : sq ] ) . and the lower one forces @xmath195 . parameters are as for fig . [ fig : vcm1 ] , but we used more force values ( @xmath256 ) . as scaling is not expected to hold for small times , we only considered @xmath257 . in agreement with fig . [ fig : vcm1 ] , we used @xmath258 and @xmath243 . we see satisfying scaling for @xmath259 , which coincides with the findings of ref . @xcite . ] , @xmath238 at time @xmath260 at three different forces representing the three regimes : under - critical , critical and over - critical . although configurations are not strictly self - affine ( pure power - law behavior in the structure factor ) there is a clear emergence of an exponent @xmath261 around the threshold force . we identify this as the `` critical '' roughness exponent . as in fig . [ fig : vcm1 ] , the system shows maximal roughness at the threshold . ] in ref . @xcite , @xmath262 has been found from studying @xmath263 and @xmath242 for rf disorder . in fig . [ fig : vcm1 ] we obtain the same result for rb disorder . additionally , we show in fig . [ fig : finscal1 ] that the data can be nicely matched using the scaling of eq . ( [ eq : finscal ] ) and @xmath264 in agreement with ref . @xcite . by definition of @xmath110 and @xmath117 , this implies @xmath265 and by means of scaling relations @xmath266 as we already pointed out in sec . [ sec : scaling ] . we do not find any evidence ( at any force ) in the structure factor supporting this roughness . we do observe the emergence of a new roughness exponent @xmath267 at higher forces as can be seen from fig . [ fig : vergl ] . we consequently conclude that this is a more plausible location of the threshold force and that @xmath267 is indeed the threshold roughness exponent . . the system parameters are as for fig . [ fig : vcm1 ] , which suggested @xmath268 . the description by a power - law fits better in the `` macroscopic '' large time regime . although we rather determined @xmath32 by demanding a consistent @xmath110 as in figs . [ fig : vcm1 ] and [ fig : finscal1 ] , we note that , for @xmath269 , the second - power law fits particularly well . ] the examination of @xmath263 at higher forces and larger times reveals that the curves that do not saturate to a finite @xmath81 for large times ( @xmath197 ) or go to zero ( @xmath195 ) seem to consist of two power - law segments , where only the first one for smaller times is consistent with @xmath262 , see fig . [ fig : v1793 ] . this is analogous to the most recent findings for the dl in ref . @xcite . in fig . [ fig : v1793 ] we show @xmath263 for @xmath269 , which we believe to be close to the threshold force @xmath32 for the system size we use . as the exponent @xmath110 in the second power - law segment is rather close to zero and we have no independent method to determine @xmath32 ( see also below in sec.[sec : middle ] ) giving a precise value for @xmath110 is difficult . as the threshold roughness does not only influence the structure factor exactly at @xmath32 , but also for deviating forces ( given that the correlation length is still noticeable large ) , we think that we can rely on our value of @xmath270 even though we do not know @xmath32 precisely . additionally , we will support the claim of a roughness exponent @xmath270 with an independent method below in sec . [ sec : parabol ] . in fig . [ fig : v1793 ] we show that @xmath271 is consistent with an exponent value @xmath272 for larger times @xmath34 , whereas @xmath262 only at smaller times . for @xmath271 we obtain , based on the numerical results @xmath272 and @xmath273 , the following set of exponents : @xmath274 to determine @xmath37 we used the scaling relation ( [ eq : delta ] ) . the value for @xmath85 then follows from the scaling relation ( [ eq : nuscalrel ] ) based on the tilt symmetry , the value for @xmath96 from the scaling relation ( [ eq : delta ] ) . we note , however , that the value for @xmath96 is inconsistent with the value of @xmath275 that we obtain numerically as explained in the next subsection . a value @xmath275 implies @xmath276 with errorbars that are consistent with the exponent @xmath277 introduced above to characterize sample - to - sample fluctuations of the free energy , see eqs . ( [ eq : nufs ] ) and ( [ eq : nufs=2 ] ) . moreover , @xmath275 is only consistent with the scaling relation @xmath278 , see eq . ( [ eq : delta ] ) , if we use @xmath277 . this suggests that the sdl in disorder is indeed characterized by two different exponents @xmath279 similar to charge density waves @xcite . therefore , we conclude that all our data are best represented by the following extended set of exponents ( cp . table [ tab : exp ] ) : @xmath280 we are not able to give sensible margins of errors and simulations of much larger systems ( such that the ratio @xmath281 becomes constant ) might be needed . and transverse size @xmath282 . the time average was computed over @xmath283 time steps @xmath284 . we interpret the smaller slope at very low force differences as the beginning of the `` single particle''-behavior @xmath285 that has been derived in ref.@xcite , whereas at higher force difference the moving line with @xmath286 becomes apparent . the best value of @xmath96 depends on the threshold force used ( here and in the inset we used @xmath287 ) , but we found consistent values of @xmath288 for various sizes and samples . inset : velocity force relation for one larger sample with @xmath237 and @xmath238 . the time average was computed over @xmath289 time steps @xmath290 . there is a larger window with visible scaling . ] we also tried to determine the velocity exponent @xmath96 directly . in the treatment of dls , there is a significant discrepancy in the reported values of @xmath96 , with either @xmath291 @xcite or @xmath292 @xcite . these were determined by means of two slightly different approaches : one can either determine ( as in @xcite ) the threshold force @xmath293 for each sample and average @xmath294 \rangle(f - f_{t,\text{sample}})}$ ] or use @xmath295 ( as in @xcite ) . we chose the first approach because @xmath296 can not be clearly extracted from short - time dynamics scaling and , thus , the sample specific threshold force has to be determined anyway . in fig . [ fig : beta ] we show data for two single samples of sizes @xmath297 and @xmath237 . we infer from our data @xmath298 this agrees with the value @xmath299 found in ref.@xcite . we have successfully confirmed the perturbative results of sec . [ sec : lf ] , cp . [ fig : feigel ] , for large driving forces . additionally , we checked that the roughness exponent @xmath36 of the sdl takes its thermal value @xmath300 for sufficiently large driving forces . an interesting ( and maybe counterintuitive ) result is that increasing the force leads to a decrease in the `` effective '' temperature . we show this in fig . [ fig : rauhvsf ] using the time averaged width ( cp . eq . ) of the line as function of the force we expect the width @xmath233 to scale as @xmath301 with an effective persistence length @xmath15 of the sdl . for purely thermal fluctuations we have @xmath302 and @xmath303 @xcite . for the static sdl in disorder , on the other hand , we found a disorder - induced persistence length which is independent of temperature in the low - temperature phase of the sdl and which is minimal at the delocalization threshold @xcite . for the dynamic depinning , on the other hand , the simulation results in fig . [ fig : rauhvsf ] show an increase of the sdl width at depinning , followed by a decrease @xmath304 at high forces , where the roughness exponent assumes its thermal value @xmath302 again . this is consistent with a behavior @xmath305 with @xmath306 and , thus , an `` effective '' temperature @xmath307 that _ decreases _ with driving force @xmath7 . this can be rationalized from the frg approach by noting that for high velocities the disorder contributes in form of an additional thermal noise corresponding to a temperature @xmath308 @xcite . at high forces the line is far from the threshold and , thus , @xmath177 should be close to the rb - disorder value of @xmath309 . the observed behavior @xmath310 implies @xmath311 because @xmath286 . for large forces in _ one _ sample . the green solid line is the perturbative result as given by eq.([eq : feigel ] ) , the brown dashed line is given by @xmath312 . a similar cross - over to the single - particle behavior has been observed for the dl @xcite . we used an euler integration scheme with @xmath289 small time steps @xmath313 to accurately simulate the system even for large forces . for steady state results the time average was performed only over the last @xmath314 time steps . the system has lateral size @xmath315 and transverse size @xmath316 . ] . the time average was performed over @xmath317 time steps . the line is maximally rough at the threshold , which is consistent with a `` shockwave '' motion , where small segments of the line move whilst the rest is blocked . at higher forces the width reduces with the force , which could be interpreted as a decrease of the effective temperature or , analogously , an increase in the effective persistence length . the solid line is a guide to the eye showing the @xmath318 trend of the width . ] a different approach @xcite to compute the threshold force and , additionally , the effective disorder correlation functions is to pull the line very slowly with a spring . this means to introduce a parabolic potential acting on each line segment according to @xmath319 and move the center @xmath242 of the parabolic potential moves with a ( small ) constant velocity @xmath320 . we call @xmath321 the strength of the potential . the underlying idea is essentially that the force @xmath322 exerted by the parabolic potential on the line as it moves forward becomes the threshold force for @xmath323 , @xmath324 . more precisely , it was found for the dl that @xmath325\rangle m^2 = f_t + c m^{2-\zeta}\end{aligned}\ ] ] with some constant @xmath326 ( from now on , we assume that @xmath327 is sufficiently small ) . for general @xmath12 , the expected corrections due to finite values of @xmath181 have to be adjusted to account for the different elasticity . the length scale @xmath328 at which the confinement through the parabolic potential becomes relevant follows from balancing the elastic and the potential energy per length @xmath329 @xmath328 is the length of independently adjusting line - segments . averaging over all @xmath233 , i.e. , averaging over disorder , typical displacements scale as @xmath330 . this leads to effective forces scaling as @xmath331 which includes the aforementioned dl result ( @xmath13 ) . a different approach @xcite leading to the same result would be to use the known scaling of the finite - size corrections to the threshold force in one sample @xmath332 together with the notion that the relevant length scale is imposed by the parabolic potential and therefore given by @xmath333 . the confinement splits the line into independent segments of length @xmath328 and , therefore , one gets @xmath334 . using eq . one finds @xmath335 and , thus , these two approaches are equivalent . in this derivation it is also clear that there will be deviations for very small @xmath12 when @xmath328 exceeds @xmath11 . as the line is constantly moved forward we chose to change the implementation of the potential . we still have a fixed amount @xmath227 of potential values ( knots for the cubic spline ) , but we update the potential `` on - the - fly '' as the line is moved forward . every time a segment of the line reaches a new quarter of @xmath227 , we update the quarter that has the greatest distance to the current location and compute the splines [ in principle , this changes the spline at the current location of the line , but the change is negligibly small if @xmath227 is large enough ( we used @xmath336 ) ] . our motivation for this scheme was to avoid finite - size - effects in the transverse direction and to be able to compute the disorder average as a time average . in fig.[fig : parabol ] we show that our data are consistent with @xmath273 . furthermore , this setup allows for a direct measurement of the effective force correlation function and hence a validation of the renormalization group solution . this is achieved via the second cumulant of @xmath337 as @xmath338 this contribution is closely related to the shape of the force correlation function @xcite , see eq . ( [ eq : zeta ] ) . the existence of a non - vanishing @xmath339 , often referred to as `` cusp '' , is a sign of the non - analyticity of the correlation function . we show our results in fig . [ fig : parabolkumul ] , which demonstrate promising agreement with the functional renormalization group fixed point function . , @xmath336 ) . the solid line is a fit for @xmath340 $ ] with @xmath261 yielding @xmath341 . the velocity of the parabolic potential was @xmath342 . the shape of the numerical results and their deviations from the analytical expectation resemble the findings for the dl in ref . the value @xmath343 deviates from the value found above because of the differing system size . ] according to ( [ eq : deltaunum ] ) with @xmath344 . the data are taken from the same simulation as for fig . [ fig : parabol ] . the dashed ( green ) line is the one - loop solution and the dotted ( black ) line is the two - loop solution , see sec.[sec : frg ] and ref . the length scale @xmath134 is determined via @xmath345 . the inset shows the relative difference between the two - loop solution @xmath346 and the numerically determined values @xmath211 , i.e. , @xmath347 for small @xmath348 . ] the finite mean velocity at arbitrarily small non - zero temperatures is only visible for very long simulation times , thus , the regime in which creep / taff behavior should occur is not accessible for us . the thermal rounding exponent @xmath349 , as defined in eq . , can be interpreted in a slightly different way , that we feel is more apt for the interpretation of numerical data . moving away from the threshold at @xmath67 and @xmath213 the velocity scales as @xmath350 or @xmath351 , respectively . thus , a finite temperature at @xmath32 can effectively be seen as a contribution to the pulling force with @xmath352 and @xmath353 . adapting our previous statements the perturbative conjecture would be @xmath354 . in fig . [ fig : thermalround ] we show that using this rescaling we can collapse data for the velocity as a function of @xmath355 at @xmath213 and for the velocity at the threshold @xmath67 as a function fo temperature . the data collapse is consistent with @xmath354 and @xmath288 . we compare our numerical results for the thermal rounding of the depinning transition for the dl and the sdl in fig . [ fig : vf_comb ] . surprisingly , we find _ no _ evidence for a qualitative change at a finite temperature that could be associated with the localization transition in the static problem . this could either mean that the change is too subtle to be apparent within our numerics or that , in terms of the frg , a finite velocity @xmath216 implies that the relevant fixed point is one featuring @xmath216 and @xmath213 , which would make the transition at finite temperature irrelevant for a moving line . or @xmath356 in the rescaling of the temperature ( black points ) . the data collapse with data for the velocity as a function of @xmath355 at @xmath213 ( blue line ) after rescaling using eq . ( [ eq : rescale ] ) . the data for the velocity as a function of @xmath355 at @xmath213 and , thus , all system parameters but the temperature are the same as for fig . [ fig : beta ] . ] ) . for a better comparison we show data for a sdl and a dl in the same disorder . note that as the same disorder was used the velocities are very similar , but not identical . all system parameters but the temperature are the same as for fig . [ fig : beta ] . ] in the study of the depinning of directed lines two important properties have been found @xcite : a ) the `` no - passing '' theorem , which states that two lines in the same disorder realization that do not cross each other at a given time will never cross each other ; b ) if each segment of the line has at some point in time a non - negative velocity the velocity will remain non - negative for all times ; this has been referred to as `` no - return '' property @xcite . the no - passing theorem is believed to hold for every convex next - neighbor elastic energy . the combination of these properties allows for a fast and precise algorithm to determine the threshold force directly @xcite . with regard to the no - passing theorem we consider the following situation : two lines @xmath357 , @xmath358 in the same medium that touch each other at exactly one point , that is there is one @xmath9 with @xmath359 and @xmath360 for @xmath361 . for the dl , it follows that @xmath362 . as both lines touch each other in @xmath9 a possibly differing velocity of the two lines is due to the elastic forces . in the discrete version we then have @xmath362 , because @xmath363 . this does not work for the sdl , because the difference in velocity @xmath364 can take any value . we visualize this in fig . [ fig : diffdlsdl ] . therefore , the sdl does not necessarily explore `` new '' regions of the disorder potential . a line that moves back and forth `` knows '' essentially the whole potential at any point . this could be an explanation for the two distinct correlation length exponents @xmath85 and @xmath115 that we found for the sdl , see eq . ( [ eq : exponents ] ) . the fact that middleton s theorems guarantee that @xmath365 is a monotonic function of time is also used in the evaluation of ambiguous vertices within the frg treatment in ref.@xcite . however , we feel that coming from a moving line with @xmath216 the quasistatic depinning limit is still well defined for the sdl , because the long time limit of the roughness is ( for finite @xmath11 ) finite and , therefore , all segments will eventually move on average with the same velocity . the agreement of the frg with our numerical results supports that there is no fundamental problem with the applicability of the frg for the sdl . still , there is definitely room and need for a more rigorous analysis . and @xmath366 are forced to move to @xmath367 and @xmath368 , whereas the ends and the line point at @xmath326 can not move . the dl ( green dashed line ) does not pass the blocked upper line and also does not cross the lower line . the sdl ( blue dash - dotted line ) seeks to reduce the curvature and , consequently , does cross both lines . ] we studied the depinning of sdls from disorder ( rf or rb ) in @xmath0 dimensions due to a driving force . using scaling arguments , analytical frg calculations , and extensive numerical simulations , we characterized the critical behavior at and around the depinning transition . our study revealed some characteristic differences in depinning behavior between sdls and dls governed by tension . the resulting equation of motion for the sdl in disorder is equivalent to the herring - mullins equation for surface growth , which is governed by surface diffusion rather than surface tension , in quenched disorder . our results also apply to semiflexible polymers with contour lengths smaller than their persistence length , which are pulled over a disordered surfaces or driven through a random medium . we show that middleton s theorems do not apply to sdls . nevertheless we find a well - defined threshold force @xmath32 for depinning . likewise , critical exponents characterizing roughness and the dynamics of depinning can be defined and numerically determined for the sdl as for the dl . the sdl represents an own dynamical universality class with a different set of exponents . our extensive numerical data is best described by the set ( [ eq : exponents ] ) of critical exponents which is also consistent with scaling relations , see section [ sec : scaling ] . we also investigated the behavior of the sdl persistence length , which exhibits a characteristic non - monotonous force - dependence through the depinning transition ( see fig . [ fig : rauhvsf ] ) . we transferred functional renormalization group results to the elasticity of stiff interfaces , which allows us to derive analytical results or bounds for critical exponents ( see section [ sec : frg ] ) . we find satisfying agreement of these analytical predictions with our numerical work . our results indirectly imply that the depinning threshold is associated with two distinct correlation length exponents @xmath85 and @xmath115 . to our knowledge this would be the first occurrence of such behavior in a non - periodic system . this result could be linked to the non - validity of middleton s no - passing theorem . our findings for the critical exponents at the threshold force disagree in parts with previous numerical work , which suggests that further work , especially on much larger systems , might be helpful to settle these exponents . for finite temperatures , the depinning of a sdl is an interesting problem because , at equilibrium ( no pulling force ) , the problem features a disorder - driven localization transition at a finite temperature . such a transition is absent for the dl , which remains in a localized disorder - dominated phase for all temperatures . surprisingly , the numerical results for a comparison of the thermal rounding of the force - driven depinning transition do not show any qualitative difference between dls and sdls , see fig . [ fig : vf_comb ] . in a renormalization group sense , this might imply that the force - driven depinning and the temperature - driven delocalization are not described by the same fixed point . we acknowledge financial support by the deutsche forschungsgemeinschaft ( ki 662/2 - 1 ) . in ref . @xcite the dynamical exponent @xmath37 has been found to be to two - loop order @xmath369 with @xmath370 , @xmath137 the same as in the main text and @xmath371 where we use @xmath372 as shorthand notation for the aforementioned correction to friction for the sdl @xmath373 and @xmath374 is the one - loop integral given by @xmath375 using @xmath376 we find the lower bound @xmath377 for an upper bound we note that in any dimension ( any value of @xmath4 ) the following inequality holds @xmath378 and @xmath379 can be evaluated to leading order in @xmath4 via laplace transforms @xmath380 substituting @xmath381 , @xmath382 gives @xmath383 @xmath384 in the second to last step we substituted @xmath385 . we have divided the integration in three terms to isolate the ( important ) divergent part . @xmath386 @xmath387 \nonumber\\ & = -\frac{1}{12}-\frac{\ln{4}}{6 } + \mathcal{o}(\varepsilon)\end{aligned}\ ] ] @xmath388 thus collecting all terms we find @xmath389 and @xmath390 which ultimately yields @xmath391 i.e. , eq . ( [ eq : zbounds ] ) in the main text . 68ifxundefined [ 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 * * ( ) link:\doibase 10.1103/revmodphys.66.1125 [ * * , ( ) ] link:\doibase 10.1103/physreve.53.414 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevlett.98.155701 [ * * , ( ) ] link:\doibase 10.1103/physrevb.75.220201 [ * * , ( ) ] @noop * * , ( ) , @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) in @noop _ _ , ( , , ) p. link:\doibase 10.1063/1.1722742 [ * * , ( ) ] link:\doibase 10.1103/physreve.61.4606 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physreve.58.r5217 [ * * , ( ) ] link:\doibase 10.1103/physreve.62.3299 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.56.1964 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevlett.66.92 [ * * , ( ) ] @noop * * , ( ) @noop * * ( ) @noop * * , ( ) http://stacks.iop.org/0022-3719/20/i=36/a=016 [ * * , ( ) ] link:\doibase 10.1103/physrevb.42.8577 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.80.849 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.36.39 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevb.79.184207 [ * * , ( ) ] @noop http://stacks.iop.org/0295-5075/81/i=2/a=26005 [ * * , ( ) ] link:\doibase 10.1103/physreve.85.021144 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevb.74.140201 [ * * , ( ) ] @noop _ _ ( , ) @noop ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop `` , ''

we estimate the impact of weak lensing by strongly nonlinear cosmological structures on the cosmic microwave background . accurate calculation of large multipoles requires n - body simulations and ray - tracing schemes with both high spatial and temporal resolution . to this end we have developed a new code that combines a gravitational adaptive particle - particle , particle - mesh ( ap3 m ) solver with a weak lensing evaluation routine . the lensing deviations are evaluated while structure evolves during the simulation so that all evolution steps rather than just a few outputs are used in the lensing computations . the new code also includes a ray - tracing procedure that avoids periodicity effects in a universe that is modeled as a 3-d torus in the standard way . results from our new simulations are compared with previous ones based on particle - mesh simulations . we also systematically investigate the impact of box volume , resolution , and ray - tracing directions on the variance of the computed power spectra . we find that a box size of mpc is sufficient to provide a robust estimate of the weak lensing angular power spectrum in the-interval ( 2,0007,000 ) . for a reaslistic cosmological model the power^{1/2}$ ] takes on values of a few in this interval , which suggests that a future detection is feasible and may explain the excess power at high in the bima and cbi observations .

1001.4991

photons from the last scattering surface of the cosmic microwave background ( cmb ) detected at redshift zero , are inevitably lensed by cosmological structures . this lensing produces a number of modifications to the cmb angular power spectrum , deviations from gaussianity , and @xmath5-polarization . while these effects have been extensively studied , and a review of their impact can be found in @xcite , we are here concerned with the lensing due to strongly nonlinear objects such as galaxy clusters and smaller - scale structures . in particular , we focus our attention on the estimation of the small angular scale ( high @xmath4 ) correlations induced by this lensing . this marks a somewhat different direction to the extensive recent efforts to extend n - body simulation techniques to all - sky modelling ( @xcite ) . n - body simulations are necessary to estimate some cmb anisotropies . this is the case for the gravitational anisotropies ( weak lensing and rees - sciama effects ) produced by strongly nonlinear cosmological structures . while simulations with sufficient resolution to model the formation of galaxies within an entire hubble volume are currently out of reach , researchers are getting close to this goal @xcite . however , until such simulations are possible in order to estimate gravitational anisotropies in the cmb it is necessary to create an artificial periodic universe where copies of the simulation box at various redshifts are replicated back to a given redshift . while numerous alternatives exist for how this is achieved all the methods propagate cmb photons along appropriate directions within these model universes . when conducting this ray - tracing , box sizes , spatial scales and ray directions must be chosen in such a way that the periodicity of the artificial universe has no appreciable effects on the resulting anisotropy . the ray - tracing procedure used here achieves these goals and was described and applied in earlier papers , where it was also compared with previous methods @xcite . two of these earlier papers dealt with weak lensing @xcite , and two others with the rees - sciama effect @xcite . the simulations discussed in these papers were conducted with the particle - mesh ( pm , @xcite ) n - body simulation algorithm . for the large simulation boxes required in our study ( see below ) the spatial resolution of pm simulations is moderate , at best . for this reason it was suggested in these papers that better n - body simulations might be necessary to estimate good cmb angular power spectra for large multipole indexes ( @xmath0 values ) . this suggestion is confirmed in the present study , where ap3 m ( adaptive particle - particle , particle - mesh ; @xcite ) simulations with high resolution are performed to build up an appropriate periodic universe , in which the cmb photons are then moved according to the aforementioned ray - tracing procedure . a cmb angular power spectrum has been estimated from @xmath6 to @xmath7 , and the accuracy of this spectrum in various @xmath4-intervals is discussed . motivated by a desire to match a number of cosmological constraints , including the five - year wmap ( wilkinson microwave anisotropy probe ) observations , high - redshift supernovae measurements and galaxy surveys , ( spergel 2007 ; vianna & liddle 1996 ; riess 1998 ; perlmutter 1999 ) we consider a cosmological model in which the universe is flat and the initial fluctuation spectrum has an inflationary origin . it is assumed that only scalar perturbations are present and that weak lensing from gravitational waves is negligible . the resulting distribution of perturbations is then gaussian and the perturbations themselves are purely adiabatic . post inflation , the power spectrum of the perturbations is almost exactly of the harrison - zeldovich @xcite form . this power spectrum is then modified during the radiation - dominated phase , as acoustic oscillations in the photon - baryon fluid damp growth , an effect accounted for by our use of a cmbfast calculation of the transfer function . the following cosmological parameters are used for consistency with the standard model @xcite : ( 1 ) a reduced hubble constant @xmath8 ( where @xmath9 is the hubble constant in units of km s@xmath10 mpc@xmath10 ) ; ( 2 ) density parameters @xmath11 , @xmath12 , and @xmath13 for the baryonic , dark matter , and vacuum energy , respectively ( the matter density parameter is then @xmath14 ) ; ( 3 ) an optical depth @xmath15 which characterizes the reionization ; and ( 4 ) a parameter @xmath16 normalizing the power spectrum of the energy perturbations . the layout of this paper is as follows : in section [ maps ] we outline salient issues in our map - making technique . we follow this in section [ nbody ] with an outline of n - body methods and our ray - tracing technique . results from our new simulations are presented in section [ results ] along with a critical comparison to earlier work . we then compare our results to current observations and conclude with a brief discussion . we begin by describing the procedure for constructing lensed maps of the cmb from small unlensed maps at the last scattering surface . note that in this section and those following , we choose units such that @xmath17 , where @xmath18 is the speed of light and @xmath19 the gravitational constant . for a quantity @xmath20 , @xmath21 and @xmath22 denote the value of @xmath20 at , respectively , the time of emission ( last scattering surface ) and the present . the scale factor is @xmath23 , where @xmath24 is the cosmological time , and its present value , @xmath25 , is assumed to be unity , which is always possible in flat universes . the unit vector @xmath26 defines the observation direction ( line of sight ) . small , unlensed maps of cmb temperature contrasts ( @xmath27 ) must be constructed to be subsequently deformed by lensing . these maps have been obtained with a method based on the fast fourier transform @xcite and lead to small squared - gaussian maps of the contrast @xmath28 . these maps are uniformly pixelised . the code designed to build up the unlensed maps ( map making procedure ) , requires the cmb angular power spectrum , which has been obtained by running the cmbfast code @xcite for the model described above . in order to deform the unlensed maps , the lens deviations corresponding to a set of directions , covering an appropriate region of the sky , must be calculated . these deviations are the quantities @xcite : @xmath29 where @xmath30 is the transverse gradient of the peculiar gravitational potential @xmath31 , and @xmath32 . the variable @xmath33 is @xmath34 in the case of weak lensing , the integral in eq . ( [ devi ] ) is usually integrated along straight paths , ignoring the small deflections associated with the lensing effect . this approximation , commonly known as the born approximation , is equivalent to keeping first order terms in a positional expansion of the transverse potential as a function of the normal ray and its lensing offset . while this approximation is known to be comparatively inaccurate at small @xmath4 ( @xcite ) , at @xmath35 detailed calculations using higher order perturbation theory suggest that corrections to the first order assumption are approximately two orders of magnitude lower than the first order term @xcite . recent n - body simulation work examining the validity of the born approximation ( @xcite ) has shown corrections only begin to become significant at the 5% level for @xmath36 . since we are interested in calculating @xmath37 values we cautiously accept the errors inherent in the first order approach . once the deviations have been calculated , they can be easily used to get the lensed maps from the unlensed ones . this is achieved using the relation @xmath38 where @xmath39 and @xmath40 are the temperature contrasts of the lensed and unlensed maps , respectively . given the unlensed map @xmath41 , and the map @xmath39 obtained from it after deformation by lensing ( the lensed map ) , the chosen power spectrum estimator can be used to get the quantities @xmath42 and @xmath43 , whose differences @xmath44 can be considered as an appropriate measure of the weak lensing effect on the cmb . moreover , a map of deformations @xmath45 can be obtained and these can be analyzed to get another angular power spectrum @xmath46 . since the maps @xmath47 and @xmath48 are not statistically independent , the spectra @xmath46 and @xmath49 appear to be very different ( see @xcite for details ) . eight hundred unlensed maps are lensed ( deformed ) by using the same @xmath50 field and the average @xmath46 and @xmath49 spectra are calculated and analyzed . both spectra are very distinct measures of the weak lensing effect under consideration . many times , only the customary oscillating @xmath49 spectrum is shown ; however , in a few appropriate cases , the @xmath46 spectra are displayed . our power spectrum estimator was described in detail in @xcite and @xcite . results obtained with this estimator were compared with those of the code anafast of the healpix @xcite package in @xcite and also in @xcite . these comparisons showed that our estimator is a very good one in the case of regularly pixelised squared maps as analyzed here . the primary simulations presented in this work were run using a parallel openmp - based implementation of the `` hydra '' code @xcite . this code uses the ap3 m algorithm to calculate gravitational forces within a simulation containing @xmath51 particles . in the ap3 m algorithm a cubic base " mesh of size @xmath52 cells per side is supplemented by a series of refined - mesh p3 m calculations to provide sub - mesh resolution . gravitational softening is implemented using the s2 softening kernel @xcite which is remarkably similar in shape to the cubic spline softening kernel used in many treecodes ( @xcite ) . the s2 softening used in the kernel is @xmath53 where @xmath54 is an equivalent plummer softening length which we quote throughout the paper to enable a simple comparison to other work . the softening length is held constant in physical coordinates subject to the resolution not falling below 0.6 of the interparticle spacing at high redshift . this technique is widely applied ( @xcite ) and is a compromise between assuring that the potential energy of clusters does not evolve significantly at low redshift , while still ensuring structures and linear perturbations at high redshift are followed with reasonable accuracy . lccccccccccccc rls_aa & ap3 m & @xmath55 & a & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 25 & 0.59 & d512a + rls_ab & ap3 m & @xmath55 & a & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 25 & 0.59 & d512b + rls_ba & ap3 m & @xmath55 & b & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 25 & 0.59 & d512a + rls_ca & ap3 m & @xmath55 & c & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 25 & 0.59 & d512a + ls_lda & ap3 m & @xmath58 & d & @xmath56 & 88 & 1024 & 24 & @xmath57 & 1024 & 6 & 50 & 0.59 & d1024a + ls_mab & ap3 m & @xmath55 & a & @xmath59 & 88 & 512 & 24 & @xmath57 & 512 & 6 & 50 & 0.59 & d512b + ls_@xmath60s1aa & ap3 m & @xmath55 & a & @xmath56 & 11 & 1024 & 24 & @xmath57 & 512 & 6 & 50 & 0.59 & d512a + ls_@xmath60s2aa & ap3 m & @xmath55 & a & @xmath56 & 11 & 1024 & 36 & @xmath57 & 512 & 6 & 75 & 0.59 & d512a + ls_@xmath601aa & ap3 m & @xmath55 & a & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 12 & 0.59 & d512a + ls_@xmath601ab & ap3 m & @xmath55 & a & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 12 & 0.59 & d512b + ls_@xmath601ba & ap3 m & @xmath55 & b & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 12 & 0.59 & d512a + ls_@xmath601bb & ap3 m & @xmath55 & b & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 12 & 0.59 & d512b + ls_@xmath602aa & ap3 m & @xmath55 & a & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 40 & 0.59 & d512a + ls_@xmath602ab & ap3 m & @xmath55 & a & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 40 & 0.59 & d512b + ls_@xmath602ba & ap3 m & @xmath55 & b & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 40 & 0.59 & d512a + ls_@xmath602bb & ap3 m & @xmath55 & b & @xmath56 & 11 & 1024 & 12 & @xmath57 & 512 & 6 & 40 & 0.59 & d512b + ls_hea & ap3 m & @xmath61 & e & @xmath56 & 1.4 & 1024 & 6 & @xmath57 & 512 & 6 & 15 & 0.29 & d256a + ls_mga & ap3 m & @xmath55 & g & @xmath59 & 88 & 512 & 24 & @xmath57 & 512 & 6 & 60 & 0.59 & d512a + ls_lha & ap3 m & @xmath55 & h & @xmath62 & 704 & 256 & 48 & @xmath57 & 512 & 6 & 120 & 0.59 & d512a + pmls & pm & @xmath61 & f & @xmath56 & 1.4 & 512 & 1000 & @xmath57 & 256 & 6 & 500 & 0.59 & d256a + initial conditions were calculated using the standard zeldovich approximation technique @xcite , and all simulations were started at a redshift of @xmath63 , which is sufficiently early to place modes in the linear regime . to account for the impact of varying box sizes , @xmath64 , we considered simulations of size @xmath65 mpc , @xmath1 mpc , and @xmath66 mpc and a full list of n - body simulation and ray - tracing parameters is given in table [ tab2 ] . at @xmath67 ( the beginning of our ray - tracing epoch in the simulation ) , the box sizes of @xmath65 mpc , @xmath1 mpc , and @xmath68 mpc correspond to square sky patches with angular sizes , @xmath69 , of @xmath70 , @xmath71 , and @xmath72 , respectively . these are then the sizes of our constructed cmb maps for each simulation . as in our previous work @xcite we divide the total lensing effect into three parts : * awl ( a weak lensing ) , namely the effect due to scales @xmath73 ( where @xmath74 mpc ) at redshifts @xmath75 . this signal is dominated by strongly nonlinear scales * bwl , the lensing signal due to scales @xmath76 which corresponds to modes that are always in the linear regime down to @xmath77 * cwl , the lensing signal due to scales @xmath78 but at redshifs @xmath79 the goal of this paper is to calculate the awl signal , using ray - tracing through n - body simulations , to @xmath4 values in the range @xmath80 . since the modes associated with the bwl component are specifically chosen to correspond to scales that are linear down to @xmath77 , which is set by wavenumbers @xmath81 mpc@xmath82 @xcite , the bwl component can be calculated with the linear approach implemented in cmbfast . the cwl component involves modes in the mildly nonlinear regime which can , nonetheless , be evolved via approximation schemes @xcite . standard semi - analytical methods designed to study weak lensing in the nonlinear regime should also apply in this case , hence , the cwl effect can be calculated without resorting to n - body techniques . this calculation can be performed using the nonlinear version of cmbfast which is based on semi - analytical approaches @xcite . for these reasons we begin ray tracing within the simulation at @xmath67 and consider nonlinear spatial scales having wavenumbers @xmath78 . other methods utilizing simulations have also included the contribution to lensing from all scales above a certain redshift using special techniques @xcite . in figure [ fig0 ] we show calculations of the lu angular power spectra for the awl , bwl and cwl lensing regimes for @xmath83 using the nonlinear lensing implementation in cmbfast . note that for @xmath84 the awl and cwl effects are too small to be calculated with cmbfast due to the fact that these spectra are obtained as differences between other spectra provided by the code . both the awl and cwl effects , as calculated by cmbfast , are clearly small with the cwl effect being virtually negligible in the @xmath83 range . as anticipated , the bwl effect is dominant . the calculation of the awl effect provides an initial estimate of the signal we wish to calculate with the simulation . we have verified that , for @xmath85 , the bwl effects calculated by using both the linear and nonlinear methods implemented in cmbfast are almost identical , which proves that the bwl component is linear , as expected . an estimate of the lens deviation integral , given in eq . ( [ devi ] ) , is computed via a numerical integration performed along the background null geodesics from the comoving distance @xmath86 corresponding to @xmath67 ( in the model under consideration ) to the observer position . the gradient of the peculiar potential @xmath87 used in the integrand of eq . ( [ devi ] ) is obtained from the simulation but is not exactly the same as that used within the n - body calculation . instead , the gradient is found by subtracting the part of the full n - body potential produced by linear spatial scales larger than @xmath88 . specifically , our algorithm for determining the potential gradient , which is proportional to the force , is as follows : 1 . decide upon the direction of the normal rays representing the geodesics ( see section [ pref ] ) 2 . assuming the born approximation and using the photon step distance @xmath89 ( see section [ pref ] ) determine all the evaluation positions and times on the geodesics within the simulation volume from @xmath67 down to the final redshift 3 . associate test particles with each of these positions and times 4 . at each time - step of the n - body simulation ( while it is running ) determine which test particles require force evaluations 5 . at each test particle position evaluate the force on the test particle using the long - range fft component and short - range pp correction as in the hydra algorithm 6 . during the fft convolution for the test particles eliminate contributions from scales larger than @xmath90 mpc by removing the signal from wavenumbers satisfying @xmath81 mpc@xmath10 7 . if the evaluation time for a point on the geodesic lies between two time - steps calculate a linear interpolation of the two forces from the time - steps that straddle the correct time 8 . resolve the force into its transverse component and hence recover the transverse component of the potential gradient once all the potential gradients are evaluated we can calculate the lensing deviation integrals . we emphasize that our algorithm ensures the potential gradient along normal rays is calculated very accurately and corresponds exactly ( modulo the removal of power from scales greater than @xmath90 mpc ) to that in the simulation . we do not resort to smoothing onto grids or the creation of lensing planes `` on the fly '' . the main drawback of our method is that since lensing is now an integral part of the n - body simulation , a different preferred direction requires a new n - body simulation . to calculate the photon propagation , the cmb photons are moved through the simulation volume along specially chosen paths to avoid repeatedly sampling the same structures . this approach uses all steps from the simulation and by using the periodicity of the box volume there are no discontinuities in the matter field anywhere . in principle this approach is similar to tiling methods that are used elsewhere ( @xcite ) . these methods are based on independent pm simulations with decreasing sizes that telescope in resolution along the line of sight . in our `` tiling '' the matter field is constantly being updated by the ap3 m code as the time - step changes and , consequently , it is not limited by the box size . we also take care to ensure paths are taken which avoid , as much as possible , periodicity effects . this is notably different from other approaches using random translations and orientations ( @xcite ) of the simulation volume which , unavoidably , have discontinuities at adjoining radial shells . while the signal from such discontinuities is likely small , our method has the advantage of avoiding it completely . choosing the directions for the ray propagation is not entirely trivial . for the directions parallel to the box edges , periodicity effects are clearly very strong . photons moving along these directions would pass close to the same structures in successive boxes and , consequently , the lensing effect of the same structure would be included a number of times ( one per crossed box ) . this repetition would lead to a false magnification of lens deviations . however , as has been emphasized in @xcite and @xcite , periodicity effects are negligible along certain directions , hereafter called _ preferred directions_. in order to define preferred directions , it can be assumed that : ( i ) the @xmath91 , @xmath92 , and @xmath93 axes are parallel to the box edges ; ( ii ) the angles @xmath94 and @xmath95 are spherical coordinates defined with respect to these axes ; ( iii ) photons moving along the direction @xmath26 , cross the @xmath96 face of a box at point @xmath97 , and the next @xmath96 face at point @xmath98 ; and ( iv ) if the segment pq is projected onto the @xmath96 plane , the length of the resulting projection is @xmath99 . then , if the condition @xmath100 is satisfied , the direction @xmath101 is assumed to be a preferred one . taking into account this definition of preferred directions and the cutoff performed at the scale @xmath102 , it is clear that photons moving along preferred directions enter successive boxes through independent uncorrelated regions . moreover , for these directions and the box sizes of interest , it can be easily verified that the cmb photons can travel from @xmath67 to @xmath77 ( @xmath103 ) through different uncorrelated regions ( without repetitions ) located in successive simulation boxes . periodicity effects can , therefore , be assumed to be negligible . the total number of crossed boxes is denoted by @xmath104 . some preferred directions used in this paper are defined in table [ tab1 ] , where the first column gives the names of all these directions . each of them has been used in universes covered by simulation boxes with a given size , as detailed in the second column . the third and fourth columns show , respectively , the angles @xmath105 and @xmath106 ( spherical coordinates ) defining the corresponding direction . the distance @xmath99 defined in the previous paragraph is given in the fifth column and , finally , in the last column the number of boxes crossed by the cmb photons from redshift @xmath107 to the present is given . from the values of table [ tab1 ] , it is clear that the distance @xmath99 is much greater than @xmath108 in all four cases , which ensures that other directions close enough to the listed ones are also preferred directions . cccccc d256a & 256 & 76.76 & 11.31 & 79.98 & 22.00 + d512a & 512 & 76.76 & 11.31 & 159.95 & 11.00 + d512b & 512 & 68.43 & 18.435 & 273.20 & 10.17 + d1024a & 1024 & 59.19 & 26.57 & 853.33 & 4.43 + we proceed as follows : any direction of table [ tab1 ] is assumed to point toward the center of a squared map , whose angular size corresponds to the box size appearing in the second column ( as noted above ) . then , these squared maps are uniformly pixelised by choosing a certain number of pixels , @xmath109 , per edge . the angular resolution is then @xmath110 . the directions of all the pixels are preferred ones and , consequently , lens deviations can be calculated for each pixel with no significant periodic effects across the full map . the parameters involved in the ray - tracing procedure are thus summarized : a number of directions , @xmath111 , per edge of the squared cmb map ( one per pixel , @xmath112 ) ; an initial redshift , @xmath113 , for the calculation of lens deviations ; a step , @xmath114 , to perform the integral in eq.([devi ] ) ( hereafter called the photon step ) ; and the angles @xmath105 and @xmath106 defining the preferred direction . hereafter , a lensing simulation ( ls ) is the calculation of the @xmath50-deviations along the pixel directions , plus the construction of the @xmath40 , @xmath115 , @xmath116 maps and the estimation of @xmath46 and @xmath49 ( angular power spectra ) . an ls is characterized by the parameters and initial conditions required by the n - body simulation together with the parameters of the ray - tracing procedure . the lensing simulations ( lss ) obtained from the parameters : @xmath117 , @xmath118 , @xmath119 , @xmath120 , @xmath121 , @xmath122 ( as in all the simulations used in the paper ) , and @xmath123 , are hereafter called reference lensing simulations ( rlss ) . the angular resolution of these simulations is @xmath124 ( @xmath125 ) . there are an infinite number of possible realizations of this type of lss corresponding to different initial conditions for the n - body simulation as well as to distinct preferred directions . we also consider the effect on the lss of changing parameters to ensure that the calculation of the power spectra @xmath46 and @xmath49 is as robust and accurate as possible . to determine the impact of changing the final ray - tracing epoch , the rls_aa lensing simulation ( based on the preferred direction @xmath126 , see tables [ tab1 ] & [ tab2 ] ) , was used to estimate lensing from @xmath122 to final redshifts between @xmath127 and @xmath128 . results are shown in fig . [ fig1 ] , where the @xmath129 angular power spectra corresponding to the final redshifts 0.3 ( dotted line ) , 0.2 ( solid line ) , and 0.0 ( dashed line ) are presented . clearly , these three spectra are very similar . we find the same result for the @xmath130 spectra ( these results are omitted since the addition of five oscillating lines would lead to a confusing plot and add little information ) . based upon these results we conclude that the signal produced between redshifts @xmath131 and @xmath128 contributes negligibly to the total lensing . given this fact , and that the cpu time required to evolve between @xmath132 and @xmath77 is comparatively lengthy for our code ( due to the absence of individual particle time - steps ) , our calculations are performed between redshifts @xmath122 and @xmath133 . the same study has been done for other rlss corresponding to different initial conditions and preferred directions . the conclusions are the same in all the cases , namely a negligible effect between @xmath132 and @xmath77 . we note that this conclusion is in close agreement with that of @xcite who showed that stopping at @xmath134 produced a deficit in the low @xmath4 signal but a negligible difference for @xmath135 . we next compare lss constructed from pm and ap3 m n - body codes to examine the impact of sub - mesh - scale resolution on the lensing signal . in fig . [ fig2 ] we plot the @xmath49 ( top panel ) and @xmath46 ( bottom panel ) for rls_ba ( solid line ) along with the same spectra for pmls ( dashed line ) , obtained from a pm code . details of the pm code can be found in @xcite ; this code has been used in previous papers ( @xcite ) . we emphasize that the same ray - tracing procedure , described in section [ nbody ] , is used in both codes . the parameters of the pmls ( see table [ tab2 ] ) are as follows : @xmath136 , @xmath118 , @xmath137 , @xmath138 , and @xmath139 , and the preferred direction is d256a ( see table [ tab1 ] ) . the effective resolution @xmath140 of the pm code is two cells ( 1 @xmath141 in this case ) ; hence , in the pmls , the photon step @xmath89 must be smaller than @xmath142 to take advantage of the pmls resolution . it has been verified that @xmath143 mpc ( half of @xmath140 ) is a good value for the photon step ( smaller values lead to very similar results ) . in keeping with other n - body work ( @xcite ) , the true effective resolution of the ap3 m simulations is estimated to be @xmath144 , which means that , in our rlss , the photon step is a little smaller than a half of the effective resolution . examination of fig . [ fig2 ] shows that the pmls traces the peaks , but the amplitudes are too small and , moreover , the @xmath49 quantities quickly tend to zero as @xmath0 increases . however , in the ap3 m case , the peaks have greater amplitudes and , furthermore , a signal of a few micro - kelvin appears for high @xmath0 values . similar conclusions follow from the @xmath129 spectra of the bottom panel , which can be directly compared with @xcite , where this kind of spectrum was used . these results show that the pmls underestimates the lensing signal we are calculating . hence , high resolution n - body simulations , be they ap3 m or computed using some alternative algorithm , are necessary to estimate cmb weak lensing from strongly nonlinear structures . having shown that the results from the ap3 m simulations are distinctly different from those of the pm simulations , we must still account for variability of results due to changing the random modes in the initial conditions of the simulation . to this end we show , in the top panel of fig . [ fig3 ] , the @xmath130 angular power spectra corresponding to three distinct rlss , namely , rls_aa , rls_ba & rls_ca , each drawn from a different realization of the modes but using the same preferred direction ( @xmath126 in table [ tab1 ] ) . the results are clearly qualitatively very similar and the only noticeable quantitative differences are below @xmath145 . this is not entirely surprising since the lower the @xmath4 value the larger the sample variance @xcite . additionally , the largest deviation occurs at the point where the signal is weakest , again an unsurprising result . these results suggest that a good average @xmath130 spectrum can be achieved from just a few chosen rlss simulations , perhaps as few as two . we have verified this assertion in the bottom panel of fig . [ fig3 ] , by plotting the average of two of the rlss as compared to the average of all three rlss . both average spectra are so similar that only two rlss suffice to get a very good average @xmath130 spectrum in the @xmath4-interval under consideration . unsurprisingly , the average signals for @xmath146 are in good agreement , as would be expected if the differences in the single realizations were due to sample variance . for @xmath147 , the spectra of the different realizations are so similar that a unique rls leads to a rather good @xmath130 spectrum of the lensing effect under consideration ( @xmath148 and @xmath149 ) . we emphasize that the sample variance we have been talking about is inherent in the lensing calculation itself , rather than the cmb maps themselves . for each rls , the spectra are calculated from @xmath150 small maps having a size of @xmath151 degrees ; which is a necessary step to understand small sample variances ( small deviations from spectrum to spectrum ) . in terms of the variance produced by different preferred directions , the ergodic nature of the ray - tracing means that we might reasonably expect the same level of variance seen for the different n - body realizations with differing modes . nevertheless , choosing a different preferred direction for a given rls means that the photons move through different regions of the simulation box and cross box faces in different places and a different number of times . so the trajectories are not fully equivalent , in turn producing lensing deviations that are subtly distinct . to evaluate this difference we plot , in fig . [ fig4 ] , two @xmath130 angular power spectra for the rlss rls_aa and rls_ab which use the same initial modes , but have preferred directions d512a and d512b . the similarity of the two curves indicates that , as anticipated , results are almost independent of the chosen preferred direction , and that our ray - tracing procedure is robust . thus far we have considered ap3 m simulations with box sizes @xmath152 mpc , and hence the @xmath130 and @xmath129 spectra have been obtained from small maps of angular size @xmath153 . we next examine whether these values are too small to make robust predictions . in order to answer this question , we conducted an ls , labeled ls_lda in table [ tab2 ] , with @xmath154 and @xmath155 , which leads to a lensed map of size @xmath156 and with the same angular resolution as in the rlss . since simulating a larger volume at fixed particle resolution requires a larger softening and the corresponding photon step , the values @xmath157 , and @xmath158 were used . the preferred direction was d1024a ( see table [ tab1 ] ) . to compare to this simulation we also ran another ls , denoted ls_mab , with parameters set to mimic the ls_lda simulation but in 1/8th the volume and using 1/8th the number of particles . the @xmath130 spectra obtained for the two simulations are shown in fig . [ fig5 ] , and are extremely similar . this is despite the fact that these two simulations have differing initial conditions , different box sizes , and different numbers of crossed boxes , @xmath104 . hence we conclude that a box size of @xmath159the same as that of the rlss is large enough to get very good angular power spectra for the range of @xmath4 considered . even smaller sizes , , @xmath160 , can be used when only large enough @xmath4 values are under consideration ( see below ) . although we do not directly calculate the 2d lensing potential or convergence we can examine the power spectrum of the angle @xmath161 as in @xcite . at our final redshift of 0.2 , the angular resolution of our rlss ( 0.59 ) corresponds to a comoving separation of 100@xmath162 kpc , which is still larger than our nominal n - body resolution of 5@xmath163 kpc . since this is the effective resolution in the ray - tracing we do not need to worry about the intrinsic resolution of our maps falling below that of the n - body simulation . in fig . [ newplot ] we plot the power spectrum of the lens deviations in the interval @xmath164 . we have quantified the noise in this signal by considering box sizes of 1024@xmath162 mpc , 512@xmath162 mpc , 256@xmath162 mpc although our smallest box has an effective angular resolution that is twice as small as the other two simulations . for comparison to semi - analytic methods we also include the _ total _ lensing signal predicted by camb in this @xmath4 range . comparing to a running average constructed by binning the average signal in @xmath165 , the rms deviations relative to the overall signal are 4.1% , 5.3% and 6.2% for the three box sizes , respectively . although we do see an increase in the relative noise with reducing box size , the overall noise is much less than the underlying signal . further , the absolute value of the signals are in close agreement as well , the largest difference being @xmath166 at @xmath167 and the qualitative shape of the power spectra agree well at least for @xmath168 . for @xmath169 we see a difference of 50% in the signals , and we are not confident of convergence in the interval @xmath170 ( see sections [ conv2 ] , [ conv ] for an extensive discussion of convergence issues ) . having examined sampling issues related to box sizes and preferred directions we now address the impact of force softening and the step size in the ray - tracing . as a first investigation , we varied parameters @xmath171 and @xmath114 , in lock - step with the remaining rls parameters held fixed . a ratio @xmath172 , identical to that of the rls simulations has been assumed in all the cases , thus maintaining a constant ratio between the photon step and the effective resolution ( @xmath173 photon steps per effective resolution interval ) . we considered the following new pairs of @xmath171 and @xmath114 : ( i ) @xmath157 and @xmath174 and , ( ii ) @xmath175 and @xmath176 , and call the lss associated with these parameters ls_@xmath60s1aa and ls_@xmath60s2aa respectively . the @xmath49 for these lss , along with rls_ba for comparison , are given in fig . [ fig6 ] . up to @xmath177 the three spectra are very similar , although for @xmath178 we see a small separation in the results . for @xmath179 , the trend with decreasing softening and photon - step appears to be systematic , a smaller softening leading to smaller @xmath49 coefficients . nevertheless , a smaller softening does not necessarily mean a more realistic simulation and , consequently , we can not decide which of the three lines of fig . [ fig6 ] is closer to the true spectrum . regardless of this issue , any of these lines gives a good estimate of the required spectrum for @xmath180 . to address the role of the photon step alone , we next held fixed all the rls parameters except the photon step , @xmath89 . along with our original choice of @xmath181 , two new values were considered : @xmath182 ( @xmath183 , corresponding to lss ls_@xmath601aa , ls_@xmath601ab , ls_@xmath601ba , ls_@xmath601bb ) , and @xmath184 ( a little smaller than @xmath185 , corresponding to lss ls_@xmath602aa , ls_@xmath602ab , ls_@xmath602ba , ls_@xmath602bb ) . for each of the @xmath89 two distinct preferred directions , namely @xmath126 and @xmath186 , were considered along with two distinct sets of initial modes , giving rise to four lss . the four spectra for each @xmath89 value were then averaged and the resulting spectra plotted in fig . note , that since the error in the lensing deviation integrals is reduced with decreasing @xmath89 , we can reasonably argue that our best estimate of the lensing signal is given by the minimum @xmath89 . we see that in all three cases the @xmath49 quantities are in close agreement up to @xmath177 . in the @xmath4-interval [ 7,00010,000 ] , the separations between the three curves increase , with a systematic trend of a lower signal with decreasing photon - step . since all these spectra have the same simulation softening it is reasonable to conclude that the trend we observed when comparing different @xmath187 and @xmath89 may actually be a function of the change in @xmath89 . given that the smallest @xmath89 exhibits the lowest signal , we thus conclude that our best estimate of the signal actually has the smallest correlations for @xmath188 values . nonetheless , for @xmath189 , the rls closely follows our best estimate , hence for @xmath189 we are very confident that the rls leads to a robust estimate of the lensing effect we are studying . even if the @xmath190 value leads to the most accurate spectrum ( which is not certain ) , the computational cost of constructing these lss is significantly greater than that of the rlss , while only providing a mild improvement in accuracy for @xmath178 . we have also attempted to reproduce results similar to that found elsewhere using interpolation methods . we ran another simulation using rls_aa parameters , except that this time the deflections were calculated using the 8 nearest neighbours . this keeps the mass resolution of the simulation fixed and does not alter the gravitational calculation , while at the same time lowering the effective resolution of the lensing methodology . the resulting power spectrum is shown in fig . [ fig10_new ] and shows a decaying signal at high @xmath4 which is similar to that found in earlier work ( @xcite ) . this shows that as we degrade the resolution of ray - tracing method we do indeed recover previous results . since our ray - tracing method samples the local gravitational field on scales down to a few tens of kpc , it is important to investigate the role of mass resolution on our simulation . while we have investigated box size earlier , which at a fixed resolution leads to a change in mass resolution , in this section we focus on changing the mass resolution in a fixed volume , and also include one additional simulation in a smaller volume at our maximum mass resolution . we ran additional simulations with @xmath62 and @xmath59 particles in the rls box size , namely @xmath191 mpc . these simulations are labeled ls_lha and ls_mga in table [ tab2 ] . notable other parameters were @xmath192 kpc , @xmath193 kpc and particle masses were @xmath194 respectively . in fig . [ fig8new ] we plot the power spectrum results for these simulations , along with an rls , another @xmath56 simulation with a shorter photon step ( ls_1ab ) , and one final simulation labeled ls_hea that uses a volume 1/8th the size of an rls but with the same number of particles . the parameters of this ls were @xmath195 mpc , @xmath196 kpc and @xmath197 kpc . note the two @xmath56 simulations have the same mass resolution but they bracket a range of photon - step values , a full estimate of the uncertainty in the signal at this mass resolution is given in figure 13 , where the spectra of these two cases appear inside the band of uncertainty . due to the different box size of the ls_hea run we could not maintain the same modes across all runs , and hence we used different initial modes at different resolutions . taken together , these simulations span a ratio in mass resolution of 512 . the plot shows that when pp corrections are included there is a resolution - dependent up - turn in the power spectrum for @xmath198 . improving the mass resolution reduces the impact of n - body discreteness on the lensing ( @xcite ) to the point where the highest resolution simulation ( ls_hea ) appears to be free of this contamination for @xmath199 . turning to lower @xmath4 values , the peak of the power spectrum around @xmath200 appears to be converged , modulo the variation due to the initial modes , for resolutions of @xmath56 and greater . the results for @xmath201 also appear to show errors consistent with the variation in initial modes observed in section [ modes ] , indicating results are converged in this range for all resolutions . to address the question of the correct form of the spectrum at @xmath202 we utilize the rlss and the ls_hea simulation . the effective resolution @xmath203 of the ls_hea simulation is half of that attained in the rlss and thus small scales should be represented more accurately . as a result of the reduced box size , the angular size of the lensed maps is @xmath204 and the angular resolution limit @xmath205 ( @xmath206 ) is half that of the rlss ( higher resolution ) . the resulting spectrum is plotted in fig . [ fig8 ] as a dotted line , and compared to our previous best estimate for the signal at @xmath202 , namely that from ls_@xmath601ab that was constructed with @xmath207 and the remaining parameters identical to those of the rlss ( the preferred direction is d512b ) . as expected , given that the new spectrum is obtained from smaller @xmath208 maps , some uncertainties in the @xmath49 quantities are observed for @xmath209 . note that for the @xmath210 maps used in the rls we expect uncertainties for @xmath211 . however , for @xmath212 , the spectrum of the smaller box ( dotted line ) is expected to be more accurate than our previous ls . although we again observe a systematic trend of better resolution producing a lower signal at high @xmath178 , the results are nonetheless in good quantitative agreement for @xmath213 . we can hence be confident that our rlss give also a rough but useful estimate of the angular power spectrum for @xmath213 ( supporting our assertion in the previous section ) . finally , for @xmath214 , the smaller box produces a decrease , whereas the previous ls increases slightly . this unfortunately suggests that the n - body simulations and the cmb lensed maps used to derive the rlss are not good enough for @xmath178 . we hence conclude that future simulations with higher resolutions are clearly necessary to go beyond @xmath215 . to bring together all that we have learned thus far , in fig . [ fig9 ] we plot all the @xmath130 angular power spectra from our simulations ( excepting simulation ls_lha with @xmath216 particles ) in the @xmath217-interval ( 3.5 , 4.0 ) , which we reemphasize corresponds to the awl effect described earlier ( scales smaller than @xmath218 at @xmath219 ) . the @xmath220 and the complementary lensing effect @xmath221 estimated with cmbfast are also included in the same plot . since we demonstrated the @xmath222 component is negligible in the interval under consideration ( see the bottom left panel of fig [ fig0 ] ) , it is not represented . the lower solid line displays the same spectrum as the dotted line of fig . [ fig8 ] , which corresponds to the ls_hea ls , which has the smallest value of @xmath171 and the highest mass resolution considered ( eight times better than that of the rlss ) . the upper solid line shows the same spectrum as the dashed line of fig . [ fig6 ] , which corresponds to the ls_@xmath60s2aa ls , which has the same mass resolution as the rlss but an @xmath54 value 3 times larger ( and 6 times larger than that used in the ls_hea ls ) . despite the large differences in the numerical parameters defining the n - body simulations , we see that from @xmath223 to @xmath224 , all the spectra lie in a narrow band ( bounded by the solid lines of fig . [ fig9 ] ) , having a width close to half a micro - kelvin . in addition to the simulation lines ( solid and dotted ) we have plotted estimates of the lensing signal based upon semi - analytical approaches implemented in the cmbfast code @xcite . by using the nonlinear version of this code with a minimum scale of @xmath225 ( as in our ls , leading to the lower solid line of fig . [ fig9 ] ) , we have estimated the awl ( dash - dots line in fig . [ fig9 ] and top left panel of fig . [ fig0 ] ) and the bwl components ( dashed line in fig . [ fig9 ] and top right panel of fig . [ fig0 ] ) of the weak lensing . finally , the cmbfast code has been also used to calculate the cmb angular spectrum , in the absence of lensing , for the model described in section [ maps ] ( dash - three - dots line ) . the complementary effect @xmath221 , not calculated within the simulation , is due to scales greater than @xmath108 . while the absence of this power has significant consequences for small @xmath4 values we will now show ( with cmbfast ) that in the interval we are examining it is small . if the scale @xmath108 is placed on the last scattering surface at ( @xmath226 ) , it subtends an angle of @xmath227 . however , in fig . [ fig9 ] we are plotting @xmath4 values corresponding to angles of @xmath228 and @xmath229 , considerably smaller than the scale set by @xmath230 on the last scattering surface . by this reason , the power of the dashed line , which is due to the linear scales greater than @xmath102 , is expected to be small in the @xmath217-interval of fig . cmbfast estimates have confirmed this smallness ( dashed line of fig . [ fig9 ] ) . nevertheless , the @xmath221 effect is very important for much smaller @xmath4 values ( see bottom right panel of fig . [ fig0 ] ) . the complementary @xmath222 effect ( see the bottom left panel of fig . [ fig0 ] ) is too small to be significant in the @xmath217-interval ( 3.5 , 4.0 ) . the @xmath220 effect obtained with cmbfast ( nonlinear version ) is shown in the dot - dashed line of fig . this effect appears to be smaller than the @xmath220 effect numerically estimated here ( with ap3 m simulations ) , whose power is found to be close to @xmath231 for @xmath232 . the total lensing would be obtained by adding this power and that of the dashed line ( @xmath221 component ) . we thus see that the total lensing effect has a power of a few @xmath233 . to compare this result with other simulation work , we turn to the work by @xcite . since our map - making method does not allow us to plot the lensed and unlensed @xmath234 we have converted their results ( specifically those in their figure 4 ) to the @xmath235^{1/2}$ ] values we use in our paper . in performing this conversion we have multiplied their power spectra by a factor of @xmath236 @xmath237k@xmath238 which includes the effective fractional sky coverage , @xmath239 . we hence find values of 1.1 @xmath237k at @xmath240 , 0.7 @xmath237k at @xmath241 , 0.6 @xmath237k at @xmath242 and 0.4 @xmath237k at @xmath243 . this compares with our own signal of @xmath244 @xmath237k across the entire range . however , we caution against over - interpreting this comparison . our cosmology and assumptions about reionization and backgrounds are slightly different from those in @xcite , however it is clear that we find a higher signal at @xmath245 when short range corrections are included . as it is apparent from fig . [ fig9 ] , the @xmath220 effect clearly dominates the primary anisotropies for @xmath246 ( the continuous and dotted lines are well above the dash - three - dots one ) . however , the sunyaev - zeldovich ( sz ) effect is also important at these scales , and foregrounds must be properly accounted for . the berkeley - illinois - maryland association ( bima ) , has reported @xcite a temperature power ( @xmath2^{1/2}$ ] ) of @xmath247 at 68 % confidence , for @xmath248 . in order to explain @xmath249 as an sz effect , one must have @xmath250 @xcite , but this is in comparatively strong disagreement with data from wmap ( see section [ sec1 ] ) that lead to @xmath251 . hence , taking into account that the sz temperature power scales as @xmath252 @xcite , the sz effect could only explain around the 44.5 % of the most likely bima value ( @xmath253 ) . the cosmic background imager ( cbi ) observations , @xcite , cover the multipole range @xmath254 and , in a recent analysis @xcite , it is argued that , if the cbi excess power observed at very small angular scales is explained as a sz effect for a certain @xmath255 value , then a power close to @xmath256 , marginally compatible with bima measurements at @xmath248 , may be explained with the same value of @xmath255 . the level of power we predict for the weak lensing ( @xmath220 plus @xmath221 components ) may thus be a missing link , simultaneously ensuring the compatibility of wmap , cbi and bima observations . we have predicted a lensing effect of a few micro - kelvin for these large angular scales , and hence a smaller sz component ( a smaller value of @xmath255 ) could lead to simultaneous compatibility with wmap , cbi , and bima observations . aside from the large errors bars on the bima measurement , we are also cautious about making firm claims since lensing and sz effects are essentially produced by the same structure distributions ( galaxy clusters and sub - structures involving dark matter and baryons ) and , consequently , these effects must be strongly correlated . this implies that the spectra of these two effects must be superposed in an unknown way ( not merely added ) . in practice , this superposition might be analyzed in detail with ray - tracing through hydrodynamical simulations ( including both baryons and dark matter ) . more work along these lines is clearly necessary . our first conclusion , in agreement with other work , is that pm simulations are inefficient for calculating cmb lensing due to strongly nonlinear structures ( see fig . [ fig2 ] and comments in section [ comparison ] ) . while it is not impossible to simulate these effects with a pm code the required resolution is so large that the additional short - scale resolution provided by ap3 m ( or other high resolution n - body technique ) is far more efficient at capturing the lensing effect . hence our development of a combined parallel ap3m - ray - tracing code is a necessary step to estimate the lensing signal at high @xmath4 . only the @xmath220 effect ( see section [ nbody ] ) produced by scales smaller than @xmath257 between redshifts @xmath258 and @xmath259 has been estimated with ap3 m simulations . this choice is appropriate for the following reasons : ( 1 ) all the strongly nonlinear scales are taken into account ; ( 2 ) omitting scales greater than @xmath257 makes our ray - tracing procedure efficient ( see section [ nbody ] ) ; ( 3 ) the effect produced by scales greater than @xmath260 ( @xmath221 in section [ nbody ] ) can be studied by using the linear version of cmbfast ; and , ( 4 ) the lensing due to scales smaller than @xmath257 , at redshift @xmath261 ( @xmath222 in section [ nbody ] ) , can be computed using standard semi - analytical methods implemented in cmbfast . given that our study has strongly focused on the @xmath220 signal , we exhaustively investigated the numerical issues involved when estimating this signal . our main conclusions can be summarized as follows : * the lensing contribution between @xmath132 and @xmath77 is negligible * for each rls , a few realizations suffice to get a good average @xmath49 spectrum and , moreover , each single simulation gives a very good spectrum rather similar to the average one * rlss , which are essentially identical except for the preferred directions , give similar spectra ( the ray - tracing procedure has little variance ) * simulation boxes of @xmath262 are not fully necessary for @xmath263 ; however , for @xmath211 , these large sizes lead to the most reliable spectra and , moreover , such sizes should lead to very good spectra in the @xmath4-interval ( @xmath264@xmath265 ) * simulations in boxes of @xmath159 lead to good @xmath49 spectra for @xmath266 * for @xmath267 , all the rlss lie in a region of width @xmath2680.5 @xmath237k , indicating that the rlss give consistent estimates of the signal in this range * the signal in the range @xmath269 is @xmath270k , which is @xmath271 @xmath237k higher than that found elsewhere ( @xcite ) despite some simulation uncertainties , our code and technique have lead to a robust estimate of the lensing effect in the @xmath4-interval @xmath272 , where it clearly dominates the primary anisotropy . moreover , the estimated power is larger than that obtained with semi - analytical methods and the cmbfast code . we thus suggest that the resulting value of a few micro - kelvin may explain the excess power at high @xmath4 in the bima and cbi observations . this conclusion is supported by recent studies based on the millennium simulation @xcite , where the authors have reported a small contribution from nonlinearity at @xmath273 . however , the methods of @xcite have been designed to build all - sky lensed maps , and do not have the resolution necessary to perform an accurate estimate of the weak lensing by strongly nonlinear structures in the @xmath4-interval where we have found our main effect . our direct estimation of the potential gradients at photon positions using pp corrections from the local dark matter particles appears to be the main origin of the difference between our results and other research relying either on planes or grid interpolations . however , we emphasize that differences only occur at the large @xmath274 values we have been investigating . our method employs extremely fine time resolution , namely that used by ap3 m simulations and also a very good angular resolution ( see above ) . due to current limitations in our method it is not possible for us to determine the precise role of temporal resolution in an accurate lensing calculation . it is well known that on small scales baryons do not follow the dark matter distribution . thus , while we have attempted to be as accurate as possible in this dark matter simulation , we are now probing scales where contributions from baryons are beginning to become significant . an investigation of the impact of baryons by @xcite considered two types of simulations , the first with non - radiative baryons while the second included dissipation and star formation . they found that for @xmath275 an effect of between 1% to 10% on the weak - lensing shear angular power spectrum , as calculated from dark matter alone , was possible . the largest difference was produced in the run with dissipation and star formation , where a 10% increase in the ( shear ) @xmath234s at @xmath169 was observed . future work is definitely necessary to determine the impact of this physics on lensing statistics . we plan to conduct simulations with baryons and feedback processes both to identify its impact on the signal we find , and also to systematically evaluate the combined impact of the sz effect and weak lensing . for calculations that are accurate to high @xmath4 it seems to be necessary to move cmb photons through the simulation box while structures are evolving , which ensures the spatial gradients are accurately calculated on the photon positions ( test particles ) . once the n - body code has been modified to compute spatial gradients at particle test positions , there are no theoretical or technical reasons to take a temporal resolution different from that defined by the simulation time step . from the theoretical point of view , the time step resolution is obviously most compatible with the n - body technique ( and thus takes into account the entire evolution of the simulation ) . from the technical point of view , the use of the time step resolution requires the minimum memory cost as the number of test particles between two successive times , although large , is minimized . because of these considerations , temporal resolution is not a parameter to be varied in our calculations . our ap3 m code adapted to cmb lensing calculations can be run for different values of the parameters defining the lss ; hence , this code allows us to see how the resulting angular power spectra depend on the parameters defining both the n - body simulation and the ray - tracing procedure . we thank the anonymous referee for suggestions that improved the clarity and content of the paper . fullana , j.v . arnau and d. sez acknowledge the financial support of the spanish ministerio de educacin y ciencia , mec - feder project fis2006 - 06062 . rjt and hmpc acknowledge funding by individual discovery grants from nserc . rjt is also supported by grants from the canada foundation for innovation and the canada research chairs program . hmpc acknowledges the support of the canadian institute for advanced research . simulations were performed at the _ computational astrophysics laboratory _ at saint mary s university . antn l. , cerd - durn , p. , quilis , v. , and sez , d. 2005 , , 628 , 1 arnau j.v . , aliaga , a. , and sez , d. 2002 , , 382 , 1138 bond , j.r . , 2005 , , 626 , 12 burigana , c. , and sez , d. 2003 , , 409 , 423 carbone , c. , springel , v. , baccigalupi , c. , bartelmann , m. , & matarrese , s. , 2008 , mnras , 388 , 1618 carbone , c. , baccigalupi , c. , bartelmann , m. , matarrese , s. , and springel , v. , 2009 , mnras , 396 , 668 cerd - durn , p. , quilis , v. , and sez , d. 2004 , , 69 , 043002 cooray , a. , and melchiorri , a. 2002 , , 66 , 083001 couchman , h. m. p. , 1991 , , 368 , 23 das , s. , & bode , p. , 2008 , apj , 682 , 1 dawson , k.s . , holzapfel , w.l . , carlstrom , j.e . , joy , m. , and laroque , s.j . , 2006 , , 647 , 13 efstathiou , g. , davis , m. , white , s. d. m. & frenk , c. , 1985 , apjs , 57 , 241 fosalba , p. , , 2008 , mnras , 391 , 435 grski , k.m . , hivon , e. , and wandelt , b.d . 1999 , in proc . mpa / eso cosmology conf . , banday , r.k . sheth and l. n. da costa ( garching : european southern observatory ) ( astro - ph/9812350 ) hamana , t. , , 2001 , a&a , 367 , 18 harrison , e. r. , 1970 , phys . d. , 1 , 2726 hernquist , l. & katz , n. , 1989 , apjs , 70 , 419 hilbert , s. , , 2009 , a&a , 499 , 31 hinshaw , g. , et al . , arxiv:0803.0732 hockney r.w . , and eastwood j.w . 1988 , computer simulations using particles ( bristol : iop ) jain , b. , seljak , u. , and white , s. 2000 , , 530 , 547 jing , y. p. , zhang , p. , lin , w. p. , gao , l. , and springel , v. , 2006 , , 640 , l119 knox , l. 1995 , , 52 , 4307 kim , j. , , 2008 , arxiv:0812.1392 komatsu , e. , & seljak , u. , 336 , 1256 komatsu , e. , et al . , arxiv:0803.0547 lewis , a. , and challinor , a. 2006 , , 429 , 1 moore , b. , governato , f. , quinn , t. , stadel , j. , & lake , g. , 1998 , , 499 , l5 moutarde , f. , alimi , j .- m . , bouchet , f.r . , pellat , r. , and ramani , a. 1991 , , 382 , 377 perlmutter , s. , 1999 , apj , 517 , 565 puchades , n. , fullana , m.j . arnau , j.v . and sez , d. 2006 , , 370 , 1849 quilis , v. , ibez , j.m . , and sez , d. 1998 , , 502 , 518 riess 1998 , aj , 116 , 1009 sez , d. , holtmann , e. , and smoot , g.f . 1996 , , 473 , 1 sez , d. , puchades , n. , arnau , j.v . and fullana , m.j . 2006 , proceedings of science , 058 sandarin , s.f . , and zeldovich , ya . b. 1989 , rev . phys . , 61,185 sato , m. , , 2009 , arxiv:0906.2237 scott , d. , srednicki , m. and white , m. 1994 , , 421 , l5 seljak , u. , and zaldarriaga , m. 1996 , , 469 , 437 seljak , u. 1996 , , 463 , 1 shapiro , c. , & cooray , a. , 2006 , jcap , 603 , 7 sievers , j.l . , , 2009 , arxiv:0901.4540 smith , r.e , 2003 , mnras , 341 , 1311 spergel , d. 2007 , apjs , 170 , 377 springel , v. , , 2005 , nature , 435 , 629 teyssier , r. , , 2009 , a&a , 497 , 335 thacker , r. j. , & couchman , h. m. p. , 2006 comm . , 174 , 540 van waerbeke , l. , , 2001 , mnras , 322 , 918 vianna , p. t. p. & liddle , a. 1996 , mnras , 281 , 323 white , m. , and hu , w. 2000 , , 537 , 1 zeldovich , ya . b. 1970 , , 5 , 84

flat directions in generic supersymmetric theories can change the thermal history of the universe . a novel scenario was proposed earlier where the vacuum expectation value of the flat directions induces large masses for all the gauge bosons and gauginos . this delays the thermalization of the universe after inflation and solves the gravitino problem . in this article we perform a detailed calculation of the above scenario . we include the appropriate initial state particle distribution functions , consider the conditions for the feasibility of the non - thermal scenario , and investigate phase space suppression of gravitino production in the context of heavy gauge bosons and gauginos in the final state . we find that the total gravitino abundance generated is consistent with cosmological constraints . physics of the early universe , supersymmetry and cosmology

1205.5408

large vacuum expectation values ( vevs ) for flat directions in supersymmetric ( susy ) theories , i.e. , directions in the field space of scalars that have a flat potential , can affect the thermal history of the universe in two ways as discussed in the literature . firstly , they may delay the thermalization of the inflaton decay products @xcite . secondly , the flat directions could potentially dominate over the inflaton decay products @xcite , such that reheating is associated with the decay of flat directions as opposed to the decay of the inflaton . below we shall consider gravitino production in the context of flat directions delaying thermalization of the decay products of the inflaton which are presumed to dominate the universe . gauge symmetries are broken along flat directions and for certain flat directions all standard model ( sm ) gauge symmetries are broken in the early universe . ( low energy phenomenology is not affected as the vevs of these flat directions ultimately disappear . ) in refs . @xcite the authors have presented an interesting scenario in which flat directions give a large mass @xmath0 to all gauge bosons and gauginos where @xmath1 is the vev of the field @xmath2 that parametrizes the flat direction and @xmath3 for gauge coupling constant @xmath4 . since processes involving gauge bosons are important for thermalization , this results in a slow rate of thermalization after the end of inflation . thus the universe enters a period of pre - thermalization after inflation . throughout this article we use the term thermalization to refer to chemical equilibration , and thermal equilibrium to refer to chemical equilibrium . in the scenario of refs . @xcite the universe reaches a state of kinetic equilibrium via @xmath5 interactions after a certain time interval has elapsed following inflaton decay . further thermal equilibrium is reached only when number violating interactions , i.e. , @xmath6 interactions , are also effective . thus the rate of inflaton decay , of kinetic equilibration and of thermal equilibration follow the relation @xmath7 where @xmath8 and we define @xmath9 , @xmath10 and @xmath11 as the times of inflaton decay , kinetic equilibration and chemical equilibration respectively . in refs . @xcite the authors argue that due to the delay in thermalization the generic problem of gravitino overproduction in susy models is also avoided . the lack of thermalization leads to a dilute plasma of high energy particles , and the low number density of particles effectively leads to low gravitino production . moreover due to a low final reheat temperature after thermalization gravitino production after thermalization is also suppressed . in this article we provide a detailed analysis of gravitino production in the non - thermal universe scenario of refs . @xcite . to find the number density of gravitinos @xmath12 , refs . @xcite use the integrated boltzmann equation of the form @xmath13 from inflaton decay till thermal equilibration for the process @xmath14 , where @xmath15 and @xmath16 are quarks or squarks and @xmath17 is a gauge boson or a gaugino . we too obtain and use this equation , but only for @xmath18 where one can describe the particle distribution function by a thermal distribution with an effective chemical potential . for @xmath19 we instead use a delta function distribution function for the incoming particles and then evaluate the collision integral for the integrated boltzman equation . as part of our analysis , we have also explicitly obtained the conditions for obtaining a non - thermal universe after inflaton decay . there is an important effect associated with large vevs of the flat directions as relevant for gravitino production that has not been considered earlier . gauge bosons and gauginos are produced along with gravitinos in the relevant processes for gravitino production . as mentioned above these get large masses @xmath20 due to the large vevs of the flat directions . consequently gravitino production ceases in cases where the energy of quarks and squarks in the universe falls below the mass of gauge bosons and gauginos due to phase space suppression . as in ref . @xcite we consider two cases , @xmath21 and @xmath22 , where @xmath23 is the condensate mass . we find that for @xmath21 phase space suppression shuts off gravitino production both before thermalization and even after thermalization , till the condensate decays . for @xmath22 , gravitino production is shut off after thermalization due to phase space suppression till the condensate decays . it should be noted here that the longevity of these flat directions has been in dispute @xcite . one of the key issues is whether or not non - perturbative effects lead to a fast decay of the condensate . in ref . @xcite it was pointed out that non - perturbative decay due to parametric resonance is suppressed for a complex scalar condensate if the real and imaginary parts of the scalar field oscillate out of phase . in that case the field does not pass through the minimum of the potential at @xmath24 and so the required non - adiabatic condition @xmath25 is not satisfied , where @xmath26 is the the frequency of the @xmath27-th mode of the @xmath28 field produced by the decay of the condensate . ( backreaction effects can also be important in suppressing non - perturbative decay . ) the condensate then decays only perturbatively and hence lasts for a long time . @xcite however argued that in the case of multiple flat directions there is a mixing of excitations along different directions which makes non - perturbative effects important . further investigations in ref . @xcite implied that even with multiple flat directions resonant decay is most likely to be suppressed ( in cases of physical interest ) because the multiple flat directions effectively reduce to one flat direction or a collection of independent single flat directions . raising certain issues related to gauge fixing , refs . @xcite worked in the unitary gauge to eliminate the nambu - goldstone boson that appears in the context of susy flat directions charged under the gauge group of the mssm and studied the circumstances under which the presence of multiple flat directions can lead to non - perturbative decay . ref . @xcite invoked charge conservation to argue that non - perturbative decay is suppressed . at the same time ref . @xcite analysed the issue with proper gauge fixing and concluded that with one flat direction there is no resonant particle production , in agreement with ref . @xcite , while disagreeing with the conclusion of ref . @xcite in the context of multiple flat directions . however refs . @xcite argue that even if non - perturbative decay occurs for multiple flat directions it leads to redistribution of energy of the condensate amongst the fields in the d flat superspace and hence to practically the same cosmological consequences ( including large gauge boson and gaugino masses and delayed thermalization ) as in the scenario with only perturbative decay . our analysis below presumes that perturbative decay is the relevant decay mechanism of the flat direction condensate and that the impact of the condensate lasts long enough to have cosmological consequences . this may be because either there is only a single flat direction that is excited , or because the multiple flat directions are excited such that parametric resonance is still suppressed , or because of the arguments of refs . @xcite , or because of backreaction effects . there are other effects that affect the evolution and decay of the flat direction condensate . thermal contributions to the condensate potential can change the mass of the condensate leading to an earlier oscillation time for the condensate @xcite . below we take the condensate mass to be of order of the susy breaking scale @xmath29 . we present our results in terms of @xmath23 which can be replaced by larger values to include these effects . @xcite obtains masses of @xmath30 and @xmath31 for @xmath32 where terms that lift the flat directions have the form @xmath33 where @xmath34 . ) another effect is that the condensate can decay due to scattering by inflaton decay products @xcite . both these effects have been discussed in appendix a.4 of ref . @xcite where it has been argued that these effects will not be relevant during the non - thermal phase . the flat direction condensate can also decay to q - balls due to inhomogeneities in the condensate field . this has been studied in the context of both gauge mediated and gravity mediated susy breaking @xcite . in addition it has been seen that condensate lumps with different properties than q - balls called q - axitons may also be formed @xcite . note however that the time scale for the formation of q - balls and q - axitons can be large . for gauge mediated susy breaking scenarios the time scale for q - ball formation @xmath35 @xcite , where @xmath36 is a characteristic mass scale for the flat direction condensate , and for gravity mediated susy breaking scenarios @xmath37 @xcite or @xmath38 @xcite . @xcite also obtains a q - ball formation time which is larger than the flat direction oscillation time by a factor of @xmath39 . the time scale for q - axiton formation is @xmath40 @xcite . q - ball formation may be avoided for certain flat directions , including those with a large stop admixture , which allow for @xmath41 , where @xmath17 is a parameter in the potential in gauge mediated susy breaking scenarios @xcite . our analysis does not include q - ball or q - axiton formation . we now present a summary of this article . in section [ sec : boltzmann ] we provide the integrated boltzmann equation and phase space distribution functions valid for a non - thermal universe . section [ sec : revproc ] provides the scattering processes relevant for gravitino production in the scenario with heavy gauge bosons and gauginos . in section [ sec : discuss ] we obtain the relevant time scales , namely , @xmath10 , @xmath11 , @xmath42 and @xmath43 where @xmath42 is the time of decay of the flat direction condensate and @xmath43 is the time in the radiation dominated era after thermalization when gravitino production could commence . in this section we also discuss the feasibility of gravitino production in the context of phase space suppression in different epochs . the cross section for gravitino production in the center - of - mass frame is obtained in section [ sec : sigmacm ] . then in section [ sec : gtabund ] we obtain the gravitino abundance generated in different time intervals in the universe . the parameters required to obtain the gravitino abundance are given in section [ sec : params ] . our results are given in section [ sec : results ] and our conclusions in section [ sec : conclusions ] . the appendices contain derivations of certain expressions used in the text . to obtain the number density of a species @xmath44 participating in reactions @xmath45 the integrated boltzmann equation is @xmath46 \label{eq : bltz2t2 } \end{aligned}\ ] ] where @xmath47 are phase space distribution functions and @xmath48 @xmath49 is the number of internal degrees of freedom of species @xmath50 . the @xmath51 includes an average over initial and final internal degrees of freedom @xcite . typically @xmath52 and @xmath53 are assumed to be in full thermal equilibrium with zero chemical potential and thus the phase space distribution function has the form @xmath54 this is because @xmath52 and @xmath53 usually have other stronger interactions than their interaction with @xmath55 . this assumption is not valid in the scenario considered in refs . @xcite as the universe is not in thermal equilibrium before @xmath56 . in fact @xmath57 have the following form [ eq : fis ] @xmath58 & = \exp \left(-\frac{e-\xi_i}{t } \right ) & \qquad & \textrm { for } { t_{kin } } < t < { t_{thr}}\label{eq : fi2 } \end{aligned}\ ] ] where , @xmath59 as derived in [ sec : phspace ] . in the above expressions @xmath60 is the mass of the inflaton , @xmath61 is the energy density of inflaton @xmath2 at time @xmath62 , @xmath63 is the scale factor and @xmath64 is the effective chemical potential . reflects the scenario that before @xmath10 the relativistic energy of fermions and sfermions ( whose interactions produce gravitinos ) is their energy at the time of their production from ( 2 body ) inflaton decay at @xmath62 scaled by @xmath65 . between @xmath10 and @xmath11 one does not have full thermalization as represented by @xmath64 in eq . . ( also see footnote 24 of ref . @xcite . ) when the number density of @xmath44 is small , as in our case where @xmath44 represents the gravitino , we can ignore the product @xmath66 in eq . . thus , eq . can be written as @xmath67 where we have defined a variable @xmath68 to denote the collision integral on the right - hand - side of the above equation . here , @xmath69 is dimensionless and lorentz invariant , and dependent only on the mandelstam variable @xmath70 . @xmath71 is given by @xmath72 which can also be written as @xmath73 where @xmath74 is @xmath75^{1/2}/(e_1 e_2)$ ] . for the completeness of the paper we provide below some details on how to obtain the collision integral . it is easier to calculate @xmath71 of eq . in the centre - of - mass ( cm ) frame and it can then be expressed as @xcite @xmath76 where @xmath77^{1/2 } \left [ s- ( m_1-m_2)^2 \right]^{1/2}}{2\sqrt{s } } \label{eq : p12}\ ] ] is the magnitude of the momentum of particle @xmath52 ( or @xmath53 ) in the center - of - mass frame of the particle pair @xmath78 . the cross section in the cm frame @xmath79 is @xmath80 where @xmath81^{1/2 } \label{eq : diffscatt } \end{aligned}\ ] ] is the differential scattering cross section in the cm frame @xcite . note that the averaging over the final state degrees of freedom in @xmath51 cancels the prefactor @xmath82 in the above equation . @xmath83 in eq . can expressed in terms of @xmath70 and @xmath84 , the angle between @xmath52 and @xmath44 in the cm frame , using @xmath85 and @xmath86 , with @xmath87 and @xmath88 @xcite . to perform the integration over @xmath89 in ( [ eq : mbltzeq ] ) we note that @xmath90 and that the volume element @xmath91 is given by @xcite , @xmath92 where @xmath93 is the angle between @xmath52 and @xmath53 . ( note that we now work in the rest frame of the universe . ) in order to simplify the integration we replace the variable @xmath94 by @xmath70 via the relation @xmath95 . the lower limit of the integral over @xmath70 should not be less than either @xmath96 or @xmath97 as dictated by the phase space factor in eq . . therefore @xmath98 where @xmath99 $ ] and @xmath100 . for @xmath101 , @xmath102 . using eq . in eq . we get @xmath103 thus , after calculating @xmath51 for each process mentioned in the next section , we follow the procedure given above to find the parameter @xmath69 and then with the @xmath47 as given in eq . we use the above equation to calculate the number density of gravitinos produced between @xmath9 and @xmath104 , and @xmath104 and @xmath56 . the production of gravitinos after inflation has been discussed in @xcite . the processes involving strong interactions denoted by a to j are @xcite * a : @xmath105 * b : @xmath106 * c : @xmath107 , @xmath108 * d : @xmath109 , @xmath110 * e : @xmath111 , @xmath112 * f : @xmath113 * g : @xmath114 , @xmath115 * h : @xmath116 , @xmath117 * i : @xmath118 * j : @xmath119 the above can be generalized for all gauge groups . we shall consider @xmath120 only . in the non - thermal scenario we only have to consider processes involving the scattering of fermions and/or sfermions the reason is as follows : the gauge bosons and gauginos have large masses induced by the flat direction vev . due to their large masses they decay quickly to lighter fermions and sfermions and therefore do not participate in the production of gravitinos . the processes not involving gauge bosons and gauginos as incoming particles are the e , i and j processes . these processes are shown in figures [ fig : e - diags ] , [ fig : i - diags ] and [ fig : j - diags ] . ( the second e process is not shown and its contribution will be the same as that of the first process . ) [ cols="^,^,^ " , ] in ref . @xcite it has been argued that the gravitino production cross section should be proportional to @xmath121 where @xmath122 is the reduced planck mass ( and @xmath123 in the expression in ref . @xcite has been replaced by the supersymmetry breaking scale @xmath23 ) . the factor @xmath124 is related to the contribution of the @xmath125 helicity states of the gravitino while the term proportional to @xmath126 represents the contribution of the helicity @xmath127 states . to obtain the contribution of the @xmath128 helicity states , for @xmath129 we may express the gravitino field as @xmath130 where @xmath131 represents a spin @xmath132 fermionic field . cross sections for spin @xmath132 gravitino production can then be obtained from the corresponding effective lagrangian for @xmath131 @xcite . the total cross section is then obtained in ref . @xcite using eq . in this section we obtain @xmath10 and @xmath11 for the cases @xmath133 and @xmath134 considered in ref . @xcite . since gravitino production is suppressed by lack of thermalization we impose the constraint that the @xmath135 at @xmath62 to study this effect . we also impose the constraint that the energy of the incoming quarks and squarks should be greater than @xmath136 to avoid phase space suppression of the gravitino production cross section by the @xmath137 factor in eq . . these constraints give lower and upper bounds on @xmath138 respectively , where @xmath138 is the flat direction vev at @xmath139 when the flat direction starts oscillating . in the case of @xmath133 , these bounds are in fact incompatible indicating that lack of thermalization effectively shuts off gravitino production through phase space suppression . surprisingly , we find that even after thermalization , for both cases @xmath140 , gravitino production is phase space suppressed till the flat direction condensate decays because of the large gluon and gluino mass . we take the inflaton mass @xmath60 to be @xmath141 and @xmath142 to be @xmath143 as applicable for the relevant energy scales for @xmath144 and @xmath145 for @xmath146 as relevant below . we presume the inflaton decays at @xmath147 via @xmath148 so that the number density of the @xmath52 or @xmath53 particles equals that of the inflaton . the gluon mass is given by @xmath149 for @xmath133 , the flat direction condensate starts oscillating after the inflaton decays . for this case we take @xmath150 and @xmath151 as in 3.1 of ref . one must separately consider the time intervals @xmath152 , @xmath153 , @xmath154 and @xmath155 . the gluon mass is given by @xmath156 . the energy of the incoming particles is @xmath157 for @xmath158 . since kinetic equilibration does not change the number of particles while redistributing their energies , the average energy of the incoming particles for @xmath159 is also given by @xmath160 . gravitino production for @xmath155 is considered in section [ ssec : ygtm0lgd ] . at @xmath161 we impose @xmath162 where we have used @xmath163 and @xmath164 @xcite . then we get @xmath165 where we have used @xmath166 . @xmath167 gives @xmath168 . note that for @xmath169 , @xmath170 is given by @xmath171 where we have used @xmath172 . since @xmath173 , we see that @xmath170 is falling faster than @xmath174 . so , if @xmath175 holds at @xmath9 , then it also holds for @xmath169 . for @xmath176 , @xmath170 is given by eq . with @xmath177 . therefore @xmath178 since @xmath174 is decreasing with time , after some time @xmath179 , @xmath174 becomes less than @xmath170 , and later it becomes less than @xmath180 . setting @xmath181 we get @xmath182 @xmath183 as @xmath184 @xcite . for @xmath185 , @xmath186 constant , as for @xmath170 . then @xmath187 . as we shall see below , there is also an upper bound on @xmath138 which is highly restrictive . taking @xmath188 , we get , @xmath189 note that @xmath190 and @xmath191 . @xmath192 . for larger @xmath138 , both @xmath10 and @xmath11 will be larger . gravitinos are produced along with gluons or gluinos in the processes discussed in [ sec : revproc ] . to avoid phase space suppression by the @xmath193 factor in the differential cross section we impose @xmath194 we apply this condition for gravitino production before and after thermalization ( taking @xmath195 in this analysis ) . + _ before thermalization : _ we first consider gravitino production in the time interval @xmath196 . . implies @xmath197 where @xmath198 and @xmath199 are the incoming energies in the cm frame . we see that the rhs falls like @xmath65 , whereas the lhs is constant . we wish to analyse whether delayed thermalization suppresses gravitino production . therefore to consider maximum gravitino production , we impose the above inequality at @xmath200 . if the above inequality holds at @xmath200 , then it will hold at earlier times till @xmath9 too . using the above inequality we find an upper bound on @xmath138 . @xmath201 thus we need @xmath202 to get gravitino production from @xmath9 till time @xmath200 . for the time interval @xmath203 , @xmath204 and the condition in eq . gives @xmath205 we see that the rhs falls like @xmath65 and the lhs falls like @xmath206 . so if the above inequality holds at @xmath200 , which implies eq . then it will hold for all time @xmath207 till @xmath11 . we see that for the time interval @xmath208 the upper bound on @xmath138 ( eq . ) is smaller than the lower bound on @xmath138 ( eq . ) . thus , gravitino production for the entire time interval @xmath208 is not possible for the @xmath133 case . + _ after thermalization : _ at @xmath209 , just after full equilibration , the average energy of particles in the universe falls which further inhibits gravitino production . gravitino production can start at @xmath43 when @xmath136 becomes less than the incoming particle energy @xmath210 , i.e. , @xmath211 where the subscript @xmath212 denotes reheating or full equilibration at @xmath11 . this implies @xmath213 then , @xmath214 ^ 2 \nonumber \\ & = 81 { \left ( \frac{g_*}{{\mathcal{a } } } \right ) } \frac{\alpha^2 \varphi_0 ^ 4 \,\gamma_d^2 } { m_\phi^4 \,m_0 ^ 3}\ , , \label{eq : tgm0lgd}\end{aligned}\ ] ] where we have used [ sssec : eqtimes ] for @xmath11 and the expression for @xmath215 is derived later in section [ ssec : tr ] . the parameter @xmath216 that enters via @xmath215 is obtained to be @xmath217 in section [ ssec : cala ] . the perturbative flat direction decay rate is @xmath218 @xcite and the flat direction decays when the decay rate is of order @xmath174 at a time @xmath42 . then @xmath219 we see that @xmath220 only for @xmath221 , which conflicts with the lower bound on @xmath138 in eq . . thus the non - equilibrium condition implies that the condensate decays before gravitino production can commence in the radiation dominated era after thermalization . subsequent to the decay of the condensate there will be gravtino production in the thermal universe as discussed later in section [ ssec : ygtm0lgd ] . if we relax the bound in eq . , we may get gravitino production for some part of the time interval @xmath222 . we first consider @xmath169 . using @xmath223 in eq . and @xmath224 we find gravitino production is not phase space suppresed for @xmath225 for @xmath188 . since this corresponds to times less than @xmath190 , it implies that one can not get gravitino production anywhere in this time interval . we now consider @xmath226 . gravitino production can start at some time after @xmath200 , when the inequality in eq . is satisfied . this is equivalent to @xmath227 for @xmath188 this implies @xmath228 since this is larger than @xmath192 , once again there is no gravitino production in this time interval . therefore the conditions required for lack of kinetic and chemical equilibration leads to a large outgoing gluon and gluino mass and thus to phase space suppression that shuts off gravitino production for the entire time interval from @xmath9 to @xmath11 in the @xmath133 case . increasing the value of @xmath138 will not affect this conclusion as this will only lower or raise the limits in eqs . and respectively . similarly , gravitino production is phase space suppressed in the radiation dominated era before the condensate decays at @xmath42 and increasing the value of @xmath138 will not alter this conclusion . ( decreasing the value of @xmath138 to avoid phase space suppression will lead to the standard scenario of reheating and gravitino production in a thermal universe . ) thus , for the parameters chosen above for the scenario under consideration , there is no gravitino production for the case @xmath229 before the condensate decays . we now consider the case @xmath134 , where the flat direction condensate starts oscillating before the inflaton decays . @xmath230 gev and we take @xmath231 gev as in section 3.1 of ref . @xcite . the absence of kinetic equilibration of the inflaton decay products requires @xmath232 . @xmath170 is given by eq . . we require @xmath170 to be much less than @xmath174 at @xmath9 to obtain a significant phase of non - equilibrium since @xmath174 decreases with time while @xmath170 remains constant . then applying the non - equilibrium condition at @xmath9 , and setting @xmath233 and @xmath234 , we get @xmath235 where we have used @xmath236 between @xmath200 and @xmath9 as relevant for the inflaton oscillating in a quadratic potential before decay . chemical non - equilibration at @xmath9 requires @xmath237 . thus , for @xmath238 the universe is in kinetic and chemical equilibrium at @xmath9 . for @xmath239 the universe is in kinetic but not in chemical equilibrium at @xmath9 . finally , for @xmath240 the universe is neither in kinetic nor chemical equilibrium at @xmath9 . to obtain @xmath10 we use eq . and @xmath241 for @xmath242 . then @xmath243 as before , @xmath244 and @xmath245 . for gravitino production not to be phase space suppressed by the @xmath246 factor for @xmath247 we again impose the condition of eq . and get eq . . to consider maximum gravitino production we assume that this condition is valid from @xmath9 . using eq . at @xmath62 gives @xmath248 where @xmath249 . therefore we get @xmath250 since the lhs of eq . falls faster than the rhs , this upper bound on @xmath138 also ensures the viability of gravitino production at times after @xmath62 . we shall take @xmath188 hereafter so that gravitino production commences at @xmath9 and there is neither kinetic nor chemical equilibration at @xmath9 . for this value of @xmath138 , @xmath251 and @xmath252 . for gravitino production after @xmath11 , eq . implies @xmath253 then @xmath254 ^ 2 \nonumber \\ & = 81{\left ( \frac{g_*}{{\mathcal{a } } } \right ) } \frac{\alpha^2 \varphi_0 ^ 4 \,\gamma_d^3 } { m_\phi^4 \,m_0 ^ 4}\ , , \label{eq : tgm0ggd}\end{aligned}\ ] ] where we have used the expression for @xmath11 from [ sssec : noeqcond - m0ggd ] and for @xmath215 as in eq . or . @xmath255 as obtained in section [ ssec : cala ] . the time when the condensate decays , @xmath42 , is given by eq . . then @xmath256 for @xmath188 , @xmath43 is @xmath257 while @xmath42 is @xmath258 . thus , there is no gravitino production in the radiation dominated era after thermalization before the flat direction condensate decays . the lower and upper bounds on @xmath138 in eqs . and are restrictive . for later use we relax the upper bound from phase space suppression . if @xmath259 it implies that gravitino production commences at a time @xmath260 . to get an expression for @xmath261 we start with eq . and solve for @xmath262 , the scale factor at which gravitino production starts . we thus get @xmath263 ^ 4 , \label{eq : tpm0gtgd}\end{aligned}\ ] ] where we have used the fact that for @xmath264 the scale factor @xmath236 and for @xmath265 , @xmath223 . in this case one will get an extra factor of @xmath266 along with @xmath9 replaced by @xmath261 in the expressions for gravitino number density generated before kinetic equilibration in eqs . , and . we remark in passing that if @xmath267 and one has kinetic equilibration of the inflaton decay products at @xmath9 without chemical equilibration , then one would obtain gravitino production using only section [ ssec : tkin2tthr ] , with @xmath170 set equal to @xmath268 . having discussed the feasibility of gravitino production in a non - thermal universe in the context of phase space availability we now calculate the cross section of gravitino production . the expressions for @xmath51 in this section include an average over spins and colors of the initial and final state particles but are for a specific flavor for quarks and squarks . the cross sections obtained below are for the dominant @xmath127 helicity states of the gravitino and will be used for obtaining the collision integral ( the @xmath68 terms ) for @xmath269 , as in [ sec : aterms ] . the differential scattering cross sections in the cm frame is as given in eq . . we use @xmath51 as provided in table 1 of ref . then @xmath270 where @xmath271 is the generator of the @xmath272 group in the fundamental representation , and @xmath273 represent the color of the incoming particles . @xmath274 above differs from ref . @xcite by a factor of 2 because we sum over both chiralities of the quarks . the expressions in table 1 of ref . @xcite do not include an average over incoming and final states and hence we have a sum over @xmath275 and an additional factor of @xmath276 in the denominators of @xmath277 . using eq . in the limit that the incoming energy @xmath278 is much larger than all masses , the differential scattering cross sections for the three processes are @xmath279 the cross section @xmath79 is obtained by integrating the differential cross section . the scattering cross sections for the e , i and j processes are @xmath280 @xmath281 @xmath282 having obtained the cross section @xmath79 we use the procedure given in section [ sec : boltzmann ] to calculate the number density of gravitinos via the e , i and j processes using the integrated boltzmann equation . for @xmath133 we only consider @xmath283 . for @xmath134 , we calculate the number density separately from @xmath9 to @xmath10 , @xmath10 to @xmath11 and for @xmath284 . parameters needed to evaluate the gravitino abundance are given in section [ sec : params ] , [ sec : bparam ] and [ sec : mgt ] . as shown in section [ ssec : tm0lgd ] there is no gravitino production until the flat direction condensate decays at @xmath42 . the temperature @xmath285 at @xmath42 is derived in section [ sec : tf ] and equals @xmath286 . the gravitino abundance generated after @xmath42 is given by the standard expression for gravitino production in a thermal universe after inflation with the reheat temperature replaced by @xmath285 . for this low value of @xmath285 the gravitino abundance will be small enough to satisfy cosmological constraints . the @xmath68 terms for the processes e , i , j that enter the rhs of the integrated boltzmann eq . for the phase space distribution in eq . are given in [ sec : aterms ] . for the process e , @xmath287 where @xmath288 the parameter @xmath289 is derived in [ sec : bparam ] . solving eq . we then get @xmath290 { \left ( \frac{t_d}{t } \right)}^{3/2 } \label{eq : ngteproc } \\ & = 144 k \left [ 1 - { \left ( \frac{a_d}{a } \right ) } \right ] { \left ( \frac{a_d}{a } \right)}^{3},\end{aligned}\ ] ] where @xmath17 is defined as @xmath291 similarly for the process i , @xmath292 eq . implies that @xmath293 { \left ( \frac{t_d}{t } \right)}^{3/2 } \label{eq : ngtiproc } \\ & = 48 k \left [ 1 - { \left ( \frac{a_d}{a } \right ) } \right ] { \left ( \frac{a_d}{a } \right)}^{3}\,.\end{aligned}\ ] ] for the process j , @xmath294 eq . implies that @xmath295 { \left ( \frac{t_d}{t } \right)}^{3/2 } \label{eq : ngtjproc } \\ & = 96 k \left [ 1 - { \left ( \frac{a_d}{a } \right ) } \right ] { \left ( \frac{a_d}{a } \right)}^{3}.\end{aligned}\ ] ] one observes that the factor in the square brackets in the above expressions is related to gravitino production while the factor @xmath296 reflects dilution due to hubble expansion . because of the fast decline with time of the source term on the rhs of the integrated boltzmann equation , the production of gravitinos shuts off quickly and then the number density falls as @xmath297 . for @xmath10 to @xmath11 we evaluate a defined in eq . with the phase space distribution function given in eq . . @xmath298 \nonumber \\ & \equiv n_1(t ) n_2(t ) { \langle \sigma v \rangle } , \label{eq : a2 } \end{aligned}\ ] ] where @xmath299 is the expression in square brackets and @xmath300 ~d^3p_i \qquad \textrm { for } i = 1,2 \nonumber \\ & = { \mathcal{a}}_i ~n_i^{eq } \nonumber \\ & = { \mathcal{a}}_i ~\frac{g_i}{\pi^2 } t^3 .\end{aligned}\ ] ] here , @xmath301 and @xmath302 and we have ignored factors of @xmath303 . our expression for @xmath304 includes @xmath49 , unlike in ref . the number density of all particles in the universe is @xmath305 @xmath306 in thermal equilibrium , where @xmath307 is the total number of relativistic degrees of freedom for a fully thermalized plasma . for the minimal supersymmetric standard model ( mssm ) since we can express @xmath216 for @xmath10 to @xmath11 in terms of the standard thermally averaged cross section we can therefore use expressions for @xmath299 from the literature . in kinetic equilibrium , temperature is well defined and we choose to re - express the integrated boltzmann eq . @xmath309 as @xmath310 where a sum over all processes e , i , j is implicit . defining @xmath311 , where @xmath312 , we can rewrite the boltzmann equation as @xmath313 since @xmath314 , @xmath315 and so @xmath316 where we have used @xmath317 and @xmath318 . expressions for @xmath299 from ref . @xcite , were used in ref . these results were for the @xmath319 helicity states of the gravitino . including the @xmath127 helicity states we get @xmath320 [ eq : sigmav ] where we have considered only the @xmath120 gauge group . the gravitino abundance just before full thermalization at @xmath11 is @xmath321 where @xmath322 and @xmath323 are the entropy density and temperature at @xmath11 just before thermal equilibration . @xmath324 and @xmath299 includes a sum over all the processes in eq . . @xmath325 is obtained using eqs . , and and the temperature at @xmath10 needed for the entropy is given by eq . . then @xmath326 & = \frac{\sum_l q_l \,k \left [ 1 - { \left ( t_d / t_{kin } \right)}^{1/2 } \right ] { \left ( t_d / t_{kin } \right)}^{3/2 } } { { \left ( 2\pi^2 /45 \right)}{\mathcal{a}}\,t_{kin}^3 } \nonumber \\[0.2 cm ] & = \frac{\sum_l q_l \,k \left [ 1 - { \left ( \gamma_{kin } / \gamma_d \right)}^{1/2 } \right ] { \left ( \gamma_{kin } /\gamma_d \right)}^{3/2 } } { { \left ( 2\pi^2 /45 \right)}{\mathcal{a}}\,{\left ( m_\phi /3 \right)}^3 { \left ( \gamma_{kin } / \gamma_d \right)}^{3/2 } } \nonumber \\[0.2 cm ] & = \frac{\sum_l q_l \,k \left [ 1 - { \left ( \gamma_{kin } / \gamma_d \right)}^{1/2 } \right ] } { { \left ( 2\pi^2 /45 \right)}{\mathcal{a}}\,{\left ( m_\phi /3 \right)}^3 } \,,\end{aligned}\ ] ] where @xmath327 for the @xmath328 processes respectively . note that @xmath329 is also process dependent , as mentioned in [ sec : bparam ] . after equilibration at @xmath11 the temperature falls to @xmath330 then , @xmath331 where @xmath332 and @xmath333 . @xmath334 and @xmath335 are associated with gravitino production between @xmath9 and @xmath10 , and between @xmath10 and @xmath11 , respectively . for @xmath336 , substituting eqs . in eq . we have @xmath337 \int_{t_{min}}^{t_{kin } } dt \nonumber \\ & = \left [ \frac{0.07}{g_*^{3/2 } } \left(\frac{g_*}{{\mathcal{a } } } \right)^{5/4 } \sum_l f_l \,\frac{\alpha}{32 m^2 } \,\left(1+\frac{m_0 ^ 2 } { 3 m_{{\widetilde{g}}}^2 } \right ) \,{\mathcal{a}}_1 g_1 \ , { \mathcal{a}}_2 g_2 m \ , \right ] \left(t_{kin } - t_{min } \right)\end{aligned}\ ] ] where @xmath338 from eq . for @xmath328 processes respectively and @xmath339 are as given in [ sec : bparam ] . we have ignored the variation of @xmath340 with temperature and shall use the value of @xmath341 . as discussed in section [ ssec : tm0ggd ] gravitino production is suppressed until the flat direction condensate decays at @xmath42 . the temperature @xmath285 at @xmath42 is given in section [ sec : tf ] and equals @xmath342 . such a low temperature will lead to a gravitino abundance consistent with cosmological constraints . the values of parameters which we require to obtain the gravitino abundances are evaluated in this section for both @xmath133 and @xmath134 . @xmath343 is the temperature at @xmath10 just after kinetic equilibration . since kinetic equilibration only re - distributes energy among particles without changing their number , the average energy per particle is the same before and after kinetic equilibration . therefore @xmath344 where the rhs represents the energy of the particles just before kinetic equilibration and we presume 2 body decay of the inflaton . the boltzmann constant @xmath345 is hereafter set to @xmath124 . therefore @xmath346 where @xmath170 is given in eqs . and . for @xmath133 , @xmath347 for @xmath134 , @xmath348 thus , the value of @xmath343 for @xmath133 is @xmath349 and for @xmath134 is @xmath350 . the temperature @xmath323 just before thermalization at @xmath11 can be obtained by @xmath351 where we have used the relation @xmath244 as discussed in section [ sec : discuss ] . to obtain @xmath216 we note that the energy density of the universe @xmath352 just after kinetic equilibration at @xmath10 is equal to the energy density @xmath353 just before kinetic equilibration . then @xmath354 thus @xmath355 . the number density of a species @xmath50 just after kinetic equilibration is @xmath357 . this equals the number density of the species just before kinetic equilibration which , in turn , is the number density of the species at @xmath9 scaled by @xmath297 . the decay rate of the inflaton to a species @xmath50 will be proportional to @xmath49 and so @xmath358 is also proportional to @xmath49 . then @xmath359 is the same for all species . since @xmath360 , @xmath361 . as stated earlier in eq . the temperature at @xmath11 is given by @xmath362 for @xmath133 , @xmath363 for @xmath134 , @xmath364 we have used eqs . and for the expression of @xmath170 for @xmath133 and @xmath134 respectively , and @xmath341 . the temperature @xmath365 at @xmath366 is given by @xmath367 for @xmath133 , using eq . to define @xmath368 , we get @xmath369 where we have used section [ sssec : eqtimes ] for @xmath11 and eq . or for @xmath215 , and @xmath370 . for @xmath134 , using eq . , the temperature at @xmath43 is @xmath371 where we have used [ sssec : noeqcond - m0ggd ] for @xmath11 and eq . or for @xmath215 , and @xmath370 . the temperature at @xmath42 is given by @xmath372 for @xmath133 , @xmath373 where the expressions for @xmath11 , @xmath42 and @xmath215 are obtained from [ sssec : eqtimes ] , eqs . and or respectively . for @xmath134 , @xmath374 where the expressions for @xmath11 , @xmath42 and @xmath215 are obtained from [ sssec : noeqcond - m0ggd ] , eqs . and or respectively for the case @xmath133 , with @xmath375 , @xmath376 and @xmath188 , @xmath9 , @xmath377 , @xmath11 and @xmath42 are @xmath378 , @xmath379 , @xmath380 , @xmath381 and @xmath382 respectively . @xmath1 is constant at @xmath383 from @xmath9 to @xmath200 and falls to @xmath384 and @xmath385 at @xmath10 and @xmath11 respectively . from [ sec : mgt ] , the gluon mass is constant from @xmath9 to @xmath200 at @xmath384 and falls to @xmath386 and @xmath387 at @xmath10 and @xmath11 respectively . the final reheat temperature @xmath215 after full equilibration is @xmath388 , which is less than the gluon and gluino mass at that time . as we have argued in section [ ssec : tm0lgd ] there is no gravitino production in the non - thermal phase , nor after thermalization till the flat direction condensate decays at @xmath42 , because of phase space suppression associated with the large mass of the outgoing gluon or gluino . the temperature @xmath285 is @xmath342 and gravitino production after @xmath42 is hence small and consistent with cosmological constraints . for @xmath134 , with @xmath375 , @xmath389 and @xmath188 , @xmath200 , @xmath9 , @xmath10 and @xmath11 are @xmath390 , @xmath391 , @xmath392 and @xmath393 . @xmath1 falls to @xmath141 at @xmath9 , @xmath394 at @xmath104 and @xmath395 at @xmath56 . the corresponding gluon mass is @xmath384 , @xmath394 , @xmath396 and @xmath397 at @xmath200 , @xmath9 , @xmath10 and @xmath11 respectively . the final reheat temperature @xmath215 is @xmath398 . the gravitino abundances generated in the different epochs are @xmath399 where @xmath334 is the abundance due to gravitino production between @xmath9 and @xmath10 and @xmath335 is the abundance due to production between @xmath10 and @xmath11 . the gravitino abundance in both epochs is less than the upper bound of @xmath400 on the gravitino abundance from various cosmological constraints @xcite . gravitino production after thermalization at @xmath11 is suppressed until the condensate decays at @xmath42 . since the temperature at @xmath42 is @xmath342 the gravitino abundance generated after @xmath42 will be small and within cosmological bounds . in this article we have studied the novel scenario proposed by refs . @xcite where the vacuum expectation values of susy flat directions generate a large mass for gauge bosons and gauginos . due to the large gauge boson masses thermalization is delayed after the end of inflation . in the non - thermal universe gravitino production in refs . @xcite is suppressed because of the dilute nature of the plasma . the delay in thermalization also leads to a low reheat temperature and so gravitino production after thermalization is also suppressed . thus it is argued that the gravitino problem of susy models is avoided in this scenario . in this article we have carried out a more careful analysis of gravitino production in the non - thermal universe scenario of refs . we have identified the initial state particle distribution functions in the gravitino production process for the eras before kinetic equilibration and after kinetic equilibration in a non - thermal universe ( the appropriate particle distribution function for the pre - kinetic equilibration era was not used in refs . these have then been included in the calculation of the collision integral in the integrated boltzmann equation . we have also explicitly obtained the conditions for obtaining a non - thermal universe after inflaton decay . we have further investigated final state phase space suppression in the context of heavy final state gluons and gluinos in both the non - thermal and thermal eras . we have investigated two cases ( i ) where the flat direction starts oscillating after inflaton decay ( @xmath133 ) and ( ii ) where the flat direction starts oscillating before inflaton decay ( @xmath401 ) . in the first case ( @xmath133 ) we find that there is no gravitino production during the non - thermal phase because of phase space suppression due to the large gluon and gluino mass . gravitino production is suppressed even after thermalization till the flat direction condensate decays at @xmath42 . however the temperature @xmath285 is low and as such any gravitino abundance produced after @xmath42 is small and is within the standard cosmological bounds . in the second case ( @xmath134 ) there is gravitino production before thermalization . the gravitino abundance generated is consistent with cosmological constraints . after thermalization , as in the case @xmath229 , there is no gravitino production before the flat direction condensate decays . after the decay of the condensate the temperature @xmath285 is low and does not lead to conflict with cosmological bounds . we would like to thank h. baer , a. joshipura , p. konar , n. mahajan , h. mishra , k. patel and p. roy for very useful discussions . we would also like to thank r. allahverdi for clarifying aspects of ref . the number of incoming particles @xmath402 for gravitino production is given by @xmath403 for the time interval @xmath404 the form of @xmath405 is given by eq . . substituting that in the above equation we get , @xmath406 where we have considered @xmath407 . now at @xmath408 , the number density of the particles is equal to the number density of the inflaton , as @xmath148 . so , @xmath409 . then for @xmath404 , the number density @xmath410 is given by @xmath411 . substituting the above expression for @xmath410 in eq . we get @xmath412 below we present the @xmath68 terms for the processes e , i and j during the interval @xmath404 as needed for the @xmath134 case using eqs . , , and , and the phase space distribution in eq . , with the expressions for @xmath79 for the @xmath127 helicity gravitino states as given in eqs . , and . the a term for the e process is @xmath413 where @xmath271 is the generator of the @xmath272 group in the fundamental representation and @xmath414 are given in eq . . @xmath415 includes a factor of @xmath416 for @xmath416 quark / squark flavors , a factor of @xmath417 for the @xmath417 squark ` chiralities ' and another factor of @xmath417 for the @xmath417 hermitian conjugate processes . ( one should actually exclude the heavy quarks and squarks associated with the flat direction . that would decrease the final abundance by a factor of @xmath418 . ) then we get @xmath419 where the parameter @xmath289 is as defined in [ sec : bparam ] . the a term for the i process is @xmath420 @xmath421 includes a factor of @xmath416 for the quark flavors . then we get @xmath422 the a term for the j process is @xmath423 @xmath424 gets a factor of @xmath416 for squark flavors and a factor of @xmath417 for the squark ` chiralities ' . then we get @xmath425 to simplify our equations we define new parameters called @xmath426 and @xmath289 for the pre - factor present in @xmath68 terms , i.e. , @xmath427 where @xmath428 . substituting the values of @xmath429 and @xmath430 from eq . in the above equation gives us @xmath431 { \left ( \frac { \sum_{a , i , j } 8\pi\,g_s^2 \,|t^a_{ji}|^2 } { 6144 \,m_{pl}^2 \,\pi^5 } \right ) } \left ( 1 + \frac{m_0 ^ 2}{3 m_{{\widetilde{g}}}^2}\right ) \left ( \frac{a_d}{a } \right)^2 \equiv b \left ( \frac{a_d}{a } \right)^2 , \end{aligned}\ ] ] where @xmath432 { \left ( \frac { \sum_{a , i , j } 8\pi \,g_s^2 \,|t^a_{ji}|^2 } { 6144 \,m_{pl}^2 \,\pi^5 } \right ) } \left ( 1 + \frac{m_0 ^ 2}{3 m_{{\widetilde{g}}}^2}\right ) \nonumber \\ & = \left[\left ( \frac{3 \,\gamma_d^2}{m_\phi^3 } \frac{(2\pi)^3}{\pi } \right)^2 { \left ( \frac{m_{pl}^2}{8\pi } \right ) } \right ] { \left ( \frac { \sum_{a , i , j } \,g_s^2 \,|t^a_{ji}|^2 } { 6144 \,g_1\,g_2 \,\pi^5 } \right ) } \left ( 1 + \frac{m_0 ^ 2}{3 m_{{\widetilde{g}}}^2}\right ) \nonumber \\ & = \frac { 3 \sum_{a , i , j } g_s^2 \,m_{pl}^2 \,|t^a_{ji}|^2 \,\gamma_d^4 } { 256 \,\pi^2 \,g_1\,g_2 \,m_\phi^6 } \left ( 1 + \frac{m_0 ^ 2}{3 m_{{\widetilde{g}}}^2}\right ) \label{eq : bparam}\end{aligned}\ ] ] and we have used @xmath163 . we suppress the index @xmath433 for @xmath289 associated with the different @xmath339 for each process . considering the spin and color of the incoming particles , @xmath434 and @xmath435 for the @xmath433 processes respectively . the mass of the gluon @xmath436 which varies as @xmath206 after the condensate starts oscillating at @xmath200 . below we derive the mass of the gluon for various epochs in time for the two cases , @xmath133 and @xmath134 . 1 . at @xmath447 , @xmath448 2 . for @xmath264 , @xmath449 where we have used the fact that the universe was matter dominated for the period @xmath264 and @xmath236 , as relevant for an inflaton oscillating in a quadratic potential after inflation . + at @xmath161 , @xmath450 3 . for @xmath451 , @xmath452 where we have used the fact that the universe is radiation dominated for the period @xmath451 and @xmath223 . 4 . for @xmath453 @xmath454 where we have used the relation @xmath455 for @xmath453 . 5 . for @xmath456 @xmath457 where we have used eqs . and , and @xmath455 for @xmath155 . r. allahverdi and a. mazumdar , _ supersymmetric thermalization and quasi - thermal universe : consequences for gravitinos and leptogenesis _ , _ jcap _ * 0610 * ( 2006 ) 008 , [ http://xxx.lanl.gov/abs/hep-ph/0512227[hep-ph/0512227 ] ] . r. allahverdi , r. brandenberger , f .- y . cyr - racine , and a. mazumdar , _ reheating in inflationary cosmology : theory and applications _ , _ ann . rev . nucl . part . * 60 * ( 2010 ) 2751 , [ http://xxx.lanl.gov/abs/1001.2600[arxiv:1001.2600 ] ] . m. postma and a. mazumdar , _ resonant decay of flat directions : applications to curvaton scenarios , affleck - dine baryogenesis , and leptogenesis from a sneutrino condensate _ , _ jcap _ * 0401 * ( 2004 ) 005 , [ http://xxx.lanl.gov/abs/hep-ph/0304246[hep-ph/0304246 ] ] . a. basboll , _ susy flat direction decay - the prospect of particle production and preheating investigated in the unitary gauge _ , _ phys . rev . _ * d78 * ( 2008 ) 023528 , [ http://xxx.lanl.gov/abs/0801.0745 [ arxiv:0801.0745 ] ] . r. allahverdi , b. a. campbell , and j. r. ellis , _ reheating and supersymmetric flat direction baryogenesis _ , _ nucl . phys . _ * b579 * ( 2000 ) 355375 , [ http://xxx.lanl.gov/abs/hep-ph/0001122 [ hep - ph/0001122 ] ] . m. dine , l. randall , and s. d. thomas , _ baryogenesis from flat directions of the supersymmetric standard model _ phys . _ * b458 * ( 1996 ) 291326 , [ http://xxx.lanl.gov/abs/hep-ph/9507453 [ hep - ph/9507453 ] ] . k. enqvist , a. jokinen , t. multamaki , and i. vilja , _ numerical simulations of fragmentation of the affleck - dine condensate _ , _ phys . * d63 * ( 2001 ) 083501 , [ http://xxx.lanl.gov/abs/hep-ph/0011134 [ hep - ph/0011134 ] ] . s. kasuya and m. kawasaki , _ q - ball formation : obstacle to affleck - dine baryogenesis in the gauge - mediated susy breaking ? _ , _ phys . * d64 * ( 2001 ) 123515 , [ http://xxx.lanl.gov/abs/hep-ph/0106119 [ hep - ph/0106119 ] ] . i. v. falomkin , g. b. pontecorvo , m. g. sapozhnikov , m. y. khlopov , f. balestra , and g. piragino , _ low - energy anti - p he-4 annihilation and problems of the modern cosmology , gut and susy models _ , _ yad . fiz _ * 39 * ( 1984 ) 990 , [ nuovo cim . a * 79 * , 193 ( 1984 ) ] . m. y. khlopov , y. l. levitan , e. v. sedelnikov , and i. m. sobol , _ nonequilibrium cosmological nucleosynthesis of light elements : calculations by the monte carlo method _ , _ yad . _ * 57 * ( 1994 ) 1466 , [ phys . at . nucl . , * 57 * , 1393 ( 1994 ) ] . m. kawasaki and t. moroi , _ gravitino production in the inflationary universe and the effects on big bang nucleosynthesis _ , _ prog . phys . _ * 93 * ( 1995 ) 879900 , [ http://xxx.lanl.gov/abs/hep-ph/9403364[hep-ph/9403364 ] ] . m. bolz , a. brandenburg , and w. buchmuller , _ thermal production of gravitinos _ , _ nucl . phys . _ * b606 * ( 2001 ) 518544 , [ http://xxx.lanl.gov/abs/hep-ph/0012052[hep-ph/0012052 ] ] . erratum - ibid . b790 ( 2008 ) 336 . r. h. cyburt , j. r. ellis , b. d. fields , and k. a. olive , _ updated nucleosynthesis constraints on unstable relic particles _ , _ phys . _ * d67 * ( 2003 ) 103521 , [ http://xxx.lanl.gov/abs/astro-ph/0211258[astro-ph/0211258 ] ] . m. kawasaki , k. kohri , and t. moroi , _ big - bang nucleosynthesis and hadronic decay of long - lived massive particles _ _ * d71 * ( 2005 ) 083502 , [ http://xxx.lanl.gov/abs/astro-ph/0408426 [ astro - ph/0408426 ] ] . j. pradler and f. d. steffen , _ constraints on the reheating temperature in gravitino dark matter scenarios _ , _ phys . lett . _ * b648 * ( 2007 ) 224235 , [ http://xxx.lanl.gov/abs/hep-ph/0612291 [ hep - ph/0612291 ] ] . r. h. cyburt , j. ellis , b. d. fields , f. luo , k. a. olive , et al . , _ nucleosynthesis constraints on a massive gravitino in neutralino dark matter scenarios _ , _ jcap _ * 0910 * ( 2009 ) 021 , [ http://xxx.lanl.gov/abs/0907.5003[arxiv:0907.5003 ] ] .

the large deviation function has been known for a long time in the literature for the displacement of the rightmost particle in a branching random walk ( brw ) , or in a branching brownian motion ( bbm ) . more recently a number of generalizations of the bbm and of the brw have been considered where selection or coalescence mechanisms tend to limit the exponential growth of the number of particles . here we try to estimate the large deviation function of the position of the rightmost particle for several such generalizations : the-bbm , the-bbm , and the cbrw ( coalescing branching random walk ) which is closely related to the noisy fkpp equation . our approach allows us to obtain only upper bounds on these large deviation functions . one noticeable feature of our results is their non analytic dependence on the parameters ( such as the coalescence rate in the cbrw ) . + + + +

1601.04652

branching brownian motions ( bbm ) and branching random walks ( brw ) are among the simplest stochastic models of a growing population in space and time . they describe particles which perform brownian motions or random walks and branch independently at random times @xcite . if one starts with a single particle , the size of the region of space occupied by the particles grows linearly with time . since the mid seventies , one has a precise understanding of the fluctuations of the size of this region @xcite . for example in the one dimensional case one knows that the probability distribution of the position of the rightmost particle of a bbm can be obtained by solving an fkpp ( fisher - kolmogorov - petrovskii - piskounov ) equation @xcite : for a bbm starting at the origin , where particles diffuse according to @xmath2 ^ 2 \rangle = \sigma^2 t\ ] ] and branch at rate @xmath3 , one can show @xcite that , at time @xmath4 , the probability @xmath5 that the rightmost particle is on the right of @xmath6 is the solution of the fkpp equation @xmath7 with a step initial condition @xmath8 ( where @xmath9 is the heaviside function ) . in the long time limit , it is known @xcite that the probability @xmath10 that the position of the rightmost particle @xmath11 is concentrated around @xmath12 . one can also show @xcite from ( [ kpp ] ) that the large deviation function @xmath13 of the position @xmath14 of the rightmost particle for @xmath15 @xmath16\ ] ] is given by @xmath17 in ( [ ld - bbm ] ) and everywhere below the symbol @xmath18 means that @xmath19 over the last decade a number of generalizations of the branching brownian motion have been considered where , due to some selection or coalescence mechanism , the density of particles generated by the bbm saturates . these extensions of the bbm are expected to be described by noisy versions of the fkpp equation @xcite . in these noisy versions , the main effect of the noise is to shift the velocity of the front @xcite and to make its position fluctuate @xcite . a phenomenological approach has been proposed in @xcite which gives a prediction for the cumulants of this position . our goal here is to understand the large _ positive _ deviations of this position . the case of large _ negative _ deviations ( studied in @xcite for branching random walks with coalescence ) would require a rather different approach and will not be discussed in this paper except for some comments in the conclusion ; in particular the large deviation function may depend on the number of particles one starts with . in the present work we try to study how ( [ psi - bbm ] ) is modified by these selection or coalescence mechanisms . we discuss three models : 1 . * the @xmath0-bbm @xcite : * + in the @xmath0-bbm , one starts at time @xmath20 with a single particle at the origin . this particle branches and diffuses like a usual branching brownian motion . the only difference with the usual bbm is that whenever a particle gets at a distance larger than @xmath0 from the rightmost particle , it is eliminated . therefore at any given time @xmath4 the system consists of a random number @xmath21 of particles at positions @xmath22 which all satisfy @xmath23 where @xmath24 . + this number of particles @xmath25 fluctuates but one can show ( see the discussion in section [ existence ] ) that the evolution of the @xmath0-bbm leads to a steady state where the event @xmath26 is recurrent . + for large @xmath4 one can also show ( see section [ existence ] ) that the probability distribution of the position @xmath14 of the rightmost particle has a large deviation form @xmath27 \ . \label{ld - lbbm}\ ] ] one of our results ( see sections [ physical ] and [ proofs ] ) is the following upper bound for @xmath28 and large @xmath0 @xmath29 with @xmath30 where @xmath31 in ( [ res - lbbm ] ) and everywhere else in this paper , the symbol @xmath32 means that @xmath33 2 . * the @xmath1-bbm @xcite : * + in the @xmath1-bbm one starts as above with a single particle at @xmath20 which diffuses and branches but the size of the population can not exceed a fixed value @xmath1 . as long as the number of particles @xmath25 is less than @xmath1 the evolution is exactly the same as for the bbm . however , when @xmath34 , as soon as a new branching event occurs , the leftmost particle is eliminated so that the total number of particles remains subsequently equal to @xmath1 . + for the @xmath1-bbm we will obtain ( see sections [ physical ] and [ proofs ] below ) for the large deviation function @xmath35 \label{ld - nbbm}\ ] ] an upper bound @xmath36 where @xmath37 where @xmath38 is given by ( [ vc - eq ] ) . in fact , as discussed in the conclusion , we believe that @xmath39 remains valid even for @xmath40 . this would follow from a conjecture ( [ conjecture ] ) that we formulate in the conclusion but that we did not succeed to prove . the cbrw ( coalescing branching random walk ) * @xcite + an important motivation in the study of the cbrw is its dual relation with the noised fkpp equation , rigorously established in @xcite . + to explain how the cbrw is defined let us first consider a branching random walk brw on a one dimensional lattice with lattice spacing @xmath41 : a particle on site @xmath6 jumps to site @xmath42 at rate @xmath43 , to site @xmath44 at rate @xmath43 and branches at rate @xmath45 to give rise to two new particles on the same site . + the trajectory of each particle is a random walk and in the long time limit the probability that such a random walk reaches a position @xmath46 is of the form @xmath47 + where @xmath48 using the fact that @xmath49 with @xmath50 the large deviation function ( [ fv ] ) can be easily obtained from the parametric form as @xmath51 + as the particles branch at rate @xmath45 , the distribution of the position @xmath52 of the rightmost particle of this brw , ( in absence of coalescence ) , is of the form @xcite @xmath53\ ] ] with @xmath54 + now in the coalescing branching random walk ( cbrw ) , in addition to the diffusion and the branching , we let each pair of particles on the same site coalesce at rate @xmath55 . we will show in section [ existence ] @xmath56\ ] ] and in sections [ physical ] and [ proofs ] that for @xmath57 , @xmath58 where @xmath59 and , where for each @xmath60 , @xmath61 is solution of @xmath62 with @xmath38 and @xmath63 given by @xmath64 ( i.e. @xmath63 is the value of @xmath60 such that @xmath65 ) . + + the general expression ( [ res1-cbrw ] ) simplifies when @xmath66 . one then has @xmath67 and in the whole range @xmath68 @xmath69 instead of ( [ g ] ) . all the other steps remain the same with @xmath70 , @xmath71 and therefore @xmath72 + if one would consider more general branching random walks , characterized by the rate @xmath73 at which a particle jumps a distance @xmath61 from the site it occupies , @xmath74 would be given by @xmath75 and all the rest ( [ parametric]-[vc - v1 ] ) would remain unchanged with only ( [ g ] ) replaced by ( [ gnew ] ) . + + _ remark : _ a way of looking for a solution @xmath61 of ( [ v ] ) is to work with the legendre transform @xmath74 related to @xmath76 by ( [ parametric ] ) . if @xmath77 and @xmath78 are defined by @xmath79 and @xmath80 , one can check that ( [ v ] ) becomes @xmath81 under this form , one can show using the convexity of @xmath74 that @xmath82 has a single minimum at some value @xmath83 , that @xmath84 and that as @xmath85 as @xmath86 or @xmath87 ( we restrict our discussions here and below to @xmath88 as @xmath89 ) , there is always a solution @xmath78 and therefore a solution @xmath61 of ( [ v ] ) . in this section we explain a heuristic way of understanding the claims in the introduction . the main idea is rather similar for the three problems ( see figure 1 ) . consider first all the possible trees of a bbm which , starting with a single particle at the origin , contain at least one particle which reaches , at time @xmath4 , a position on the right of @xmath90 at time @xmath4 . here we focus on velocities @xmath91 ( for the bbm one knows that @xmath92 ) . the probability that the tree has at time @xmath4 at least one particle on the right of @xmath90 is ( [ ld - bbm],[psi - bbm ] ) for @xmath91 @xmath93 = \exp \left[t \left(1- { v^2 \over v_c^2 } \right ) \right ] \ .\ ] ] for each such tree event , we will call red particles all the particles which end up on the right of @xmath90 . given its position at time @xmath4 , the trajectory of a red particle is , up to a shift ( linear in time ) , a brownian bridge ( in fact it is more like a brownian excursion @xcite but this has no incidence on the discussion below ) . when one goes from the bbm to the @xmath0-bbm , a red particle will survive if between time @xmath94 and time @xmath4 no other particle of the bbm overtakes it by a distance @xmath0 . any tree of the bbm for which a red particle survives contributes to the event that the the rightmost particle of @xmath0-bbm is on the right of @xmath90 . so the probability that a tree of the bbm reaches position @xmath95 and that at least one red particle is never overtaken by any other particle of the bbm by a distance @xmath0 is a lower bound for the probablity that a @xmath0-bbm reaches position @xmath90 . this is why in the following , by estimating the survival probability of a red particle of a bbm , we will get an upper bound on the large deviation function ( [ ld - lbbm ] ) of the @xmath0-bbm . as a red particle is moving on average faster than @xmath38 the only possibility for it to be killed is that for a relatively short time interval @xmath96 , i.e. a time @xmath97 , either this red particle moves slower than @xmath60 , or one of the other particles of the tree moves sufficiently fast to overtake it by a distance @xmath0 or both . so the picture is the following . a red particle moves at velocity @xmath60 . along its trajectory , branching events occur which give rise to subtrees . this red particle is then killed if , shortly after one of these branching events , the red particle slows down and one of the particles of the subtree overtakes it by a distance @xmath0 . let us now be quantitative . the discussion below will hold for more general random walks , where the probability ( [ pbbm ] ) would be replaced by @xmath98 where @xmath76 is the large deviation function of the position of the random walk . in this general case @xmath38 is given by @xmath99 the case of the branching brownian motion will then be recovered by taking @xmath100 one can show that , conditioned on the fact that a red particle moves at velocity @xmath60 , the probability @xmath101 that during a relatively short time interval @xmath102 ( here @xmath103 ) it moves a distance @xmath6 is @xmath104 \ . \label{peq}\ ] ] now the probability @xmath105 that at least one particle of the subtree created at time @xmath106 moves a distance @xmath107 during the time interval @xmath96 is given by @xmath108 \right\ } \ . \label{qeq}\ ] ] therefore the probability @xmath109 that such a subtree will kill the red particle is @xmath110 _ remark : _ it is rather easy to establish ( [ peq ] ) . if a random walk has a large deviation function @xmath76 , the probability that during the time interval @xmath102 it moves from a position @xmath61 to a position @xmath111 , conditioned on the fact that during on a time @xmath4 it moves a distance @xmath90 ( with @xmath112 ) is given by @xmath113 \over \exp[t f(v ) ] } \ .\ ] ] optimizing over @xmath61 gives @xmath114 \over \exp[t f(v)]}\ ] ] and this leads to ( [ peq ] ) when @xmath97 . + depending on which term realizes the minimum in the rhs of ( [ qeq ] ) one has to distinguish two cases : + * if 1 dominates in ( [ qeq ] ) this means that the particle of the subtree moves at velocity @xmath38 . in this case @xmath6 and @xmath96 are related by @xmath115 because for @xmath116 , @xmath105 would remain @xmath117 but @xmath101 would get smaller . + + one can then see that the value of @xmath96 which maximizes ( [ peq ] ) is solution of @xmath118 + this condition takes the form @xmath119 where @xmath120 and this gives ( [ peq ] ) @xmath121 very much like in the remark at the end of the introduction , assuming as above that @xmath88 as @xmath122 , one can show that ( [ cond1 ] ) has always a solution . + + as the number @xmath123 of branching events along the red trajectory is of order @xmath4 ( for a rigorous justification , see chauvin and rouault @xcite ) the survival probability of the red particle is @xmath124 therefore @xmath125 and this implies that @xmath126 in the particular case where @xmath127 the solution of ( [ cond1 ] ) is @xmath128 and this leads to the announced result ( [ res - lbbm],[res2-lbbm ] ) . + * when the second alternative dominates in ( [ qeq ] ) one needs to find the maximum over @xmath96 and @xmath6 of @xmath129 \ .\ ] ] + this implies that @xmath130 and @xmath96 are solutions of @xmath131 after some algebra which uses ( [ cond2 ] ) one ends up with the same expression ( [ peq2 ] ) , the only difference being that @xmath61 is now solution of ( [ cond2 ] ) instead of ( [ cond1 ] ) . + as @xmath38 is solution of ( [ vc ] ) , one can check that the solution @xmath61 of ( [ cond2 ] ) reduces to the solution of ( [ cond1 ] ) when @xmath132 , meaning that the rightmost particle of the subtree moves at the velocity @xmath38 . + in the particular case where @xmath133 the solution of ( [ cond2 ] ) is @xmath134 ( where @xmath135 ) and this leads to the second line of ( [ res2-lbbm ] ) . in the @xmath1-bbm , the picture is rather similar and one has to estimate the probability @xmath109 that a subtree will kill a red particle . to do so one needs the red particle to slow down so that the subtree produces @xmath1 particles ahead of the red particle to eliminate it . the probability that the red particle moves a distance @xmath6 during time @xmath96 is still given by ( [ peq ] ) . we now need to estimate the probability @xmath105 that the subtree produces , at time @xmath96 , @xmath1 particles on the right of postion @xmath6 . we do not have an expression for @xmath105 ( see the discusion in the conclusion for a conjecture ) . one can however obtain an easy upper bound ( using the markov inequality ) @xmath136 where @xmath137 is the number of particles of a subtree ( of age @xmath96 ) on the right of position @xmath6 . one has @xmath138\ ] ] so that @xmath139 \label{expression - justifiee}\ ] ] which , as for the @xmath0-bbm , we can write for more generality @xmath140 \label{q - nbbm}\ ] ] to treat the case of an arbitrary @xmath1-bbm . now we need to find a bound for @xmath109 given by ( [ peq ] ) and the discussion is very similar to what we did for the @xmath0-bbm : + * if @xmath3 dominates in ( [ q - nbbm ] ) , then @xmath141 where @xmath96 and @xmath61 are related by @xmath142 + the optimisation of ( [ peq ] ) under the constraint ( [ s - nbbm ] ) leads to @xmath143}\ ] ] where @xmath61 is solution of @xmath144 + [ a solution @xmath145 exists for @xmath146 for the same reason as in . ] one gets after some algebra @xmath147 + for the @xmath1-bbm , one has @xmath133 the solution of ( [ ber1 ] ) is @xmath148 ; so @xmath149 , and @xmath150 + this agrees with the first line of ( [ res1-nbbm ] ) . + + + * in the second alternative of ( [ q - nbbm ] ) @xmath151 \big ) \label{ber2}\ ] ] given that @xmath152 there is also the natural condition @xmath153 ( because it is highly unlikely to have more than @xmath154 particles in a time @xmath155 , @xmath156 ) so that @xmath157 + the expression in the exponential being linear in @xmath96 , the maximum in @xmath96 is achieved at one of the two boundaries in . + if the maximum is realized by the condition @xmath158 , one recovers the results and . on the other hand , if the maximum is realized by @xmath159 , the optimal value of @xmath61 in ( [ ber2 ] ) is solution of @xmath160 and this leads to @xmath161 one can check that the range of validity of ( [ bber1 ] ) is @xmath162 and for ( [ bber4 ] ) is @xmath163 where @xmath164 is the value of @xmath60 where ( [ ber1 ] ) and ( [ bber3 ] ) have a common solution @xmath61 . it is remarkable to notice that for @xmath165 , both ( [ bber1 ] ) and ( [ bber4 ] ) coincide to give @xmath166 . + for @xmath133 the solution of ( [ bber3 ] ) is @xmath167 , which leads to @xmath168 + comparing with , one can check that holds for @xmath169 , while is valid for @xmath170 , as announced in ( [ res1-nbbm ] ) . for a branching random walk on a lattice , the probability that a red particle reaches the position @xmath171 with @xmath91 at time @xmath4 is of the form @xmath172 for example if the random walk is characterized by the probability @xmath73 that the walker jumps a distance @xmath61 from the site it occupies , @xmath76 is given in a parametric form as @xmath173 with @xmath74 given by ( [ gnew ] ) . given that the red particle moves on average at velocity @xmath60 during time @xmath4 , the probability @xmath101 that it moves a distance @xmath6 during a time interval @xmath103 is as before ( [ peq ] ) by @xmath174 \ .\ ] ] on the other hand the number of particles produced by the subtree at position @xmath6 at time @xmath96 is @xmath175 . therefore the probability that the red particle is killed by a subtree of age @xmath96 is @xmath176 \ . \label{qexp}\ ] ] as for the @xmath0-bbm , one needs to distinguish two cases : + * if @xmath3 dominates in ( [ qexp ] ) this means that @xmath177 satisfies the relation @xmath178 then one has to maximize @xmath101 given by ( [ peq ] ) over @xmath96 and @xmath6 given the constraint ( [ constraint ] ) . + this leads to the fact that @xmath179 where @xmath61 is solution of @xmath180 and after some algebra to @xmath181 . this leads to ( [ v ] ) . + * the other case , when ( [ qexp ] ) is dominated by @xmath182 , is much easier . the optimum over @xmath96 gives @xmath183 and therefore @xmath184 . in this section we prove the existence of the large deviation functions ( [ ld - lbbm],[ld - nbbm],[ld - cbrw ] ) . we first establish two elementary properties of the @xmath0-bbm if one starts at time @xmath94 with @xmath1 surviving particles . in view of the statement , we can assume @xmath185 . for any @xmath186 , let @xmath187 be the number of surviving particles of the @xmath0-bbm at time @xmath96 ( so that @xmath188 ) . [ l : premiere_propriete_l - bbm ] let @xmath189 then @xmath190 : { \cal n}(s ) < n^\lambda \right ] > 1 - { 3 \over n^\mu } \label{p1}\ ] ] where @xmath191 and @xmath55 are constants which satisfy some conditions ( [ a - b - lambda - nu ] ) . for example , @xmath192 , @xmath193 , @xmath194 and @xmath195 will work . [ l : seconde_propriete_l - bbm ] there exist constants @xmath196 and @xmath197 , depending only on @xmath198 , such that @xmath199 : \ , { \cal n}(s ) = 1 ) \ge c_2\ .\ ] ] in words , lemma [ l : premiere_propriete_l - bbm ] says that with a probability close to 1 when @xmath1 is large , the number of surviving particles @xmath200 will be greatly reduced within a very short time @xmath106 ( defined in ( [ tau - def ] ) ) , whereas lemma [ l : seconde_propriete_l - bbm ] ensures that no matter how large @xmath1 is , within a time independent of @xmath1 ( but which may depend on @xmath0 for example @xmath201 ) , the total number of surviving particles will have become 1 , at least once . in lemma [ l : seconde_propriete_l - bbm ] , it is possible to get moment estimates of the first time when the system has exactly a single particle ; see @xcite . _ proof of lemma [ l : premiere_propriete_l - bbm ] . _ it suffices to establish the following upper bound @xmath202 : { \cal n}(s)\ge n^\lambda \right ] < { 3 \over n^\mu } \ . \label{p1-bis}\ ] ] let us write @xmath203 without loss of generality , one can choose the origin to be the position of the rightmost particle of the @xmath0-bbm at time @xmath94 . so all the initial positions are in @xmath204 $ ] . if we assume that @xmath205 at all times @xmath206 , we want to follow the trajectories @xmath207 of @xmath208 surviving particles between time @xmath183 and time @xmath106 . at time @xmath183 we choose any set of @xmath208 different particles among the @xmath1 present at time @xmath94 . let @xmath209 be their positions at time @xmath94 . these particles move , branch and can get killed according to the rule of the @xmath0-bbm ( they get killed as soon as their distance to the leading particle of the full @xmath0-bbm exceeds @xmath0 ) . when one of these @xmath208 particles gets killed , one replaces it immediately by any of the remaining @xmath210 . on the other hand , when one of them branches , one just keeps one of the two branches in our list of @xmath208 particles and ignore the other branch . we obtain this way @xmath208 trajectories . let us denote @xmath211 the positions of these particles . these @xmath208 trajectories are those of brownian particles , except that whever one of these particles gets killed , it is replaced by one of the surviving @xmath212 particles of the @xmath0-bbm ( i.e. the corresponding trajectory makes a jump to its right ) . let us consider also @xmath208 regular brownian motions which start at time @xmath183 at the same positions as the above @xmath208 particles of the @xmath0-bbm . we denote by @xmath213 the positions of these @xmath208 brownian particles at time @xmath96 . by a simple coupling argument it is clear that at any time @xmath214 and for @xmath215 , one has @xmath216 so that @xmath217 therefore the probability @xmath218 that there exists at least one surviving particle of the full @xmath0-bbm on the right of some fixed position @xmath219 is bound from below by @xmath220 \\ & \ge & \p \left[\max_{1 \le i \le m } y_i(s ) > b l \right ] \ge 1 - \left[\int_{-\infty}^{(b+1 ) l /\sqrt{2 \tau \sigma^2 } } { e^{-u^2 } du \over \sqrt{\pi } } \right]^m \ .\end{aligned}\ ] ] using the fact that for @xmath221 @xmath222\ ] ] and that for @xmath223 @xmath224 one gets that @xmath225 to complete the proof of ( [ p1-bis ] ) , we now show that there is a small probability that the number @xmath226 of particles of the @xmath0-bbm on the right of position @xmath227 at time @xmath106 exceeds @xmath208 . to do so , we first notice that @xmath228 < \p[\widetilde{\cal n } > m ] \ ] ] where @xmath229 is the number of particles on the right of @xmath227 at time @xmath106 generated by @xmath1 independent bbm s ( with no selection ) starting all at time 0 at position @xmath0 . one can calculate the expectation @xmath229 @xmath230 = n e^\tau \int_{{(b-1 ) l \over \sqrt{2 \sigma^2 \tau}}}^\infty { e^{-u^2 } du \over \sqrt { \pi } } < 2 n^{1-{(b-1)^2 \over a}}\ ] ] where we have used that for @xmath231 @xmath232 therefore by the markov inequality one gets @xmath228 < 2 n^{1- \lambda -{(b-1)^2 \over a } } .\ ] ] now we know that , at time @xmath106 , there is a probability @xmath218 close to @xmath3 that there is at least one particle on the right of @xmath219 and a probability also close to @xmath3 that @xmath233 . therefore , because when there is at least one particle on the left of @xmath219 and no more than @xmath208 particles on the right of @xmath234 , one knows that the total number of surviving particles of the @xmath0-bbm does not exceed @xmath208 . consequently , @xmath235 < 1 - q + \p [ \widehat{\cal n } > m ] < 3 n^{-\mu}\ ] ] if we choose @xmath236 this completes the proof of ( [ p1]).@xmath237 _ proof of lemma [ l : seconde_propriete_l - bbm ] . _ let @xmath238 be a large constant independent of @xmath1 . it suffices to prove that if one starts with an arbitrary number @xmath1 of particles of the @xmath0-bbm , there is , uniformly in @xmath1 , a positive probability @xmath239 that the number of particles will be less than or equal to @xmath240 at least once before a time of order 1 . to prove this statement , we use @xmath241 times the result ( [ p1 ] ) : the number @xmath242 of steps needed is such that @xmath243 according to ( [ p1 ] ) , one has @xmath244 if the constant @xmath240 is chosen sufficiently large such that @xmath245 ; on the other hand , the time needed ( [ tau - def ] ) for this to happen will be less than @xmath246 this proves lemma [ l : seconde_propriete_l - bbm].@xmath237 now that we have proved lemmas [ l : premiere_propriete_l - bbm ] and [ l : seconde_propriete_l - bbm ] , it is quite easy to deduce the existence of the large deviation function for the @xmath0-bbm . let @xmath247 , and let @xmath248 [ clearly , @xmath249 depends on @xmath60 , @xmath4 and @xmath1 . ] the existence of the large deviation function we need to prove means the existence of @xmath250 . we prove this by considering @xmath251 where @xmath25 denotes as before the number of particles in the @xmath0-bbm at time @xmath4 . clearly , @xmath252 as such , the function @xmath253 is superadditive on @xmath254 , and as @xmath4 goes to infinity , @xmath255 $ ] . the existence of @xmath256 implies the existence of @xmath250 ; indeed , we trivially have @xmath257 because @xmath258 . conversely , by lemma[l : seconde_propriete_l - bbm ] , @xmath259 the last two inequalities together yield the existence of @xmath250 , which equals @xmath256 . for the @xmath1-bbm , we start with two simple but useful monotonicity properties , which are borrowed from @xcite . we include the elementary proof for the sake of self - containedness . we say that @xmath260 dominates @xmath261 if @xmath262 for all @xmath263 ( so in particular , @xmath264 ) . [ l : premiere_propriete_de_monotonie_n - bbm ] * ( first monotonicity property for the @xmath1-bbm ) * let @xmath265 and @xmath266 be such that @xmath267 for all @xmath268 . there exists a coupling for two @xmath1-bbm systems on a same probability space , starting at positions @xmath269 and @xmath270 respectively , such that the first system dominates the second at all time . _ consider two @xmath1-bbm systems , the first starting at positions @xmath269 , and the second at @xmath270 . we attach the same brownian motion to particles starting at @xmath271 and @xmath272 ( for @xmath268 ) respectively in the two systems , and also attach the same poisson process which determines the branching times along the paths . as such , the first branching time is identical in the two systems , and before this time , the @xmath6-system obviously dominates the @xmath61-system . it is also easy to check that right after the first branching time , the @xmath6-system still dominates the @xmath61-system . then by attaching as before the same brownian motions and the same poissonian clocks to the @xmath6- and the @xmath61-particles , the @xmath6-system will continue to dominate the @xmath61-system . and so on . the procedure leads to the desired coupling.@xmath237 [ l : seconde_propriete_de_monotonie_n - bbm ] * ( second monotonicity property for the @xmath1-bbm ) * let @xmath273 . let @xmath274 and @xmath266 be such that @xmath267 for all @xmath268 . there exists a coupling for an @xmath275-bbm and an @xmath1-bbm on a same probability space , with initial positions @xmath276 and @xmath270 respectively , such that the @xmath275-bbm dominates the @xmath1-bbm all time . _ if @xmath277 , this amounts to the previous lemma . so let us assume @xmath278 . then , as in the proof of the previous lemma , if initially the @xmath1 rightmost particles of the system with @xmath275 particles dominates the other system , this remains true subsequently . the remaining @xmath279 particles can only reinforce this domination.@xmath237 let us now turn to the proof of the existence of the large deviation function for the @xmath1-bbm . let @xmath280 . consider the following event for the @xmath1-bbm : @xmath281 to prove the existence of the large deviation function , we need to show that the limit @xmath250 exists . we prove this by an argument of superadditivity . by removing all particles at time @xmath4 except the rightmost one , the second monotonicity property stated in lemma [ l : seconde_propriete_de_monotonie_n - bbm ] tells us that @xmath282 so the function @xmath283 is superadditive on @xmath254 . in particular , @xmath284 \ , , \ ] ] exists . the existence of the large deviation function of the cbrw is very similar . as in lemma [ l : seconde_propriete_de_monotonie_n - bbm ] for the @xmath1-bbm , the probability of the large deviation event increases with the number of initial particles . consequently , by removing all particles except the rightmost one at time @xmath4 , on sees that if @xmath249 denotes the event that in the cbrw , there exists a particle lying in @xmath285 at time @xmath4 , @xmath286 from which the existence of @xmath250 follows immediately . we first describe the strategy for the @xmath0-bbm . the strategy for the @xmath1-bbm will be along similar lines , with a few appropriate modifications indicated below . the output of this paragraph has been described in section [ physical ] . let @xmath287 denote as before the event that there exists at least one particle in the @xmath0-bbm whose position at time @xmath4 lies in @xmath288 . to bound from below @xmath289 , we consider the following event of the bbm ( without selection):-bbm , we use the superscript in @xmath290 to remind us that it will serve to study the large deviation function for the @xmath0-bbm . a similar remark applies to the forthcoming events @xmath291 and @xmath292 . ] @xmath293 here , @xmath25 denotes , as before , the number of particles at time @xmath4 . leaning to the left means that the path of the particle lies in @xmath294 $ ] for all @xmath295 $ ] . is arbitrary ; anything in @xmath296 will do the job . ] we say that a particle with trajectory @xmath297)$ ] is @xmath0-dominated if at some time @xmath298 $ ] there is a particle lying in @xmath299 . clearly , if @xmath290 is realized , then one can construct an @xmath0-bbm such that the large deviation event @xmath287 is realized . therefore , @xmath300 we estimate @xmath301 which will serve as a lower bound for @xmath289 . to bound @xmath301 from below , let us write @xmath302 by the cauchy schwarz inequality , we have @xmath303 ^ 2}{\e [ ( \ # \widetilde{e}_t^{\mathrm{lbbm}})^2 ] } \ , .\ ] ] a fortiori , @xmath304 ^ 2}{\e [ ( \ # \widetilde{e}_t^{\mathrm{lbbm}})^2 ] } \ , . \label{cauchy - schwarz}\ ] ] we need to bound @xmath305 from below , and bound @xmath306 $ ] from above . the main estimates for the @xmath0-bbm which we obtain below are as follows : @xmath307 \ , l } \big ) t \right ] \ , , \label{l - bbm : e(et):lb } \\ \e [ ( \ # \widetilde{e}_t^{\mathrm{lbbm}})^2 ] & \lesssim & \exp \left [ -\big ( \frac{v^2}{2\sigma^2 } -1\big ) t \right ] , \label{l - bbm : e(et):ub}\end{aligned}\ ] ] with the value of @xmath308 given in . as before , the notation @xmath309 or @xmath310 means that @xmath311 , whereas @xmath312 denotes a term not depending on @xmath4 , such that @xmath313 . in view of the cauchy schwarz inequality , it is clear that and together will imply the upper bound stated in for the large deviation function @xmath314 of the @xmath0-bbm . the next subsection is devoted to the proof of ( [ l - bbm : e(et):lb ] ) . the proof of ( [ l - bbm : e(et):ub ] ) , which is identical for all the three models , is postponed to subsection [ 2nd_moment ] . we write @xmath315)$ ] for the trajectory of the particle @xmath316 in the definition of @xmath317 , and write @xmath318\ } \ , \label{at}\ ] ] which stands for the event that the particle @xmath316 leans to the left . then @xmath319 where , for all @xmath320 $ ] , @xmath321 stands for the event that the subtree of bbm branched at time @xmath322 on the path of @xmath323 does not produce any descendant going beyond @xmath323 by distance @xmath324 at any time during @xmath325 $ ] . here , @xmath326 is a rate-2 poisson process . the identity above , which is intuitively clear ( except , maybe , for the rate being 2 instead of 1 which is a property of the poisson process ; we mention that the rate of the poisson process plays no role in the final result ) , follows immediately from the chauvin rouault spinal decomposition theorem @xcite . it is easily guessed that the essential contribution to the integral @xmath327 on the right - hand side comes from the neighbourhood of @xmath328 . in any case , we can limit ourselves to the neighbourhood of @xmath328 to pretend that it only gives a lower bound : @xmath329 by conditioning upon @xmath330)$ ] and @xmath331 , we have @xmath332 where @xmath333 denotes conditional probability given @xmath323 . as such , writing @xmath334 for expectation with respect to @xmath335 , we have @xmath336 \ , \mathrm{d } u } \ , , \end{aligned}\ ] ] the second identity being a consequence of the fact that @xmath337 is a rate-2 poisson process . accordingly , @xmath338 \ , \mathrm{d } u } \ , \big| \ , x_t = vt\big\ } \ , .\ ] ] given @xmath339 , the process @xmath340)$ ] is a brownian bridge of length @xmath4 ; it can be realized as @xmath341 , where @xmath342 is a standard brownian motion ( of variance @xmath3 ) . thus @xmath343 \ , \mathrm{d } u } \big\ } . \label{pf - lb-1}\ ] ] where @xmath344\ } .\ ] ] we will see that the indicator @xmath345 brings no significant difference to the expectation . writing the conditional probability @xmath346 and @xmath347 for the associated expectation , we obtain : @xmath348 \ , \mathrm{d } u } \big ) \ , .\ ] ] by scaling , @xmath349\}$ ] , which converges to @xmath3 when @xmath350 . so in our notation for @xmath351 " , we have @xmath352 \ , \mathrm{d } u } \big ) \\ & \ge & \exp \big\ { -(\frac{v^2}{2\sigma^2 } -1)t - 2\int_0^t \e^t [ 1-\p_x ( d_t^{\mathrm{lbbm}}(u ) ) ] \ , \mathrm{d } u \big\}\ , , \end{aligned}\ ] ] the last line following from jensen s inequality . by definition , @xmath353 & = & \frac{\e \ { [ 1-\p_x ( d_t^{\mathrm{lbbm}}(u ) ) ] \ , { \bf 1}_{a_t^{(w)}}\}}{\p(a_t^{(w ) } ) } \\ & \le & \frac{\e [ 1-\p_x ( d_t^{\mathrm{lbbm}}(u))]}{\p(a_t^{(w ) } ) } .\end{aligned}\ ] ] we have already seen that @xmath354 , @xmath350 . so for all sufficiently large @xmath4 ( which will be taken for granted from now on ) , we have on the right - hand side is arbitrary ; anything in @xmath355 will do the job . ] @xmath356 as such , @xmath357 \ , \mathrm{d } u \big\}\ , . \label{l - bbm : jensen}\ ] ] [ so the presence of the indicator function @xmath345 in indeed has no significant influence . ] for all @xmath358 , let us write @xmath359 for the maximal position at time @xmath96 of a bbm independent of @xmath323 . [ this was denoted by @xmath360 in the introduction . ] by definition of @xmath321 , @xmath361 : \ , m(s ) \ge l+x_{s+u}-x_u \big ) \nonumber \\ & \le & \int_0^{t - u } \p_x \big ( m(s ) \ge l+x_{s+u}-x_u \big ) \ , \mathrm{d } s \ , . \label{premier_moment_l - bbm : estimation_technique}\end{aligned}\ ] ] [ the inequality in is heuristic ; it would be trivially true if @xmath96 were an integer ( in which case we would have a sum over @xmath96 instead of an integral on the right - hand side ) . however , we can easily make it rigorous by arguing that @xmath362 : \ , m(s ) \ge l+x_{s+u}-x_u ) \le \sum_{i=1}^{\lfloor t - u\rfloor+1 } \p_x ( \sup_{s\in [ i-1 , \ , i ] } m(s ) \ge l+ \inf_{s\in [ i-1 , \ , i ] } ( x_{s+u}-x_u ) ) $ ] . the rest of the argument will go through , by noting that the tail probability of @xmath363 } m(s)$ ] behaves like the tail probability of @xmath364 ( in the sens of `` @xmath32 '' ) , and that in the estimates of @xmath365 and @xmath365 , instead of using the exact gaussian distribution of @xmath366 , we can use the fact that the negative tail distribution of @xmath367 } ( w_{s+u}-w_u -\frac{s}{t } w_t)$ ] is bounded by the gaussian tail . the same argument applies to the @xmath1-bbm . for the cbrw , the situation is slightly different due to the fact that the space is discrete , but some obvious modifications to the argument readily make it rigorous . ] by the markov inequality , @xmath368 is bounded by the @xmath335-expectation of the number of particles located beyond @xmath369 at time @xmath96 ; this @xmath335-expectation is bounded by @xmath370 . of course , this bound is interesting only when @xmath371 ; otherwise , we use the trivial inequality @xmath372 . as a consequence , @xmath373 \ , \mathrm{d } u \ , .\end{aligned}\ ] ] with the notation @xmath341 , we have @xmath374 . assembling these pieces yields that @xmath375 \ , \mathrm{d } s \big ) \ , \mathrm{d } u \big\ } \ , , \ ] ] where @xmath376 ^ 2}{2\sigma^2 s } \big ) \big].\end{aligned}\ ] ] the random variable @xmath366 has the gaussian @xmath377 law . some elementary but tedious computations lead to the following conclusion : in case @xmath378 , the subtrees move forward faster than the usual speed @xmath379 ( i.e. , the integral of @xmath380 dominates ) , whereas if @xmath381 , these subtrees make no particular effort : they only need , in this case , to wait for the occasions when the red particle makes some fluctuations toward the left ( which happens with some frequency ) . letting @xmath350 and then @xmath382 ( in this order ) , we obtain : @xmath383 \ , , \label{premier - moment : lb}\ ] ] where @xmath308 is given in . this is the desired lower bound . the proof for the @xmath1-bbm is similar to the proof for the @xmath0-bbm , so we present only an outline , indicating the places where modifications are needed . we fix @xmath384 , and write @xmath385 . consider @xmath386 let us explain the definition of @xmath291 . the meaning of `` leans to the left '' is as for the @xmath0-bbm : the path of the particle lies in @xmath387 $ ] for all @xmath295 $ ] . by `` does not split much '' , we mean and @xmath388 are arbitrary : they can be replaced by @xmath389 and @xmath390 with two sufficiently large constants @xmath391 and @xmath392 . ] that the number of branchings ( from the path of the particle @xmath316 ) at each of the time intervals @xmath393 $ ] , for @xmath394 , is bounded by @xmath388 . by `` @xmath208-dominated '' , we mean the existence of a time @xmath320 $ ] such that either there are at least @xmath208 particles branching at time @xmath322 from the path of the particle @xmath316 lying in @xmath395 at some time @xmath396 $ ] , or there is a particle branching at time @xmath322 from the path of the particle @xmath316 lying in @xmath395 at some time @xmath397 $ ] ( if the interval is not empty ) . the event @xmath291 is the analogue , for the @xmath1-bbm , of the event @xmath290 in . the probability @xmath398 will serve as a lower bound for the probability of the large deviation event for the @xmath1-bbm , because by definition , @xmath291 implies the large deviation event for the @xmath1-bbm . write as before @xmath399 for the number of @xmath316 satisfying the conditions in @xmath291 . the main estimates for the @xmath1-bbm we are going to prove are : @xmath400 \ , , \label{n - bbm : e(et):lb } \\ \e[(\ # \widetilde{e}_t^{\mathrm{nbbm}})^2 ] & \lesssim & \exp\left [ - \big(\frac{v^2}{2\sigma^2 } -1 \big)t\right ] \ , , \label{n - bbm : e(et):ub}\end{aligned}\ ] ] where @xmath401 is defined in , and @xmath402 stands for a term not depending on @xmath4 such that @xmath403 . since @xmath404 can be as small as possible , and together with the cauchy schwarz inequality will yield the upper bound stated in for the large deviation function for the @xmath1-bbm . the proof of ( [ n - bbm : e(et):ub ] ) , which is identical for all the three models , is postponed to subsection [ 2nd_moment ] . the rest of this subsection is devoted to the proof of . writing @xmath315)$ ] again for the trajectory of the red particle @xmath316 , and @xmath405\}$ ] as in , we have @xmath406 where , for all @xmath320 $ ] , @xmath407 stands for the event that the subtree of bbm branched at time @xmath322 on the path of @xmath323 does not produce @xmath208 descendants going beyond @xmath323 at any time during @xmath408 $ ] and does not produce any descendant going beyond @xmath323 at any time during @xmath409 $ ] ( if the interval is non empty ) . here , @xmath326 is as before the atoms of a rate-2 poisson process , and for each @xmath242 , @xmath410 is the event that the number of atoms @xmath326 lying in @xmath393 $ ] is bounded by @xmath388 . once again , the essential contribution to the integral @xmath327 on the right - hand side comes from the neighbourhood of @xmath328 ; we write @xmath411 compared to the discussions for the @xmath0-bbm in the previous subsection , we have a new factor @xmath412 ; conditionally on the path of @xmath323 , the probability of @xmath413 is at least @xmath414 ( for some constant @xmath415 ) , which is greater than or equal to @xmath416 ( for some constant @xmath417 ) . as such , using again @xmath335 to denote the conditional probability given @xmath323 , we have @xmath418 we have @xmath419 which equals @xmath420\ , \mathrm{d } u\}$ ] . we can now carry out the same computations as in the case of the @xmath0-bbm , to see that @xmath421 \ , \mathrm{d } u \big\}\ , .\ ] ] [ this is the analogue for the @xmath1-bbm , of the inequality in . ] for all @xmath358 and @xmath422 , let us write @xmath423 for the number of particles lying in @xmath424 at time @xmath96 in an bbm independent of @xmath323 , and @xmath359 the maximal position at time @xmath96 of the bbm . by definition of @xmath407 , @xmath425 : \ , m(s ) \ge x_{s+u}-x_u \big ) + \\ & & + \p_x \big ( \exists s\in ( 0 , \ , t - u ] : \ , { \cal n}(x_{s+u}-x_u , \ , s ) \ge m \big ) \ .\end{aligned}\ ] ] we argue that this implies @xmath426 even though the rigorous meaning of the inequality should be formulated as in the paragraph following . the first probability expression on the right - hand side @xmath427 is bounded by @xmath428 $ ] . the probability @xmath429 was denoted by @xmath430 in section [ subs : n - bbm ] ( with @xmath208 in place of @xmath1 ) , and we have seen in that @xmath431 \ , .\ ] ] as such , @xmath432 \ , \mathrm{d } u \\ & & + \int_{\ln m}^{t - u } \big [ { \bf 1}_{\ { x_{s+u}-x_u < [ ( 2\sigma^2 s)(s-\ln m)]^{1/2 } \ } } \\ & & \hskip-40pt + { \bf 1}_{\ { x_{s+u}-x_u \ge [ ( 2\sigma^2 s)(s-\ln m)]^{1/2 } \ } } \exp\big ( s - \frac{(x_{s+u}-x_u)^2}{2\sigma^2 s } \big ) \big ] \ , \mathrm{d } u \ , .\end{aligned}\ ] ] with the notation @xmath341 ( where @xmath342 denotes again a standard brownian motion with variance @xmath3 , we have @xmath433 . the random variable @xmath366 has the gaussian @xmath377 law . as for the @xmath0-bbm , some elementary computations yield that , in case @xmath434 , the subtrees move forward faster than the usual speed @xmath379 , whereas if @xmath435 , these subtrees make no particular effort , and wait only for the occasions when the red particle makes some fluctuations toward the left . , this picture is probably inaccurate , and is only due to the fact that our upper bound for @xmath436 is not optimal . we conjecture that regardless of the value of @xmath60 , the subtrees never make any particular effort in the @xmath1-bbm , which would be in complete contrast with the @xmath0-bbm . ] letting @xmath350 and then @xmath437 , we obtain : @xmath438 \ , , \ ] ] where @xmath401 is defined in , and @xmath402 stands for a term not depending on @xmath4 such that @xmath403 . note that @xmath439 is negligible compared to @xmath440 . this yields the desired lower bound . the proof for the cbrw is along the lines of the proof for the @xmath0-bbm and for the @xmath1-bbm . let @xmath441 the meaning of `` leans to the left '' is as before : the path of the particle lies in @xmath387 $ ] for all @xmath295 $ ] . by `` does not coalesce '' , we mean that at no time during @xmath442 $ ] does the particle coalesce with any other particle . let @xmath443 denote the number of @xmath316 satisfying the conditions in @xmath292 . the main estimates for the cbrw are : @xmath444 \ , , \label{cbrw : e(et):lb } \\ \e[(\ # \widetilde{e}_t^{\mathrm{cbrw}})^2 ] & \lesssim & \exp\left [ -(f(v)-r)t\right ] \ , , \label{cbrw : e(et):ub}\end{aligned}\ ] ] where @xmath445 and @xmath76 are defined in and respectively , and @xmath446 stands for a term not depending on @xmath4 such that @xmath447 . equations and together with the cauchy schwarz inequality will yield the upper bound stated in for the large deviation function for the cbrw . the proof of ( [ cbrw : e(et):ub ] ) , which is identical for all the three models , is postponed to subsection [ 2nd_moment ] . the rest of this subsection is devoted to the proof of . writing @xmath315)$ ] again for the trajectory of the red particle @xmath316 , and @xmath405\}$ ] as in , we have @xmath448 where , @xmath449 is the probability that a random walk is at position @xmath450 at time @xmath4 , and for all @xmath320 $ ] , @xmath451 stands for the event that none of the particles in the subtree of bbm branched at time @xmath322 on the path of @xmath323 coalesces with the red particle . here , @xmath326 is as before the atoms of a rate-2 poisson process . for @xmath350 , @xmath452 where @xmath453 is as in ( [ fv0 ] ) , and the essential contribution to the sum on the right - hand side comes from @xmath454 ; we treat @xmath455 as an integer , and write @xmath456 the same computations as for the @xmath0-bbm ( see ) give that @xmath457 \ , \mathrm{d } u \big\}\ , .\ ] ] as for the @xmath0-bbm , we argue that @xmath458 where @xmath459 denotes the event that there exists a particle branched at time @xmath322 that coalesces with the red particle at time @xmath460 . [ for a rigorous meaning of this inequality , see the paragraph following . ] by the markov inequality , @xmath461 is bounded by the @xmath335-expected number of particles branched at time @xmath322 that coalesce with the red particle at time @xmath460 , and this @xmath335-expected number is approximately @xmath462 $ ] . on the other hand , @xmath463 . so @xmath464 \big ] \ , .\ ] ] taking expectation with respect to the law of the red particle , we arrive that @xmath465 \big ] \big\}\ , .\end{aligned}\ ] ] from here , we can use the computations presented at the end of section [ subs : heuristic_cbrw ] ( those leading to and ) . this yields . we use a common proof for and , for the @xmath0-bbm and the @xmath1-bbm , respectively . the proof of , for the cbrw , is along similar lines , and is omitted . it suffices to prove that @xmath466 if @xmath467 denotes the number of particles in the bbm ( without selection ) at time @xmath4 lying in @xmath288 and leaning on the left ( i.e. , whose trajectories are in @xmath468 $ ] for all @xmath295 $ ] ) . by definition , @xmath469 [ it is an inequality because the trajectories are not required to lean on the left , but only lie in @xmath470 $ ] at time @xmath106 , when they split . ] we have @xmath471 . it is convenient to split @xmath472 into the sum of @xmath473 and @xmath474 . since @xmath475 is non - decreasing on @xmath476 $ ] ( for all @xmath477 and @xmath478 ) , it follows for the first integral that @xmath479 using again our notation @xmath480 meaning that @xmath481}{\ln t } \le 0 $ ] . a few more lines of elementary computations show that the extra integral @xmath474 leads to an upper bound @xmath482 . therefore , we get the claimed upper bound for @xmath483.@xmath237 in the present paper we have tried to estimate the large deviation function for the position of the rightmost particle of three generalizations of the branching brownian motion and of the branching random walks , subjected to selection or coalescence mechanisms . we have proved the existence of a large deviation function ( [ ld - lbbm],[ld - nbbm],[ld - cbrw ] ) for positive deviations of the position of the rightmost particle of these three models : the @xmath0-bbm , the @xmath1-bbm and the cbrw . for large @xmath0 for the @xmath0-bbm , large @xmath1 for the @xmath1-bbm , and for small @xmath55 for the cbrw , we obtain upper bounds for these large deviation functions ( [ res - lbbm],[res2-lbbm ] ) , ( [ res - nbbm],[res1-nbbm ] ) , ( [ res - cbrw],[res1-cbrw ] ) . our results are limited to velocities larger than the typical velocity @xmath38 of the rightmost particle of the bbm or of the brw . our approach does not allow us to give lower bounds for these large deviation functions . it has been shown by duality @xcite that the coalescence branching random walk is closely related to the noisy version of the f - kpp equation . of course it would be interesting to see whether a direct analysis of the noisy f - kpp equation could confirm our result ( [ ld - cbrw],[res - cbrw],[1.22 ] ) . as recalled in the introduction , the f - kpp equation gives the evolution the the probability distribution of the position of the rightmost particle of a bbm and the large deviation function of this position is ( [ ld - bbm],[psi - bbm ] ) . one question we tried ( without success ) to solve and that we would like to raise in this conclusion is how to obtain the probability @xmath484 of finding @xmath1 particles on the right of position @xmath485 for @xmath486 . we could only get the following lower bound @xmath487 \label{conjecture}\ ] ] by considering the events where a single particle moves first a distance @xmath61 during a time @xmath96 and then gives rise to a regular tree which produces @xmath1 particles on the right of @xmath6 , i.e. @xmath488 \right]\ ] ] where @xmath61 and @xmath96 are related by @xmath489 the question of negative large deviations of the position of the rightmost particle ( as considered in @xcite for the cbrw ) would also be interesting to attack . in this case the result might strongly depend on whether one starts with a single particle or more than one particle ( in @xcite it was assumed that the initial number of particles is large and even infinite ) . if one starts with a single particle , one would get already for the large deviation function ( [ ld - bbm ] ) of the bbm @xmath491 ( where @xmath135 ) and the events which would dominate the contributions to the large deviations of the @xmath0-bbm , @xmath1-bbm and the cbrw would be rather different from those considered in the present paper . chauvin , b. , rouault , a. ( 1988 ) . kpp equation and supercritical branching brownian motion in the subcritical speed area . application to spatial trees . probability theory and related fields , 80 , 299 - 314 . doering , c. r. , mueller , c. , smereka , p. ( 2003 ) . interacting particles , the stochastic fisher - kolmogorov - petrovsky - piscounov equation , and duality . physica a : statistical mechanics and its applications , 325 , 243 - 259 . rouault , a. ( 2000 ) . large deviations and branching processes . proceedings of the 9th international summer school on probability theory and mathematical statistics ( sozopol , 1997 ) . pliska studia mathematica bulgarica 13 , 15 - 38 .

the presence of massive , compact , quiescent galaxies at presents a major challenge for theoretical models of galaxy formation and evolution . using one of the deepest large public near - ir surveys to date , we investigate in detail the correlations between star formation and galaxy structural parameters ( size , stellar mass , and surface density ) from to the present . at all redshifts , massive quiescent galaxies ( i.e. those with little or no star formation ) occupy the extreme high end of the surface density distribution and follow a tight mass - size correlation , while star - forming galaxies show a broad range of both densities and sizes . conversely , galaxies with the highest surface densities comprise a nearly - homogeneous population with little or no ongoing star formation , while less dense galaxies exhibit high star - formation rates and varying levels of dust obscuration . both the sizes and surface densities of quiescent galaxies evolve strongly from ; we parameterize this evolution for both populations with simple power law functions and present best - fit parameters for comparison to future theoretical models . higher - mass quiescent galaxies undergo faster structural evolution , consistent with previous results . interestingly , star - forming galaxies sizes and densities evolve at rates similar to those of quiescent galaxies . it is therefore possible that the same physical processes drive the structural evolution of both populations , suggesting that `` dry mergers '' may not be the sole culprit in this size evolution .

0906.4786

ample evidence now exists for the presence of massive galaxies at @xmath0 with little or no ongoing star formation , suggesting that a non - negligible fraction of the local early - type galaxy population was effectively in place only a few gyr after the big bang , and that massive red galaxies constitute a significant fraction of ( or possibly even dominate ) the stellar mass density at @xmath3 ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . although their masses and star formation rates ( sfrs ) are nominally similar , it has since become clear that these quiescent galaxies at @xmath4 have dramatically different structures compared to local ellipticals : specifically , the high - redshift galaxies sizes are much smaller ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? their surface densities likewise evolve strongly with redshift , as does the `` threshold '' surface density above which galaxies are predominantly quiescent @xcite . these phenomena pose several new challenges for studies of galaxy formation . indeed , if massive quiescent galaxies at @xmath3 were almost universally compact and dense , why are similar objects in the local universe practically nonexistent ( e.g. * ? ? ? * ; * ? ? ? several mechanisms for this transformation have been proposed , including a scenario whereby local early - type galaxies are built through a series of `` dry '' ( i.e. , gas - poor ) mergers of these compact progenitors @xcite . such mergers , being largely non - dissipative , would over time tend to `` puff up '' compact galaxies . some observational estimates of the major dry merger rate since @xmath5 indicate that these are indeed a factor in the buildup of massive galaxies @xcite , but the remnants of such major mergers would also have correspondingly larger masses and so it is unclear whether major dry mergers can entirely solve the size - mass discrepancy ( though they are likely to mitigate it to some degree ; e.g. * ? ? ? minor mergers may be a more important process both from a simple virial argument @xcite and because massive compact `` cores '' at high redshift may accrete material in their outskirts to form the massive ellipticals seen today @xcite . however , theoretical models have yet to converge on a definitive explanation some proposed models predict galaxy sizes dramatically different from those observed @xcite , while others match observed sizes at @xmath1 and @xmath6 @xcite but not at @xmath5 @xcite . whatever the underlying mechanism may be , observations of large samples of galaxies over a broad redshift range are crucial to adequately test current and upcoming models . however , many of the aforementioned galaxy structure studies employ hst - nicmos , which provides accurate size measurements but can typically image only one target at a time and is thus inadequate for observing very large samples . optical imagers are much larger , but at @xmath3 near - ir data are more robust for size determinations ; these bands fall in the rest - frame optical and therefore better trace the distribution of stellar mass than the rest - frame uv ( observed optical ) . furthermore , spectroscopy of quiescent galaxies ( both to determine their redshifts and confirm their quiescence ) requires large amounts of observing time due to the absence of strong emission lines ; again , while this can be performed for a few bright galaxies at @xmath3 with current facilities @xcite , it is infeasible for the large samples of fainter objects needed to provide a comprehensive view of the high - redshift galaxy population . fortunately , although spectroscopy and space - based imaging give the most precise picture of star formation and galactic structure for individual objects , the average properties of large samples can be accurately investigated with less `` expensive '' data . for example , both @xcite and @xcite successfully based their galaxy size measurements on ground - based imaging and photometric redshifts . although there are inherently greater uncertainties on the size and mass of any individual object than with nicmos imaging and spectroscopic redshifts , the _ average _ structural parameters of large populations can be accurately determined . furthermore , even without spectroscopy there are several ways to determine which galaxies are quiescent and which are actively forming stars . one straightforward method is to identify galaxies on the `` red sequence '' ( using , e.g. , the rest - frame @xmath7 color ) , which at low redshift is primarily composed of quiescent galaxies with strong 4000 breaks . however , star - forming galaxies containing large amounts of dust can have similarly red colors , so `` red and dead '' galaxy samples selected through this color cut are likely to be contaminated with substantial numbers of `` red and dusty '' starbursts with increasing redshift ( e.g. * ? ? ? * ) . with photometric observations in a suitable set of filters , the shape of the broadband spectral energy distribution ( sed ) can be used to distinguish between the sharp 4000 break characteristic of old stellar populations and the more gradual reddening caused by dust absorption ; the best - fit sfrs and dust column densities from sed modeling codes can then be used to define quiescent galaxy samples . empirical selection techniques using multiple rest - frame colors , such as the one employed by @xcite , are also effective at separating dusty from `` dead '' galaxies up to @xmath1 ( and even higher ; i. labb et al . , in preparation ) with little dependence on input templates or models . coupled with the large , deep multiwavelength surveys and improved photometric redshift techniques that have emerged in the past few years , these diagnostics provide an unprecedented view of star formation and its relation to other physical properties in massive galaxy populations . using the largest such publicly - available data set with sufficient depth , the ukidss ultra - deep survey ( and overlapping imaging from the subaru - xmm deep survey and spitzer wide - area extragalactic survey ) , we analyze the structural evolution of quiescent galaxies up to @xmath1 , specifically the correlations between star formation activity , galaxy size , and stellar mass surface density . first in [ sec_data ] we review the data and describe the size and mass measurements employed for this study . next in [ sec_ssfr ] we investigate how the specific star - formation rate ( ssfr ) is related to galaxy mass , size , and surface density , and in [ sec_rfsize ] present quantitative constraints on the evolution of these structural properties for both star - forming and quiescent galaxies . finally , the uniformity of the connections between star formation and structure is discussed in [ sec_uniformity ] along with some novel ways to interpret these correlations . ab magnitudes and a concordance cosmology ( @xmath8 , @xmath9 , @xmath10 ) are assumed throughout . the high - redshift galaxy sample analyzed herein is based on an updated version of the @xmath11-selected galaxy catalog described in detail by @xcite ; a brief summary follows . this catalog comprises near - infrared @xmath12 and @xmath11 data taken from the ukidss ultra - deep survey data release 1 ( uds ; * ? ? ? * ; * ? ? ? * ) , with overlapping @xmath13 imaging from the subaru - xmm deep survey ( sxds ; * ? ? ? * ) , and @xmath14 m data from the spitzer wide - area infrared extragalactic ( swire ) survey @xcite . using the positions and shapes of bright but unsaturated stars in the different bands , all mosaics were astrometrically matched and convolved to consistent point - spread functions ( psfs ) . using the sextractor software @xcite , `` color '' fluxes were measured in fixed 175 apertures and total @xmath11 fluxes from flexible elliptical @xcite apertures . in addition to the original @xcite catalog , we incorporate the @xmath15-band mosaic from the ukidss data release 3 ( to be described by s. j. warren et al . , in preparation ) with psf- and astrometric matching and flux measurements performed in the same manner as with the other bands . photometric redshifts were calculated with the @xmath15 data included using the publicly - available code eazy @xcite . the new photometric redshifts differed very little from the original ones presented by @xcite , but inclusion of the @xmath15 band should in principle improve both the @xmath16 values and stellar mass estimates at @xmath17 . at the adopted flux limit of @xmath18 , this catalog contains nearly @xmath19 galaxies . because of its depth and @xmath20deg@xmath21 area , the uds is well - suited for studies of large galaxy samples at @xmath22 ; at lower redshifts , however , the comoving volume probed by the uds is too small to provide meaningful samples ( and may also be severely affected by cosmic variance ) . thus , for comparison to the uds we also include the same @xmath23 sdss sample used by @xcite . to summarize , this sample was originally defined by @xcite , and @xcite performed minor corrections in the mass - to - light ratios ( accounting for the fact that the galaxy centers , where the spectroscopic fibers are placed , are typically redder than the outskirts ) as well as total flux corrections . the redshift range of @xmath24 was chosen to avoid selection effects on galaxies with high mass ( at the low-@xmath25 end ) and small angular sizes ( at higher redshift ) . size measurements were performed in a manner similar to that by @xcite . we use the galfit software package @xcite to fit @xcite models to all bright ( @xmath18 ) sources detected in the unconvolved ( i.e. , before psf matching , to ensure the highest possible angular resolution ) uds @xmath11 image . this image had a typical seeing fwhm of 07 . first a square postage stamp 21 pixels ( 42 ) on a side was made around each galaxy to be fitted . initial guesses for the effective radius @xmath26 , magnitude , ellipticity , and position angle were taken from the sextractor catalog . the psf used by galfit to deconvolve the galaxy images is the average psf over the image ( taken as the median of 300 bright , unsaturated stars ; also used in the psf - matching step described in williams et al . the srsic parameter @xmath27 was allowed to vary between 1 and 4 and the effective radius between 001 ( effectively a point source ) and 15 . since galaxies in general may exhibit color gradients , it is important to measure sizes at approximately the same rest - frame wavelength when comparing samples at different redshifts . the procedure described above was thus repeated for the unconvolved @xmath12 and @xmath15 images ( both exhibiting 08 seeing ) , and we define the size of a galaxy as its circularized effective radius ( @xmath28 ) at a rest - frame wavelength of 8000 , interpolated from the size measurements of the two adjacent bands . in other words , when rest - frame 8000 falls between two observed bands @xmath29 and @xmath30 , the interpolated size is @xmath31\ ] ] in practice the interpolated sizes are quite similar to those computed by , e.g. , simply taking the size measured from the observed band closest to 8000 . however , the interpolation method smoothes out possible discretization effects due to the choice of the measurement filter , and also in principle helps to mitigate failed size measurements in a given band . previous studies have shown that reliable galaxy sizes can be measured from ground - based data , provided the signal - to - noise ratio is sufficiently high and the psf across the image is stable ( e.g. * ? ? ? * ; * ? ? ? for instance , @xcite use simulations to investigate systematic effects , finding them to be minimal down to @xmath32fwhm(psf ) , though individual measurements exhibit a large degree of scatter . the effective radius is typically the most robust structural parameter measured ; other variables defining the profile shape ( e.g. sersi index and axis ratio ) are far more susceptible to systematic uncertainties . it is nonetheless instructive to investigate whether the specific characteristics of the uds data and our fitting techniques have introduced systematic biases ; we thus performed simulations analogous to @xcite to test our effective radius measurements . model galaxy profiles over a range of magnitudes @xmath33 , @xmath26 from 0.1 to 1 arcsec , and @xmath34 were created ; these were chosen to span typical galaxy angular sizes at @xmath35 . blank postage stamps ( containing only noise ) were then cut from random positions in the uds @xmath11-band image , and the model galaxies added to them . the identical fitting procedure used on the real images was then applied to 15000 such simulated postage stamps to derive the output effective radii . figure [ fig_sims ] shows the distribution of input and output effective radii , and the fractional difference between the two , as a function of input effective radius . brighter galaxies are plotted as larger points . as expected , relatively faint galaxies exhibit larger random uncertainties in their size measurements . while systematic deviations are seen for both the faintest and largest galaxies , these are relatively small ( with a median of @xmath36% ) compared to the random errors ; additionally , since we only include galaxies with @xmath37 in our analysis , objects with such large angular sizes are quite rare . these observed offsets are comparable to the offsets reported by @xcite ; most importantly , the average effective radii of galaxies with sizes @xmath38 appear to be reasonably reliably recovered . given the large scatter , however , it should be noted that individual measurements of galaxy sizes are highly uncertain even if ensemble averages are accurate . the simulations described above provide a realistic view of some of the random and systematic uncertainties in this analysis . however , to some extent these are idealized ; for example , the evident ability to measure effective radii as small as one - half of a pixel may be true in a numerical sense , but in real images additional effects such as psf variations will limit the precision that can be achieved . we therefore supplement the simulations with a set of purely empirical tests to better assess the reliability of the galaxy size measurements . specifically , we check the following three important assumptions : ( 1 ) the simple interpolation of galaxy sizes between different bands is robust ; ( 2 ) a single psf is sufficient to model the entire image ; and ( 3 ) the fitting box size of 42 is large enough to obtain reasonable fits . first , to test the interpolation assumption , we estimated the @xmath15-band sizes by averaging the measured @xmath12 and @xmath11 sizes ; the estimated and measured @xmath15 sizes matched very well with no noticeable systematic offset . the `` single psf '' assumption ( 2 ) was tested by generating four different psfs , each composed of bright stars from the four quadrants of the @xmath11 mosaic , and re - measuring the sizes of galaxies within the central @xmath39 of the @xmath11 mosaic . while there were small systematic differences in the sizes measured with the different psfs , these were on the order of 002 , which is effectively negligible compared to typical measured galaxy sizes of 02 - 10 . similarly , the median galaxy sizes show no evidence of systematic deviations with position in the image , indicating that the psf does not significantly vary . finally , the box - size assumption ( 3 ) was tested by re - fitting a subset of galaxies using postage stamp sizes of 52 and 62 ; the new sizes were fully consistent with those measured in the original 42 cutouts . the exception to this is the largest objects in this sample with @xmath40 ( about 10kpc radius at @xmath35 ) ; these objects exhibit 30% larger sizes when a larger fitting box is used , but such objects comprise only a tiny fraction ( @xmath41% ) of our @xmath37 sample and so do not affect our results . re - fitting the same subset of galaxies using different psfs and box sizes also gives an estimate of the measurement uncertainties ( i.e. , the degree to which a given size measurement is reproducible given different assumed input parameters ) . between 02 and 10 the measured sizes were consistent with each other within @xmath42% ; at 01 the fractional error rapidly increases to about 10%@xmath43% , and at smaller effective radii sizes can no longer be reliably inferred . to determine the efficiency to which galfit can distinguish between point sources and extended objects in these data , we also measured the radii of stars ( found via color selection ) using the same fitting parameters . interestingly , while essentially all of the stars have best - fit radii of zero ( as is expected for point sources ) , galfit measures non - zero radii for the vast majority of compact galaxies even those with best - fit radii much smaller than the psf . thus , these simulations and empirical tests confirm that effective radius measurements are likely free of major systematic effects down to @xmath44 . even if some offsets are present ( as seen in our simulations and by * ? ? ? * ) , they are comparable to or smaller than the systematic uncertainties on other parameters ( e.g. , in mass , redshift , and sfr determinations ) . we therefore conclude that measurements of the average sizes of large galaxy samples down to small radii are robust given these data . this is similar to what @xcite found by fitting their ground - based data and comparing it to advanced camera for surveys ( acs ) imaging . the stellar masses of galaxies in this sample were calculated with the _ fitting and assessment of synthetic templates _ ( fast ) code @xcite . this code uses @xmath45 minimization to fit @xcite stellar population evolution models to the observed broadband photometry . a @xcite initial mass function ( imf ) and solar metallicity were assumed ; masses were then shifted by a factor of @xmath46dex for consistency with the @xmath47 sloan digital sky survey ( sdss ) masses ( which were computed with a @xcite imf ) . we re - fitted a subset with @xcite models and both imfs to verify that this factor is essentially constant and does not vary systematically with galaxy color or mass . redshifts were fixed to the @xmath16 values derived by eazy , and a variety of evolutionary histories were allowed in the fitting , including exponentially - declining ( sfr@xmath48 ) models with @xmath49 ranging from @xmath50gyr . the computed masses are consistent with those calculated using other standard methods , for example using hyperz as a fitting engine and scaling the model amplitudes to estimate masses ( as done by , e.g. , * ? ? ? note that @xcite models result in galaxy masses approximately @xmath51dex _ lower _ than @xcite models ; however , unlike the imf correction factor , this discrepancy varies somewhat with galaxy mass . other parameters , such as the degree of dust extinction @xmath52 , average stellar population ages , exponential factor @xmath49 , and sfr are also computed by fast during the sed fitting procedure . most of these are highly uncertain when based on broadband data @xcite ; however , the _ ssfr _ ( ssfr@xmath53sfr@xmath54 ) is somewhat more robust . this quantity , taking the mass of the host galaxy into account , provides a more meaningful characterization of star - formation activity ( e.g. , an sfr of 10@xmath55/year is far more significant in a dwarf galaxy than a giant elliptical ) . furthermore , since the total sfr and mass both exhibit a similar dependence on the assumed imf , the ssfr is less sensitive to the choice of imf and input stellar population model(s ) than the absolute sfr . rest - frame colors , particularly the combination of @xmath7 and @xmath56 , are invaluable for distinguishing quiescent galaxies from those actively forming stars @xcite . although such colors can be derived directly from the templates used for photometric redshift fitting , by definition these colors are confined to the range of colors spanned by the template set ( which may be narrower than the intrinsic range of the galaxy population ) . instead , we interpolate rest - frame @xmath57 , @xmath58 , and @xmath12 fluxes directly from the observed photometry and photometric redshifts using the same method as @xcite but with the new @xmath15-band data included . this is accomplished using the interrest utility @xcite , which in turn is an implementation of the method described by @xcite . the reddest observed band in our data set is irac @xmath59 m , corresponding to rest - frame @xmath12 at @xmath60 ; reasonably robust fluxes can thus be interpolated up to approximately this redshift . in general , the mass completeness limit of a flux - limited galaxy sample depends strongly on color : since red galaxies have higher mass - to - light ratios , mass - limited red galaxy samples become incomplete at brighter magnitudes than for blue galaxies . completeness limits derived for strongly heterogeneous ( mixed red and blue ) samples are therefore dominated by blue galaxies , and such samples can still exhibit serious incompleteness effects among massive red galaxies . computing mass limits from red galaxies alone therefore provides a more conservative estimate of the overall sample completeness . we derive mass completeness limits for the uds sample by comparing the masses of red galaxies ( @xmath61 ) to their observed @xmath11 magnitudes . the 75% completeness limit is then defined as the mass at which about 25% of the galaxies in the sample fall below the adopted @xmath11 flux limit . this is performed for the @xmath62 redshift bin , thus ensuring that our analysis does not suffer from significant incompleteness effects up to @xmath1 . to ensure accurate size measurements , the size and surface density analysis is restricted to galaxies with @xmath63 ( approximately 1 magnitude brighter than the formal @xmath64 survey limit , also used in * ? ? ? * ) , corresponding to a mass limit of @xmath65 . our newly constructed sample allows us to study the mass - size relation and its evolution to @xmath1 . this relation , in three redshift bins from the uds , is plotted in figure [ fig_smassbw ] . to supplement the higher - redshift data , the same relation from sdss at @xmath47 is also shown . two notable trends stand out : first , there is a correlation between radius and mass , whereby the most massive galaxies out to @xmath66 have on average larger sizes than less massive galaxies ; this effect is weaker or nonexistent at @xmath67 . a second and related point is that there are effectively no massive , compact galaxies ( @xmath68m@xmath69 , @xmath70kpc ; in the lower right - hand region of the size - mass plot ) at @xmath6 , but at higher redshifts this area of parameter space becomes progressively more populated . this simply reflects the strong size evolution of massive galaxies found in several previous studies and introduced in 1 . @xcite and @xcite showed that the broad range in galaxy effective radii , at a given mass , is tightly correlated with ssfr out to @xmath71 . these authors concluded that ssfr is a tight function of stellar mass surface density ( @xmath72 ) , or velocity dispersion ( @xmath73 ; * ? ? ? * ) . put differently , these results imply that the size - mass relation is different for galaxies with different ssfrs . our sample is ideally suited to study this aspect at higher redshifts , as it covers an area @xmath74 times larger than that used by @xcite . we show in figure [ fig_smass ] how the ssfr depends on mass and size . it is clear that the dependence is very strong : at a given mass , galaxies with low ssfrs ( binned and plotted as red / yellow squares in this figure ) have small sizes , while those with large sizes have high ssfrs ( green / blue squares ) . the effect is not only strong at very low redshift , but extends to the highest redshift bins . at the same time , the sizes of galaxies with low ssfrs ( red in the figure ) are very small at high redshift , consistent with results obtained by others on smaller samples @xcite . table [ tab_smass ] lists the best - fit slopes and normalizations of the _ quiescent galaxy _ size - mass relations seen in figure [ fig_smass ] . for this we defined `` quiescent '' as galaxies exhibiting ssfr@xmath75 , where @xmath76 is the age of the universe at each galaxy s redshift , effectively picking out the red and yellow points in figure [ fig_smass ] ( see also [ sec_rfsize ] and the appendix ) . in each redshift interval the median log effective radius was calculated in mass bins of width 0.2dex , only including galaxies above the mass completeness limit at each redshift , and a least - squares fit was performed to these median points . for convenience , the normalizations of the power - law fits in table [ tab_smass ] are defined as the log effective radius of a typical galaxy with @xmath77 . uncertainties were computed using bootstrap resampling . note that any comparisons between the mass - size relations seen in sdss and uds should be treated with caution , as systematic differences in the size and mass determinations between these two samples may exist . nonetheless , strong evolution in the normalization of the size - mass relation is evident ; the slope of the relation , however , does not significantly change between @xmath78 . as the strong mass - size correlation of quiescent galaxies implies , ssfr does not depend strictly on size or mass alone , but rather on some combination of the two . the bottom panels of figure [ fig_smass ] instead show how ssfrs depend on stellar mass surface density ( @xmath79 ) and mass . galaxies with the highest surface densities have weak star formation , with very little dependence on mass . in addition , we can define a `` threshold surface density '' between quiescent and star - forming galaxies ( similar to that defined by * ? ? ? * ) at each redshift ; in figure [ fig_smass ] , the sharp color division ( between green and yellow ) at logssfr@xmath80 could be taken as a simple example of one such threshold . this clearly increases with increasing redshift . note that this particular threshold depends only weakly on mass ( i.e. is nearly horizontal in the @xmath81-mass plot ) , confirming that surface density is more fundamentally related to star formation activity in galaxies than mass @xcite . furthermore , this figure clearly shows at which epoch galaxies of a given size , mass , and density were forming stars : for example , galaxies with @xmath82m@xmath69 were generally star - forming at @xmath3 but by @xmath83 most of their star formation activity had ceased . however , the most compact massive galaxies ( the lower right - hand region of the radius - mass plot ) are quiescent at all redshifts considered here , and must have been quenched at @xmath0 . most intriguingly , the compact quiescent galaxies visible at high redshift do not maintain their small sizes or high surface densities to lower redshifts ; instead , the quiescent galaxies progressively evolve to larger radii and lower surface densities with decreasing redshift . this phenomenon is apparent from the uds data alone in figure [ fig_smass ] , but is most starkly illustrated by the @xmath47 sdss panel where essentially no massive , compact galaxies are visible . this re - illustrates the phenomenon pointed out by @xcite , @xcite , and @xcite , but with multiple redshift bins the depletion of these galaxies is now easily seen . since these galaxies are unlikely to be losing more than a small fraction of their mass ( via mass loss from evolved stars ; * ? ? ? * ) , size evolution must be the primary driver behind their disappearance . lcc sdss & @xmath84 & @xmath85 + @xmath86 & @xmath87 & @xmath88 + @xmath89 & @xmath90 & @xmath91 + @xmath92 & @xmath93 & @xmath94 the trends shown in figure [ fig_smass ] could in principle be affected by certain assumptions we have made . first , a minimum ssfr of @xmath95 yr@xmath96 has been imposed as an lower limit to what can be measured with sed fitting ; however , varying this limit only serves to shift the apparent `` quiescence threshold '' in this figure , without affecting the overall trends . again , the absolute radius , mass , and ssfr values may exhibit some systematic offsets between the sdss and uds samples due to different fitting and calculation methods , but the _ relative _ values within each frame ( and between the three uds redshift bins ) are generally consistent . it is also possible that some degeneracies may exist between the mass and ssfr estimates used here since both stem from the same template fits ( and hence are nt entirely independent quantities ) . to test the robustness of these results , we reproduced figure [ fig_smass ] using the swire 24@xmath97 m data in this field as a fully independent ssfr diagnostic . a relation between the ssfr and @xmath98 ( with a minor correction using @xmath99 to account for low - mass blue galaxies ) is presented in ( * ? ? ? * appendix ) . using the @xmath100 m data in place of the best - fit sed ssfr does not significantly change the results shown in figure [ fig_smass ] ( though it is somewhat noisier , due to the shallowness of the swire 24@xmath97 m data ) . the strong size and surface density evolution of the `` red and dead '' galaxy population , and the apparent `` disappearance '' of the most compact objects , presents an intriguing puzzle for observers and theorists alike . in order to take full advantage of the unprecedented sample size of the uds and provide constraints for present and future models , we now turn to quantifying these trends . figure [ fig_sizesd ] shows the sizes and surface densities of @xmath101 m@xmath69 galaxies as a function of redshift . this mass limit is equivalent ( given the difference in imf ) to the limits employed by @xcite and @xcite . quiescent galaxies ( selected via the quiescent red sequence method described in the appendix and by * ? ? ? * ) are shown in red , and star - forming galaxies are shown in blue . @xcite note that any proposed models for the size evolution of quiescent galaxies ( in particular those relying on dry mergers ) are additionally constrained by the co - moving mass and/or number densities of these galaxies as a function of redshift . the right - hand panel of this figure thus shows the number density evolution of quiescent galaxies with @xmath102 ; this was calculated at @xmath37 with galaxy counts in three redshift bins from the uds , and at @xmath47 by integrating the sdss value added galaxy catalog mass function tabulated by @xcite . the plotted uds number densities only include poisson uncertainties ; cosmic variance and selection effects have not been taken into account . since the uds sample is more than 99% complete at these somewhat higher masses , no correction to the best - fit number densities is necessary . lcccccc all & @xmath103 & @xmath104 & @xmath105 & @xmath106 & & + quiescent red sequence & @xmath107 & @xmath108 & @xmath109 & @xmath110 & @xmath111 & @xmath112 + non-(quiescent rs ) & @xmath107 & @xmath113 & @xmath114 & @xmath115 & & + ssfr@xmath75 & @xmath116 & @xmath117 & @xmath118 & @xmath119 & @xmath120 & @xmath121 + ssfr@xmath122 & @xmath123 & @xmath124 & @xmath125 & @xmath126 & & + @xmath127 & @xmath128 & @xmath129 & @xmath130 & @xmath131 & @xmath132 & @xmath133 + @xmath134 & @xmath135 & @xmath136 & @xmath137 & @xmath138 & & lcccccc @xmath139 & @xmath140 & @xmath141 & @xmath142 & @xmath143 & @xmath144 & @xmath145 + @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath151 & @xmath152 + @xmath153 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & @xmath159 the trends seen in these three quantities are quite well - fitted with power law functions of the form @xmath160 , where @xmath161 , 2.38 , and @xmath162 for @xmath163 , @xmath164 , and @xmath27 respectively ( best - fit @xmath165 and normalization values are also listed in table [ tab_fit ] for this and other samples ) . these fits were performed only to individual galaxies between @xmath166 and @xmath167 , excluding the sdss data and points that are major outliers in effective radius ( @xmath168 or @xmath169 ) . however , the sdss data point was included in the fit to the number density evolution . for comparison , @xcite find that effective radii of early - type galaxies with @xcite masses @xmath170m@xmath69 ( equivalent to our mass limit ) are @xmath171 times smaller at @xmath172 than @xmath6 and @xmath173 times smaller at @xmath1 . in both size and surface density the evolution they infer is somewhat slower than what we find ; this might be a result of different sample selection methods . they also derive a comoving _ mass _ density equal to @xmath174% of the local value at @xmath172 and @xmath175% at @xmath176 , while our _ number _ density evolution is somewhat shallower than this ( although roughly consistent between @xmath6 and 1 ) , implying significant mass growth in the existing quiescent galaxy population over these redshifts . the above fits include galaxies spanning a somewhat wide range of masses ( @xmath177 ) , but with this large sample it is possible to investigate the growth of galaxies as a function of stellar mass . table [ tab_massfit ] lists size evolution power - law indices ( @xmath178 ) and normalizations ( @xmath179 ) for these same quiescent and star - forming subsamples ( as well as the total sample ) in three narrower mass bins . a strong trend is immediately apparent : more massive galaxies undergo significantly faster size evolution with redshift . for the total sample @xmath178 steepens from @xmath180 to @xmath181 between stellar masses of @xmath182 and @xmath170@xmath183 , and from about @xmath184 to @xmath185 over the same mass interval for both the quiescent and star - forming subsamples . a similar trend was also noted by @xcite . although this could in principle be due to systematic galaxy mass overestimates at high redshift , such offsets would need to be quite large and redshift - dependent ( e.g. , @xmath186dex too high at @xmath3 , but correct at @xmath187 ) . we therefore conclude that there is strong evidence for differential size evolution with mass . as noted in the appendix , a number of other physically - meaningful definitions of `` quiescence '' are possible and may be simpler to define within certain models . we thus repeated the power law fits to other @xmath177 quiescent galaxy samples defined through the other two selection methods described in the appendix , i.e. based on low best - fit ssfr ( @xmath188 ) and red rest - frame colors ( @xmath61 ) . for completeness we performed the same fits to the size and surface density evolution of the complementary samples ( star - forming and blue ) , as well as all massive galaxies in this sample . errors on the power - law slopes were computed using bootstrap resampling . the results of these fits are tabulated , along with the reference `` quiescent red sequence '' sample , in table [ tab_fit ] . the sizes and surface densities of the different samples evolve at similar rates ( differing by no more than @xmath189 for any two `` equivalent '' samples , e.g. low - ssfr and quiescent red sequence ) . the observed structural evolution thus appears to be more or less independent of the exact technique used to select quiescent galaxies . interestingly , massive galaxies overall exhibit slower evolution than either of the star - forming or `` dead '' subsamples ; this is simply a consequence of the increasing fraction of massive quiescent galaxies at lower redshifts . from the preceding sections it is clear that star formation and galaxy structure are tightly coupled . while it was already established that such correlations exist @xcite , the order - of - magnitude larger galaxy sample provided by the uds allows us to not only observe and quantitatively measure said correlations , but also to investigate how uniformly the galaxy population at high redshift is described by them . in this final section we present some alternate tests of the connection between structure and star formation , with the specific goal of qualitatively understanding the degree to which these relations are `` universal . '' average seds are an invaluable tool for ascertaining both the underlying spectral shapes of large galaxy samples and the scatter ( or lack thereof ) in their seds . to determine whether dense galaxies are universally quiescent , we separate all @xmath35 , @xmath18 uds galaxies into four _ surface density _ bins and construct average rest - frame seds using their observed , de - redshifted photometric data points . figure [ fig_avgsed ] shows these photometric data ( normalized at 8000 ) in each bin ; median and 75% dispersion values of the individual galaxy points in bins of rest - frame wavelength are overplotted . to guide the eye , four @xcite stellar population models ( from bottom to top , one single - burst model with age 1gyr , and three constant star - forming with increasing levels of dust obscuration ) are also shown in each panel ; note that these are _ not _ fits to the data . this figure provides a nearly model - independent confirmation of the result shown in figure [ fig_smass ] : namely , that the densest galaxies exhibit an average spectral shape that is well - represented by an evolved stellar population , with remarkably low dispersion . interestingly , there appears to be a smooth progression between average surface density and dust properties : galaxies with the lowest @xmath190 are typically blue and have low dust obscuration , but with increasing surface density the seds become dustier . it is especially notable that this does not appear to be caused by the addition of progressively more quiescent galaxies to the blue galaxy population ; at least in the lowest two @xmath191 bins , the median seds are unambiguously within the `` dusty '' regime . thus , it appears that there is not only an anticorrelation between ssfr and surface density , but also a positive correlation between _ dust _ and surface density for star - forming galaxies . however , since the ssfr is a mass - normalized quantity and the ( qualitative ) dust obscuration seen in figure [ fig_avgsed ] is not , we caution that this last point may be entirely due to more massive galaxies containing more total dust . more importantly , the sed dispersion of the highest - density galaxies is at least a factor of 2 lower than for less - dense galaxies . this strongly implies that the highest density galaxies are overwhelmingly represented by a single model with relatively few outliers ; at lower surface densities , on the other hand , galaxies show larger dispersions in their spectra and thus span a range of spectral shapes . the comparatively low scatter in the high density galaxies further reinforces the idea that they are by and large quiescent : if they instead were predominantly red due to dust obscuration , they would most likely exhibit large dispersions comparable to the seds in the lower - density bins . a quiescent galaxy model also provides a somewhat closer representation of the 4000 break region than the @xmath192 model , but the models underpredict the observed uv flux in these galaxies , indicating that there may be some contribution from very dusty starbursts or low levels of residual star formation in the `` quiescent '' galaxies . up until now we have primarily investigated the star - formation ( and sed ) properties of galaxies as a function of structural parameters . we now consider the converse question : what are the structural parameters of galaxies as a function of their star - formation properties ? @xcite describe the use of the rest - frame @xmath7 vs. @xmath56 ( hereafter @xmath193 ) diagram as a powerful diagnostic of both star - formation activity and dust obscuration . figure [ fig_z12 ] shows this diagram for all @xmath35 galaxies in the uds . in short , quiescent galaxies fall in a discrete clump in the upper left - hand region of this plot ( above the dashed line ) , while star - forming galaxies form a `` dust sequence '' extending from the lower - left ( blue ) to upper - right ( red ) region . the utility of this plot as a star - formation diagnostic was confirmed using stacked @xmath100 m data ; a more thorough discussion is provided in the appendix . in each panel of figure [ fig_z12 ] galaxies have been binned and color - coded by their median ( a ) ssfr , ( b ) mass , ( c ) effective radius , and ( d ) surface density . in the first panel we confirm with our sed - based ssfr estimates what was previously shown by @xcite with @xmath100 m data the `` quiescent red sequence '' ( above the dashed diagonal line ) is indeed overwhelmingly populated by quiescent galaxies , while red galaxies that do not lie on this sequence have significant star formation , and are thus dusty starbursts . panels ( b ) and ( c ) illustrate how most quiescent galaxies have large masses and small effective radii , but some systematic variation is clearly present ( due to the size - mass relation ) . however , in the final panel it is evident that quiescent galaxies almost uniformly exhibit high _ surface densities _ ; there appears to be little variation in surface density with either mass or size along the quiescent red sequence . the converse also appears to hold true : the highest - density galaxies almost entirely fall within the `` quiescent '' region , confirming the result shown in figure [ fig_avgsed ] ( although some of the reddest star - forming galaxies exhibit densities approaching those of quiescent galaxies ) . though useful for illustrating the utility of the @xmath193 diagram , figure [ fig_z12 ] spans a fairly wide redshift range and may also contain systematic effects due to incompleteness at low masses . we therefore re - plot panels ( c - d ) , the size and surface density as a function of color , in figure [ fig_rfcol_sd ] for four redshift bins . a mass limit of @xmath65 ( the 75% red - galaxy mass completeness limit at @xmath67 ) is also imposed ; thus , this figure is complete only up to @xmath1 and we caution that the @xmath194 bin likely suffers from some incompleteness effects as well as larger uncertainties in the masses and rest - frame colors . the color - color plots shown in figure [ fig_rfcol_sd ] present a wealth of information about the structural evolution of massive galaxies , and indeed provide a novel view of the results presented thus far . first , at all redshifts , quiescent galaxies ( above the dashed line in each plot ) exhibit small radii and large surface densities , though their surface densities are far more uniform than their sizes . however , while most massive compact galaxies are quiescent , the same is not always true for massive , high - surface density galaxies : at @xmath3 there are many dense , star - forming galaxies , but the number of such objects drops dramatically with decreasing redshift , suggesting that galaxies become quiescent at progressively lower surface densities with time . even with this effect taken into account , it remains clear that the quiescent galaxy population undergoes dramatic structural changes from @xmath1 to 0 , so another mechanism must be at work to bring about this transformation . by combining size measurements of a very large @xmath11selected galaxy sample with sed fitting to determine sfrs and masses , we find that quiescent galaxies at all redshifts are far more compact than star - forming galaxies of comparable mass . similar results have been found by other authors at high redshift ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but with the larger sample size considered here , detailed investigations into the correlations between galaxy structure and star formation ( and their evolution with redshift ) are now possible . as shown in figure [ fig_smass ] , a well - defined mass - size relation for quiescent galaxies exists up to at least @xmath1 , but evolves to larger sizes with decreasing redshift . this figure also shows a clear anticorrelation between star formation activity and stellar mass surface density ( nearly independent of mass ) , and an evolving surface density `` threshold '' above which galaxies are predominantly quiescent , confirming the findings of @xcite . the effective radii of @xmath177 quiescent galaxies in this sample evolve as @xmath195 between @xmath196 , with the power - law index varying only slightly depending on the exact definition of quiescence ( see table [ tab_fit ] ) ; this is in excellent agreement with the @xmath197 `` upsizing '' observed by @xcite in a smaller sample from @xmath198 , and the factor @xmath199 size evolution since @xmath200 seen by @xcite . however , the slope of the size evolution varies strongly with _ mass _ , such that the most massive galaxies show the fastest mass evolution ( see table [ tab_massfit ] ) . @xcite found evidence for similar mass - dependent evolution of `` disk - like '' galaxies between @xmath201 ; the sample presented here confirms that this result extends to higher redshifts and applies to the entire quiescent galaxy population . the evolution in the uds sample also agrees well with other previous studies specifically focusing on early - type galaxies ; for example , @xcite find a factor of @xmath202 growth in size from @xmath172 to the present and a factor of @xmath203 since @xmath1 , which agrees with the observed evolution in our `` quiescent red sequence '' sample . on the other hand , we find all galaxies above @xmath204 exhibit size evolution with a power - law index of @xmath103 , slightly faster than the @xcite index of @xmath205 ; however , this may simply be due to the different redshift ranges considered . the total massive galaxy sample grows more slowly in size than either quiescent or star - forming subsamples of similar mass , highlighting the increasing dominance of `` dead '' ellipticals over starbursts at lower redshifts . in the highest redshift bin , @xmath67 , we see a population of ultra - compact , high - density quiescent galaxies that are no longer present in the local universe , or even ( for the most part ) at @xmath206 . studies with higher - resolution data find that the typical sizes of @xmath68m@xmath69 quiescent galaxies at @xmath207 are @xmath208kpc @xcite , and at @xmath0 no quiescent galaxies with @xmath209kpc are seen @xcite . although 1kpc is smaller than what we can confidently measure for individual galaxies , this size is consistent with our measurements . figure [ fig_smass ] does , however suggest the presence of quiescent galaxies at @xmath210 with somewhat extended radii ( @xmath211kpc ) . a few such objects are also seen at similar redshifts ( and with space - based imaging ) by @xcite ; thus , the difference may simply be due to mild evolution of these galaxies from @xmath207 to @xmath66 . follow - up observations of candidate _ extended _ quiescent galaxies at @xmath3 would confirm or rule out their existence , and perhaps shed some light on the mechanism behind the size evolution . the presence of ultra - compact massive galaxies with extreme surface densities ( and velocity dispersions ; * ? ? ? * ) at @xmath0 is puzzling , since the prevalence of such objects is already diminished substantially by @xmath212 @xcite and they effectively no longer exist at @xmath6 @xcite . major and minor mergers , in particular those involving only gas - poor galaxies ( `` dry mergers '' ) , likely play a significant role in both the mass and size evolution of these objects @xcite . however , as suggested by @xcite , other processes are also likely to be important , in particular the quenching of progressively larger star - forming galaxies at lower redshift . indeed , this may account for the seemingly incongruous result that the quiescent and star - forming samples each evolve faster than the overall galaxy population . in other words , if the compact , high - surface - density `` tail '' of the star - forming galaxy population is preferentially quenched at any given redshift , this will accelerate the apparent evolution of both subsamples . the strong size growth of both massive star - forming and quiescent galaxies by itself suggests that major gas - poor ( `` dry '' ) mergers are not the only process driving massive galaxies strong structural evolution . galaxies which are undergoing significant star - formation must contain substantial amounts of gas , and therefore by definition can not undergo dry mergers . however , the sizes and densities of massive star - forming galaxies are seen to evolve at a rate similar to the quiescent galaxies . it is therefore possible that at least one of the mechanisms acting on compact quiescent galaxies similarly affects star - forming galaxies . another proposed mechanism is adiabatic expansion , whereby mass loss from evolved stars decreases the potential well depth and therefore increases effective radii as galaxies age . as discussed by @xcite , however , such a mechanism does not seem feasible because only a small fraction of the total mass is expected to be lost in passively - evolving stellar populations between @xmath212 and @xmath6 . the size growth may also be explained by many minor mergers or the accretion of relatively low - mass satellites . @xcite point out that minor mergers are more efficient at increasing the radii of the primary galaxy than major mergers , per unit mass of the secondary galaxies . this scenario is also particularly attractive because it would depend only weakly ( if at all ) on whether or not the central galaxy is forming stars , and therefore could explain the size growth of both massive and quiescent galaxies . in a forthcoming paper we will investigate the role of major and minor mergers on massive galaxy evolution ( r. j. williams et al . , in preparation ) . although it is yet unclear what causes the observed size and surface density evolution of massive galaxies , the results presented herein illustrate how large , ground - based , near - ir surveys can provide large statistical samples , with sufficiently accurate size measurements , for comparison with the variety of theoretical models now under development . upcoming instruments like vista and wfc3 will greatly enhance the volume and quality of available data ; however , the proliferation of complementary observational results from existing surveys already present an important challenge to , and constraints on , these models . by applying accurate size measurements and broadband sed fitting to the largest photometric sample of massive galaxies from @xmath78 to date , we have investigated in detail the interplay between star formation , galaxy structure , and mass , and its evolution with redshift . our main conclusions are as follows : 1 . galaxies with low ssfrs follow a well - defined mass - size relation up to at least @xmath1 , and this relation moves to larger sizes at lower redshifts . the anticorrelation between stellar mass surface density and star formation activity is much stronger than the size - ssfr relation ; this confirms the result of @xcite that surface density and star formation are tightly connected at all redshifts . 3 . even with the far larger galaxy sample studied here , the densest , most massive galaxies seen at @xmath3 have essentially disappeared by @xmath206 . 4 . the sizes and surface densities of massive quiescent and star - forming galaxies evolve smoothly with time ( following simple power - law behavior in @xmath213 ) , with more massive galaxies exhibiting faster evolution . although galaxies with low surface densities exhibit a wide range of dust and star - formation properties , at the highest surface densities their seds are well - represented by a single quiescent galaxy model ; thus , these `` dense '' galaxies comprise a relatively homogeneous population . we thank natascha frster schreiber for helpful discussions and the anonymous referee for suggestions that helped improve the manuscript . r.j.w . acknowledges support from the netherlands organization for scientific research ( nwo ) and the leids kerkhoven - bosscha fonds . is supported by a nova postdoctoral fellowship and s.t . gratefully acknowledges support from the lundbeck foundation . the primary goal of this work is to investigate the structural properties and evolution of quiescent galaxies , defined as those falling below some ssfr threshold . in practice there are a number of ways to define such a sample , particularly when only broadband photometric data are available . in this appendix we describe the primary selection technique employed for this analysis ( the @xmath193 color - color selection described by * ? ? ? * ) , as well as two alternate methods considered in [ sec_rfsize ] . average seds of the quiescent galaxies selected via these three techniques are shown in figure [ fig_asedred ] . note that these three average seds appear quite similar , suggesting that the three techniques are effective at selecting quiescent galaxies ( though we caution that very dusty galaxies would not clearly stand out in these plots ; see the models in figure [ fig_avgsed ] ) . for the most part , galaxies which have ceased their star formation activity appear red due to their evolved stellar populations , and up to @xmath212 or higher ( depending on the data quality ) form a well - defined `` red sequence '' when plotted in a color - magnitude diagram . however , at higher redshifts progressively larger numbers of dusty starbursts are present , and these often exhibit optical colors which mimic `` red and dead '' galaxies , making the red sequence methods less reliable ( especially with only broadband photometric data ) . as shown by @xcite , incorporating rest - frame near - ir data into the analysis makes it possible to distinguish between dead and dusty galaxies . in particular , quiescent and star - forming galaxies occupy distinct regions of the rest - frame @xmath7 vs. @xmath56 ( @xmath193 ) color space ; with the ukidss uds and overlapping data , @xcite found that a `` _ quiescent _ red sequence '' is clearly visible ( and distinct from dusty star - forming galaxies ) in the @xmath193 distribution up to @xmath1 , and criteria based on these colors provide an effective physical basis for separating quiescent from star - forming galaxies . figure [ fig_z12 ] ( left ) shows the @xmath193 diagram for all galaxies between @xmath214 in the uds with @xmath18 . the quiescent galaxy selection criterion defined in @xcite is shown as a dashed line , where quiescent galaxies lie above and to the left of the line and star - forming galaxies are below . for reference , the criterion at @xmath35 is : @xmath215 at lower redshifts the criterion changes slightly , shifting upwards by 0.1 dex ( i.e. the @xmath216 becomes @xmath217 ) . additional constraints of @xmath218 and @xmath219 are imposed to prevent too many star - forming galaxies from scattering into the selection region . we note that some of the galaxies defined via this method as `` passive '' may nonetheless still have emission in the mid - ir from dusty star formation or active galactic nuclei ( agns ) . when we compare this technique to deep @xmath100 m imaging in the chandra deep field - south , we find that 4 out of 29 `` quiescent red sequence '' galaxies at @xmath1 have significant @xmath100 m emission . hence this classification is not a guarantee that a given galaxy has no star formation . on average , however , as shown by @xcite for this field , the @xmath100 m emission is very low for these galaxies . this empirical `` quiescent red sequence '' method is preferred to , e.g. , sed fitting because photometric redshifts ( as well as interpolated rest - frame colors based on these redshifts and observed fluxes ) are typically the best - constrained parameter in broadband sed fitting . by providing a directly empirical criterion , the rest - frame color separation is also less subject to template - dependent systematics that may affect sed - based ssfrs . we therefore adopt this color - color cut as our primary method for selecting quiescent galaxy samples . since the @xmath193 technique is relatively new and some theoretical models may better predict sfrs , we consider a sample ( listed as ssfr@xmath75 in table [ tab_fit ] ) based on an evolving cut in galaxy ssfrs . in this case , quiescent galaxies are defined as those exhibiting ssfrs less than 0.3/@xmath220 , where @xmath220 is the age of the universe for each galaxy at redshift @xmath221 , analogous to the quiescent galaxy definitions employed by @xcite and @xcite . the factor 0.3 is somewhat arbitrary , but is comparable to the ssfrs of red sequence galaxies in our sdss subsample . finally , one of the simplest criteria that can be defined is a single rest - frame color cut , i.e. assuming that all galaxies redder than a certain value are quiescent . as noted before this method is somewhat flawed because sufficiently dusty starburst galaxies can mimic the colors of truly `` red and dead '' galaxies . nonetheless , for completeness we construct a third sample with rest - frame @xmath222 irrespective of ssfr . bell , e. f. , et al . 2006 , , 640 , 241 bertin , e. , & arnouts , s. 1996 , a&as , 117 , 393 bezanson , r. , van dokkum , p. g. , tal , t. , marchesini , d. , kriek , m. , franx , m. , & coppi , p. 2009 , , 697 , 1290 brammer , g. b. , van dokkum , p. g. , & coppi , p. 2008 , , 686 , 1503 bruzual , g. & charlot , s. 2003 , , 344 , 1000 buitrago , f. , trujillo , i. , conselice , c. j. , bouwens , r. j. , dickinson , m. , & yan , h. 2008 , , 687 , l61 bundy , k. , et al . 2009 , , 697 , 1369 cappellari , m. , et al . 2009 , , 704 , l34 cenarro , a. j. & trujillo , i. 2009 , , 696 , l43 cimatti , a. , et al . 2008 , , 482 , 21 daddi , e. , et al . 2005 , , 626 , 680 damjanov , i. , et al . 2009 , , 695 , 101 de ravel , l. , et al . 2009 , , 498 , 379 feldmann , r. , carollo , c. m. , mayer , l. , renzini , a. , lake , g. , quinn , t. , stinson , g. s. , & yepes , g. 2009 , , 709 , 218 fontana , a. , et al . 2009 , , 501 , 15 frster schreiber , n. m. , et al . 2004 , , 616 , 40 franx , m. , van dokkum , p. g. , frster schreiber , n. m. , wuyts , s. , labb , i. , & toft , s. 2008 , , 688 , 770 hopkins , p. f. , bundy , k. , murray , n. , quataert , e. , lauer , t. , & ma , c .- p . 2009a , , 398 , 898 hopkins , p. f. , hernquist , l. , cox , t. j. , keres , d. , & wuyts , s. 2009b , , 691 , 1424 joung , m. r. , cen , r. , & bryan , g. l. 2009 , , 692 , l1 kauffmann , g. , et al . 2003 , , 341 , 54 kauffmann , g. , et al . 2006 , , 367 , 1394 khochfar , s. & silk , j. 2006 , , 648 , l21 khochfar , s. & silk , j. 2009 , , 397 , 506 kriek , m. , et al . 2006 , , 649 , l71 kriek , m. , van der wel , a. , van dokkum , p. g. , franx , m. , & illingworth , g. d. 2008a , , 682 , 896 kriek , m. , et al . 2008b , , 677 , 219 kriek , m. , van dokkum , p. g. , labb , i. , franx , m. , illingworth , g. d. , marchesini , d. , & quadri , r. f. 2009 , , 700 , 221 kron , r. 1980 , , 43 , 305 kroupa , p. 2001 , , 322 , 231 labb , i. , et al . 2005 , , 624 , l81 lawrence , a. , et al . 2007 , , 379 , 1599 lonsdale , c. j. , et al . 2003 , , 115 , 897 mccarthy , p. j. , et al . 2004 , , 614 , l9 maier , c. , et al . 2009 , , 694 , 1099 maraston , c. 2005 , , 362 , 799 marchesini , d. , et al . 2007 , , 656 , 42 naab , t. , johansson , p. h. , & ostriker , j. p. 2009 , , 699 , l178 peng , c. y. , ho , l. c. , impey , c. d. , & rix , h .- w . 2002 , , 124 , 266 rudnick , g. , et al . 2003 , , 599 , 847 rudnick , g. , et al . 2006 , , 650 , 624 salpeter , e. e. 1955 , , 121 , 161 sekiguchi , k. , et al . 2004 , baas , 36 , 1478 srsic , j. l. 1968 , atlas de galaxias australes ( cordoba , argentina : observatorio astronomico ) stutz , a. m. , papovich , c. , & eisenstein , d. j. 2008 , , 677 , 828 taylor , e. n. , et al . 2009 , , 694 , 1171 taylor , e. n. , et al . 2009 , , submitted ( arxiv:0907.4766 ) toft , s. , et al . 2007 , , 671 , 285 toft , s. , et al . 2009 , , 705 , 255 trujillo , i. , et al . 2006a , , 650 , 18 trujillo , i. , et al . 2006b , , 373 , l36 trujillo , i. , conselice , c. j. , bundy , k. , cooper , m. c. , eisenhardt , p. , & ellis , r. s. 2007 , , 382 , 109 trujillo , i. , cenarro , a. j. , de lorenzo - cceres , a. , vazdekis , a. , de la rosa , i. g. , & cava , a. 2009 , , 692 , l118 van der wel , a. , holden , b. p. , zirm , a. w. , franx , m. , rettura , a. , illingworth , g. d. , & ford , h. c. 2008 , , 688 , 48 van der wel , a. , bell , e. f. , van den bosch , f. c. , gallazzi , a. , & rix , h .- w . 2009 , , 698 , 1232 van dokkum , p. g. , et al . 2008 , , 677 , l5 van dokkum , p. g. , kriek , m. , & franx , m. 2009 , nature , 460 , 717 van dokkum , p. g. , et al . 2010 , , 709 , 1018 warren , s. j. , et al . 2007 , , 375 , 213 williams , r. j. , quadri , r. f. , franx , m. , van dokkum , p. , & labb , i. 2009 , , 691 , 1879 wuyts , s. , et al . 2007 , , 655 , 51 yang , x. , mo , h. j. , & van den bosch , f. c. 2009 , , 695 , 900 zirm , a. w. , et al . 2007 , , 656 , 66

hamiltonian bifurcations in the context of noncanonical hamiltonian matter models are described . first , a large class of 1 + 1 hamiltonian multi - fluid models is considered . these models have linear dynamics with discrete spectra , when linearized about homogeneous equilibria , and these spectra have counterparts to the steady state and hamiltonian hopf bifurcations when equilibrium parameters are varied . examples of fluid sound waves and plasma and gravitational streaming are treated in detail . next , using these 1 + 1 examples as a guide , a large class of 2 + 1 hamiltonian systems is introduced , and hamiltonian bifurcations with continuous spectra are examined . it is shown how to attach a signature to such continuous spectra , which facilitates the description of the continuous hamiltonian hopf bifurcation . this chapter lays the groundwork for kren - like theorems associated with the chh bifurcation that are more rigorously discussed in our companion chapter .

1308.3807

a common bifurcation to instability , one that occurs in so - called natural hamiltonian systems that have hamiltonians composed of the sum of kinetic and potential energy terms , happens when under a parameter change the potential energy function changes from positive to negative curvature . in such a bifurcation , pairs of pure imaginary eigenvalues corresponding to real oscillation frequencies collide at zero and transition to pure imaginary , corresponding to growth and decay . this behavior , which can occur in general hamiltonian systems and is termed the steady state ( ss ) bifurcation , is depicted in the complex frequency @xmath0 plane in fig . alternatively , the hamiltonian hopf ( hh ) bifurcation is the generic bifurcation that occurs in hamiltonian systems when pairs of nonzero eigenvalues collide in the so - called kren collision @xcite between eigenmodes of positive and negative signature , as depicted in fig . [ hbif]b . such bifurcations occur in a variety of mechanical systems @xcite ; however , hh bifurcations also occur in infinite - dimensional systems with discrete spectra . in fact , one of the earliest such bifurcations was identified in the field of plasma physics @xcite for streaming instabilities , where signature was associated with the sign of the dielectric energy , and this idea made its way into fluid mechanics @xcite . streaming instabilities were interpreted in the noncanonical hamiltonian context in @xcite , where signature was related to the sign of the oscillation energy in the stable hamiltonian normal form @xcite ( see [ stabnf ] below ) . the purpose of this chapter and its companion @xcite is to describe hamiltonian bifurcations in the noncanonical hamiltonian formalism ( see @xcite ) , which is the natural form for a large class of matter models including those that describe fluids and plasmas . particular emphasis is on the continuum hamiltonian hopf ( chh ) bifurcation , which is terminology we introduce for particular bifurcations that arise in hamiltonian systems when there exists a continuous spectrum . there also exist a continuum steady state ( css ) bifurcation , but this will only be mentioned in passing . a difficulty presents itself when attempting to generalize kren s theorem , which states that a necessary condition for the bifurcation to instability is that the colliding eigenvalues of the hh bifurcation have opposite signature , to systems with continuous spectra . this difficult arises because ` eigenfunctions ' associated with the continuous spectrum are not normalizable , in the usual sense , and consequently obstacles have to be overcome to define signature for the continuous spectrum . this was done first in the context of the vlasov equation in @xcite and for fluid shear flow in @xcite . given this definition of signature , it become possible in @xcite to define the chh , a meaningful generalization of the hh bifurcation . in the present chapter we motivate and explore aspects of the chh , which are picked up in our companion chapter @xcite . to this end we describe in secs . [ sec : discrete ] and [ sec : theories ] large classes of hamiltonian systems that possess discrete and continuous spectra when linearized about equilibria . these classes are noncanonically hamiltonian , as is the case in general for matter models in terms eulerian variables . for a general field variable @xmath1 that represents the state of such a system , a noncanonical hamiltonian dynamical system has the form _ t=\ { , h}= , where @xmath2 $ ] is the hamiltonian functional and @xmath3 the poisson bracket defined by \{f , g}=d [ ] . [ genpb ] in general one may consider a @xmath4 multicomponent theory , i.e. , @xmath5 , with @xmath6 being an operator that makes ( [ genpb ] ) a lie algebra realization on functionals ( observables ) . because the operator @xmath6 need not have the canonical form , may depend on @xmath1 , and may possess degeneracy this structure was referred to in @xcite as noncanonical . because of the degeneracy , the poisson bracket of ( [ genpb ] ) possesses casimir invariants @xmath7 $ ] that satisfy \{c , f}0 f. we refer the reader to @xcite for further details . in sec . [ sec : discrete ] we consider a class of 1 + 1 multi - fluid theories , that possess discrete spectra when linearized about homogeneous equilibria . the linearization procedure along with techniques for canonization and diagonalization , i.e. , transformation to conventional canonical form and transformation to the stable normal form , respectively , are developed . then , specific examples are considered that display both ss and hh bifurcations . in sec . [ sec : theories ] we consider a class of 2 + 1 theories . the class is described and the chh bifurcation for the particular case of the vlasov - poisson system is discussed . relationship to the results of sec . [ sec : discrete ] is shown by introducing the waterbag model , which is one way of discretizing the continuous spectrum , and motivates our definition of the chh bifurcation . finally , in sec . [ sec : conclu1 ] , we summarize and introduce the material that will be treated in @xcite . we first describe a class of hamiltonian theories of fluid type that have equilibria with discrete spectra . three examples are considered that demonstrate the occurrence of hamiltonian bifurcations like those of finite - dimensional systems . in the last example of sec . [ sssec : jeans ] , the hh bifurcation is seen to arise in the context of streaming . for our purposes here it sufficient to consider a class of 1 + 1 theories of hamiltonian fluid type . these theories have space - time independent variables @xmath8 , where @xmath9 , where @xmath10 , on which we assume spatial periodicity for dependent variables of fluid type , @xmath11 , where @xmath12 and @xmath13 are the density and velocity fields , respectively , with @xmath14 . these fields will be governed by a coupled set of ideal fluid - like equations generated by a hamiltonian with a noncanonical poisson bracket . the noncanonical poisson bracket for the class is obtained from that for the ideal fluid @xcite reduced to one spatial dimension , \{f , g}= _ = 1^m_dx ( - ) , [ mfpb ] where the shorthand @xmath15 is used and @xmath16 and @xmath17 are the usual functional ( variational ) derivatives ( see e.g. @xcite ) . we consider hamiltonian functionals of the following form : h[_,u_]=_=1^m_dx(12 _ u^2 _ + _ u _ ( _ ) + 12 _ ) [ mfb ] where the internal energy per unit mass , @xmath18 , is arbitrary but often taken to be @xmath19 where @xmath20 and the polytropic index @xmath21 are positive constants . the coupling between the fluids is included by means of a field @xmath22 that satisfies ( x , t)=_=0^m [ _ ] [ mfh ] where @xmath23 is a symmetric pseudo - differential operator , @xmath24= \int_{\mathbb{t}}dx\ , g\p[f]$ ] , and an arbitrary constant term @xmath25 has been included on the right hand side of ( [ mfh ] ) . from ( [ mfh ] ) we obtain = + h _ + = _ u _ [ funcdr ] where the enthalpy @xmath26 and the pressure of each fluid is given by @xmath27 . using ( [ funcdr ] ) with ( [ mfpb ] ) , gives & = & \ { _ , h}= - ( _ u _ ) , + & = & \{u _ , h}= - u _ u _ - p_/ _ - , which are a system of fluid equations coupled through @xmath22 alone . the noncanonical bracket of ( [ mfpb ] ) is degenerate and possesses the following casimir invariants c^_=_dx _ c^u_=_dx u _ , = 1 , , m. [ mcas ] these invariants satisfy @xmath28 for all functionals @xmath29 . the physical significance of these casimirs can be traced back to the liouville theorem of kinetic theory @xcite ( cf . sec . [ sssec : wb ] ) . because of the existence of the casimir invariants , the hamiltonian is not unique and , consequently , equilibria possess a variational principle since @xmath30 \ , \frac{\de h}{\de \psi}= { \j}[\psi ] \ , \frac{\de f}{\de \psi}\,.\ ] ] where @xmath31 . that is , @xmath32 . in the present context this amounts to @xmath33 , with lagrange multipliers @xmath34 , or = + h _ + + ^_=0 = _ u _ + ^u_=0 . [ mfe1 ] equations ( [ mfe1 ] ) have the equilibrium solution @xmath35 , and @xmath36 and @xmath37 . expansion around such equilibria give a linear dynamical system . because the equilibria of interest are homogeneous we can use the following expression en route to linearization : @xmath38 where the equilibrium constants @xmath39 could be absorbed into the sum by redefinition of the @xmath40 terms . for linearization we expand in the smallness of @xmath41 . functionals of @xmath42 can be mapped onto functions of the fourier components by insertion of the fourier series , i.e. , @xmath43 = f(\rho_0^{\al},\rho^{\al}_{\pm 1 } , \rho^{\al}_{\pm 2 } , \dots ; u_0^{\al},u^{\al}_{\pm 1 } , u^{\al}_{\pm 2},\dots)\ ] ] and this transformation ( for our purposes ) can be considered invertible upon using @xmath44 functional derivatives can also be expanded , e.g. , = _ k ( ) _ k e^ikx ( ) _ -k= 12 , [ fnder ] where the second equality follows from the chain rule ( see , e.g. , @xcite ) . using ( [ fnder ] ) and its counterpart for @xmath45 , the bracket of ( [ mfpb ] ) becomes = _ k_=1^m ( - ) [ fpb ] observe that in the poisson bracket of ( [ fpb ] ) the casimir invariants of ( [ mcas ] ) , the @xmath40 components of @xmath46 , have been removed , i.e. , the bracket has become nondegenerate in terms of the ostensible dynamical variables @xmath41 . geometrically , the choice of equilibrium selects the symplectic leaf on which the dynamics takes place . it remains to determine the hamiltonian . this is done by inserting the fourier expansions of @xmath46 into ( [ mfb ] ) . from this one obtains the full nonlinear dynamics in terms of fourier amplitudes , but since our interest is in bifurcations of the linear dynamics , we expand in the smallness of the amplitudes to obtain a quadratic form . although this can be done in general terms , we prefer to explore particular cases of equilibrium and stability in sec . [ ssec : mfexamples ] . however , before doing so , we makes some general comments about canonization and diagonalization . the bracket of ( [ fpb ] ) is not yet in canonical form . to canonize we rewrite the sums as follows : & = & _ k_=1^m [ pbraw ] + & = & _ k_m=1 ^ 2 m ( - ) , [ pbcan ] where in the second equality the canonical fields @xmath47 are obtained as particular real linear combinations of @xmath48 . thus modes , i.e. , degrees of freedom , are indexed by the wavenumber @xmath49 and an index @xmath50 that takes two values for every value of the species index @xmath51 . in terms of a choice of canonical fields , the hamiltonian for the linear dynamics is given by the second variation as @xmath52 and takes the form h_l=_k , _ m , n=1 ^ 2 m z_k^m _ mn^k z_^n , [ mfham ] where @xmath53 and the matrix @xmath54 depends on the specific values of the equilibrium parameters @xmath39 . given the system with hamiltonian in the form of ( [ mfham ] ) , it remains to effect a canonical transformation to a normal form . for example , if the system is linearly stable then there exists a canonical transformation @xmath55 , from real variables to real variables , where the hamiltonian becomes the following in terms of the new canonical coordinates h_l^s=12_k _ m=1 ^ 2 m _ k^m _ k^m ( ( p^m_k)^2 + ( q^m_k)^2 ) [ stabnf ] where the frequencies @xmath56 and the signature @xmath57 . thus , the stable normal form is just an infinite sum over simple harmonic oscillators . those for which @xmath58 are negative energy modes , stable oscillations with negative energy ( hamiltonian ) . it is important to emphasize that even though the energy is negative , the system is stable . for finite - dimensional systems , the method for constructing the canonical transformation to normal coordinates @xmath59 is treated in standard texts and this method carries over . however , when negative energy modes exist the method is complicated somewhat and , although well - known in hamiltonian dynamics lore , is not usually treated in physics texts . an accessible treatment is given in @xcite , where it is applied in a plasma physics context . in order to make the ideas discussed in secs . [ sssec : eqstab ] and [ sssec : candia ] more concrete we consider a few examples that demonstrate explicitly canonization , diagonalization , and hamiltonian bifurcations to instability in the context of multi - fluid models ; in particular , the hh bifurcation will emerge for particular modes indexed by @xmath60 , just as it appears in finite - dimensional systems . consider first the case of a single fluid with an equilibrium state given by @xmath61 some positive constant and @xmath62 . the linear hamiltonian is evidently h_l= 12 _ dx ( _ e ( u)^2 + c_s^2 ( ) ^2 /_e ) = _ k ( _ e |u_k|^2 + c_s^2 ( |_k|^2 /_e ) , where @xmath63 is the sound speed . the appropriate poisson bracket is is that of ( [ pbraw ] ) with a single @xmath51-term . with some thought canonization and diagonalization is possible in a single step , but we will proceed in a direct manner by assuming the canonical coordinates are @xmath64 with corresponding momenta @xmath65 a simple chain rule calculation takes ( [ pbraw ] ) into the following : = _ k _ m=1 ^ 2 ( - ) . [ cansnd ] observe in ( [ cansnd ] ) that indexing a degree of freedom by @xmath66 requires a multiplicity index @xmath50 . each mode , which is described by an amplitude and a phase , constitutes a single degree of freedom ; a single degree of freedom is thus two dimensional and , consequently each mode corresponds to two eigenvalues . for a stable degree of freedom these eigenvalues correspond to two frequencies , one the negative of the other . here we have multiplicity , the reason for which will become mentioned when we diagonalize . now , using @xmath67 valid with @xmath68 , in the hamiltonian @xmath69 gives @xmath70= \frac1{4}\sum_{k\in\n}\big ( \rho_e|q_k^1|^2 + 4k^2 c_s^2\ , |p_k^1|^2/{\rho_e } + c_s^2 |q_k^2|^2/{\rho_e } + 4k^2\rho_e\ , |p_k^2|^2 \big)\,.\ ] ] the normal form is achieved upon substitution of the following canonical transformation : @xmath71 i.e. , @xmath69 becomes h_l[q , p]= 12_k _ m=1 ^ 2 kc_s((q^m_k)^2 + ( p^m_k)^2 ) . [ soundnorm ] this is the sought after normal form where the frequency of all modes is @xmath72 as appropriate for sound waves . we close this example with a few comments . first , for a given wavelength as determined by @xmath49 there are in fact two modes , one that propagates to the right and one to the left . this is accounted for by the multiplicity index , @xmath50 . in obtaining this normal form we have assumed @xmath73 , which can be traced back to a property of @xmath74 and is in essence le chtelier s principle of thermodynamics , viz . , that pressure increases upon compression . if we had some exotic fluid for which this was not the case , then the the system would be unstable and the normal form of ( [ soundnorm ] ) would not be achievable . imagine that the equilibrium parameter @xmath61 can be varied and that at some critical value @xmath75 makes a transition from positive to negative . since a mode frequency @xmath76 it is evident that this transition happens at zero frequency and , consequently , is a ss bifurcation ( see fig . [ moreover , because of the multiplicity this is a degenerate bifurcation where , for each fixed @xmath77 , four pure imaginary eigenvalue collide at zero frequency and then transition to four pure real eigenvalues of growing and decaying pairs . the situation is completely degenerate since this happens for all @xmath77 values simultaneously . in sec . [ ssec : comp ] we will see that the hh bifurcation , as depicted in fig . [ hbif]b , can be transformed to a similar collision with four eigenvalues at zero frequency , but it differs in that after bifurcation one obtains the hamiltonian quartet , four eigenvalues with both real and imaginary parts , a situation that is sometimes called over stability . next we consider a simple one - dimensional multi - fluid plasma configuration consisting of two cold counterstreaming ion beams in a neutralizing isothermal electron background . a detailed linear , nonlinear , and numerical analysis of this problem , from a hamiltonian perspective , can be found in @xcite and we refer the reader there for further details . the dynamical system of interest in dimensionless form is given by & & + u _ + = 0 , + & & + ( _ u_)=0 , + & & ^2 = e^- _ + - _ - , [ ions : poi ] where @xmath78 labels each ion stream with velocity @xmath79 . here @xmath45 represents a dimensionless number density instead of mass density . equation ( [ ions : poi ] ) , poisson s equation , is a constraint and in principle the electrostatic potential can in principle solved as @xmath80 , so that the entire system is described in terms of the dynamical variables @xmath81 and @xmath82 . as usual the electric field is given by @xmath83 . thus , this system is of the class described in sec . [ ssec : mulitftheo ] , with ( [ ions : poi ] ) a specific case of ( [ mfh ] ) . it has the hamiltonian functional @xmath84 and poisson bracket of ( [ mfpb ] ) . homogeneous equilibria follow from @xmath85 for @xmath86 and @xmath87 , which are consistent with an equilibrium of ion streams of equal density and speed , @xmath88 that we assume for simplicity . thus , we have a one parameter family of equilibria controlled by @xmath89 . linearizing about this equilibrium state gives the following hamiltonian for the linear dynamics : h_l&=&12 _ dx ( ( u_+)^2 + ( u_-)^2 + 2u_e_+u_+ + & & -2u_e_-u_-+()^2+()^2 ) . observe , the sign of @xmath69 may be either positive or negative , depending on the perturbation ; thus we may have instability , or negative energy modes in the system . expansion in a fourier series as in sec . [ sssec : eqstab ] , including the expansion @xmath90 , and using the linearized poisson equation ( [ ions : poi ] ) gives @xmath91 , where @xmath92 . with this expression , the energy @xmath69 becomes h_l&=&_k ( |u_k^+|^2+|u_k^-|^2 [ del2f2 ] + & & + 2u_e(_k^+u_-k^+-_k^-u_-k^-+ ) + 2 ) , where c.c . denotes complex conjugate . under the transformation _ k^+&= & ( p_2- i q_1),u_k^+= ( p_1- i q_2 ) + _ k^-&= & ( p_4-i q_3 ) , u_k^-=(p_3-i q_4 ) , [ nvtoqp ] with @xmath93 and @xmath94 , the poisson bracket becomes that of ( [ pbcan ] ) with @xmath95 and the linear hamiltonian ( [ del2f2 ] ) becomes @xmath96 where ^k&=&. + ^k&=&thus , in terms of the canonical coordinates of ( [ nvtoqp ] ) the system is diagonal in @xmath77 , but it remains to transform the @xmath97 block structure , the part corresponding to the multiplicity , to normal form . for values of @xmath89 for which the system is stable , the diagonalizing canonical transformation is given explicitly in an appendix of @xcite . the reader is directed there to see how to obtain h_l^s&=&12_k ( _ k^+((p_k^1)^2+(q_k^1)^2 ) - _ k^-((p_k^2)^2+(q_k^2)^2 ) [ del2f4 ] + & & + _ k^+((p_k^3)^2+(q_k^3)^2 ) - _ k^-((p_k^4)^2 + ( q_k^4)^2 ) ) . evidently for each value of @xmath77 there exist four modes , two positive energy modes and two negative energy modes . the symmetry of the equilibrium facilitates the calculation of the frequencies , which are given by _ ^@xmath98>0 , [ freqs ] which can be obtained from the plasma fluid dielectric ( dispersion ) function ( k,)= 1 + 1k^2 - 12 ( 1(-k u_e)^2 + 1(+k u_e)^2 ) = 0 . [ dielectric ] from ( [ freqs ] ) it is evident that all bifurcations to instability occur through zero frequency as depicted in fig . [ hbif]a and in fact are degenerate ; i.e. , if we fix @xmath77 and vary @xmath89 , then there is a value of @xmath89 at which @xmath99 , the slow mode , vanishes and then becomes unstable with pure imaginary eigenvalues , two representing growth and two decay . thus , this is another example of a ss bifurcation that is forced to be degenerate because of the imposed symmetry . in the next section we will break this symmetry and obtain the hh bifurcation , but for variety we do so in a physically different , yet mathematically similar , context . the widely studied jeans instability occurs in newtonian gravitational matter models . for the present example , we suppose matter is governed by our 1 + 1 fluid model with two interpenetrating streams . we refer the reader to @xcite for background material and further details . the model is the same as that of sec . [ sssec : counterstreaming ] except poisson s equation is replaced by ^2 = _ + + _ _ _ , [ eq:3dgrav ] where we incorporate einstein s device of introducing a cosmological repulsion term , which in the newtonian setting amounts to introducing a negative constant gravitational mass of density @xmath100 . the sign change in ( [ eq:3dgrav ] ) accounts for gravitational attraction . the equilibrium for this case is similar to that of sec . [ sssec : counterstreaming ] , except we allow for asymmetry and , like the equilibrium of sec . [ sssec : sound ] , we allow for pressure in each stream . specifically , we have the equilibrium constant densities @xmath101 and @xmath102 , such that @xmath103 and @xmath104 , the two stream velocities @xmath105 and @xmath106 , chosen in opposite directions , and two sound speeds @xmath107 . upon scaling , these can be reduced to four independent equilibrium parameters : @xmath108 , @xmath109 , @xmath110 , and @xmath111 . from the results of secs . [ sssec : counterstreaming ] and [ sssec : sound ] we can immediately write down the linearized hamiltonian h_l&=&12 _ dx ( ^+_e(u_+)^2 + ^-_e(u_-)^2 + 2u^+_e_+u_+ - 2u^-_e_-u_- + & & + ( c_s^+)^2 ( _ + ) ^2 /^+_e + ( c_s^-)^2 ( _ -)^2 /^-_e - ( ) ^2 ) . fourier expansion and canonization proceeds in the same manner as in the previous examples . in the case where the equilibrium parameters indicate stability , then diagonalization can be shown to give a hamiltonian of the form of ( [ stabnf ] ) with @xmath95 . the frequencies are roots of the following ` diagravic ' function ( k , ) = 1 + + = 0 [ diagravic ] with two fast modes being positive energy modes and two slow modes being negative energy modes ( cf . figs . 1 and 3 of @xcite ) . in general all four modes are distinct , but if we symmetrize parameters as in sec . [ sssec : counterstreaming ] , then the quartic obtained from ( [ diagravic ] ) becomes biquadratic and is easily solved , indicating degenerate modes of each sign as before . evidently , this system possesses a rich parameter space , and various bifurcations to instability for various @xmath77-values are possible . in addition to the four parameters above , one can use also use @xmath77 as a control parameter : we have scaled the system size to @xmath112 , but upon reinstatement this translates into varying @xmath77 . the jeans instability is a long wavelength instability , and one can observe the transition to instability as @xmath77 decreases . this is immediate if @xmath113 and @xmath114 , in which case ( [ diagravic ] ) implies @xmath115 . using @xmath77 as the control parameter , as the wavelength is increased one sees the instability set in as a degenerate ss bifurcation . the situation is complicated with the presence of two streams , the subject of this subsection , and the hh bifurcation as depicted in fig . [ hbif]b is clearly present ( cf . fig . 2 of @xcite ) . this is quite generally the case for fluid systems with steaming equilibria . in sec . [ sssec : wb ] , we will see how multi - fluid streaming relates to the waterbag distribution of kinetic theory , and we will discuss explicitly the hh bifurcation in this context . it is evident from the discussion of sec . [ ssec : mfexamples ] that a requisite for determining a hh bifurcation is the identification of the energy for the linear system . in the context of noncanonical hamiltonians systems this naturally comes from second variation @xmath116 , the hamiltonian for the linear system . sometimes ` energy ' expressions are obtained by direct manipulation of the linear equations of motion , as done , for example , in the original mhd energy principle paper @xcite , but this procedure can obscure the notion of signature . for example , a system of two simple harmonic oscillators conserves @xmath117 for both signs and either might be obtained by manipulation of the equations of motion . the unambiguous sign for the correct energy is uniquely given by @xmath116 ; this is important because this sign can drastically affect the behavior of the system when dissipation or nonlinearity is considered . for example , a system with a negative energy mode can become unstable to arbitrarily small deviations from the equilibrium when nonlinearity is added . ( see cherry s example as described in @xcite . ) in the plasma literature , other definitions of energy are usually considered , e.g , in the context of streaming instabilities the dielectric energy , which is proportional to @xmath118 , where @xmath119 is the electric field amplitude , is incorporated . this expression was originally derived by von laue @xcite for the energy content in a dielectic medium by tracking the energy input due to an external agent . however , we have seen how it arises from @xmath116 , and only then can one be assured that it represents a quantity conserved by the linear dynamics . in fact , for our general multi - fluid model the @xmath120 takes the form @xmath121 , with a contribution @xmath122 from each fluid , e.g. , that for counterstreaming and jeans are ( [ dielectric ] ) and ( [ diagravic ] ) , respectively , ( also cf . ( [ wbvareps ] ) below ) and it can be seen in general that @xmath123 . for neutral modes embedded in the continuous spectrum of vlasov theory ( cf . @xcite ) the formula @xmath124 remains valid @xcite , but this formula is incorrect for excitation of the continuous spectrum as shown in @xcite , where the correct alternative formula was first derived , and the notion of signature for the continuous spectrum was defined . sometimes energy is defined in terms of the lagrangian displacement variable , as was done by manipulation of the linear equations of motion in @xcite . such expressions can also be obtained by expansion of an appropriate hamiltonian , @xmath125 . it was shown in @xcite that this procedure gives an expression that is essentially equivalent to @xmath116 . see @xcite for a recent general discussion in the context of mhd . in conventional kren theory , the signature is defined in terms of the lagrange bracket ( see , e.g. , @xcite ) . however it is a simple matter to see that this corresponds to the normal form definition @xcite , which follows by comparison of terms in the diagonalization procedure ( cf . @xcite ) . in @xcite it was argued that all these definitions of signature , that using , the dielectric energy , @xmath116 , @xmath125 , and the lagrange bracket , are essentially the same when they are meaningful . one ostensible difference between the hh and ss bifurcations is that the latter occurs at zero frequency . however , one can effect a time - dependent canonical transformation so that all four hh eigenvalues of fig . [ hbif]b collide at zero frequency . to see this consider one of the stable degrees of freedom , which has a contribution to the hamiltonian in action - angle variables @xmath126 given by @xmath127 , where @xmath128 depends on the bifurcation control parameter and takes the value @xmath129 at the bifurcation point . using the mixed variable generating function @xmath130 to transform to new canonical variables @xmath131 , f_2(_f,_f)= ( _ f- ^ * t ) , [ shift ] we obtain @xmath132 which amounts to moving into a rotating coordinate system with new hamiltonian @xmath133 thus , in the new frame , the frequency @xmath134 vanishes at the bifurcation point . at bifurcation the companion mode @xmath135 has the same value @xmath129 , consequently , a similar transformation will bring this mode to zero frequency at bifurcation . at first glance one might think this has made the hh bifurcation identical to a degenerate ss bifurcation , but the behavior of the two beyond the bifurcation point is different . the degenerate ss bifurcation transitions to two purely growing and two purely decaying eigenvalues , while the hh transforms to over stability , i.e. , it obtains a quartet structure immediately upon bifurcation . ( note , one could argue that the frame shift could be a function of control parameter , but with this line of reasoning all bifurcations could be make to look like ss bifurcations , even into the nonlinear regime . ) the frame shift of ( [ shift ] ) is identical to a galilean shift that can be done for fluid and plasma theories in order to bring modes to zero frequency at bifurcation . this artifice is used in the development of the single - wave model @xcite and will be considered in @xcite . the connection between degeneracy and symmetry is well - known and there is a very large literature on bifurcations in finite - dimensional hamiltonian systems with symmetry ( see , e.g , @xcite for an entryway ) . in our examples above we have seen , as one might expect , that this is also the case for infinite systems with discrete spectra . in fact , it is quite common for the dispersion relation to factor as a consequence of symmetry @xcite . however , systems with symmetry and continuous spectra are less well - studied , but counterparts exist , e.g. , the degeneracy of the ss bifurcation of jeans inability with @xmath114 of sec . [ sssec : jeans ] has a css counterpart when described by the vlasov system ( see sec . iiid of @xcite ) . now we turn to the general class of 2 + 1 hamiltonian mean - field theories in which the linear theory around equilibria possess a continuous spectrum . this is followed by the exposition of the two - stream instability in the vlasov - poisson equation , which is a standard example of the chh bifurcation . next we introduce the waterbag reduction of the vlasov - poisson equation and use it to connect the two - stream instability to kren - bifurcations in the corresponding waterbag model , linking this section to sec . [ sec : discrete ] . we begin with the class of 2 + 1 hamiltonian field theories introduced in @xcite , which have with a single dynamical variable , @xmath136 , a time - dependent density on the phase space variables @xmath137 . the density satisfies a transport equation , + [ f , ] = 0 , [ eq : den.eq ] where the bracket @xmath138= f_{q } g_{p } - g_{q } f_{p}$ ] is the poisson bracket for a single particle with particle energy @xmath139 , which will depend globally on @xmath140 . equation ( [ eq : den.eq ] ) is therefore a mean field theory , where @xmath140 is a density of particles in phase space that generates @xmath139 and is advected along the single particle trajectories that result from @xmath139 . the resulting equations are typically quasilinear partial integrodifferential equations . we assume that the particle energy arises from a hamiltonian functional of the form @xmath141= h_1 + h_2 + h_3 + \dots$ ] , where generally @xmath142 is the @xmath143-point energy , e.g. , h_1 [ ] = _ d^2z h_1(z ) f(z ) , h_2 [ ] = 12 _ d^2z _ d^2z f(z ) h_2(z , z ) f(z ) , with @xmath144 and @xmath145 being interaction kernels . here we will only consider hamiltonian systems with up to binary interactions , and we will assume that that @xmath146 possesses the symmetry @xmath147 . if @xmath139 is obtained from the field energy by functional differentiation : @xmath148 then @xmath141 $ ] is a constant of motion for ( [ eq : den.eq ] ) . equation ( [ eq : den.eq ] ) with @xmath149 is a hamiltonian field theory @xcite in terms of the noncanonical lie - poisson bracket of @xcite \{f , g}=_d^2z f . [ eq : pb ] this bracket depends explicitly upon @xmath140 , unlike usual poisson brackets that only depend on ( functional ) derivatives of the canonical variables . the bracket of ( [ eq : pb ] ) is antisymmetric and satisfies the jacobi identity , though it is degenerate , unlike canonical brackets . the equations of motion may be written : = \{f , h } = - = - [ f , ] , [ eq : eom0 ] where @xmath150 . as mentioned in sec . [ sec : intro ] , degeneracy of the poisson bracket gives rise to casimir invariants , quantities that are conserved for _ any _ hamiltonian . for the bracket of ( [ eq : pb ] ) the casimir invariants are @xmath151= \int_{\calz } d^2z \ , \calc(f)$ ] , where @xmath152 is an arbitray function . the existence of casimir invariants leads to a foliation of phase space ( in this case a function space ) with symplectic leaves , which are the level sets of the casimir invariants and which inherit a symplectic structure from the lie - poisson bracket . the evolution of @xmath140 is restricted to one of these symplectic leaves , and the equations on a single leaf are canonical . in addition to the casimir invariants and the total energy , there may be conserved momenta @xmath153 $ ] generally arising from translation symmetries of the interaction kernels @xmath154 . the system conserves momentum if there exists a canonical transformation of the phase space @xmath155 , @xmath156 , such that in the new particle coordinates @xmath157 , the interactions @xmath144 , @xmath145 , etc . have upon composition with @xmath158 one of the following two forms : h_1z= = |h_2(i , i,|@xmath159-@xmath159| ) [ eq : vpform ] or h_1z= 0,h_2(z , z ) = |h_2(|i - i|,|@xmath159-@xmath159| ) . [ euform ] in the first case @xmath160= \int_{\calz}\!\!d^2z \ , i\ , f(z)\,.\ ] ] is conserved , while in the second case we have two kinds of translation invariance and thus two components of the momentum @xmath161= \int_{\calz}\!\!d^2z \ , i\ , f(z ) \ , \quad { \rm and } \quad p_2[f]= \int_{\calz}\!\!d^2z \,\ , { \ensuremath{\theta}}\ , f(z)\,.\ ] ] these momenta can be very useful ( cf . , e.g. , @xcite ) , but they will not be discussed further here . equilibrium states have @xmath140 a function of the single particle constants of motion only , i.e. , the single particle energy @xmath139 and possibly momenta . the example treated here has an equilibrium that only depends on @xmath162 , where @xmath163 are the action - angle variables corresponding to a given @xmath139 . for this reason we set @xmath164 and then when a choice of @xmath165 is made , @xmath166 represents the main dynamical variable . the phase space is @xmath167 , i.e. , periodic in @xmath168 , and @xmath169 where here @xmath170 . upon substitution of @xmath171 into @xmath139 , both of the forms of ( [ eq : vpform ] ) and ( [ euform ] ) can be written as follows : @xmath172=\cale[f_0 ] + \cale[\ze]=:h(i ) + \phi({\ensuremath{\theta}},i)\,,\ ] ] with @xmath173 where @xmath174 and @xmath175 are determined by @xmath176 and @xmath146 . thus the governing equation is _ t + [ f_0 , ] + [ , h+]=0 , [ eq : eom ] where @xmath138=f_{\th } g_{i } - g_{\th } f_{i}$ ] and @xmath177 . equation ( [ eq : eom ] ) will serve as a starting point for our subsequent linear analyses . all of the models described in sec . [ ssec : badass ] possess chh bifurcations ; however , here we will concentrate on the vlasov - poisson system . first we describe it , then we make connection to the multi - fluid results of sec . [ ssec : mulitftheo ] and in this way relate the chh to the ordinary hh bifurcation . the vlasov - poisson equation arises out of ( [ eq : den.eq ] ) through definition of the single particle energy @xmath139 and potential @xmath178 , where @xmath179 and f_t=-[f,]=-pf_q+_q f_p _ qq=1-_f dv . the interaction kernels for this model are : @xmath180 and @xmath181 . the function @xmath140 represents the density of a positive charge species in phase space , under the assumption that there is neutralizing background with uniform negative charge density . the particles interact with each - other through electrostatic forces , which are included by the poisson equation . under the identification @xmath182 , @xmath183 , we recover ( [ eq : vpform ] ) . arbitrary functions of @xmath184 alone , @xmath185 , form an important class of solutions to this model , the spatially homogeneous equilibria . the analog of ( [ eq : eom ] ) is : @xmath186 upon linearizing the vlasov - poisson system around a homogeneous stable equilibrium , i.e. , dropping the nonlinear term @xmath187 , and then supposing @xmath188 to eliminate @xmath189 ( which is @xmath159 in the previous language ) in lieu of the wavenumber @xmath77 , @xmath190 becomes our dynamical variable that satisfies @xmath191 which simplifies to the following integro - differential equation for @xmath192 : _ k , t=- i kp_k+ if_0 k^-1 _ d|p _ k(|p , t)=:-_k _ k. [ zek ] here we have introduced the time evolution operator @xmath193 , whose spectrum under changes in @xmath165 we will study to understand the chh bifurcation . the linearized equations inherit a hamiltonian structure . because of the noncanonical form , linearization requires expansion of the poisson bracket as well as the hamiltonian . in terms of the variables @xmath192 and @xmath194 the hamiltonian is : @xmath195 = \frac{1}{2}\sum_{k\in \n}\left(-\int_{\mathbb{r}}\!dp\ , \frac{p}{f_0'}\ , |\ze_k|^2 + @xmath196 observe from ( [ pbk ] ) that @xmath49 and thus @xmath192 and @xmath194 are independent variables that are almost canonically conjugate . thus the complete system is @xmath197 whence it can be shown directly that the spectrum is hamiltonian . now we consider properties of the evolution operator @xmath198 defined by ( [ zek ] ) . we suppose @xmath192 varies as @xmath199 , where @xmath200 is the frequency and @xmath201 is the eigenvalue . for convenience we also use @xmath202 , where we can view @xmath203 by varying the system size . the system is said to be spectrally stable if the spectrum of @xmath198 is less than or equal to zero or the frequency is always in the closed lower half plane . since the system is hamiltonian , the question of stability reduces to deciding if the spectrum is confined to the imaginary axis . the solutions of a spectrally stable system are guaranteed to grow at most sub - exponentially . the operator @xmath193 is the sum of a multiplication operator and an integral operator , and the multiplication operator causes the continuous spectrum to be composed of the entire imaginary axis except possibly for some discrete points . instability comes from the point spectrum . the linearized vlasov poisson equation is not spectrally stable when the time evolution operator has an element of the point spectrum away from the imaginary axis ( implying a doublet or quartet of modes with non - trivial real part ) . the point spectrum are the roots of the plasma dispersion function : @xmath204 here @xmath205 . the one - dimensional linearized vlasov - poisson system with homogeneous equilibrium @xmath165 is spectrally unstable if for some @xmath206 and @xmath207 in the upper half plane , the plasma dispersion function vanishes . using the nyquist method that relies on the argument principle of complex analysis , penrose @xcite was able to relate the vanishing of @xmath208 to the winding number of the closed curve determined by the real and imaginary parts of @xmath209 as @xmath207 runs along the real axis . such closed curves are called penrose plots . the crucial quantity is the integral part of @xmath209 as @xmath207 approaches the real axis from above : @xmath210(u)+if_0'(u)\,,\ ] ] where @xmath211 $ ] denotes the hilbert transform , @xmath211=\frac{1}{\pi}\dashint\!dp\,{f_0'}/({p - u})$ ] , where @xmath212 indicates the cauchy principle value , leading to the following expression for the contour , parametrized by @xmath213 , in the complex plane : @xmath214(u)-i \pi k^{-2 } f_0'(u)\,.\ ] ] the image of the real line under this mapping is the penrose plot , and its winding number about the origin is the number of members of the point spectrum of @xmath215 in the upper half plane . figure [ gaus1 ] shows the derivative of the distribution function , @xmath216 , for the case of a maxwellian distribution @xmath217 and fig . [ penrosegaus ] shows the contour @xmath218-if_0'(u)$ ] that emerges from the origin in the complex plane at @xmath219 , descends , and then wraps around to return to the origin at @xmath220 . from this figure it is evident that the winding number of the @xmath208-plot is zero for any fixed @xmath221 , and as a result there are no unstable modes . here we take the value of @xmath77 to be fixed . for a maxwellian distribution function . ] penrose plots can be used to visually determine spectral stability . as described above , the maxwellian distribution function is stable , as the resulting @xmath222-plot does not encircle the origin . however , it is not difficult to construct unstable distribution functions . in particular , the superposition of two displaced maxwellian distributions , @xmath223 , is such a case . as @xmath224 increases the distribution goes from stable to unstable . this instability is known as the two - stream instability . figures [ 2max]a and [ 2max]b demonstrate how the transition from stability to instability is manifested in a penrose plot . at the bifurcation point the penrose plot crosses the origin , indicating the vanishing of the dispersion relation on the real axis and therefore the presence of a member of the point spectrum . this eigenmode will be stable because @xmath225 , and will be embedded within the continuous spectrum . thus , the two - stream instability is an example of the chh bifurcation . the description of the chh bifurcation requires that one be able to assign an energy signature to the continuous spectrum . because eigenfunctions associated with continuous spectra are not normalizable , this requires some delicacy . this was first done in the vlasov context in @xcite , where a comparison to the usual energy signature for discrete modes was given , and followed by a rigorous treatment of signature in @xcite . in the context of shear flow , signature was defined in @xcite , in magnetofluids in @xcite , and for the general system described in the present subsection in @xcite . a rigorous version of kren s theorm for the chh bifurcation was given in @xcite . we shall give a general description of this energy signature for the continuous spectrum in @xcite , but we motivate it here first by treating the analogous version of this instability in the context of the waterbag model , which will have the advantage of only possessing a discrete spectrum . one important feature of the system ( [ eq : den.eq ] ) is that its solution is a symplectic rearrangment of the initial condition @xmath226 , i.e. , its solution has the form @xmath227 where @xmath228 is a canonical transformation . the rearrangement comes from the solution of the ordinary differential equation for a single particle in the self - consistent potential @xmath178 . this implies that the level set topology of the initial condition is preserved , which can be leveraged to simplify the equations in the case of certain types of initial conditions . one such simplification is known as the waterbag reduction ( see , e.g. , @xcite ) , in which it is assumed that the initial condition @xmath229 is a sum of characteristic functions . this property is preserved under composition with the symplectic map @xmath230 , so that the solution remains a sum of characteristic functions . the equations simplify to equations for the locations of the contours separating different regions of constant @xmath140 . piecewise constant initial conditions lead to a fluid closure that is exact for waterbag initial conditions , and the @xmath231 theories in the previous section can be seen to arise from such an ansatz . we will exploit the reduction by using a layered waterbag or onion - like initial condition to closely approximate a continuous distribution function that undergoes the bifurcation to linear instability we are interested in . in this way we will be able to connect the hh bifurcation with the chh bifurcation which we describe later . we begin by assuming @xmath140 to be piecewise constant between @xmath232 curves @xmath233 , i.e. , @xmath234 where @xmath235 is a positive constant . the equations for the curves @xmath236 come from the equations of single particle motion for a particle at @xmath237 : @xmath238 and this system is hamiltonian , with hamiltonian function being the classical energy : @xmath239 here @xmath240 , is the change in the distribution function when crossing the @xmath241th waterbag layer . the poisson bracket is similar to those seen in hamiltonian fluid theories @xcite : @xmath242 the equilibria of the waterbag model that we are interested in studying are charge neutral and spatially homogeneous , @xmath243 constant , such that the electric potential @xmath244 . we chose such a state and linearize about it , yielding the equations of motion @xmath245 moving to fourier space and eliminating the dependence on @xmath189 in favor of the wavenumber @xmath77 gives @xmath246 the equations of motion for the fourier coefficients . in terms of the fourier coefficients , the hamiltonian of the linearized system is @xmath247 here the term @xmath248 arises from the term @xmath249 in the linearized vlasov equation , which indicates the signature of the continuous spectrum . the bracket is the bracket of the original nonlinear system written in terms of the fourier modes : @xmath250=\sum_{k\in \n}\sum_{\al}\frac{ik}{\delta f_{\al}}\ , \left ( \frac{\p f}{\p p_k^{\al}}\frac{\p g}{\p p_{-k}^{\al } } - \frac{\p g}{\p p_k^{\al}}\frac{\p f}{\p p_{-k}^{\al } } \right)\,.\ ] ] this bracket is nondegenerate and therefore the system is nearly canonical in terms of the new variables . in particular , for a given pair @xmath251 , the linear equations form a finite - dimensional canonical hamiltonian system with scaling similar to that of sec . [ ssec : mfexamples ] . the dispersion relation for this system , for a given wave number @xmath77 , and @xmath205 , is derived by multiplying the @xmath252th equation by @xmath253 and summing , which is analogous to that for the vlasov system , @xmath254 this dispersion relation can be analyzed graphically in terms of @xmath207 . there are poles of the dispersion function where @xmath255 . for @xmath256 , the dispersion function always has a zero if @xmath257 has the same sign as @xmath253 , because @xmath258 will converge to the opposite value of infinity at each end of the interval . therefore there will be at least one zero in each interval that has this property . in intervals where @xmath257 and @xmath253 have different signs there are either @xmath259 zeros or an even number of zeros , because @xmath260 must converge to the same value of infinity . the reader may have noticed a similarity between the above formulas , and the multi - fluid formulas of sec . [ ssec : mulitftheo ] . in fact , the waterbag models are examples of multi - fluid models , which are thus exact fluid closures of the vlasov - poisson system . this can be seen by writing the waterbag model in terms of new variables @xmath45 and @xmath79 given by @xmath261 where @xmath262 is a fluid density , and @xmath263 is a fluid velocity . under this change of variables the equations governing the waterbag model take the following form : & & _ , t+ ( u _ _ ) _ q = 0 , _ qq=- _ _ + & & ( _ u_)_t+ ( u_^2_+ _ ^3f_^2/12)_q=-__q . evidently , under the identification @xmath264 ( or @xmath265 , the above equations are identified as a multi - fluid hamiltonian system . the dispersion function can also be rewritten in terms of the new variables , so that it resembles that analogous expressions ( i.e. , the diagravic or dielectric functions of the multi - fluid section ) . after linearizing around an equilibrium state with @xmath266 , @xmath267 , and then performing some algebraic manipulations yields ( k,)= 1 - _ , [ wbvareps ] where @xmath268 is a thermal velocity that measures the width in velocity space of a waterbag . thus bifurcations in the waterbag model , the vlasov - poisson system , and hamiltonian multi - fluid equations are all described using similar mathematical expressions . because the waterbag system is a finite - dimensional canonical linear hamiltonian system , the standard results of that theory apply , including kren s theorem . we can therefore determine whether there are any unstable modes by counting the total number of neutral eigenvalues . if it is equal to the number of degrees of freedom of the system , then we can expect stability , otherwise , due to the fact that eigenvalues off the imaginary axis come in quartets , we can expect instability . now we determine the signature of each of the stable discrete modes of the waterbag model . beginning with the linearized equations , and assuming the normalization condition , @xmath269 , we find the fourier eigenvector @xmath270 $ ] . using this in the expression for the energy , we get a formula for the energy of a discrete mode , viz . [ hwb ] next , replacing @xmath271 in the numerator of ( [ hwb ] ) by @xmath272 leads to h=_(+)+ = _ = u , where in the last expression we obtain the dielectric energy ( with the electric field amplitude dependence scaled away ) . the energy of a discrete mode is proportional to the derivative of the dispersion function at the frequency corresponding to the mode . as mentioned previously , this familiar formula is also true for embedded modes in the vlasov equation @xcite , and is particularly convenient for use in the waterbag model because it allows geometric evaluation of the signature of modes in the waterbag model . suppose at first that @xmath273 . then the signature of a mode is positive if the dispersion function is increasing at the mode , and negative if it is decreasing at the mode . if @xmath253 does not change sign from one interval to the next , and there is one mode in the corresponding interval , the mode will have signature @xmath274 . similarly , any modes in the same interval must have opposite signature ( or one must have zero signature ) , because the dispersion function must cross the axis in opposite directions at each discrete mode . an example of such a waterbag distribution function is plotted in fig . [ waterbag]a , and the dispersion relation is plotted in fig . [ waterbag]b , where we have marked the zeros with colored circles that indicate their signature . as noted , in the general case there is exactly @xmath275 mode in intervals where @xmath253 does not change sign , and either @xmath259 or @xmath276 modes in intervals where it does change sign . using the waterbag model we can replicate the most important instabilities of the vlasov - poisson equation , in particular the two stream instability and bump on tail instability . both of these instabilities can be emulated by a waterbag model with only a few layers ( fluids ) . in particular , we will consider the special case of a waterbag with 5 layers as depicted in fig . [ twostreamf0 ] . observe , the outermost two have vanishing distribution function , i.e. , @xmath277 , while we choose @xmath278 , @xmath279 , and @xmath280 , so that the distribution has two peaks , one large and one small , separated by a valley . the stability of this model depends on the various parameters involved in defining the equilibrium . for a very large separation of the two peaks , the two - stream distribution function will be stable as depicted in fig . [ 2stream]a ; as the peaks are moved closer together the two modes in the valley of the distribution function between the two peaks move closer together , eventually colliding as depicted in fig . [ 2stream]b , and leaving the axis to become a pair of exponentially growing and decaying modes as depicted in fig . [ 2stream]c . this transition here is identical to that which occurs in the two stream instability of the vlasov equation ( or the corresponding bump on tail instability ) . in the waterbag case there is a positive energy mode that collides with a negative energy mode in the valley of the distribution function . in this chapter we have described bifurcations in general classes of noncanonical hamiltonians systems that describe , e.g. , matter as fluid or kinetic theories . in the multi - fluid systems of sec . [ sec : discrete ] we showed how to linearize , canonize and , for stable systems with discrete spectra , diagonalize to obtain a normal form . hamiltonian bifurcations to instability were described , examples of ss bifurcations were given , but the emphasis was on the hh bifurcation . from the normal form , signature was identified , and it was seen that kren s theorem applies , just as for finite - dimensional systems . next , the class of 2 + 1 hamiltonian theories of sec . [ sec : theories ] were defined and considered . these theories generically posses continuous spectra when linearized , but the specific case of the vlasov - poisson systems was treated in detail . in particular , penrose plots , which allow one to describe transitions to instability , via embedded modes in a continuous spectrum , were described . the technique here is of general utility , e.g. , it was worked out also in detail for shear flow in @xcite . it was also shown how to canonize the linearization of these 2 + 1 theories . next , in order to understand the relationship between discrete bifurcations and the css and chh bifurcations , we introduced the waterbag model , which is a reduction of the 2 + 1 class to a class of systems with a countable number of degrees of freedom , in which the continuous spectrum is discretized . the identification of the waterbag models with the multi - fluid models of sec . [ sec : discrete ] was made and , consequently , the procedure for canonization and diagonalization of the waterbag models was established . a main motivation for studying hamiltonian systems is their universality , i.e. , one is interested in understanding features of one system that apply to all . in this chapter we have shown how infinite - dimensional noncanonical hamiltonian systems enlarge this universality class . it is clear that the same bifurcations occur in a variety of systems that describe different physical situations . any specific system within our classes of systems may possess ss bifurcations , positive and negative energy modes , and kren s theorem for hh bifurcations . our aim is show that an analogous situation transpires for css and chh bifurcations . however , continuous spectra are harder to deal with mathematically and functional analysis is essential , but the existence of analogous behavior in the cases considered here guides us to develop a theory . for example , one can interpret the chh bifurcation as an hh bifurcation with the second mode coming from the continuous spectrum . as stated before , the contents of this chapter are to set the stage for the explicit treatment of bifurcations with the continuous spectrum of @xcite , to which we direct the reader . * acknowledgements . * hospitality of the gfd summer program , held at the woods hole oceanographic institution , is greatly appreciated . pjm and gih were supported by usdoe grant nos . de - fg02 - 04er54742 and de - fg02-er53223 , respectively . ifxundefined [ 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 in @noop _ _ , ( , ) in @noop _ _ ( , ) in @noop _ _ , , vol . , ( , , ) pp . @noop _ _ , lecture notes in mathematics 1160 ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) in @noop _ _ , ( , , ) pp . @noop * * , ( ) @noop * * , ( ) @noop ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) in @noop _ _ , ( , , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , , ) in @noop _ _ , vol . , ( , ) pp . @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop ( ) @noop * * , ( ) in @noop _ _ , ( , , ) pp . @noop _ _ ( , , ) @noop ( ) @noop * * , ( ) @noop * * , ( ) in @noop _ _ , ( , ) pp . @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( )

i review our understanding of the structure and kinematics of the large magellanic cloud ( lmc ) , with a particular focus on recent results . this is an important topic , given the status of the lmc as a benchmark for studies of microlensing , tidal interactions , stellar populations , and the extragalactic distance scale . i address the observed morphology and kinematics of the lmc ; the angles under which we view the lmc disk ; its in - plane and vertical structure ; the lmc self - lensing contribution to the total microlensing optical depth ; the lmc orbit around the milky way ; and the origin and interpretation of the magellanic stream . our understanding of these topics is evolving rapidly , in particular due to the many large photometric and kinematic datasets that have become available in the last few years . it has now been established that : the lmc is considerably elongated in its disk plane ; the lmc disk is thicker than previously believed ; the lmc disk may have warps and twists ; the lmc may have a pressure - supported halo ; the inner regions of the lmc show unexpected complexities in their vertical structure ; and precession and nutation of the lmc disk plane contribute measurably to the observed line - of - sight velocity field . however , many open questions remain and more work is needed before we can expect to converge on a fully coherent structural , dynamical and evolutionary picture that explains all observed features of the lmc .

astro-ph0404192

the large magellanic cloud ( lmc ) is one of our closest neighbor galaxies at a distance of @xmath0 . the sagitarrius dwarf is closer at @xmath1 , but its contrast with respect to the milky way foreground stars is so low that it was discovered only about a decade ago . the lmc is therefore the closest , big , easily observable galaxy from our vantage point in the milky way . as such , it has become a benchmark for studies on various topics . it is of fundamental importance for studies of stellar populations and the interstellar medium ( ism ) , it is being used to study the presence of dark objects in the galactic halo through microlensing ( e.g. , alcock et al . 2000a ) , and it plays a key role in determinations of the cosmological distance scale ( e.g. , freedman et al . for all these applications it is important to have an understanding of the structure and kinematics of the lmc . this is the topic of the present review . for information on other aspects of the lmc , the reader is referred to the book by westerlund ( 1997 ) . the book by van den bergh ( 2000 ) discusses more generally how the properties of the lmc compare to those of other galaxies in the local group . the small magellanic cloud ( smc ) at a distance of @xmath2 is a little further from us than the lmc , and is about 5 times less massive . its structure is more irregular than that of the lmc , and it is less well studied and understood . recent studies of smc structure and kinematics include the work by hatzidimitriou et al . ( 1997 ) , udalski et al . ( 1998 ) , stanimirovic et al . ( 1999 , 2004 ) , kunkel , demers & irwin ( 2000 ) , cioni , habing & israel ( 2000b ) , zaritsky et al . ( 2000 , 2002 ) , crowl et al . ( 2001 ) and maragoudaki et al . however , our overall understanding of smc structure and kinematics has not evolved much since the reviews by westerlund and van den bergh . the present review is therefore restricted to the lmc . the lmc consists of an outer body that appears elliptical in projection on the sky , with a pronounced , off - center bar . the appearance in the optical wavelength regime is dominated by the bar , regions of strong star formation , and patchy dust absorption . the lmc is generally considered an irregular galaxy as a result of these characteristics . it is in fact the prototype of the class of galaxies called `` magellanic irregulars '' ( de vaucouleurs & freeman 1973 ) . detailed studies of the morphological characteristics of the lmc have been performed using many different tracers , including optically detected starlight ( bothun & thompson 1988 ; schmidt - kaler & gochermann 1992 ) , stellar clusters ( lynga & westerlund 1963 ; kontizas et al . 1990 ) , planetary nebulae ( meatheringham et al . 1988 ) and non - thermal radio emission ( alvarez , aparici & may 1987 ) . recent progress has come primarily from studies of stars on the red giant branch ( rgb ) and asymptotic giant branch ( agb ) and from studies of hi gas . = 0.6 recently , two important near - ir surveys have become available for studies of the magellanic clouds , the two micron all sky survey ( 2mass ; e.g. , skrutskie 1998 ) and the deep near - infrared southern sky survey ( denis ; e.g. , epchtein et al . cross - correlations of the data from these surveys and from other catalogs are now available as well ( delmotte et al . 2002 ) . the surveys are perfect for a study of lmc morphology and structure . near - ir data is quite insensitive to dust absorption , which is a major complicating factor in optical studies ( zaritsky , harris & thompson 1997 ; zaritsky 1999 ; udalski et al . 2000 ; alcock et al . the surveys have superb statistics with of the order of a million stars . also , the observational strategy with three near - ir bands ( @xmath3 , @xmath4 and @xmath5 in the 2mass survey ; @xmath6 , @xmath3 and @xmath5 in the denis survey ) allows clear separation of different stellar populations . in particular , the data are ideal for studies of evolved rgb and agb stars , which emit much of their light in the near - ir . this is important for studies of lmc structure , because these intermediate - age and old stars are more likely to trace the underlying mass distribution of the lmc disk than younger populations that dominate the light in optical images . figure [ f : cmds ] shows the @xmath7 color - magnitude diagram ( cmd ) for the lmc region of the sky . several finger - like features are visible , each due to different stellar populations in the lmc or in the foreground ( nikolaev & weinberg 2000 ; cioni et al . 2000a ; marigo , girardi & chiosi 2003 ) . the lmc features of primary interest in the present context are indicated in the figure , namely the red giant branch ( rgb ) , the tip of the rgb ( trgb ) , the oxygen - rich agb stars , and the carbon - rich agb stars ( `` carbon stars '' ) . van der marel ( 2001 ) used the color cut shown in the figure to extract a sample from the 2mass and denis datasets that is dominated by rgb and agb stars . these stars were used to make the number - density map of the lmc shown in figure [ f : maps ] . kontizas et al . ( 2001 ) showed the distribution on the sky of @xmath8 carbon stars identified by eye from optical objective prism plates . their map does not show all the rich detail visible in figure [ f : maps ] , but it is otherwise in good agreement with it . = 0.6 the near - ir map of the lmc is surprisingly smooth . the morphology is much less irregular than it is for the the younger stellar populations that dominate the optical light . apart from the central bar there is a hint of some spiral structure , as discussed previously by , e.g. , de vaucouleurs & freeman ( 1973 ) . however , the spiral features all have very low contrast with respect to their surroundings , and there is certainly no well organized spiral pattern in the lmc . quantitative analysis can be performed on the basis of ellipse fits to the number density contours . this yields a surface number density profile that can be fit reasonably well fit by an exponential with a scale length of @xmath9 ( @xmath10 ) . the radial profiles of the ellipticity @xmath11 ( defined as @xmath12 , where @xmath13 is the axial ratio ) and the major axis position angle @xmath14 both show pronounced variations as function of distance from the lmc center , due to the presence of the central bar . however , at radii @xmath15 the contour shapes converge to an approximately constant position angle @xmath16 and ellipticity @xmath17 . = 0.6 to study the distribution of hi in the lmc on small scales requires many pointings to cover the lmc at high angular resolution . kim et al . ( 1998 ) used the australia telescope compact array ( atca ) to obtain a map that is sensitive on scales of 15@xmath18 . on these scales , the morphology is dominated by hi filaments with numerous shells and holes . the turbulent and fractal nature of the ism on these scales is the result of dynamical feedback into the ism from star formation processes . in the context of the present review we are more interested in the large scale distribution of hi gas this issue has been studied for decades , including work by mcgee & milton ( 1966 ) , rohlfs et al . ( 1984 ) and luks & rohlfs ( 1992 ) . most recently , staveley - smith et al . ( 2003 ) obtained the map shown in figure [ f : himap ] , using a multibeam survey with the parkes telescope . this map is sensitive to spatial structure in the range @xmath19 to @xmath20 . the hi distribution is very patchy ( see also kim et al . 2003 , who combined the data of kim et al . [ 1998 ] and staveley - smith et al . [ 2003 ] into a single map that is sensitive to structures on all scales down to @xmath21 ) . the hi in the lmc is not centrally concentrated and the brightest areas are several degrees from the center . the overall distribution is approximately circular and there is no sign of a bar . all of these characteristics are strikingly different from the stellar distribution shown in figure [ f : maps ] . figure [ f : hioverlay ] shows contours of the hi distribution on a somewhat larger scale , from the hipass data presented in putman et al . ( 2003 ) , overlaid on the lmc near - ir star count map from figure [ f : maps ] . this comparison suggests that collisionless tracers such as rgb and agb stars are best suited to study the structure and mass distribution of the lmc disk , whereas hi gas may be better suited to study the effects of tidal interactions . the stream of gas towards the right of figure [ f : hioverlay ] is the start of the magellanic bridge towards the smc , which provides one of the many pieces of evidence for strong tidal interactions between the lmc , the milky way and the smc . figure [ f : stream ] shows a wide - area hi map of the magallanic system . the most prominent feature is the magellanic stream , a @xmath22 wide filament with @xmath23 of neutral hydrogen that spans more than @xmath24 across the sky ( e.g. , westerlund 1997 ; putman et al . it consists of gas that trails the magellanic clouds as they orbit the milky way . a less prominent leading gas component was recently discovered as well ( lu et al . 1998 ; putman et al . 1998 ) , the start of which is seen at the left in figure [ f : stream ] . = many detailed theoretical models have been constructed for the magellanic stream ( e.g. , murai & fujimoto 1980 ; lin & lynden - bell 1982 ; shuter 1992 ; liu 1992 ; heller & rohlfs 1994 ; gardiner , sawa & fujimoto 1994 ; moore & davis 1994 ; lin , jones & klemola 1995 ; gardiner & noguchi 1996 ; yoshizawa & noguchi 2003 ; mastropietro et al . 2004 ; conners et al . models in which tidal stripping is the dominating process have been particularly successful . the most sophisticated recent calculations in this class are those by yoshizawa & noguchi ( 2003 ) and conners et al . ( 2004 ) , both of which build on earlier work by gardiner & noguchi ( 1996 ) . in these models the lmc and smc form a gravitationally bound system that orbits the milky way . the magellanic stream and the leading arm represent material that was stripped from the smc @xmath25 ago . this was the time of the previous perigalactic passage , which coincided with a close encounter between the clouds . the models successfully reproduce many properties of the magellanic stream , including its position , morphology , width variation , and the velocity profile along the stream . the models also explain the presence of the leading arm , and why it is less prominent than the trailing stream . given the successes of tidal models , it has always been surprising that no population of stars associated with the stream has ever been found ( e.g. , irwin 1991 ; guhathakurta & reitzel 1998 ; majewski et al . 2003 ) . in a tidal model where the stripping is dominated by gravity , one might naively expect that both stars and gas are stripped equally . however , galaxies generally have hi gas disks that are more extended than the stellar distribution . since material is preferentially stripped from the outskirts of a galaxy , this can explain why there may not be any stars associated with the stream ( yoshizawa & noguchi 2003 ) . alternatively , it has been argued that the lack of stars in the stream may point to important contributions from other physical processes than tidal effects . for example , moore & davis ( 1994 ) and mastropietro et al . ( 2004 ) suggest that the stream consists of material which was ram - pressure stripped from the lmc during its last passage through a hot ( @xmath26 k ) ionized halo component of the milky way . another alternative was proposed by heller & rohlfs ( 1994 ) , who suggested that hydrodynamical forces ( rather than gravitational / tidal forces ) during a recent lmc - smc interaction are responsible for the existence of the stream . models of the magellanic stream have traditionally used the properties of the stream to estimate the orbit of the lmc , rather than to base the calculations on estimates of the lmc proper motion . an important characteristic of the orbit is the present - day tangential velocity in galactocentric coordinates , for which values have been inferred that include @xmath27 ( lin & lynden - bell 1982 ) , @xmath28 ( shuter 1992 ) , @xmath29 ( heller & rohlfs 1994 ) , @xmath30 ( murai & fujimoto 1980 ) , @xmath31 ( liu 1992 ) and @xmath32 ( gardiner et al . 1994 ; gardiner & noguchi 1996 ) , respectively . proper motion measurements have improved significantly over time , and it is now known that @xmath33 ( as discussed in detail in section [ s : orbit ] ) . this is consistent with the lower range of the values predicted by models for the stream . the data are most consistent with the models of gardiner & noguchi ( 1996 ) and their follow - ups , which also provide some of the best fits to many other properties of the magellanic stream . the observational error on @xmath34 is almost small enough to start ruling out some of the other models . it is possible that future proper motion measurements may yield a much more accurate determination of the lmc orbit . models of the magellanic stream then hold the promise of providing important constraints on the mass , shape , and radial density profile of the milky way s dark halo ( e.g. , lin et al . to understand the tidal interactions in the magellanic system it is important to know the orbit of the lmc around the milky way . this requires knowledge of all three of the velocity components of the lmc center of mass . the line - of - sight velocity can be accurately determined from the doppler velocities of tracers ( see section [ s : kinematics ] ) . by contrast , determination of the velocity in the plane of the sky is much more difficult . for the lmc , proper motion determinations are available from the following sources : kroupa et al . ( 1994 ) , using stars from the ppm catalogue ; jones , klemola & lin ( 1994 ) , using photographic plates with a 14 year epoch span ; kroupa & bastian ( 1997 ) , using hipparcos data ; drake et al . ( 2002 ) , using data from the macho project ; and anguita , loyola & pedreros ( 2000 ) and pedreros , anguita & maza ( 2002 ) , using ccd frames with an 11 year epoch span . some of the measurements pertain to fields in the outer parts in the lmc disk , and require corrections for the orientation and rotation of the lmc disk . the measurements are all more - or - less consistent with each other to within the error bars . the exception to this is the anguita et al . result , which almost certainly suffers from an unidentified systematic error . when this latter result is ignored , the weighted average of the remaining measurements yields proper motions towards the west and north of ( van der marel et al . 2002 ) @xmath35 transformation of the proper motion to a space velocity in @xmath36 requires knowledge of the lmc distance @xmath37 . many techniques have been used over the years to estimate the lmc distance , but unfortunately , there continue to be systematic differences between the results from different techniques that far exceed the formal errors . it is beyond the scope of the present review to address this topic in detail . instead , the reader is refered to the recent reviews by , e.g. , westerlund ( 1997 ) , gibson et al . ( 2000 ) , freedman et al . ( 2001 ) and alves ( 2004b ) . freedman et al . adopt a distance modulus @xmath38 on the basis of a review of all published work . this corresponds to @xmath39 . at this distance , a proper motion of @xmath40 corresponds to @xmath41 and 1 degree on the sky corresponds to @xmath42 . combination of the distance and proper motion of the lmc yields velocities @xmath43 and @xmath44 towards the west and north , respectively , with errors of @xmath45 in each direction . this can be combined with the observed line - of - sight velocity , @xmath46 ( see section [ ss : carbonstars ] ) , to obtain the three - dimensional motion of the lmc with respect to the milky way . it is usual to adopt a cartesian coordinate system @xmath47 , with the origin at the galactic center , the @xmath48-axis pointing towards the galactic north pole , the @xmath49-axis pointing in the direction from the sun to the galactic center , and the @xmath50-axis pointing in the direction of the sun s galactic rotation . the observed lmc velocities must be corrected for the reflex motion of the sun , which is easily done with use of standard estimates for the position and velocity of the sun with respect to the galactic center . the @xmath47 position of the lmc is then found to be ( van der marel et al . 2002 ) @xmath51 and its three - dimensional space velocity is @xmath52 this corresponds to a distance of @xmath53 from the galactic center , and a total velocity of @xmath54 in the galactocentric rest frame . the motion has a radial component of @xmath55 pointing away from the galactic center , and a tangential component of @xmath56 . the proper motion of the smc is known only with errors of @xmath57 in each coordinate ( kroupa & bastian 1997 ) , which is five times less accurate than for the lmc . however , to within these errors the smc is known to have a galactocentric velocity vector that agrees with that of the lmc . the combination of a small but positive radial velocity and a tangential velocity that exceeds the circular velocity of the milky way halo implies that the lmc must be just past pericenter in its orbit . the calculation of an actual orbit requires knowledge of the three - dimensional shape and the radial profile of the gravitational potential of the milky way dark halo . gardiner et al . ( 1994 ) and gardiner & noguchi ( 1996 ) calculated orbits in a spherical milky way halo potential with a rotation curve that stays flat at @xmath58 out to a galactocentric distance of at least @xmath59 . such a potential is consistent with our present understanding of the milky way dark halo ( kochanek 1996 ; wilkinson & evans 1999 ; ibata et al . the calculations properly take into account that the lmc and smc orbit each other while their center of mass orbits the milky way . however , since the lmc is more massive than the smc , its motion is not too different from that of the lmc - smc center of mass . for an assumed present - day lmc velocity @xmath60 consistent with that given in equation ( [ lmcvelres ] ) the apocenter to pericenter ratio is inferred to be @xmath61 . the perigalactic distance is @xmath62 and the orbital period around the milky way is @xmath25 . the kinematical properties of the lmc provide important clues to its structure . observations have therefore been obtained for many tracers . the kinematics of gas in the lmc have been studied primarily using hi ( e.g. , rohlfs et al . 1984 ; luks & rohlfs 1992 ; kim et al . 1998 ) . discrete lmc tracers which have been studied kinematically include star clusters ( freeman , illingworth & oemler 1983 ; schommer et al . 1992 ) , planetary nebulae ( meatheringham et al . 1988 ) , hii regions and supergiants ( feitzinger , schmidt - kaler & isserstedt 1977 ) , and carbon stars ( kunkel et al . 1997b ; graff et al . 2000 ; alves & nelson 2000 ; van der marel et al . a common result from all these studies is that the line - of - sight velocity dispersion of the tracers is generally at least a factor @xmath63 smaller than their rotation velocity . this implies that the lmc is kinematically cold , and must therefore to lowest approximation be a disk system . to understand the kinematics of an lmc tracer population it is necessary to have a general model for the line - of - sight velocity field that can be fit to the data . all studies thus far have been based on the assumption that the mean streaming ( i.e. , the rotation ) in the disk plane can be approximated to be circular . however , even with this simplifying assumption it is not straightforward to model the kinematics of the lmc , because it is so near to us . its main body spans more than @xmath64 on the sky and one therefore can not make the usual approximation that `` the sky is flat '' over the area of the galaxy . spherical trigonometry must be used , which yields the general expression ( van der marel et al . 2002 ) : @xmath65 with @xmath66^{1/2 } } } } .\ ] ] in this equation , @xmath67 is the observed component of the velocity along the line of sight . the quantities @xmath68 identify the position on the sky : @xmath69 is the angular distance from the center and @xmath70 is the position angle with respect to the center ( measured from north over east ) . the kinematical center is at the center of mass ( cm ) of the galaxy . the quantities @xmath71 describe the velocity of the cm in an inertial frame in which the sun is at rest : @xmath72 is the systemic velocity along the line of sight , @xmath73 is the transverse velocity , and @xmath74 is the position angle of the transverse velocity on the sky . the angles @xmath75 describe the direction from which the plane of the galaxy is viewed : @xmath76 is the inclination angle ( @xmath77 for a face - on disk ) , and @xmath78 is the position angle of the line of nodes , as illustrated in figure [ f : drawviewang ] . the line - of - nodes is the intersection of the galaxy plane and the sky plane . the velocity @xmath79 is the rotation velocity at cylindrical radius @xmath80 in the disk plane . @xmath37 is the distance to the cm , and @xmath81 is a geometrical factor . the quantity @xmath82 is the ` spin sign ' that determines in which of the two possible directions the disk rotates . = 0.6 the first term in equation ( [ vfield ] ) corresponds to the internal rotation of the lmc . the second term is the part of the line - of - sight velocity of the cm that is seen along the line of sight , and the third term is the part of the transverse velocity of the cm that is seen along the line of sight . for a galaxy that spans a small area on the sky ( very small @xmath69 ) , the second term is simply @xmath72 and the third term is zero . however , the lmc does not have a small angular extent and the inclusion of the third term is particularly important . it corresponds to a solid - body rotation component . given the lmc transverse velocity implied by equation ( [ weightb ] ) , it rises to an amplitude of @xmath83 at @xmath84 , which significantly exceeds the amplitude of the intrinsic rotation contribution ( first term of eq . [ [ vfield ] ] ) at that radius . the fourth term in equation ( [ vfield ] ) describes the line - of - sight component due to changes in the inclination of the disk with time , as are expected due to precession and nutation of the lmc disk plane as it orbits the milky way ( weinberg 2000 ) . this term also corresponds to a solid - body rotation component . the general expression in equation ( [ vfield ] ) appears complicated , but it is possible to gain some intuitive insight by considering some special cases . along the line of nodes one has that @xmath85 and @xmath86 , so that @xmath87 .\ ] ] here it has been defined that @xmath88 . the quantity @xmath89 is the component of the transverse velocity vector in the plane of the sky that lies along the line of nodes ; similarly , @xmath90 is the component perpendicular to the line of nodes . perpendicular to the line of nodes one has that @xmath91 and @xmath92 , and therefore @xmath93 here it has been defined that @xmath94 . this implies that perpendicular to the line of nodes @xmath95 is linearly proportional to @xmath96 . by contrast , along the line of nodes this is true only if @xmath79 is a linear function of @xmath80 . this is not expected to be the case , because galaxies do not generally have solid - body rotation curves ; disk galaxies tend to have flat rotation curves , at least outside the very center . this implies that , at least in principle , both the position angle @xmath78 of the line of nodes and the quantity @xmath97 are uniquely determined by the observed velocity field : @xmath78 is the angle along which the observed @xmath95 are best fit by a linear proportionality with @xmath96 , and @xmath97 is the proportionality constant . = 0.8 among the discrete tracers in the lmc that have been studied kinematically , carbon stars have yielded some of the largest and most useful datasets in recent years . van der marel et al . ( 2002 ) fitted the general velocity field expression in equation ( [ vfield ] ) to the data for 1041 carbon stars , obtained from the work of kunkel et al . ( 1997a ) and hardy , schommer & suntzeff ( unpublished ) . the combined dataset samples both the inner and the outer parts of the lmc , although with a discontinuous distribution in radius and position angle . figure [ f : fit ] shows the data , with the best model fit overplotted . overall , the model provides a good fit to the data . = 0.6 as discussed in section [ ss : generalexp ] , the line - of - nodes position angle is uniquely determined by the data ; the model yields @xmath98 . the lmc inclination can not be determined kinematically , but it is known reasonably well from other considerations ( see section [ s : viewing ] ) . with both viewing angles and the lmc proper motion known , the rotation curve @xmath79 follows from equation ( [ vlosalong ] ) . the result is shown in figure [ f : rotcurve ] . the inferred rotation curve @xmath79 rises linearly in the central region and is roughly flat at @xmath99 for @xmath100 . the negative value at the innermost radius @xmath101 has limited significance ; it is probably affected by the sparse coverage of the data at these radii ( see figure [ f : fit ] ) , as well as potential non - circular streaming motions in the region of the bar . the error bars on @xmath79 in figure [ f : rotcurve ] reflect the random errors due to the sampling of the data . however , there are also contributions from the errors in the line - of - nodes position angle @xmath78 , the lmc proper motion and the inclination @xmath76 . in particular , @xmath79 scales as @xmath102 ; and if the component @xmath103 of the transverse velocity vector along the line of nodes is larger than given in equation ( [ weightb ] ) , then @xmath79 will go up . when all uncertainties are properly accounted for , the amplitude of the flat part of the rotation curve and its formal error become @xmath104 . when asymmetric drift is corrected for , the circular velocity in the disk plane can be calculated to be @xmath105 . the implied total mass of the lmc inside the last measured data point is therefore @xmath106 . by contrast , the total stellar mass of the lmc disk is @xmath107 and the mass of the neutral gas in the lmc is @xmath108 ( kim et al . the combined mass of the visible material in the lmc is therefore insufficient to explain the dynamically inferred mass . the lmc must therefore be embedded in a dark halo . this is consistent with the fact that the observed rotation curve amplitude is relatively flat as a function of radius . the lmc tidal radius can be calculated to be @xmath109 , which corresponds to an angle on the sky of @xmath110 . the uncertainty in the tidal radius is due primarily to our ignorance of how far the lmc dark halo extends . either way , the tidal radius extends beyond the region for which most observations of the main body of the lmc are available . however , it should be kept in mind that the tidal radius marks the position beyond which material becomes unbound . the structure of a galaxy can be altered well inside of this radius ( weinberg 2000 ) . as discussed in section [ ss : generalexp ] , the line - of - sight velocity field constrains the value of @xmath111 . the carbon stars yield @xmath112 . given our knowledge of the proper motion and distance of the lmc ( see section [ s : orbit ] ) this implies that @xmath113 . the lmc is the first galaxy for which it has been possible to measure @xmath114 . the @xmath115-body simulations by weinberg ( 2000 ) show that the tidal torques from the milky way are expected to induce precession and nutation in the symmetry axis of the lmc disk plane . although a detailed data - model comparison is not possible at the present time , it is comforting to note that the observed @xmath114 is of the same order as the rate of change of the disk orientation seen in the simulations . hi gas provides another powerful method to study the kinematics of the lmc . high quality data is available from , e.g. , kim et al . unfortunately , the kinematical analysis presented by kim et al . was not as general as that discussed above for carbon stars . they did not leave @xmath97 as a free parameter in the fit . instead , they corrected their data at the outset for the transverse motion of the lmc using the proper motion measured by jones et al . ( 1994 ) , @xmath116 , and assumed that @xmath117 . this fixes @xmath118 , which is inconsistent with the value inferred from the carbon star velocity field . kim et al . obtained a kinematic line of nodes that is both twisting with radius and inconsistent with the value determined from the carbon stars . in addition , they inferred a rotation curve for which the amplitude exceeds that in figure [ f : rotcurve ] by @xmath119 . it is likely that these results are affected by the imposed value of @xmath97 . a more general analysis of the hi kinematics is therefore desirable , but unfortunately , is not currently available . the same limitations apply to many of the other published studies of lmc tracer kinematics cited at the start of section [ s : kinematics ] . independent of how the data are analyzed , it is likely that collisionless tracers provide a more appropriate means to study the structure of the lmc using equilibrium models than does hi gas . the dynamical center of the carbon star velocity field is found to be consistent ( to within the @xmath120 per - coordinate errors ) with both the center of the bar and the center of the outer stellar contours . however , it has long been known that the center of the hi rotation velocity field does not coincide with the center of the bar , and it also does nt coincide with the center of the outer contours of the stellar distribution . it is offset from both by @xmath121 . this indicates that the hi gas may not be in equilibrium in the the gravitational potential . added evidence for this comes from the fact that the lmc and the smc are enshrouded in a common hi envelope , and that they are connected by a bridge of hi gas ( see figure [ f : stream ] ) . even at small radii the lmc gas disk appears to be subject to tidal disturbances ( see figure [ f : hioverlay ] ) that may well affect the velocity field . a disk that is intrinsically circular will appear elliptical in projection on the sky . the viewing angles of the disk ( see figure [ f : drawviewang ] ) are then easily determined : the inclination is @xmath122 , where @xmath11 is the apparent ellipticity on the sky , and the line - of - nodes position angle @xmath78 is equal to the major axis position angle @xmath123 of the projected body . the viewing angles of the lmc have often been estimated under this assumption , using the projected contours for many different types of tracers ( de vaucouleurs & freeman 1973 ; bothun & thompson 1988 ; schmidt - kaler & gochermann 1992 ; weinberg & nikolaev 2001 ; lynga & westerlund 1963 ; kontizas et al . 1990 ; feitzinger et al . 1977 ; kim et al . 1998 ; alvarez et al . 1987 ) . however , it now appears that this was incorrect . the kinematics of carbon stars imply @xmath98 ( see section [ ss : carbonstars ] ) , whereas the near - ir morphology of the lmc implies @xmath16 ( see section [ s : morphology ] ) . the result that @xmath124 implies that the lmc can not be intrinsically circular . the value of @xmath14 is quite robust ; studies of other tracers have yielded very similar results , although often with larger error bars . the result that @xmath125 therefore hinges primarily on our confidence in the inferred value of @xmath78 . there have been other kinematical studies of the line of nodes , in addition to that described in section [ ss : carbonstars ] . these have generally yielded values of @xmath78 that are both larger and twisting with radius ( e.g. , kim et al . 1998 ; alves & nelson 2000 ) . however , the accuracy of these results is suspect because of the important simplifying assumptions that were made in the analyses ( see section [ ss : hikin ] ) . no allowance was made for a potential solid - body rotation component in the velocity field due to precession and nutation of the lmc disk , which is both predicted theoretically ( weinberg 2000 ) and implied observationally by the carbon star data ( see section [ ss : carbonstars ] ) . arguably the most robust way to determine the lmc viewing angles is to use geometrical considerations , rather than kinematical ones . for an inclined disk , one side will be closer to us than the other . tracers on that one side will appear brighter than similar tracers on the other side . this method does not rely on absolute distances or magnitudes , which are notoriously difficult to estimate , but only on relative distances or magnitudes . to lowest order , the difference in magnitude between a tracer at the galaxy center and a similar tracer at a position @xmath68 in the disk ( as defined in section [ ss : generalexp ] ) is @xmath126 where the angular distance @xmath69 is expressed in degrees . the constant in the equation is @xmath127 magnitudes . hence , when following a circle on the sky around the galaxy center one expects a sinusoidal variation in the magnitudes of tracers . the amplitude and phase of the variation yield estimates of the viewing angles @xmath75 . = 0.5 van der marel & cioni ( 2001 ) used a polar grid on the sky to divide the lmc area into several rings , each consisting of a number of azimuthal segments . the data from the denis and 2mass surveys were used for each segment to construct near - ir cmds similar to that shown in figure [ f : cmds ] . for each segment the modal magnitude ( magnitude where the luminosity function peaks ) was determined for carbon - rich agb stars selected by color , as had been suggested by weinberg & nikolaev ( 2001 ) . figure [ f : banddep ] shows the inferred variation in magnitude as function of position angle @xmath70 for the radial range @xmath128 . the expected sinusoidal variations are confidently detected . the top panel shows the results for stars selected from the denis survey with the color selection criterion @xmath129 . the bottom panel shows the results from the 2mass survey with the same color selection . the same sinusoidal variations are seen , indicating that there are no relative calibration problems between the surveys . also , the same variations are seen in the @xmath6 , @xmath3 , @xmath4 and @xmath5 bands , which implies that the results are not influenced significantly by dust absorption . the middle panel shows the variations in the trgb magnitudes as a function of position angle , from the denis data . rgb stars show the same variations as the agb stars , suggesting that the results are not influenced significantly by potential peculiarities associated with either of these stellar populations . the observed variations can therefore be confidently interpreted as a purely geometrical effect . the implied viewing angles are @xmath130 and @xmath131 . the @xmath78 value thus inferred geometrically is entirely consistent with the value inferred kinematically ( see section [ ss : carbonstars ] ) . moreover , there is an observed drift in the center of the lmc isophotes at large radii which is consistent with both estimates , when interpreted as a result of viewing perspective ( van der marel 2001 ) . the aforementioned analyses are sensitive primarily to the structure of the outer parts of the lmc . several other studies of the viewing angles have focused mostly on the region of the bar , which samples only the central few degrees . many of these studies have been based on cepheids . their period - luminosity relation allows calculation of the distance to each individual cepheid from a light curve . the relative distances of the cepheids in the sample can then be analyzed in similar fashion as discussed above to yield the lmc viewing angles . cepheid studies in the 1980s did nt have many stars to work with . caldwell & coulson ( 1986 ) analyzed optical data for 73 cepheids and obtained @xmath132 and @xmath133 . laney & stobie ( 1986 ) obtained @xmath134 and @xmath135 from 14 cepheids , and welch et al . ( 1987 ) obtained @xmath136 and @xmath137 from 23 cepheids , both using near - ir data . the early cepheid studies have now all been superseded by the work of nikolaev et al . they analyzed a sample of more than 2000 cepheids with lightcurves from macho data . through use of photometry in five different bands , including optical macho data and near - ir 2mass data , each star could be individually corrected for dust extinction . from a planar fit to the data they obtained @xmath138 and @xmath139 . other recent work has used the magnitude of the red clump to analyze the relative distances of different parts of the lmc . olsen & salyk ( 2002 ) obtained @xmath140 and @xmath141 , also from an analysis that was restricted mostly to the the inner parts of the lmc . there is one caveat associated with all viewing angle results for the central few degrees of the lmc . namely , it appears that the stars in this region are not distributed symmetrically around a single well - defined plane , as discussed in detail in section [ ss : nonplanar ] . in the present context we are mainly concerned with the influence of this on the inferred viewing angles . olsen & salyk ( 2002 ) perform their viewing angle fit by ignoring fields south - west of the bar , which do not seem to agree with the planar solution implied by their remaining fields . by contrast , nikolaev et al . ( 2004 ) fit all the stars in their sample , independent of whether or not they appear to be part of the main disk plane . clearly , the @xmath75 results of olsen & salyk and nikolaev et al . are the best - fitting parameters of well - posed problems . however , it is somewhat unclear whether they can be assumed to be unbiased estimates of the actual lmc viewing angles . for a proper understanding of this issue one would need to have both an empirical and a dynamical understanding of the nature of the extra - planar structures in the central region of the lmc . only then is it possible to decide whether the concept of a single disk plane is at all meaningful in this region , and which data should be included or excluded in determining its parameters . this is probably not an issue for the outer parts of the lmc , given that the agb star results of van der marel & cioni ( 2001 ) provide no evidence for extra - planar structures at radii @xmath142 . in summary , all studies agree that @xmath76 is in the range @xmath143@xmath144 . at large radii , @xmath78 appears to be in the range @xmath145@xmath146 . by contrast , at small radii all studies indicate that @xmath78 is in the range @xmath147@xmath148 . as mentioned , it is possible that the results at small radii are systematically in error due to the presence of out - of - plane structures . alternatively , it is quite well possible that there are true radial variations in the lmc viewing angles due to warps and twists of the disk plane . many authors have suggested this as a plausible interpretation of various features seen in lmc datasets ( van der marel & cioni 2001 ; olsen & salyk 2002 ; subramaniam 2003 ; nikolaev et al . moreover , numerical simulations have shown that milky way tidal effects can drive strong warps in the lmc disk plane ( mastropietro et al . the inferred lmc viewing angles can be used to deproject the observed morphology that is seen in projection on the sky ( van der marel 2001 ) . this yields an in - plane ellipticity @xmath11 in the range @xmath149@xmath150 , depending somewhat on the adopted viewing angles ; e.g. , the nikolaev et al . ( 2004 ) angles give @xmath151 and the van der marel & cioni ( 2001 ) angles give @xmath152 . the conclusion that the lmc is elongated is in itself not surprising . the dark matter halos predicted by cosmological simulations are generally triaxial ( e.g. , dubinski & carlberg 1991 ) , and the gravitational potential in the equatorial plane of such halos does not have circular symmetry . so it is generally expected that disk galaxies are elongated rather than circular . furthermore , it is possible to construct self - consistent dynamical models for elliptical disks ( e.g. , teuben 1987 ) . what is surprising is that the lmc ellipticity is fairly large . studies of the apparent axis ratio distribution of spiral galaxy disks ( binney & de vaucouleurs 1981 ; lambas , maddox & loveday 1992 ) of the structure of individual spiral galaxies ( rix & zaritsky 1995 ; schoenmakers , franx & de zeeuw 1997 ; kornreich , haynes & lovelace 1998 ; andersen et al . 2001 ) and of the scatter in the tully - fisher relation ( franx & de zeeuw 1992 ) indicate that the average ( deprojected ) ellipticity of spiral galaxies is only 510% . so while spiral galaxies are generally elongated , their elongation is usually smaller than inferred here for the lmc . of course , galaxies of type sm and i m are ( by definition ) more irregular and lopsided than spirals . so it is not a priori clear whether or not the lmc is atypically elongated for its hubble type . it is interesting to ask what may be the cause of the large in - plane ellipticity of the lmc . the prime candidate is distortion by the milky way tidal field . the present - day tidal force on the lmc by the milky way exceeds that from the smc . moreover , the milky way tidal field is responsible for other well - known features of the magellanic system , such as the magellanic stream ( see section [ s : stream ] ) . n - body simulations have shown that the structure of the lmc can be altered significantly by the milky way tidal force ( weinberg 2000 ) and the simulations of mastropietro et al . ( 2004 ) indeed predict a considerable in - plane elongation for the lmc . also , the lmc elongation in projection on the sky points approximately towards the galacic center and is perpendicular to the magellanic stream ( van der marel 2001 ) , as predicted naturally by simulations of tidal effects ( mastropietro et al . however , a very detailed data - model comparison is not possible at the present time . that would require accurate knowledge of the past history as a function of time of the lmc orbit , of the disk - plane orientation due to precession and nutation , and of the lmc - smc distance . such knowledge is not available at the present time . a consequence of the ellipticity of the lmc disk is that one can not expect the streamlines of tracers in the disk to be perfectly circular , by contrast to what has been assumed in all kinematical studies to date . the effect of this is probably not large , because the gravitational potential of a mass distribution is always rounder than the mass distribution itself . one effect of ellipticity is an apparent offset between the kinematical line of nodes and the true line of nodes ( schoenmakers , franx & de zeeuw 1997 ) . this effect is presently not at observable levels , given that kinematical analysis of carbon stars ( see section [ s : kinematics ] ) yields a line - of - nodes position angle @xmath78 that is in adequate agreement with geometrical determinations ( see section [ s : viewing ] ) . however , it should be kept in mind that , as the sophistication of the studies of lmc kinematics increases , it might become necessary to account for the effect of non - circularity on the observed kinematics . the scale height of the lmc disk can be estimated from the observed line - of - sight velocity dispersion @xmath153 . for carbon stars , the scatter of the velocity measurements around the best - fitting rotating disk model ( figure [ f : fit ] ) yields @xmath154 . the ratio of rotation velocity to velocity dispersion is therefore @xmath155 . for comparison , the thin disk of the milky way has @xmath156 and its thick disk has @xmath157 . in a relative sense one might therefore expect the lmc disk to be similar , but somewhat thicker than the milky way thick disk . weinberg ( 2000 ) argued from @xmath115-body simulations that such considerable thickness could be the result of milky way tidal effects on the lmc . the radial profile of the velocity dispersion contains information on the lmc scale height as function of radius . the carbon star velocity dispersion is close to constant as function of radius , and this is not what is expected for a disk with a constant scale height . to fit this behavior one must assume that the scale height increases with radius in the disk ( alves & nelson 2000 ) . this can arise naturally as a result of tidal forces from the milky way , which become relatively more important ( compared to the lmc self - gravity ) as one moves to larger radii . alves & nelson considered an isothermal disk with a vertical density profile proportional to @xmath158 , where @xmath159 can vary with disk radius . application to the carbon star data of van der marel et al . ( 2002 ) yields @xmath160 at the lmc center , rising to @xmath161 at a radius of @xmath162 . the lmc carbon stars are part of the intermediate - age population which is believed to be fairly representative for the bulk of the mass in the lmc . in this sense , the results inferred for the carbon star population are believed to be characteristic for the lmc as a whole . however , it is certainly not true that all populations have the same kinematics . as in the milky way , younger populations have a smaller velocity dispersion ( and hence a smaller scale height ) than older populations . a summary of measurements for various populations is given by gyuk , dalal & griest ( 2000 ) . the youngest populations ( e.g. , supergiants , hii regions , hi gas ) have dispersions of only @xmath163 . old long - period variables have dispersions @xmath164 ( bessell , freeman & wood 1986 ) and so do the oldest star clusters ( freeman et al . 1983 ; schommer et al . these values are considerably below the lmc circular velocity ( see section [ ss : carbonstars ] ) . this has generally been interpreted to mean that the lmc does not have an old pressure supported halo similar to that of the milky way . the first possible evidence for the presence of a pressure supported halo was presented recently by minniti et al . they measured a dispersion @xmath165 for a sample of 43 rr lyrae stars within @xmath166 from the lmc center . this value is consistent with what would be expected for a pressure supported halo in equilibrium in the gravitational potential implied by the circular velocity ( alves 2004a ) . the rr lyrae stars make up @xmath63% of the visible mass of the lmc , similar to the value for the milky way halo . however , it is surprising that the surface density distribution of the lmc rr lyrae stars is well fit by an exponential with the same scale length as inferred for other tracers known to reside in the disk ( alves 2004a ) . this is very different from the situation for the milky way halo , where rr lyrae stars follow a power - law density profile . this suggests that maybe the rr lyrae stars in the lmc did form in the disk , instead of in a halo . in this scenario they might simply have attained their large dispersions by a combination of disk heating and milky way tidal forces ( weinberg 2000 ) . to discriminate between halo and disk origins it will be important to determine whether the velocity field of the rr lyrae stars has a rotation component . this will require observations at larger galactocentric distances.=-2 one of the most important reasons for seeking to understand the vertical structure of the lmc is to understand the results from microlensing surveys . the observed microlensing optical depth towards the lmc is @xmath167 with an additional 2030% of systematic error ( alcock et al . it has generally been found that equilibrium models for the lmc do not predict enough lmc self - lensing events to account for the observed optical depth ( gyuk , dalal & griest 2000 ; jetzer , mancini & scarpetta 2002 ) . for the most favored set of lmc model parameters in the gyuk et al . study the predicted self - lensing optical depth is @xmath168 , a factor of @xmath169 less than the observed value . to account for the lenses it is therefore necessary to assume that some @xmath170% of the milky way dark halo is made up of lenses of mass @xmath171@xmath172 ( alcock et al . however , it is a mystery what the composition of this lensing component could be . a large population of old white dwarfs has been suggested , but this interpretation is not without problems ( e.g. , fields , freese , & graff 2000 ; flynn , holopainen , & holmberg 2003 ) . it is therefore worthwhile to investigate whether the models that have been used to estimate the lmc self - lensing might have been oversimplified . this is particularly important since there is evidence from the observed microlensing events themselves that many of the lenses may reside in the lmc ( sahu 2003 ) . to lowest order , the self - lensing optical depth in simple disk models of the lmc depends exclusively on the observed velocity dispersion and not separately on either the galaxy mass or scale height ( gould 1995 ) . one way to increase the self - lensing predicted by lmc models is therefore to assume that a much larger fraction of the lmc mass resides in high velocity dispersion populations than has previously been believed . salati et al . ( 1999 ) showed that the observed optical depth can be reproduced if one assumes that 70% of the mass in the lmc disk consists of objects with @xmath153 ranging from @xmath173 to @xmath174 . however , this would seem difficult to explain on the basis of present data . although rr lyrae stars have now been observed to have a high velocity dispersion , these old stars make up only 2% of the visible mass . so their influence on self - lensing predictions is negligible . a better way to account for the observed optical depth might be to assume that the vertical structure of the lmc is more complicated than for normal disk galaxies . the self - lensing optical depth might then have been considerably underestimated.=-2 one way to explain the observed microlensing optical depth is to assume that there might be stellar populations outside of the main lmc disk plane . for example , there might be a population of stars in front of or behind the lmc that was pulled from the magellanic clouds due to milky way tidal forces ( zhao 1998 ) or there might be a non - virialized shroud of stars at considerable distances above the lmc plane due to milky way tidal heating ( weinberg 2000 ; evans & kerins 2000 ) . if microlensing source stars belong to such populations , then this would yield observable signatures in their characteristics . in particular , if microlensing source stars are behind the lmc , then they should be systematically fainter ( zhao , graff & guhathakurta 2000 ) and redder ( zhao 1999 , 2000 ) than lmc disk stars . hst / wfpc2 cmds of fields surrounding microlensing events show no evidence for this , although the statistics are insufficient to rule out specific models with strong confidence ( alcock et al . other tests of the spatial distribution and properties of the lmc microlensing events also do not ( yet ) discriminate strongly between different models for the location and nature of either the lenses or the sources ( alcock et al . 2000a ; gyuk , dalal & griest 2000 ; jetzer , mancini & scarpetta 2002 ) . this is primarily because only 1317 lmc microlensing events are known . so to assess whether the lmc has out - of - plane structures it is best to focus on other types of tracers that might yield better statistics . zaritsky & lin ( 1997 ) studied the optical cmd of a field @xmath175 north - west of the lmc center and found a vertical extension at the bright end of the red clump . they suggested that this feature is due to stars 15@xmath176 above the lmc disk . however , this interpretation has been challenged for many different reasons ( e.g. , bennett 1998 ; gould 1998 , 1999 ) . most seriously , beaulieu & sackett ( 1998 ) pointed out that a vertical red clump ( vrc ) extension is naturally expected from stellar evolution due to young helium core burning stars . the same feature is seen in the fornax and sextans a dwarfs ( gallart 1998 ) . it remains somewhat open to discussion whether the number of stars in the lmc vrc is larger than expected given our understanding of the lmc star formation history , so a foreground population is not strictly ruled out ( zaritsky et al . however , it certainly does nt appear to be a favored interpretation of the data . also , the kinematics of the vrc stars is indistinguishable from that of lmc red clump stars ( ibata , lewis & beaulieu 1998 ) . more recently , zhao et al . ( 2003 ) performed a detailed radial velocity survey of 1300 stars of various types within @xmath177 from the lmc center . they found no evidence for stars with kinematics that differ significantly from that of the main lmc disk . this rules out a significant foreground or background population of stars that reside in a tidal stream that is seen superposed onto the lmc by chance ( zhao 1999 ) . any foreground or background population must be physically associated with the lmc , and share its kinematics . sub - populations within the lmc with subtly different kinematics have been suggested ( graff et al . 2000 ) , but only at low statistical significance . other arguments for stars at large distances from the lmc disk also have not been convincing . kunkel et al . ( 1997b ) argued on the basis of carbon star velocities that the lmc has a polar ring . however , the carbon star velocity field shown in figure [ f : fit ] seems to be well fit by a single rotation disk model . it is possible that the kunkel et al . study was affected by the use of an lmc transverse velocity value of only @xmath178 , which is considerably below the value indicated by presently available proper motion data ( @xmath179 ; see section [ s : orbit ] ) . weinberg & nikolaev ( 2001 ) found a tail of relatively faint stars in luminosity functions of agb stars selected by @xmath180 color from 2mass data . they suggested that this is not due to dust extinction but might indicate stars behind the lmc . on the other hand , the distribution of these stars on the sky ( van der marel , unpublished ) bears a strong resemblance to the far infrared iras map of lmc dust emission ( schwering 1989 ) . this casts doubt on the interpretation that the brightnesses of these stars have not been affected by dust . weinberg & nikolaev ( 2001 ) also found a slightly non - gaussian tail in their agb star luminosity functions towards brighter magnitudes . however , this need not indicate stars in the foreground , given that there is no a priori reason why the agb star luminosity function would have to be gaussian . in another study , alcock et al . ( 1997 ) found no evidence for unexpected numbers of rr lyrae stars at distances beyond @xmath181 from the lmc disk plane . the studies discussed in sections [ s : kinematics ] and [ s : viewing ] indicate that the overall structure of the lmc is that of a ( thick ) disk . from section [ ss : foreback ] it follows that there is no strong evidence for unexpected material far from the disk plane . however , this does not mean that there may not be non - planar structures in the disk itself . for example , it was already discussed in section [ s : viewing ] that there might be warps and twists in the disk plane . early evidence for non - planar material came from hi observations . one of the most prominent optical features of the lmc is the star forming 30 doradus complex , located just north of the eastern tip of the bar . this region is a very strong source of uv radiation , yet the hi gas disk of the lmc does not show a void in this part of the sky . this indicates that the 30 doradus complex can not be in the plane of the lmc disk , but must be at least @xmath182@xmath183 away from it ( luks & rohlfs 1992 ) . hi channel maps show that there is a separate hi component , called the `` l - component '' , that is distinct from the main lmc disk . it has lower line - of - sight velocities than the main disk by @xmath184@xmath185 , contains some 19% of the hi gas in the lmc , and does not extend beyond @xmath186@xmath187 from the lmc center ( luks & rohlfs 1992 ) . absorption studies indicate that this component is behind the lmc disk ( dickey et al . the 30 doradus complex is spatially located at the center of the l - component . it is probably directly associated with it , given that the l - component shows a hole of hi emission at the 30 doradus position as expected from ionization . these results suggest that in the central few degrees of the lmc an important fraction of the gas and stars may not reside in the main disk . = 0.6 it has now proven possible to test some of these ideas more directly by using large stellar databases . section [ s : viewing ] already mentioned the study of more than 2000 cepheids by nikolaev et al . they obtained accurate reddening - corrected ( relative ) distances to each of the cepheids . the distance residuals with respect to the best - fitting plane do not follow a random gaussian distribution . instead they show considerable structure , as shown in figure [ f : cepheids ] . the two - dimensional distribution of the residuals on the sky was interpreted as a result of two effects , namely a symmetric warp in the disk , and the fact that the bar is located @xmath188 in front of the main disk . interestingly , an offset between the lmc bar and disk had previously been suggested by zhao & evans ( 2000 ) as an explanation for the observed lmc microlensing optical depth . if the disk and the bar of the lmc are not dynamically connected , then this may also explain why in projection the lmc bar appears offset from the center of the outer isophotes . the variation of the reddening - corrected red clump magnitude over the face of the lmc has also been used to study vertical structures in the lmc disk . olsen & salyk observed 50 randomly selected fields in the central @xmath189 of the lmc at ctio . they found that fields between @xmath186@xmath189 south - west of the lmc center are @xmath190 mag brighter than expected from the best plane fit . they interpreted this to indicate that stars in these fields lie some @xmath191 above the lmc disk , and argued for warps and twists in the lmc disk plane . subramaniam ( 2003 ) used red clump magnitudes determined from stars in the ogle database to address the same issue . this yielded a map of the red clump magnitude along the length and width of the lmc bar . this map shows considerable structure and clearly can not be fit as a single plane . subramaniam ( 2004 ) suggested that the residuals can be interpreted as the result of a misaligned secondary bar inside the primary bar . eclipsing binaries have also provided interesting information on this subject . these binaries can be modeled in detail to yield a fairly accurate distance . the distances of sources studied so far seem to indicate a slightly lower lmc distance modulus of @xmath192 ( ribas et al . 2002 ) than has been inferred from other tracers ( see section [ s : orbit ] ) . however , it is possible to obtain somewhat higher values with a slightly different analysis ( groenewegen & salaris 2001 ; clausen et al . 2003 ) so this is no great cause for concern ( alves 2004b ) . what is interesting though is that four eclipsing binaries analyzed by the same team with the same method show a considerable spread in distance . this has been interpreted to mean than one of the binaries lies @xmath193 behind the lmc disk plane ( ribas et al . 2002 ) and that another one lies @xmath194 in front of it ( fitzpatrick et al . 2003 ) . the above studies indicate that there is considerable and complicated vertical structure in the central few degrees of the lmc disk . however , this is a rapidly developing field , and many important questions remain open . in particular , it is unclear whether the features reported in the various studies are actually the same or not . qualitative comparison is not straightforward . different authors study different areas of the lmc , they plot residuals with respect to different planes , and they present their results in figures that plot different types of quantities . however , cursory inspection of the various papers shows very few features that are obviously in common between the studies . quantitative comparison is therefore needed . direct comparison to the results for the outer parts of the lmc , where there is little evidence for extra - planar structures ( van der marel & cioni 2001 ) , is also important . if discrepancies emerge from such comparisons , then this can mean two things . either some studies are in error ( e.g. , due to use of inaccurate dust corrections , or by incorrect interpretation of stellar evolutionary variations as distance variations ) or different tracers do not trace the same structure . the latter might well the case . cepheids are young stars with ages less than a few times @xmath195 years whereas stars on the rgb and agb are typically older than 1 gyr . since the structure of the lmc and its gaseous component vary with time as a result of tidal interactions with the milky way and the smc , one would nt necessarily expect stars formed at different epochs to trace identical structures . another open question is what the physical and dynamical interpretation is of the extra - planar structures that are being detected . many authors have used the term `` warp '' . however , the residuals that have been reported do not much resemble the smoothly varying residuals that are expected from a single plane with a global warp . the structures appear to be both different and more complicated than a single warp . if a decomposition into different components is attempted , then such a decomposition must use entities that can be understood dynamically ( disk , bar , bulge , warp , etc . ) and that make sense in the context of the overall evolutionary history of the magellanic system . components with holes or sharp edges ( e.g. , subramaniam 2004 ) are not particularly realistic from a dynamical viewpoint . phase mixing of stellar orbits quickly removes such sharp discontinuities . also , if the vertical structures detected in the inner region of the lmc disk are due to components that are not connected to the main disk plane , then why does the projected image of the lmc ( e.g. , figure [ f : maps ] ) look so smooth ? and why is there so little evidence from stellar kinematics for components with decoupled kinematics ( e.g. , figure [ f : fit ] ; zhao et al . 2003 ) ? clearly , many questions remain to be answered before we can come to a full understanding of the vertical structure of the lmc . the structure and kinematics of the lmc continue to be active areas of research . as outlined in this review , much progress has been made recently . improved datasets have played a key role in this , most notably the advent of large stellar datasets of magnitudes in many bands , lightcurves , and line - of - sight kinematics , and also the availability of sensitive hi observations over large areas . as a result we now have a fairly good understanding of the lmc morphology and kinematics . the proper motion of the lmc is reasonably well measured and the global properties of the lmc orbit around the milky way are understood . the angles that determine how we view the lmc are now known much more accurately than before and this has led to the realization that the lmc is quite elliptical in its disk plane . we are starting to delineate the vertical structure of the lmc and are finding complexities that were not previously expected . despite the excellent progress , many questions on lmc structure still remain open . why is the bar offset from the center of the outer isophotes of the lmc ? why is the dynamical center of the hi offset from the center of the bar , from the center of the outer isophotes , and from the dynamical center of the carbon stars ? why do studies of the inner and outer regions of the lmc yield differences in line - of - nodes position angle of up to @xmath143 ? does the lmc have a pressure supported halo ? are there populations of stars at large distances from the lmc plane ? what is the origin and dynamical nature of the non - planar structures detected in the inner regions of the lmc ? do different tracers outline the same non - planar structures ? it might be necessary to answer all of these open questions before we can convince ourselves that the optical depth for lmc self - lensing has been correctly estimated . this seems to be the most critical step in establishing whether or not the milky way halo contains hitherto unknown compact lensing objects ( machos ) . the open questions about lmc structure are important also in their own right . the tidal interaction between the magellanic clouds and the milky way provides one of our best laboratories for studying the processes of tidal disruption and hierarchical merging by which all galaxies are believed to grow . a better understanding of lmc structure may also provide new insight into the origin of the magellanic stream , which continues to be debated . and with improved proper motion measurements of the magellanic clouds , the stream may become a unique tool to constrain the shape and radial density distribution of the milky way halo at radii inaccessible using other tracers . skrutskie , m. 1998 , in the impact of near - infrared sky surveys on galactic and extragalactic astronomy , proc . of the 3rd euroconference on near - infrared surveys , n. epchtein . , ed . , astrophysics and space science library , vol . 230 , 11 ( dordrecht : kluwer )

we present a scattering theory description for the inelastic current noise in the presence of electron - vibration interactions . in this description , we specify elastic and inelastic scattering contributions to the shot noise by examining charge transfers between scattering states and energy exchange between electrons and vibrations . the elastic and inelastic scattering processes are further decomposed into current correlations of electrons at the same energy and those of electrons at different energies . focusing on the inelastic noise signals defined as steps in the voltage derivative of the shot noise , we show that single - channel systems have two ranges of transmission at which the inelastic noise signals exhibit the crossover between positive and negative signs . in a high transmission regime , even and odd vibrational modes of mirror - symmetric systems provide upper and lower bounds to the ratio of the inelastic noise signal to the conductance step . this can be a theoretical justification for models used to understand the recent noise experiment [ phys . rev . lett . * 108 * , 146602 ( 2012 ) ] and numerical calculations on gold atomic chains [ phys . rev . b * 86 * , 155411 ( 2012 ) ] .

1309.0061

since pioneering experiments measuring inelastic electron tunneling spectroscopy ( iets ) signals @xcite , intense efforts have been made to understand inelastic transport properties when conducting electrons interact with local vibrations @xcite . the iets signals are identified as steps in the differential conductance at a threshold bias voltage equal to a vibrational energy @xcite . the same steps also appear in the voltage derivative of the shot noise @xcite . these steps indicate opening of inelastic transport channels , where electrons can pass through the junction by losing energy . it is known that the conductance variation undergoes a crossover from an increase to a decrease when a bare transmission @xmath0 evolves from zero to one @xcite . the similar crossover behavior in the shot noise has been reported in a recent experiment of the shot noise measurement @xcite . in ref . , it is observed that inelastic noise corrections are negative when a measured zero - bias conductance is approximately below @xmath1 . in contrast , the noise corrections are exclusively positive for the zero - bias conductance close to @xmath2 . to understand this crossover , several model systems have been theoretically investigated @xcite . the representative system is a single - level model with a single vibrating scatterer @xcite . it is shown that the conductance crossover occurs when @xmath3 for the single level symmetrically connected to single - channel electrodes @xcite . considering the current noise , the single - level model symmetrically coupled to electrodes exhibits two crossovers in the inelastic noise correction at @xmath4 @xcite . reference studied a two - site tight - binding model symmetrically connected to electrodes in order to analyze density functional theory ( dft ) calculations on inelastic signals of au atomic point contacts . for the out - of - phase longitudinal vibrational mode that gives a dominant contribution to inelastic signals , it is found that the two - site model exhibits the crossover in the conductance at @xmath3 and the noise signal crossover at @xmath4 , as the symmetric single - level model does @xcite . although these simple models can give some qualitative hint on understanding the inelastic shot noise , it has not yet been clearly understood why the models exhibit the same crossover behaviors . furthermore , performing dft calculations for a variety of au atomic chains , ref . calculated the ratios of the steps in the inelastic noise signal to the conductance changes for the chains , and compared them with the ratio of the two - site model . reference concluded that the ratio predicted by the two - site model might be an upper bound to those of the atomic chains . however , there is no theoretical justification for why the two - site model , which is originally designed to study atomic point contacts , can provide an upper bound to the computed ratios of the atomic chains varying in configurations and lengths . thus , it is needed to find out a general picture that can explain the aforementioned issues . recently we presented a scattering - state description of inelastic electron transport in a weak electron - vibration ( el - vib ) coupling regime , which is established by converting non - equilibrium green s functions ( negf ) to scattering states @xcite . in this description , we clarified elastic and inelastic contributions to conductance variations in terms of eigenchannel scattering states and associated scattering matrices . doing so , we obtained a general expression for the conductance jump of single - channel systems . while the inelastic contribution is proportional to @xmath5 , the elastic one is always negative . it leads to conclusion that the conductance crossover can generally occur below @xmath3 for single - channel systems . in this paper , we extend our scattering theory approach to the inelastic current noise , especially in order to unveil an unified picture that can comprehensively explain the existing results in the literature @xcite . we specify elastic and inelastic contributions to the current noise by examining charge transfers between scattering states and energy exchange between electrons and vibrations . different from the conductance , such contributions to the current noise can be further decomposed into current correlations of electrons at the same energy and those of electrons at different energies . considering single - channel systems , we show that the crossover for the inelastic noise signal can occur in two ranges of transmissions , _ i.e. _ , between @xmath6 and @xmath7 or between @xmath8 and @xmath9 . in particular , for mirror - symmetric junctions , even parity vibrational modes lead to two crossovers of the noise correction at @xmath4 , while there is only one crossover at @xmath3 for odd parity modes . when @xmath0 is close to one , our scattering - state description shows that the ratios of noise correction steps to conductance changes for odd and even parity modes are lower and upper bounds for those of general cases , respectively . this paper is organized as follows . in sec . [ general_picture ] , we present a general argument about how elastic and inelastic scattering processes compete with each other for transport properties such as conductance and shot noise . in sec . [ negf_method ] , we summarize current and shot noise expressions in the negf formalism @xcite . in sec . [ scattering_theory ] , we present our scattering - theory description of inelastic transport . first , we briefly summarize the current correction due to el - vib scatterings discussed in our previous work @xcite . second , we explain the scattering - state description of the inelastic noise correction , focusing on identification of scattering processes in the noise correction . in sec . [ discussion ] , we apply our theory to single - channel systems . here we discuss the crossover of the noise correction in general situations including the systems investigated in the literature @xcite . in sec . [ conclusion ] , we make a final conclusion . technical details and derivations are discussed in appendixes [ derivation ] and [ multichannel_systems ] . before presenting our scattering - theory description for the shot noise , we first discuss the competition between elastic and inelastic scattering processes in a general perspective @xcite . here we define that electron transport is elastic ( inelastic ) if the total energy of conducting electrons is conserved ( not conserved ) after electrons pass through the vibrating region . in ref . , we clarified elastic and inelastic scattering contributions to the current in the second order of el - vib couplings , and used their interplay to explain the conductance crossover . in fact , the competition between elastic and inelastic scattering processes is not limited only to the current correction @xcite , but it is generally respected in other transport properties such as the current noise @xcite . when a zero temperature regime is considered for simplicity , a many - body state of electrons and vibrons perturbed by el - vib interactions can be written as @xmath10 where @xmath11 is a ground state of no el - vib coupling at zero temperature , _ i.e. _ , @xmath12 where @xmath13 and @xmath14 denote electronic and vibronic ground states respectively . note that the vibronic state @xmath15 represents that there are @xmath16 vibrons excited . @xmath17 is a perturbed state in the @xmath16th order of el - vib couplings . since electrons emit one vibron when electrons interact with vibrations once , the first - order perturbed state is written as @xmath18 , where @xmath19 denotes the electronic state after losing its energy by one vibronic quantum @xmath20 . at the second order of the el - vib coupling , two cases are possible : the first case is that electrons regain the vibronic energy quantum from vibrations , and the other is that electrons emit another vibron . thus , the second - order perturbed state is expressed as a superposition of these two possibilities , _ i.e. _ , @xmath21 . here @xmath22 and @xmath23 denote electronic wavefunctions associated with the first case ( emission - reabsorption ) and the second one ( double - vibron emission ) respectively . for any electronic operator @xmath24 that does not change the vibronic occupation , it is shown that the expectation value of @xmath24 with respect to @xmath25 is expanded as @xmath26 where @xmath27 , and @xmath28.\end{aligned}\ ] ] here eq . ( [ inelastic_general ] ) is the inelastic correction since it includes only @xmath29 where electrons lose energy by emitting one vibron . since the vibronic state @xmath30 associated with @xmath31 is orthogonal to @xmath14 of the ground state , @xmath31 does not contribute to @xmath32 . thus , eq . ( [ elastic_general ] ) is reduced to @xmath33 $ ] , thereby implying that it is the elastic process . since the elastic correction eq . ( [ elastic_general ] ) is a form of interference between @xmath13 and @xmath22 , it is expected to contain some phase information . in fact , the elastic corrections to the current and the noise depend on phases of scattering matrix elements and transition amplitude between scattering states as shown later . based on this general structure that elastic and inelastic scattering processes compete with each other in the nontrivial lowest order of el - vib couplings , we will discuss the inelastic current noise in detail in the following sections . the system in which we are interested is the two - terminal geometry with local vibrations located in a conductor region . the corresponding hamiltonian is written as @xmath34 where @xmath35 and @xmath36 are the electronic hamiltonian describing the two - terminal setup and the vibronic one of local vibrations respectively . @xmath37 is the coupling term between electronic and vibronic hamiltonians . the electronic hamiltonian @xmath35 is further divided into two electrodes ( left and right ) @xmath38 , a central part @xmath39 connected to leads , and a coupling @xmath40 between leads and the central device . in the second quantized form , it is @xmath41 where @xmath42 here @xmath43 ( @xmath44 ) and @xmath45 ( @xmath46 ) represent electronic creation ( annihilation ) operators for the electrodes and the device part , respectively . note that a spin index is not explicitly shown . @xmath36 describes local vibrations confined in the device region , which consists of a collection of harmonic oscillators , @xmath47 where @xmath48 ( @xmath49 ) is a vibronic creation ( annihilation ) operator for the @xmath50th mode . the coupling hamiltonian between electrons and local vibrations is given by @xmath51 next we consider the non - equilibrium transport theory based on negfs in a weak el - vib coupling regime @xcite . in a zero temperature limit and a regime where a damping rate of vibrations is much larger than a heating rate , the current correction due to el - vib interactions is @xmath52 \label{qel_current } \\ & & + \frac{2e}{h } \sum_{\lambda } \int_{\mu_{r}+\hbar\omega_{\lambda}}^{\mu_{l}}d\varepsilon\textrm{tr}\left[b\mathcal{m}^{\lambda}a_{r}^{- } \mathcal{m}^{\lambda}\right ] \\ & & + \frac{2e}{h } \sum_{\lambda } \int^{\mu_{l}}_{\mu_{r}+\hbar\omega_{\lambda } } d\varepsilon \textrm{im } \textrm{tr } \left [ c \mathcal{m}^{\lambda } a_{r}^{- } \mathcal{m}^{\lambda } \right ] \\ & & + \frac{2e}{h } \sum_{\lambda } \int^{\mu_{l}-\hbar\omega_{\lambda}}_{\mu_{r } } d\varepsilon \textrm{im } \textrm{tr } \left [ c \mathcal{m}^{\lambda } a_{l}^{+ } \mathcal{m}^{\lambda } \right],\end{aligned}\ ] ] where @xmath53 , @xmath54 , and @xmath55 . @xmath56 is the retarded ( advanced ) green s function @xcite of the conductor part without el - vib interactions , which is expressed as @xmath57^{-1}$ ] . here @xmath58 denotes the retarded ( advanced ) lead self - energy of the electrode @xmath59 . @xmath60 $ ] is the coupling function to leads , and @xmath61 is the spectral function of the conductor region originating from the electrode @xmath59 @xcite . superscripts @xmath62 indicate that energy argument is @xmath63 and without @xmath62 , the argument is @xmath64 . we assume that the left chemical potential @xmath65 is bigger than the right one @xmath66 . the first term [ eq . ( [ qel_current ] ) ] is interpreted as a quasi - elastic correction to a bare transmission @xcite . since eq . ( [ qel_current ] ) does not contribute to the conductance step at the threshold bias voltage in the regime of our interest , it is neglected in the following discussion . to the second order of el - vib interactions , the shot noise correction @xmath67 has been derived by using the counting field method in the full - counting statistics @xcite . the inelastic noise correction in the single - level model was first investigated by several groups @xcite , and later it was extended to general cases where multiple electronic levels and many vibrational modes are involved in a regime of equilibriated vibrations @xcite . recently the backaction effect of nonequilibrium vibrational populations on the shot noise has been investigated @xcite . following ref . , we consider situations where multiple electronic levels are coupled to many vibrational modes in the regime of zero temperature and equilibriated vibrons . the noise correction @xmath67 has two contributions @xmath68 and @xmath69 , which are classified as _ mean - field _ and _ vertex _ corrections in the diagrammatic representation @xcite : @xmath70 \\ \label{noise_mf1 } & & + \sum_{\lambda}\int_{\mu_{r}+\hbar\omega}^{\mu_{l}}d\varepsilon\textrm{tr}\left[\left(1 - 2t\right)g^{r}_{0}\mathcal{m}^{\lambda}a_{r}^{-}\mathcal{m}^{\lambda}g^{a}_{0}\gamma_{l}\right ] \nonumber \\ \\ \label{noise_mf2 } & & + \sum_{\lambda}\int_{\mu_{r}+\hbar\omega}^{\mu_{l}}d\varepsilon\textrm{im}\textrm{tr}\left[\left(1 - 2t\right)c\mathcal{m}^{\lambda}a_{r}^{-}\mathcal{m}^{\lambda}\right ] \nonumber \\ \\ \label{noise_mf3 } & & + \sum_{\lambda}\int_{\mu_{r}}^{\mu_{l}-\hbar\omega}d\varepsilon\textrm{im}\textrm{tr}\left[\left(1 - 2t\right)c\mathcal{m}^{\lambda}a_{l}^{+}\mathcal{m}^{\lambda}\right ] \nonumber \\\end{aligned}\ ] ] and @xmath71 \\ \label{noise_vc2 } & & -\sum_{\lambda}\int_{\mu_{r}+\hbar\omega}^{\mu_{l}}d\varepsilon2\textrm{re}\textrm{tr}\left[c\mathcal{m}^{\lambda}c^{-}\mathcal{m}^{\lambda}\right ] \nonumber \\ \\ \label{noise_vc3 } & & -\sum_{\lambda}\int_{\mu_{r}+\hbar\omega}^{\mu_{l}}d\varepsilon2\textrm{im}\textrm{tr}\left[c\mathcal{m}^{\lambda}d^{-}\mathcal{m}^{\lambda}\right ] \nonumber \\ \\ \label{noise_vc4 } & & + \sum_{\lambda}\int_{\mu_{r}}^{\mu_{l}-\hbar\omega}d\varepsilon2\textrm{im}\textrm{tr}\left[c\mathcal{m}^{\lambda}d^{+}\mathcal{m}^{\lambda}\right ] , \nonumber \\\end{aligned}\ ] ] where @xmath72 . similarly to the current , eq . ( [ qel_noise ] ) representing the noise correction due to the quasi - elastic current [ eq . ( [ qel_current ] ) ] does not contribute to steps in the noise signal . thus it will be ignored in our discussion . we note that another contribution showing an asymmetric behavior with respect to the applied bias voltage is also ignored @xcite . this asymmetric contribution leads to a logarithmic divergent correction at the threshold bias voltage in the zero temperature limit @xcite . in fact , it indicates that the lowest - oder perturbation theory is not valid at the threshold bias voltage @xcite . this divergece might be regularized by introducing damping of vibrational modes , or using a resummation scheme of diagrams @xcite . away from the threshold bias voltage , the asymmetric term becomes much smaller than the symmetric contribution . in addition , it is known that the asymmetric term is negligible for symmetric junctions or for cases where the conductor is close to or far from resonances @xcite . we also remark that the hartree diagram is not taken into account since it does not lead to step behaviors at the threshold voltage @xcite . ( color online ) schematic illustration of left - incident and right - incident eigenchannel scattering states . the scattering state consists of incident , transmitted and reflected waves , which are represented by arrows . ] the current and noise corrections based on negfs can be expressed in terms of scattering states . in the landauer - bttiker formalism @xcite , coherent transport is described by scattering states @xmath73 and a scattering matrix @xmath74 associated with the states . here indices @xmath75 denote transverse modes called quantum channels @xcite . the scattering matrix @xmath74 is written in the form of a block matrix , @xmath76 where @xmath77 and @xmath78 ( @xmath79 and @xmath80 ) are reflection and transmission submatrices for left - incident ( right - incident ) waves . note that the energy normalization @xmath81 is used for scattering states @xcite . scattering states @xmath82 are related to green s functions of the conductor part @xmath83 in the following way @xcite : @xmath84 where @xmath85 and @xmath86 is the coupling hamiltonian between the conductor and the electrode @xmath87 . @xmath88 is the scattering state when @xmath89 , which is just a sum of the incident wave and the totally reflected one . @xmath90 is written as @xmath91 . then the scattering matrix @xmath92 @xcite is @xmath93 in this work , instead of using the scattering states @xmath94 , we choose the transmission eigenchannel representation @xcite @xmath95 , in which there is no inter - channel mixing ( see fig . [ scattering_states ] ) . as seen below , el - vib interactions lead to mixing between different eigenchannel states . the transmission eigenchannel representation is achieved by diagonalizing the submatrices @xmath77 , @xmath78 , @xmath79 , and @xmath80 . when the system respects the time - reversal symmetry , it can be done by choosing unitary matrices @xmath96 and @xmath97 as follows : @xmath98_{mn } = r_{m } \delta_{mn}$ ] , @xmath99_{mn } = t_{m } \delta_{mn}$ ] , @xmath100_{mn } = r_{m}^\prime \delta_{mn}$ ] , and @xmath101_{mn } = t_{m}^\prime \delta_{mn}$ ] . see ref . for details . consequently , the original scattering matrix @xmath74 is transformed to the scattering matrix @xmath92 which is decomposed into a collection of @xmath102 block scattering matrices @xmath103 for eigenchannel states @xmath95 . @xmath104 is related to @xmath105 via @xmath106_{mn } |\psi_{\alpha n}\rangle$ ] . if the system respects the time - reversal symmetry , one can prove following relations by using eq . ( [ identity1 ] ) : @xmath107 where @xmath108 is the time - reversal operator . note that these relations are true not only far from the scattering region , but also inside the conductor . ( color online ) schematic explanation of scattering processes contributing to current and noise corrections . while solid lines represent parts of the scattering states that electrons follow when they transport to the right electrode , dashed lines are the rest of the scattering states . blue and red lines indicate vibronic emission and absorption processes , respectively . gray solid lines represent electron transport without interacting with vibrations , which gives a bare transmission . orange wiggly lines indicate vibronic energy transfer . on the left side , scattering states and energies corresponding to arrows are indicated . scattering states at the same energy are vertically shifted for clarity . ( a ) inelastic contribution to current [ eq . ( [ inelastic ] ) ] . ( b ) one - electron scattering process in the elastic contribution to current [ eq . ( [ elastic ] ) ] . ( c ) elastic scattering process involving two electrons ( green and brown circles ) in eq . ( [ elastic ] ) . ( 1 ) and ( 2 ) indicate the order of scattering events . transmission corrections for ( a ) , ( b ) , and ( c ) scattering processes are indicated on the right side . ( d ) and ( e ) represent inelastic noise corrections where conducting electrons at two energies differing by @xmath20 are correlated . ( d ) and ( e ) correspond to eqs . ( [ inelastic_noise2 ] ) and ( [ elastic_noise2 ] ) , respectively . note that ( a ) and ( d ) correspond to @xmath109 , while ( b ) , ( c ) , and ( e ) describe @xmath110 of @xmath111 in sec . [ general_picture ] . ] in ref . , using eqs . ( [ identity1])-([identity2 - 2 ] ) , we obtained the current correction @xmath112 in terms of scattering states : @xmath113 where @xmath114 and @xmath115 \nonumber\\ & & + \frac{2e}{h}\left(2\pi\right)^{2}\sum_{\lambda}\sum_{m , n}\int_{\mu_{r}}^{\mu_{l}-\hbar\omega_{\lambda}}d\varepsilon \nonumber \\ & & \textrm{re}\left[r_{m}^{\prime}t_{m}^{*}\langle\phi_{rm}|\mathcal{m}^{\lambda}|\phi_{ln}^{+}\rangle\langle\phi_{ln}^{+}|\mathcal{m}^{\lambda}|\phi_{lm}\rangle\right ] , \nonumber \\\end{aligned}\ ] ] where @xmath116 and @xmath117 . equation ( [ inelastic ] ) represents the inelastic scattering process where a conducting electron is scattered off by emitting one vibron @xmath20 ( see fig . [ current_noise](a ) ) . in contrast , eq . ( [ elastic ] ) describes the elastic scattering contribution to the current with a vibrational emission - reabsorption process . the first term of eq . ( [ elastic ] ) is an one - electron scattering process where only one electron is involved in the vibrational emission - reabsorption process as depicted in fig . [ current_noise](b ) . in this case , energy of each conducting electron does not change . the second term of eq . ( [ elastic ] ) involves successive scatterings of two electrons with local vibrons ( see fig . [ current_noise](c ) ) . one vibrational energy quantum @xmath118 is transferred from one electron to the other via the vibrational emission - reabsorption process . while energies of the two electrons involved in the two - electron process change ( one increases and the other decreases ) , the total energy of conducting electrons does not change , so it is regarded as the elastic process . in fact , eqs . ( [ inelastic ] ) and ( [ elastic ] ) correspond to the first and second born approximations in a standard scattering theory @xcite . see ref . for this specification based on charge transfers between scattering states . for the inelastic process , the right scattering state @xmath119 should be empty , when the electron is incident at @xmath64 from the left side . otherwise , the inelastic scattering is prohibited due to the pauli exclusion principle . the inelastic scattering , therefore , can occur when @xmath120 $ ] . two elastic scattering processes take place at different energy ranges . let us consider elastic corrections at @xmath64 . similar to the inelastic process , the one - electron process requires an empty right scattering state @xmath119 that a left - incident electron at @xmath64 can occupy after emitting one vibron . in contrast , the two - electron process assumes a left - incident electron occupying @xmath121 , which is first scattered off to an empty right scattering state @xmath122 @xcite . considering these , one can realize that the one - electron and two - electron elastic processes can occur at @xmath120 $ ] and @xmath123 $ ] respectively , as shown in eq . ( [ elastic ] ) . these scattering processes can be written as transmission corrections to a bare transmission @xmath124 when there is no el - vib interaction . considering scattering with the @xmath50th vibrational mode , the inelastic scattering process [ eq . ( [ inelastic ] ) ] gives two transmission corrections at @xmath64 and @xmath125 : @xmath126 where @xmath127.\end{aligned}\ ] ] here @xmath128 denotes the step function . @xmath129 accounts for the energy window in which the inelastic scattering is possible as discussed above . when interacting with the @xmath50th vibrational mode , the electron at the left scattering state @xmath130 is inelastically scattered off to the right scattering state @xmath131 with the scattering probability @xmath132 . in comparison with the case where there is no el - vib interaction , the transmission of the @xmath133th eigenchannel at @xmath64 is reduced by eq . ( [ inelastict1 ] ) in which the total scattering probability is taken into account by summing over all right - incident eigenchannels . for electrons scattered to the right scattering state @xmath131 , the transmission probability to move to the right electrode is given by the reflection amplitude @xmath134 . thus , by considering the scattering probability @xmath132 together , the transmission correction of scattered electrons from the @xmath133th eigenchannel to the @xmath16th eigenchannel is given by @xmath135 . summing over all incident states @xmath130 , the total transmission correction to the eigenchannel @xmath16 is given by eq . ( [ inelastict2 ] ) . in contrast , the elastic scattering process gives corrections to the transmission at @xmath64 , @xmath136 where @xmath137\mathcal{f } ( \varepsilon ) \\ \label{elastict2 } \delta \mathcal t^{\textrm{el},\lambda}_{2,m}(\varepsilon ) & = & ( 2\pi)^2 \sum_{n}\textrm{re}\left[r_{m}^{\prime}t_{m}^{*}\langle\phi_{rm}|\mathcal{m}^{\lambda}|\phi_{ln}^{+}\rangle\langle\phi_{ln}^{+}|\mathcal{m}^{\lambda}|\phi_{lm}\rangle\right]\mathcal{f } ( \varepsilon + \hbar\omega_{\lambda } ) . \end{aligned}\ ] ] equations ( [ elastict1 ] ) and ( [ elastict2 ] ) corresponds to corrections due to one - electron and two - electron elastic processes respectively . in terms of these transmission corrections , the current corrections @xmath138 and @xmath139 are written as @xmath140 using eqs . ( [ identity1])-([identity2 - 2 ] ) , one can obtain the scattering state description of the noise correction : @xmath141 where @xmath142 @xmath143 @xmath144 \nonumber \\ & & + \left(2\pi\right)^{2}\sum_{\lambda , m , n}\int_{\mu_{r}}^{\mu_{l}-\hbar\omega_\lambda}d\varepsilon\left(1 - 2\mathcal{t}_{m}\right)\textrm{re}\left[r_{m}^{\prime}t_{m}^{*}\langle\phi_{rm}|\mathcal{m}^{\lambda}|\phi_{ln}^{+}\rangle\langle\phi_{ln}^{+}|\mathcal{m}^{\lambda}|\phi_{lm}\rangle\right]\end{aligned}\ ] ] @xmath145 here eqs . ( [ inelastic_noise1 ] ) and ( [ inelastic_noise2 ] ) are interpreted as inelastic processes , while eqs . ( [ elastic_noise1 ] ) and ( [ elastic_noise2 ] ) are elastic ones . we first consider eqs . ( [ inelastic_noise1 ] ) and ( [ elastic_noise1 ] ) . they can be understood as current correlations of electrons at one energy @xmath64 . when there is no el - vib interaction , the shot noise at zero temperature is written as @xmath146 if the transmission @xmath124 is slightly changed by @xmath147 , _ i.e. _ , @xmath148 , the shot noise is corrected as @xmath149 , where @xmath150 in the previous subsection , it is discussed that the inelastic scattering process leads to transmission corrections @xmath151 [ eq . ( [ inelastict1 ] ) ] and @xmath152 [ eq . ( [ inelastict2 ] ) ] to bare transmissions @xmath124 and @xmath153 , respectively . for the elastic contribution that consists of one - electron and two - electron processes as discussed before , @xmath154 [ eq . ( [ elastict ] ) ] is added to a bare transmission @xmath124 . thus , when these transmission corrections @xmath151 , @xmath152 , and @xmath154 are plugged into eq . ( [ noise_correction ] ) . one can check that eqs . ( [ inelastic_noise1 ] ) and ( [ elastic_noise1 ] ) are recovered . further , eqs . ( [ inelastic_noise1 ] ) and ( [ elastic_noise1 ] ) are identified as inelastic and elastic corrections to the shot noise respectively , considering that they are derived from the transmission corrections due to inelastic and elastic scattering processes . in contrast , eqs . ( [ inelastic_noise2 ] ) and ( [ elastic_noise2 ] ) accounts for current correlations of electrons at two different energies @xmath64 and @xmath125 . to be specific , eq . ( [ inelastic_noise2 ] ) describes the current correlation between an electron of @xmath130 that is not scattered by vibrons , and another electron that initially occupies @xmath130 , but is transferred to @xmath131 by emitting one vibron ( see fig . [ current_noise](d ) ) . on the contrary , eq . ( [ elastic_noise2 ] ) is the current correlation of two electrons exchanging a vibronic energy @xmath20 ( see fig . [ current_noise](e ) ) . the electron of @xmath130 is first scattered off to the right scattering state @xmath131 by emitting one vibron . after that , another electron of @xmath155 absorbs the vibron , thereby being excited to @xmath156 . considering whether the total electronic energy is conserved or not , it is obvious that eqs . ( [ inelastic_noise1 ] ) and ( [ elastic_noise2 ] ) are inelastic and elastic current correlations , respectively . note that eq . ( [ elastic_noise2 ] ) is written in the interference - like form of eq . ( [ elastic_general ] ) , which is identified as elastic in sec . [ general_picture ] . we remark how eqs . ( [ inelastic_noise1])-([elastic_noise2 ] ) are related to the mean - field correction @xmath68 and the vertex correction @xmath69 @xcite . inelastic corrections eqs . ( [ inelastic_noise1 ] ) and ( [ inelastic_noise2 ] ) are derived from both @xmath68 and @xmath69 . while eq . ( [ elastic_noise1 ] ) arises solely from the mean - field correction @xmath68 , eq . ( [ elastic_noise2 ] ) is derived from the vertex correction @xmath69 . see appedix [ derivation ] for detailed derivations . this scattering - state description enables to quantitatively calculate intra - channel and inter - channel scattering contributions to the inelastic signal as well as elastic and inelastic ones . it can be complementary to the existing analysis based on scattering rates @xcite that helps qualitatively figure out what scattering process is dominant in the inelastic signal , which is known as the propensity rule @xcite . for example , it is possible that some vibrational modes , which are minor in the conductance steps , can give visible contributions to the inelastic noise signals , as reported in ref . . different visibilities of those modes in the conductance and the noise signal are not understood by calculating scattering rates , which are in the form of fermi s golden rule @xcite . instead , our description can give a quantitative explanation to those visibilities by clarifying the interplay of intra - channel and inter - channel scattering contributions , or that of elastic and inelastic scattering processes . see appendix [ multichannel_systems ] for further discussions . ( color online ) crossover transmissions for the conductance and the shot noise . purple dotted line and blue solid line represent possible crossover transmissions of conductance and noise corrections , respectively . ] we apply our scattering - theory description of the current and noise corrections to single - channel systems . we further adopt the approximation known as the _ extended wide - band limit _ ( ewbl ) , in which green s functions and coupling functions are replaced by those at fermi energy @xmath157 : @xmath158 and @xmath159 @xcite . in this approximation , scattering states and their scattering matrices are replaced by those at @xmath157 . the ewbl can be valid when the density of states of the system is slowly varying over a few vibrational energies around @xmath157 . in ref . , we show that the conductance change at the threshold voltage @xmath160 is written as @xmath161 where @xmath162 and @xmath163 $ ] . by solving @xmath164 , it is shown that the crossover transmission @xmath165 for the conductance satisfies the following relation : @xmath166 since @xmath167 generally depends on @xmath0 ( and other system parameters ) , eq . ( [ crossover ] ) does not provide the analytic expression of the crossover transmission @xmath165 . one can nonetheless find out a possible range of the crossover transmission @xmath165 from eq . ( [ crossover ] ) . considering @xmath168 , it can be shown that @xmath169 , where the equality holds when @xmath170 . when @xmath171 , @xmath0 becomes negative , and it implies that there is no crossover ( see fig . [ fig_crossover ] ) @xcite . following ref . , the _ inelastic noise signal _ is defined as @xmath172 where @xmath173 stands for a small positive value . the inelastic noise signal in the ewbl is @xmath174 \left| \mathbb{m}^{\lambda}_{rl } \right|^2.\end{aligned}\ ] ] @xmath175 is also decomposed into inelastic and elastic scattering contributions , @xmath176 , where @xmath177\left| \mathbb{m}^{\lambda}_{rl } \right|^2.\end{aligned}\ ] ] here eqs . ( [ inelastic_noise1 ] ) and ( [ inelastic_noise2 ] ) [ eqs . ( [ elastic_noise1 ] ) and ( [ elastic_noise2 ] ) ] give rise to @xmath178 [ @xmath179 . while @xmath180 is determined only by @xmath0 , @xmath181 depends on the phase information of the system via @xmath182 , as discussed in sec . [ general_picture ] . in contrast to the conductance variation where the elastic contribution is always negative , the elastic term @xmath181 can be either positive or negative . from eq . ( [ noise_single_channel ] ) , it is readily shown that the crossover transmission @xmath183 for the inelastic noise signal satisfies @xmath184 we plot two solutions of eq . ( [ crossover_noise ] ) ( blue solid lines ) in fig . [ fig_crossover ] . as shown , the crossover for the noise signal can occur between @xmath6 and @xmath7 or between @xmath8 and @xmath9 . the crossover transmission @xmath183 can not exceed @xmath9 . ( color online ) the ratio @xmath185 in a high transmission regime . purple dashed line and blue solid line indicate the ratios @xmath185 of @xmath186 and @xmath187 , respectively . when @xmath188 , the ratio is located in the green region . ] now we focus on systems with the mirror - reflection symmetry along the transport direction . mirror - symmetric systems are of particular interest because bridge junctions in many experiments roughly respect this symmetry . as discussed in ref . , one can analytically calculate the crossover transmissions since @xmath189 is just a number for mirror - symmetric systems . for mirror - symmetric systems , vibrational modes are either even or odd mirror - symmetric . when @xmath190 denotes the mirror - reflection operator , even and odd mirror - symmetric modes satisfy the following relations @xcite : @xmath191 moreover , left - incident and right - incident scattering states are related to each other as @xmath192 @xcite . using these relations , it is found that @xmath170 @xmath193 for even ( odd ) modes @xcite . it means that the elastic contribution to the conductance variation , which is @xmath194 in eq . ( [ conductance_single_channel ] ) , vanishes for even vibrational modes , while it becomes @xmath195 for odd modes . from eq . ( [ crossover ] ) , it is shown that even vibrational modes exhibit the crossover of the conductance step at @xmath3 , while odd modes do not show any crossover in the conductance step ( see fig . [ fig_crossover ] ) . the inelastic noise signal also exhibits different crossover behaviors , depending on whether vibrational modes are even or odd . for even vibrational modes @xmath196 ) , the inelastic noise signal changes its sign at @xmath197 . in contrast , odd vibrational modes @xmath198 ) have only one transition between positive and negative noise signals at @xmath3 . this result is consistent with the fact that sign changes in the inelastic noise signal take place at @xmath197 in the single - level model @xcite and the two - site model interacting with the even - parity mode @xcite , both of which are symmetrically coupled to electrodes . while the aforementioned models deal only with even modes , the parity effect on the crossover behavior of the noise signal is observed in tight - binding models of @xmath199-site atomic chains @xcite . for chains with an even number of sites @xmath200 where even vibrational modes dominantly contribute to conductance and noise corrections , it is reported that the noise signal crossover occurs at @xmath4 . by contrast , the crossover of the noise signal at @xmath3 is observed for chains of an odd number of sites @xmath201 where odd modes are dominant in conductance and noise corrections . our theory shows that the crossover behaviors reported in the previous studies @xcite are not coincident , but they are generally expected in any single - channel system respecting the mirror symmetry . reference also performed dft calculations on gold atomic chains varying in lengths and configurations . the computed ratios of @xmath175 to @xmath202 for the atomic chains are compared with the ratio predicted by the two - site model @xcite . the calculated ratios do not exactly match that of the two - site analytic model . rather , it seems that the ratio of the analytic model might be an upper bound to the ratios calculated from dft results . from eqs . ( [ conductance_single_channel ] ) and ( [ noise_single_channel ] ) , the ratio of @xmath175 to @xmath203 is given by @xmath204 where @xmath205 . since the out - of - phase longitudinal mode in the two - site model of ref . is even mirror - symmetric , it is readily noticed that @xmath206 . thus eq . ( [ inelastic_ratio ] ) gives the same expression as the ratio obtained from the two - site model in ref . , @xmath207 our theory also predicts the ratio @xmath208 for odd vibrational modes , which is not investigated in ref . : @xmath209 in fig . [ fig_ratio ] , purple dashed line and blue solid one represent eqs . ( [ even_ratio ] ) and ( [ odd_ratio ] ) respectively in a high transmission regime ( @xmath210 ) that the atomic chains of ref . belong to . for general cases where the mirror symmetry is broken , @xmath167 can have a nonzero value less than unity , _ i.e. _ , @xmath211 . in this case , the ratio @xmath208 of eq . ( [ inelastic_ratio ] ) is located between eqs . ( [ even_ratio ] ) and ( [ odd_ratio ] ) ( see the green region in fig . [ fig_ratio ] ) . our theory confirms that eq . ( [ even_ratio ] ) is indeed an upper bound to the ratio @xmath208 as speculated in ref . . furthermore , eq . ( [ odd_ratio ] ) is predicted to be a lower bound to the ratio @xmath208 in a high transmission regime . ( color online ) schematic illustration of crossover transmissions for the non - symmetric single - level model . equation ( [ slm_cos ] ) ( yellow dashed lines ) for @xmath212 , @xmath213 , @xmath8 , and @xmath214 is plotted with eq . ( [ crossover ] ) ( purple dotted line ) and eq . ( [ crossover_noise ] ) ( blue solid lines ) . for @xmath215 , eq . ( [ slm_cos ] ) meets the upper branch of eq . ( [ crossover_noise ] ) at @xmath3 . ] note that the ratios @xmath208 for some chains simulated in ref . are out of the region determined by eq . ( [ inelastic_ratio ] ) . it might be because the gold atomic chains are not strictly single - channel systems , but are multichannel with one dominant channel and a few minor channels . if it is true , our single - channel result [ eq . ( [ inelastic_ratio ] ) ] can not be applied . in fact , three configurations [ @xmath216 , @xmath217 , and @xmath218 of ref . have bare transmissions larger than unity , implying that they are multichannel . further , some other cases such as @xmath219 and @xmath220 of ref . have the ratios slightly outside the region determined by eq . ( [ inelastic_ratio ] ) . for these cases , our main results eqs . ( [ elastic_noise1])-([elastic_noise2 ] ) , which are generalized for multichannel systems , can be used to verify possible effects of minor channels on such deviations . ( color online ) phase diagram of inelastic noise signals @xmath175 for the single - level model . ( a ) blue dashed line and purple dotted line represent @xmath221 [ eq . ( [ inelastic_noise_signal ] ) ] and @xmath222 [ eq . ( [ elastic_noise_signal ] ) ] , respectively . the first and second signs in parenthesis denote those of @xmath180 and @xmath181 respectively . phase boundaries @xmath223 ( green solid lines ) are located in the region where @xmath180 and @xmath181 have opposite signs . ( b ) the phase diagram drawn together with @xmath224 ( black dashed line ) . blue and purple regions indicate @xmath225 and @xmath226 , respectively . @xmath227 and the upper boundary of @xmath223 meet at @xmath228 . ] in the previous subsection , we discussed possible values of @xmath229 and @xmath229 without specifying @xmath167 that is generally a function of @xmath0 , and we considered the mirror - symmetric cases where @xmath167 is either @xmath6 or @xmath230 . in this subsection , we consider general situations where the mirror symmetry is broken . when @xmath167 is specified for a given system , crossover transmissions @xmath165 and @xmath231 can be obtained by calculating points where eqs . ( [ crossover ] ) and ( [ crossover_noise ] ) intersect with @xmath167 respectively . we take two simple models , ( 1 ) the single - level model and ( 2 ) the @xmath232 atomic chain , to illustrate how @xmath233 depends on @xmath0 and other parameters , and how crossover transmissions are determined . first we revisit the single - level model coupled to a single local vibration @xcite . it is shown that @xmath234 is given by @xmath235 where @xmath236 is a dimensionless parameter measuring a relative strength of couplings to left and right electrodes . note that the maximum transmission is given by @xmath237 . in fact , the constraint @xmath238 of eq . ( [ slm_cos ] ) leads to @xmath239 . crossing points of eqs . ( [ crossover ] ) and ( [ crossover_noise ] ) with eq . ( [ slm_cos ] ) are crossover transmissions for the conductance and the noise signal , respectively . figure [ fig_slm ] illustrates how eq . ( [ slm_cos ] ) intersects with eqs . ( [ crossover ] ) and ( [ crossover_noise ] ) for @xmath240=@xmath241 , @xmath213 , @xmath8 and @xmath214 . note that eqs . ( [ slm_cos ] ) and ( [ crossover ] ) cross once , thereby implying that there is only one crossover for the conductance step . in contrast , the number of the crossovers for the inelastic noise signal can change , depending on @xmath240 . when @xmath242 , there are two crossing points of eqs . ( [ crossover_noise ] ) and ( [ slm_cos ] ) . on the other hand , when @xmath243 , eq . ( [ slm_cos ] ) intersects once only with the lower branch of eq . ( [ crossover_noise ] ) . all of these observations are well consistent with phase diagrams of the single level model @xcite . here we re - draw the phase diagram for the inelastic noise signal @xmath175 in fig . [ slm_phase ] , emphasizing signs of elastic and inelastic contributions @xmath181 and @xmath180 . for @xmath244 , it is shown that there are two phase boundaries ( green solid lines ) separating @xmath225 and @xmath226 in fig . [ slm_phase](b ) . the upper phase boundary of @xmath223 ends at @xmath228 where it meets @xmath227 . for @xmath245 , there is only one phase boundary in a low transmission regime . ( color online ) schematic illustration of the @xmath199=@xmath230 atomic chain model . three orbitals @xmath246 , @xmath247 , and @xmath248 constitute the device region . @xmath249 is a hopping parameter bewteen nearest neighbors . the central atom is allowed to vibrate . ] ( color online ) crossover transmissions of the conductance and the shot noise for the @xmath199=@xmath230 chain model : ( a ) @xmath250 , ( b ) @xmath251 , and ( c ) @xmath252 . purple dotted line , blue solid line and yellow dashed line denote trajectories of eqs . ( [ crossover ] ) , ( [ crossover_noise ] ) , and ( [ atc_cos ] ) , respectively . ] as the second example , we consider the tight - binding model for the shortest atomic chain , _ i.e. _ , the @xmath199=@xmath230 chain , which can be a simple model with odd mirror - symmetric vibrations . the central region consists of three orbitals : @xmath246 and @xmath248 are located at outermost atoms of electrodes and @xmath247 is at a central atom vibrating between electrodes . the corresponding hamiltonian is @xmath253,\ ] ] where @xmath46 and @xmath254 are electronic annihilation and creation operators of the @xmath255th site , respectively . the hopping parameter @xmath249 is taken to be real for simplicity . considering a longitudinal motion of the central atom , its vibrational mode is odd mirror - symmetric . the el - vib coupling hamiltonian is written as @xmath256 \left(a+a^{\dagger}\right)\ ] ] when left and right coupling functions are simply given by @xmath257 and @xmath258 respectively , one can show that @xmath259 and @xmath260 \right)\right|^2 } , \nonumber \\ \end{aligned}\ ] ] where @xmath261 , @xmath262 , and @xmath263 . @xmath167 ( [ atc_cos ] ) ] implicitly depends on @xmath0 [ eq . ( [ atc_t ] ) ] via system parameters such as @xmath264 , @xmath265 , and @xmath240 . for the symmetric coupling @xmath266 , eq . ( [ atc_cos ] ) becomes unity ( @xmath167=@xmath230 ) as predicted in the previous subsection . note that @xmath167 is indeterminate at @xmath267 . since @xmath268 for @xmath267 , @xmath182 is not defined . in this perfect transmission case , one can directly calculate @xmath269 and @xmath270 by using original expressions , eqs . ( [ elastict1 ] ) , ( [ elastict2 ] ) , and ( [ inelastic_noise1])-([elastic_noise2 ] ) . next we consider non - symmetric cases , _ i.e. _ , @xmath271 . for simplicity , we assume that non - symmetric electrode couplings do not alter the hopping parameter @xmath249 and the vibrational mode in the conductor region . [ fig_atc ] , trajactories of eqs . ( [ atc_cos ] ) and ( [ atc_t ] ) ( yellow dashed lines ) are plotted by tuning the parameter @xmath264 for @xmath250 , @xmath8 , and @xmath272 . here @xmath265 is @xmath8 . figure [ fig_atc ] illustrates how crossover transmissions for noise signals qualitatively change as @xmath240 varies . for @xmath250 , @xmath167 intersects four times only with the lower branch of eq . ( [ crossover_noise ] ) . when we increase @xmath240 to @xmath8 , @xmath167 meets both upper and lower branches of eq . ( [ crossover_noise ] ) . as @xmath240 approaches @xmath230 , crossing points with the lower branch disappear , and crossover transmissions of the upper branch become closer to @xmath8 , which is the crossover transmission for the odd mirror - symmetric case . we also notice that @xmath273 , which implies the particle - hole symmetry , is of particular interest for non - symmetric junctions ( @xmath271 ) . in this case , @xmath167 always vanishes , irrespective of @xmath274 and @xmath265 , and thus it is expected that @xmath275 and @xmath276 for any non - symmetric coupling . in this paper , we have presented the scattering - state description of the inelastic shot noise in a regime of a weak el - vib coupling and equilibrated vibrons . as discussed in the inelastic current , the inelastic shot noise is determined by the interplay of elastic and inelastic scattering processes . the elastic and inelastic contributions to the current noise are further decomposed into current correlations of electrons at the same energy and those of electrons at two energies that differ by the vibrational energy @xmath118 . applied to single - channel systems , our description enables to find out two ranges of transmission at which the crossover in the inelastic noise signal can take place . in particular , for mirror - symmetric systems , we have shown that even parity modes lead to the crossover at @xmath4 , while the crossover occurs at @xmath3 for odd parity ones . considering the ratio @xmath277 , we have confirmed that the ratio @xmath277 of the even parity mode is indeed an upper bound to ratios of general cases in a high transmission regime as speculated in ref . , and further we have predicted that the ratio @xmath277 of the odd parity mode is a lower bound . our scattering - state description is formulated for general situations involving many electronic states , many vibrational modes , and multiple conducting channels , so that it can be used to analyze first - principle calculation results , especially when specification of inter - channel and intra - channel scattering processes is crucial to understand the results . we thank prof . mahn - soo choi for helpful discussions . this work is supported by the cac of kias . using eqs . ( [ identity1])-([identity2 - 2 ] ) , one can show that @xmath278 \left|\langle \phi_{rn}^{-}|\mathcal{m}^{\lambda}|\phi_{lm}\rangle \right|^2 \\ & & -\left(2\pi\right)^{2}\sum_{\lambda , m , n}\int_{\mu_{r}+\hbar\omega_{\lambda}}^{\mu_{l}}d\varepsilon\left(1 - 2\mathcal{t}_{m}\right)\textrm{re}\left[r_{m}^{\prime}t_{m}^{*}\langle\phi_{rm}|\mathcal{m}^{\lambda}|\phi_{rn}^{-}\rangle\langle\phi_{rn}^{-}|\mathcal{m}^{\lambda}|\phi_{lm}\rangle\right ] \\ & & + \left(2\pi\right)^{2}\sum_{\lambda , m , n}\int_{\mu_{r}}^{\mu_{l}-\hbar\omega_{\lambda}}d\varepsilon\left(1 - 2\mathcal{t}_{m}\right)\textrm{re}\left[r_{m}^{\prime}t_{m}^{*}\langle\phi_{rm}|\mathcal{m}^{\lambda}|\phi_{ln}^{+}\rangle\langle\phi_{ln}^{+}|\mathcal{m}^{\lambda}|\phi_{lm}\rangle\right]\end{aligned}\ ] ] and @xmath279 the prefactors of transition amplitudes @xmath280 in @xmath68 and @xmath68 can be written as follows : @xmath281 and @xmath282 thus , by canceling out terms proportional to @xmath283 in @xmath68 and @xmath68 , one can finally obtain eqs . ( [ inelastic_noise1])-([elastic_noise2 ] ) . for multi - channel systems , corrections to the conductance and the noise are decomposed into _ intra_-channel and _ inter_-channel scattering contributions , @xmath284 the intra - channel scattering contributions are simply a collection of each eigenchannel contribution , which is the same form of the single - channel result : @xmath285\left|\mathbb{m}^{\lambda m}_{rl } \right|^2,\end{aligned}\ ] ] where @xmath286 and @xmath287 $ ] . unfortunately , the inter - channel scattering terms are not simplified as the intra - channel ones . when the mirror symmetry along the transport direction is respected , the inter - channel scattering contributions can be written in the form of fermi s golden rule : @xmath288 where the upper ( lower ) sign is for even ( odd ) mirror - symmetric modes . here @xmath289 $ ] , @xmath290 $ ] , @xmath291 , and @xmath292 $ ] . note that @xmath293 for the unitarity of the scattering matrix . it implies that @xmath294 is purely imginary , and @xmath295=\pm 1 $ ] . r. c. jaklevic and j. lambe , phys . * 17 * , 1139 ( 1966 ) . b. c. stipe , m. a. rezai and w. ho , science * 280 * , 1732 ( 1998 ) . j. r. hahn , h. j. lee , and w. ho , phys . rev 85 * , 1914 ( 2000 ) . r. h. m. smit _ et al . _ , nature ( london ) * 419 * , 906 ( 2002 ) . n. agrat , c. untiedt , g. rubio - bollinger , and s. vieira , phys . rev . lett . * 88 * , 216803 ( 2002 ) w. wang , t. lee , i. kretzschmar , and m. a. reed , nano . lett . * 4 * , 643 ( 2004 ) d. djukic , k. s. thygesen , c. untiedt , r. h. m. smit , k. w. jacobsen , and j. m. van ruitenbeek , phys . b * 71 * , 161402 ( 2005 ) w. h. a. thijssen , d. djukic , a. f. otte , r. h. bremmer , and j. m. van ruitenbeek , phys . lett . * 97 * , 226806 ( 2006 ) o. tal , m. krieger , b. leerink and j. m. van ruitenbeek , phys . * 100 * , 196804 ( 2008 ) . b. n. j. persson and a. baratoff , phys . rev . lett . * 59 * , 339 ( 1987 ) . j. bonca and s. a. trugman , phys . lett . * 75 * , 2566 ( 1995 ) . h. ness and a. j. fisher , phys . lett . * 83 * , 452 ( 1999 ) . e. g. emberly and g. kirczenow , phys . b * 61 * , 5740 ( 2000 ) . n. lorente and m. persson , phys . * 85 * , 2997 ( 2000 ) . n. lorente , m. persson , l. j. lauhon , and w. ho , phys . lett . * 86 * , 2593 ( 2001 ) . m. j. montgomery , j. hoekstra , t. n. todorov , and a. p. sutton , j. phys : condens . matter * 15 * , 731 ( 2003 ) . m. j. montgomery and t. n. todorov , j. phys : condens . matter * 15 * , 8781 ( 2003 ) . y. c. chen , m. zwolack and m. di ventra , nano lett . * 3 * , 1691 ( 2003 ) . y. c. chen , m. zwolack and m. di ventra , nano lett . * 4 * , 1709 ( 2004 ) . y. c. chen , m. zwolack and m. di ventra , nano lett . * 5 * , 621 ( 2005 ) . j. jiang , m. kula and w. lu and y. luo , nano lett . * 5 * , 1551 ( 2005 ) . l. c. davis , phys . b * 2 * , 1714 ( 1970 ) . c. caroli , d. saint - james , r. combescot , and p. nozieres , j. phys . c * 5 * , 21 ( 1972 ) . m. galperin , m. a. ratner , and a. nitzan , j. chem . phys . * 121 * , 11965 ( 2004 ) . h. ueba , t. mii , and s. g. tikhodeev , surf . sci . * 601 * , 5220 ( 2007 ) . j. k. viljas , j. c. cuevas , f. pauly , and m. hfner , phys . b * 72 * , 245415 ( 2005 ) . l. de la vega , a. martin - rodero , n. agrat and a. levy yeyati , phys . rev . b * 73 * , 075428 ( 2006 ) . m. paulsson , t. frederiksen , h. ueba , n. lorente , and m. brandbyge , phys . * 100 * , 226604 ( 2008 ) . m. paulsson , thomas frederiksen , and mads brandbyge , phys . b @xmath296 , 201101(r ) ( 2005 ) and _ references therein_. t. frederiksen , m. paulsson , m. brandbyge , and a .- jauho , phys . b * 75 * , 205413 ( 2007 ) and _ references therein_. t. frederiksen , m. brandbyge , n. lorente , and a .- jauho , phys . lett . * 93 * , 256601 ( 2004 ) . t. frederiksen , n. lorente , m. paulsson , and m. brandbyge , phys . rev . b * 75 * , 235441 ( 2007 ) . i. s. kristensen , m. paulsson , k. s. thygesen , and k. w. jacobsen , phys . b * 79 * , 235411 ( 2009 ) . s. kim and n. marzari , phys . rev . b * 87 * , 245407 ( 2013 ) s. kim and y .- w . son , phys . rev . b * 87 * , 195423 ( 2013 ) a. gagliardi _ et al . _ , b * 75 * , 174306 ( 2007 ) . a. troisi and m. a. ratner , j. chem . phys . * 125 * , 214709 ( 2006 ) . chen and m. di ventra , phys . 95 * , 166802 ( 2005 ) m. tsutsui , m. taniguchi and t. kawai , nat * 1 * , 138 ( 2010 ) t .- h . park and m. galperin , phys b * 84 * , 205450 ( 2011 ) d. f. urban , r. avriller , and a. levy yeyati , phys . rev . b * 82 * , 121414 ( 2010 ) t. novotn , f. haupt , and w. belzig , phys . rev . b * 84 * , 113107 ( 2011 ) r. avriller and a. levy yeyati , phys . rev . b * 80 * , 041309 ( 2009 ) t. l. schmidt and a. komnik , phys . rev . b * 80 * , 041307 ( 2009 ) f. haupt , t. novotn , and w. belzig , phys . rev . lett . * 103 * , 136601 ( 2009 ) m. kumar , r. avriller , a. l. yeyati , and j. m. van ruitenbeek , phys . lett . * 108 * , 146602 ( 2012 ) r. avriller and t. frederiksen , phys . b * 86 * , 155411 ( 2012 ) . f. haupt , t. novotn , and w. belzig , phys . b * 82 * , 165441 ( 2010 ) . r. landauer , ibm j. res . * 1 * , 223 ( 1957 ) . m. bttiker , ibm j. res . dev . * 32 * , 63 ( 1988 ) . s. datta , _ electronic transport in mesoscopic systems _ ( cambridge university press , 1995 ) . t. martin and r. landauer , phys . b * 45 * , 1742 ( 1992 ) . lee and c. s. kim , phys . b * 63 * , 075306 ( 2001 ) . m. paulsson and m. brandbyge , phys . b @xmath297 , 115117 ( 2007 ) . j. wang and h. guo , phys . b * 79 * , 045119 ( 2009 ) .

often times , individuals working together as a team can solve hard problems beyond the capability of any individual in the team . cooperative optimization is a newly proposed general method for attacking hard optimization problems inspired by cooperation principles in team playing . it has an established theoretical foundation and has demonstrated outstanding performances in solving real - world optimization problems . with some general settings , a cooperative optimization algorithm has a unique equilibrium and converges to it with an exponential rate regardless initial conditions and insensitive to perturbations . it also possesses a number of global optimality conditions for identifying global optima so that it can terminate its search process efficiently . this paper offers a general description of cooperative optimization , addresses a number of design issues , and presents a case study to demonstrate its power . [ section ] [ section ] [ section ] [ section ]

cs0701057

optimization is a core problem both in mathematics and computer science . it is a very active research area with many international conferences every year , a large amount of literature , and many researchers and users across many fields for a wide range of applications . combinatorial optimization @xcite is a branch of optimization where the set of feasible solutions of problems is discrete , countable , and of a finite size . the general methods for combinatorial optimization are 1 ) local search @xcite , 2 ) simulated annealing @xcite , 3 ) genetic algorithms @xcite , 5 ) ant colony optimization @xcite , 4 ) tabu search @xcite , 5 ) branch - and - bound @xcite 6 ) dynamic programming @xcite . the successful applications of different combinatorial optimization methods have been reported in solving a large variety of optimization problems in practice . optimization is important in the areas of computer vision , pattern recognition , and image processing . for example , stereo matching is one of the most active research problems in computer vision @xcite . the goal of stereo matching is to recover the depth image of a scene from a pair of 2-d images of the same scene taken from two different locations . like many other problems from these areas , it can be formulated as a combinatorial optimization problem , which is np - hard @xcite in computational complexity in general . the researchers in computer vision have developed a number of search techniques which have been proven effective in practice for finding good solutions for combinatorial optimization problems . two well - known ones are the cooperative algorithm proposed by d. marr and t. poggio in @xcite for stereo matching and the probabilistic relaxation proposed by a. rosenfield et al @xcite for scene labeling . recently , there are some remarkable progresses in discovering new optimization methods for solving computer vision problems . graph cuts @xcite is a powerful specialized optimization technique popular in computer vision . it has the best known results in energy minimization in the two recent evaluations of stereo algorithms @xcite , more powerful than the classic simulated annealing method . however , graph cuts has a limitation in its scope because it is only applicable when the energy minimization of a vision problem can be reduced into a problem of finding the minimum cut in a graph @xcite . the second optimization method is so called the sum - product algorithm @xcite , a generalized belief propagation algorithm developed in ai @xcite . the sum - product algorithm is the most powerful optimization method ever found so far for attacking hard optimization problems raised from channel decoding in communications . the min - sum algorithm and max - product algorithm @xcite are its variations . it has also been successful applied to solve several computer vision problems with promising experimental results @xcite . the third method proposed recently is so called max - product tree - reweighted message passing @xcite . it is based on a lower bounding technique called linear programming relaxation . its improvement has been proposed recently and its successful applications in computer vision have been reported @xcite . the cooperative optimization is a newly discovered general optimization method for attacking hard optimization problems @xcite . it has been found in the experiments @xcite that cooperative optimization has achieved remarkable performances at solving a number of real - world np - hard problems with the number of variables ranging from thousands to hundreds of thousands . the problems span several areas , proving its generality and power . for example , cooperative optimization algorithms have been proposed for dna image analysis @xcite , shape from shading @xcite , stereo matching @xcite , and image segmentation @xcite . in the second case , it significantly outperformed the classic simulated annealing in finding global optimal solutions . in the third case , its performance is comparable with graph cuts in terms of solution quality , and is twice as faster as graph cuts in software simulation using the common evaluation framework for stereo matching @xcite . in the fourth case , it is ten times faster than graph cuts and has reduced the error rate by two to three factors . in all these cases , its memory usage is efficient and fixed , its operations are simple , regular , and fully scalable . all these features make it suitable for parallel hardware implementations . this paper is organized in three major themes as 1 ) a formal presentation for cooperative optimization , 2 ) design issues , and 3 ) a case study . they are the generalization and the extension of the previous papers on cooperative optimization . in the case study , another cooperative optimization algorithm for stereo matching besides the one proposed before @xcite is offered to demonstrate the power and flexibility of cooperative optimization . compared with the previous one for stereo matching , the new one lowers the energy levels of solutions further and is more than ten times faster . just like the previous one , the new one is also simple in computation and fully parallel in operations , suitable for hardware implementations . different forms of cooperative optimization can be derived from different cooperation schemes . the basic form defines an important collection of cooperative optimization algorithms . there are two different ways to derive it ; namely , 1 ) as a cooperative multi - agent system for distributed optimization and 2 ) as a lower bounding technique for finding global optimums . each way offers its own inspirations and insights to understand the algorithms . this section describes the first way . the following section offers the description for the second way . readers who are not interested in them can directly jump to section [ sec_gc_form ] for a general description of cooperative optimization . those three sections are relatively independent to each other . team playing is a common social behavior among individuals of the same species ( or different ) where the team members working together can achieve goals or solve hard problems which are beyond the capability of any member in the team . often times , team playing is achieved through competition and cooperation among the members in a team . usually , competition or cooperation alone can hardly lead to good solutions either for a team or for the individuals in the team . without competition , individuals in a team may lose motivation to pursue better solutions . without cooperation , they might directly conflict with each other and poor solutions might be reached both for the team and themselves . through properly balanced competition and cooperation , individuals in a team can find the best solutions for the team and possibly good solutions for themselves at the same time . in the terms of computer science , we can view a team of this kind as a cooperative system with multiple agents . in the system , each agent has its own objective . the collection of all the agent s objectives form the objective of the system . we can use a cooperative system to solve a hard optimization problem following the divide - and - conquer principle . we first break up the objective function of the optimization problem into a number of sub - objective functions of manageable sizes and complexities . following that , we assign each sub - objective function to an agent in a system as the agent s own objective function and ask those agents in the system to optimize their own objective functions through competition and cooperation . ( throughout this paper , we use the term `` objective '' and `` objective function '' interchangeably since the objective of an optimization problem is defined by an objective function and this paper focuses only on optimizing objective functions . ) specifically , the competition is achieved by asking each agent to optimize its own objective function by applying problem - specific optimization methods or heuristics . however , the objectives of agents may not be always aligned with each other . in other words , the best solutions of the agents for optimizing their own objective functions may conflict with each other . to resolve the conflicts , each agent passes its solution to its neighbors through local message passing . after receiving its neighbor s solutions , each agent compromises its solution with the solutions of its neighbors . the solution compromising is achieved by modifying the objective function of each agent to take into account its neighbors solutions . it is important to note that solution compromising among agents is a key concept for understanding the cooperation strategy introduced by cooperative optimization . let the objective of the individual @xmath0 be @xmath1 . let the solution of the individual @xmath0 at time @xmath2 be @xmath3 . let the collection of solutions of the neighbors of the individual @xmath0 at time @xmath2 be @xmath4 . the basic operations of a cooperative system are organized as a process shown in figure [ fig_cooperative_system ] . 123456789012345678901 + for each individual @xmath0 in the system , find the initial solution , + @xmath5 ; + _ iteration _ + for each individual @xmath0 in the system , + modify its original objective by including its neighbors solutions , + @xmath6 + @xmath7 ; + find solutions of the modified objective , + @xmath8 ; the process of a cooperative system of this kind is iterative and self - organized and each agent in the system is autonomous . the system is also inherently distributed and parallel , making the entire system highly scalable and less vulnerable to perturbations and disruptions on individuals than a centralized system . despite of its simplicity , it has many interesting emerging behaviors and can attack many challenging optimization problems . in light of the cooperative multi - agent system for distributed optimization described in fig . [ fig_cooperative_system ] , we can derive the basic form of cooperative optimization now . it is based on a direct way for defining the solution of each agent and a simple way to modify the objective of each agent . the derivation can be generalized further in a straightforward way to any other definitions of solutions and modifications of objectives . given a multivariate objective function @xmath9 of @xmath10 variables , or simply denoted as @xmath11 , where each variable @xmath12 is of a finite domain @xmath13 of size @xmath14 . assume that @xmath11 can be decomposed into @xmath10 sub - objective functions @xmath15 , denoted as @xmath16 , satisfying 1 . @xmath17 [ condition_1 ] , 2 . @xmath15 , for @xmath18 , contains at least variable @xmath12 , 3 . the minimization of @xmath15 , for @xmath18 , is computationally manageable in complexity . let us assign @xmath15 as the objective of agent @xmath0 , @xmath19 there are @xmath10 agents in the system , one agent for each sub - objective function . let the initial solution of agent @xmath0 be the minimization result of @xmath15 defined as follows , @xmath20 where @xmath21 is the set of variables contained in @xmath15 , and @xmath22 stands for minimizing with respect to all variables in @xmath21 excluding @xmath12 . the solution is an unary function on variable @xmath12 , denoted as @xmath23 . assume that the system takes discrete - time with iteration step @xmath24 . to simplify notations , let @xmath25 be the modified objective function of agent @xmath0 at iteration @xmath26 , i.e. , @xmath27 it is also referred to as the @xmath0-th modified sub - objective of the system . the agent s solution at the iteration is defined as @xmath28 the solution is an unary function on variable @xmath12 , denoted as @xmath29 . it is the state of agent @xmath0 at iteration @xmath26 . it can be represented as a vector of real values of size @xmath14 , the domain size of variable @xmath12 . the @xmath0-th equation in ( [ solution_i ] ) defines the dynamics of agent @xmath0 . all the @xmath10 equations define the dynamics of the system . as described in the previous subsection , the cooperation among the agents in the system is introduced by solution compromising via modifying the objective of each agent . let agent @xmath0 define its modified objective function @xmath25 at iteration @xmath26 as a linear combination of its original objective @xmath15 and the solutions of its neighbors at the previous iteration @xmath30 as follows , @xmath31 where @xmath32 and @xmath33 are coefficients of the linear combination . agent @xmath34 is the neighbor of agent @xmath0 if variable @xmath35 of the same index @xmath34 is contained in the agent @xmath0 s objective function @xmath15 . ( based on this definition , the agent @xmath0 is also a neighbor of itself . such a generalization is necessary because there is no restriction to have agent @xmath0 modify its objective using its own solution . ) the neighbors of agent @xmath0 is denoted as @xmath36 , i.e. , @xmath37 . specifically , it is defined as the set of indices as @xmath38 substituting eq . ( [ modified_objective_function ] ) into eq . ( [ solution_i ] ) and letting @xmath39 if @xmath40 , the dynamics of the cooperative system can be written as the following @xmath10 difference equations , @xmath41 such a set of difference equations defines a basic cooperative optimization system ( algorithm ) for minimizing an objective function of the form @xmath42 . at iteration @xmath26 , variable @xmath12 , for @xmath18 , has a value in the solution for minimizing the @xmath0-th modified sub - objective function @xmath25 . it is denoted as @xmath43 , i.e. , @xmath44 from ( [ solution_i ] ) , we have @xmath45 the agent @xmath0 is responsible for assigning that value to variable @xmath12 . the assignments of other variables are taken care of by other agents . all these values together form a solution of the system at iteration @xmath26 , denoted as @xmath46 . putting everything together , we have the pseudo code of the algorithm is given in figure [ fig_cooperative_optimization_algorithm ] . the global optimality condition mentioned in the line @xmath47 will be discussed in detail later in this paper . * procedure * basic cooperative optimization algorithm + 123456789012345678901 1 initialize the soft assignment function @xmath23 , for each @xmath0 ; + 2 @xmath48 to @xmath49 * do * + 3 each @xmath0 * do * + / * modify the @xmath0-th sub - objective function @xmath15 * / + 4@xmath50 ; + / * minimize the modified sub - objective function * / + 5@xmath51 ; + / * find the best value for @xmath12 * / + 6@xmath52 ; + 7 @xmath46 is a global optimal solution * return * @xmath46 ; + 8 @xmath46 ; / * as an approximate solution * / the coefficient @xmath32 in ( [ cooperative_optimization ] ) controls the level of the cooperation among the agents at iteration @xmath26 . it is so called the cooperation strength , satisfying @xmath53 . from ( [ cooperative_optimization ] ) we can see that , for each agent , a high value for @xmath32 will weigh the solutions of the other agents more than its own objective @xmath15 . in other words , the agents in the system tend to compromise more with their solutions . as a consequence , a strong level of cooperation is reached in this case . if the cooperation strength @xmath32 is of a small value , the cooperation among the agents is weak . particularly , if it is equal to zero , there is no cooperation among the agents and each agent minimizes its own objective function independently ( see ( [ cooperative_optimization ] ) ) . the coefficients @xmath33 control the propagation of solutions @xmath54 , for @xmath55 , as messages among the agents in the system . all @xmath33s together form a @xmath56 matrix called the propagation matrix . to have @xmath42 as the objective function to be minimized , it is required @xcite that the propagation matrix @xmath57 is non - negative and @xmath58 to have solutions @xmath54 uniformly propagated among all the agents , it is required @xcite that the propagation matrix @xmath59 is irreducible . a matrix @xmath59 is called reducible if there exists a permutation matrix @xmath60 such that @xmath61 has the block form @xmath62 the role of propagation matrices in basic cooperative optimization algorithms is exactly same as the one of transition matrices in markov chains ( or random walks over directed graphs ) . in a markov chain , a transition matrix governs the distribution of states over time . in a basic cooperative optimization algorithm , a propagation matrix governs the distribution of solutions among agents . the mathematical foundation for analyzing markov chains has been well established . they can be directly applied to analyze the message propagation of cooperative optimization . as mentioned before , the solution @xmath29 of agent @xmath0 at iteration @xmath26 is an unary function on @xmath12 storing the solution of minimizing the @xmath0-th modified sub - objective function @xmath25 ( see ( [ solution_i ] ) ) . given a value of @xmath12 , @xmath29 is the minimal value of @xmath25 with the variable @xmath12 fixed to that value . to minimize @xmath25 , the values of @xmath12 which have smaller function values @xmath29 are preferred more than those of higher function values . the best value for assigning the variable @xmath12 is the one of the minimal function value @xmath29 ( see ( [ cop_solution ] ) ) . therefore , @xmath29 is inversely related to the preferences over different values of @xmath12 for minimizing @xmath25 . it is so called the assignment constraint on variable @xmath12 , an algorithm introduced constraint on the variable . it can also be viewed as a soft decision made by the agent for assigning the variable @xmath12 at iteration @xmath26 . in particular , a soft decision of agent @xmath0 falls back to a hard decision for assigning the variable @xmath12 when the agent accept only one value and reject all the rest values . such a hard decision can be represented by the assignment constraint @xmath29 as @xmath63 , for some @xmath64 , and @xmath65 for any @xmath66 . with that insight , it can be understood now that the messages propagated around among the agents in a basic cooperative optimization system are the soft decisions for assigning variables . an agent can make a better decision using soft decisions propagated from its neighbors than using the hard ones instead . it is important to note that soft decision making is a critical feature of cooperative optimization , which makes it fundamentally different from many classic optimization methods where hard decisions are made for assigning variables . given an objective function of the following form @xmath67 where each variable is of a finite domain . the goal is to seek values ( labels ) of the five variables such that the objective function is minimized . let us simply denote the function as @xmath68 to design a basic cooperative optimization algorithm to minimize the objective function , we first decompose it into the following five sub - objective functions , @xmath69 a propagation matrix @xmath59 of dimensions @xmath70 can be chosen as @xmath71 with the decomposition and the propagation matrix , substituting them into ( [ cooperative_optimization ] ) we have a basic cooperative optimization algorithm with five difference equations for minimizing the five sub - objective functions in an iterative and cooperative way . replacing @xmath72 by @xmath73 in the difference equations ( [ cooperative_optimization ] ) , we have the basic canonical form of cooperative optimization as @xmath74 the basic form of cooperative optimization ( [ cooperative_optimization ] ) has its cooperation strength @xmath32 restricted to @xmath53 . it is because its difference equations ( [ cooperative_optimization ] ) do not make sense when @xmath75 . however , such a restriction can be relaxed to @xmath76 for the basic canonical form ( [ cooperative_optimization_canonical ] ) . often in times in practice , the basic canonical form is preferred over the basic one because the cooperation strength @xmath32 in the former has a broader range to choose from to maximize performance . in principle , a basic cooperative optimization algorithm can be understood as a lower bounding technique for finding global minima . it first initializes a function of some form as a lower bound function to an objective function . one may intentionally choose a form for the lower bound function such that the minimization of the function is simple in computation . following that , the algorithm progressively tightens the lower bound function until its global minimum touches the global minimum of the original objective function . the latter is then found by searching the former instead ( see the illustration in fig . [ fig_lowerbound ] ) . specifically , let the objective function to be minimized be @xmath11 . assume that the initial lower bound function be @xmath77 , @xmath78 . from @xmath77 , assume that the algorithm progressively tightens the function in an iterative way such that @xmath79 where @xmath26 is the iteration number . let the global minimum of the lower bound function @xmath80 at iteration @xmath26 be @xmath46 . finding @xmath46 is simple in computation due to the simple form of the lower bound function @xmath80 . at iteration @xmath26 , if the algorithm found that the lower bound function @xmath80 at the solution @xmath46 has the same function value as the original objective function @xmath11 , i.e. , @xmath81 in other words , the two functions touch each other at the point where @xmath82 in the search space . then @xmath46 must also be the global minimum of @xmath11 simply because @xmath83 such a condition implies that the lower bound function @xmath80 has been tightened enough such that its global minimum @xmath46 touches the global minimum of the original objective function @xmath11 . the latter is thus found by searching the former instead . such a lower bounding technique is so called the bound function tightening technique for optimization . there are other lower bounding techniques based on principles different from this one . examples are lagrangian relaxation techniques , cutting plane techniques , branch - and - bound algorithms , and branch - and - cut algorithms . in light of the bound function tightening technique described in the previous subsection , we can derive the basic form of cooperative optimization based on a simple form of lower bound functions . the derivation can be generalized further in a straightforward way to any other forms of lower bound functions . given an objective function of @xmath10 variables , @xmath9 , or simply denoted as @xmath11 . assume that @xmath84 is a lower bound function of @xmath11 defined on the same set of variables . obviously the linear combination of the two functions , @xmath85 defines a new lower bound function of @xmath11 if the parameter @xmath86 satisfying @xmath87 . let us choose a simple form for the lower bound function as @xmath88 where @xmath89 is an unary component function defined on variable @xmath12 , for @xmath90 . its global minimum , denoted as @xmath91 , can be easily found by minimizing the unary component functions @xmath89 independently as @xmath92 assume that the objective function @xmath11 can be decomposed into @xmath10 sub - objective functions , @xmath93 the lower bound function @xmath84 can also be easily decomposed into @xmath10 sub - functions as follows @xmath94 based on the two decompositions , the new lower bound function ( [ new_lbf ] ) can be rewritten as @xmath95 to put the above function in a simple form , let @xmath96 then it can be rewritten simply as @xmath97 in the above sum , let @xmath98 be the set of variables contained in the @xmath0-th component function @xmath99 . if we minimize the function with respect to all variables in @xmath98 except for @xmath12 , we obtain an unary function defined on @xmath12 , denoted as @xmath100 , i.e. , @xmath101 the sum of those unary functions defines another lower bound function of @xmath11 , denoted as @xmath102 , i.e. , @xmath103 this new lower bound function has exactly the same form as the original one @xmath104 . therefore , from a lower bound function @xmath105 of the form @xmath106 , we can compute another lower bound function @xmath102 of the same form . such a process can be repeated and we can have an iterative algorithm to compute new lower bound functions . rewriting eq . ( [ compute_psi_i ] ) in an iterative format , we have @xmath107 where @xmath26 is the iteration step , @xmath24 . the above @xmath10 difference equations define a basic cooperative optimization algorithm for minimizing an objective function @xmath11 of the form @xmath42 . the solution at iteration @xmath26 , denoted as @xmath46 , is defined as the global minimal solution of the lower bound function @xmath80 at the iteration , i.e. , @xmath108 which can be easily obtained as @xmath109 if @xmath110 at some iteration @xmath26 , then the solution @xmath46 must be the global minimum of the original objective function @xmath11 . without loss of generality , we assume in the following discussions that all sub - objective functions @xmath15 are nonnegative ones . one may choose the initial condition as @xmath111 , for any value of @xmath12 and @xmath90 . the parameter @xmath32 can be varied from one iteration to another iteration . if it is of a constant value and the above initial condition has been chosen , cooperative optimization theory @xcite tells us that the lower bound function @xmath80 is monotonically non - decreasing as shown in ( [ monotonically_nondecreasing ] ) . it has been shown that a basic cooperative optimization algorithm ( [ cooperative_optimization ] ) has some important computational properties @xcite . given a constant cooperation strength @xmath86 , i.e. , @xmath112 for all @xmath26s , the algorithm has one and only one equilibrium . it always converges to the unique equilibrium with an exponential rate regardless of initial conditions and perturbations . the two convergence theorems proved in @xcite are very important and so they are listed here again . one formally describes the existence and the uniqueness of the equilibrium of the algorithm , and the another reveals the convergence property of the algorithm . a basic cooperative optimization algorithm with a constant cooperation strength @xmath86 ( @xmath87 ) has one and only one equilibrium . that is , the difference equations ( [ cooperative_optimization ] ) of the algorithm have one and only one solution ( equilibrium ) , denoted as a vector @xmath113 , or simply @xmath114 . [ theorem_7 ] a basic cooperative optimization algorithm with a constant cooperation strength @xmath86 ( @xmath87 ) converges exponentially to its unique equilibrium @xmath114 with the rate @xmath86 with any choice of the initial condition @xmath115 . that is , @xmath116 where @xmath117 is the maximum norm of the vector @xmath118 defined as @xmath119 [ theorem_8 ] the two theorems indicate that every basic cooperative optimization algorithm ( [ cooperative_optimization ] ) is stable and has a unique attractor , @xmath114 . hence , the evolution of the algorithms is robust , insensitive to perturbations . the final solution of the algorithms is independent of their initial conditions . in contrast , the conventional algorithms based on iterative local improvement of solutions may have many local attractors due to the local minima problem . the evolution of those local optimization algorithms are sensitive to perturbations , and the final solution of those algorithms is dependent on their initial conditions . furthermore , the basic cooperative optimization algorithms ( [ cooperative_optimization ] ) possess a number of global optimality conditions for identifying global optima . they know whether a solution they found is a global optimum so that they can terminate their search process efficiently . however , this statement does not imply that _ np = p _ because a basic cooperative optimization algorithm can only verify within a polynomial time whether a solution it found is a global optimum or not . it can not decide the global optimality for any given solution other than those it found . it is important to note that a basic canonical cooperative optimization algorithm ( [ cooperative_optimization_canonical ] ) may no longer possess the unique equilibrium property when its cooperation strengths at some iterations are greater than one , i.e. , @xmath120 for some @xmath26s . in this case , the algorithm may have multiple equilibriums . it can evolve into any one of them depending on its initial settings of the assignment constraints @xmath23 @xmath121 . as described before , a basic cooperative optimization algorithm is defined by the @xmath10 difference equations ( [ cooperative_optimization ] ) . the @xmath0-th equation defines the minimization of the @xmath0-th modified sub - objective function @xmath25 ( defined in ( [ modified_objective_function ] ) ) . given any variable , say @xmath12 , it may be contained in several modified sub - objective functions . at each iteration , @xmath12 has a value in the optimal solution for minimizing each of the modified sub - objective functions containing the variable . those values may not be the same . if all of them are of the same value at some iteration , we say that the cooperative optimization algorithm reach a consensus assignment for that variable . moreover , if a consensus assignment is reached for every variable of the problem at hand at some iteration , we call the minimization of the @xmath10 modified sub - objective functions reaches a solution consensus . that is , there is no conflict among the solutions in terms of variable assignments for minimizing those functions . in this case , those consensus assignments form a solution , called a consensus solution , and the algorithm is called reaching a consensus solution . to be more specific , given @xmath10 modified sub - objective functions , @xmath99 , for @xmath90 ( to simplify notation , let us drop the superscript @xmath26 temporarily ) . let the optimal solution of the @xmath0-th modified sub - objective function be @xmath122 , i.e. , @xmath123 assume that variable @xmath12 is contained in both @xmath34-th and @xmath26-th modified sub - objective functions @xmath124 , @xmath125 . however , it is not necessary that @xmath126 given a variable @xmath12 , if the above equality holds for any @xmath34 and @xmath26 where @xmath124 and @xmath125 contain @xmath12 , then a consensus assignment is reached for that variable with the assignment value denoted as @xmath127 . moreover , if the above statement is true for any variable , we call the minimization for all @xmath99s reaches a solution consensus . the solution @xmath91 with @xmath127 as the value of variable @xmath12 is called a consensus solution . as defined before , @xmath98 stands for the set of variables contained in the function @xmath99 . @xmath98 is a subset of variables , i.e. , @xmath128 . let @xmath129 stand for the restriction of a solution @xmath91 on @xmath98 . another way to recognize a consensus solution @xmath91 is to check if @xmath129 , for any @xmath0 , is the global minimum of @xmath99 , i.e. , @xmath130 simply put , a solution is a consensus one if it is the global minimum of every modified sub - objective function . consensus solution is an important concept of cooperative optimization . if a consensus solution is found at some iteration or iterations , then we can find out the closeness between the consensus solution and the global optimal solution in cost . the following theorem from @xcite makes these points clearer . let @xmath131 given any propagation matrix @xmath59 , and the general initial condition @xmath132 , for each @xmath0 , or @xmath133 . if a consensus solution @xmath91 is found at iteration @xmath134 and remains the same from iteration @xmath134 to iteration @xmath135 , then the closeness between the cost of @xmath91 , @xmath136 , and the optimal cost , @xmath137 , satisfies the following two inequalities , @xmath138 @xmath139 where @xmath140 is the difference between the optimal cost @xmath137 and the lower bound on the optimal cost @xmath141 obtained at iteration @xmath142 . in particular , if @xmath143 , for @xmath144 , when @xmath145 , @xmath146 that is , the consensus solution @xmath91 must be global minimum of @xmath11 , i.e.,@xmath147 . [ theorem_2 ] consensus solution is also an important concept of cooperative optimization for defining global optimality conditions . the cooperative optimization theory tells us that a consensus solution can be the global minimal solution . as mentioned in the previous subsection that a basic cooperative optimization algorithm has one and only one equilibrium given a constant cooperation strength . if a cooperative optimization algorithm reaches an equilibrium after some number of iterations and a consensus solution is found at the same time , then the consensus solution must be the global minimal solution , guaranteed by theory . the following theorem ( with its proof in the appendix ) establishes the connection between a consensus solution and a global optimal solution . [ sufficient_condition ] assume that a basic cooperative optimization ( [ cooperative_optimization ] ) reaches its equilibrium at some iteration , denoted as @xmath114 . that is , @xmath114 is a solution to the difference equations ( [ cooperative_optimization ] ) . if a consensus solution @xmath148 is found at the same iteration , then it must be the global minimum of @xmath11 , i.e.,@xmath147 . besides the basic global optimality condition given in the above theorem , a few more ones are offered in @xcite for identifying global optimal solutions . the capability of recognizing global optimums is a critical property for any optimization algorithm . without any global optimality condition , it will be hard for an optimization algorithm to know where to find global optimal solutions and whether a solution it found is a global optimum . finding ways of identifying global optimums for any optimization algorithm is of both practical interests as well as theoretical importance . the convergence theorem [ theorem_2 ] can be generalized further to any initial conditions for @xmath115 and @xmath149 , and to any cooperation strength series @xmath150 . dropping the restriction on the initial conditions @xmath115 and @xmath149 in the theorem , from the difference equations ( [ cooperative_optimization ] ) , we have @xmath151 it is obvious from the above equation that @xmath152 still approaches @xmath137 exponentially with the rate @xmath86 when the cooperation strength @xmath32 is of a constant value @xmath86 ( @xmath87 ) . when the cooperation strength @xmath32 is not of a constant value @xmath86 , the convergence to the global optimum is still guaranteed as long as the cooperation strength series @xmath153 is divergent . let @xmath150 be a sequence of numbers of the interval @xmath154 . 1 . if @xmath155 , then @xmath156 2 . if @xmath157 , then @xmath158 [ lemma_infinite_product ] the proof of the lemma is offered in appendix . from the above lemma and eq . ( [ generalized_convergence ] ) , the convergence theorem [ theorem_2 ] can be generalized further as follows . given any initial conditions , assume that a consensus solution @xmath91 is found by a basic cooperative optimization algorithm at some iteration @xmath26 and remains the same in the following iterations . if the series @xmath159 is divergent , then @xmath160 that is , the consensus solution @xmath91 must be the global minimal solution @xmath161 , @xmath162 . [ convergence_theorem_2 ] if @xmath163 , for instance , the series ( [ series_lambda ] ) is the harmonic series , @xmath164 the harmonic series is divergent . hence , with the choice of @xmath165 , if a consensus solution @xmath91 is found at some iteration and it remains the same in the following iterations , it must be the global minimal solution @xmath161 . if @xmath166 , as another example , is a convergent sequence of a positive limit , then @xmath167 is divergent . in this case , a consensus solution is also the global minimal solution . this statement can be generalized further to cauchy sequences . every convergent sequence is a cauchy sequence , and every cauchy sequence is bounded . thus , if @xmath168 is a cauchy sequence of a positive bound , a consensus solution is the global minimal solution . to maximize the performance of a cooperative optimization algorithm , it is popular in the experiments to progressively increase the cooperation strength as the iteration of the algorithm proceeds . a weak cooperation level at the beginning leads to a fast convergence rate ( see theorem [ theorem_8 ] ) . a strong cooperation level at a later stage of the iterations increases the chance of finding a consensus solution . theorem [ convergence_theorem_2 ] offers us some general guidance and justification for choosing a variable cooperation strength . it tells us that the increment of the cooperative strength should not be too fast if we want the guarantee of a consensus solution being the global optimal one . by combining different forms of lower bound functions and different ways of decomposing objective functions , we can design cooperative optimization algorithms of different complexities and powers for attacking different optimization problems . the basic canonical form of cooperative optimization ( [ cooperative_optimization_canonical ] ) can be generalized further in a straightforward way to the general canonical one as follows . given a multivariate objective function @xmath9 of @xmath10 variables , or simply denoted as @xmath11 , where each variable is of a finite domain . assume that @xmath11 can be decomposed into @xmath169 sub - objective functions @xmath15 which may satisfy the condition @xmath170 one may define another function @xmath105 , on the same set of variables as @xmath11 , as the composition of @xmath169 component functions as follows , @xmath171 where @xmath172 is a component function defined on a subset of variables @xmath173 , @xmath174 , for @xmath175 . @xmath176 is the restriction of @xmath118 on @xmath173 , denoted as @xmath177 . a cooperative optimization algorithm of the general canonical form is defined as minimizing the @xmath169 sub - objective functions @xmath15 in the following iterative and cooperative way , @xmath178 for @xmath179 . in the equations , @xmath180 is the iteration step ; @xmath21 is the set of variables contained in the functions at the right side of the @xmath0-th equation ; @xmath32 is a real value parameter at iteration @xmath26 satisfying @xmath181 ; and @xmath182 are also real value parameters satisfying @xmath183 . the solution at iteration @xmath26 is defined as @xmath184 moreover , @xmath46 is called a consensus solution if it is the conditional optimum of all the @xmath169 minimization problems defined in ( [ cooperative_optimization_general ] ) . that is , @xmath185 when @xmath186 and @xmath175 . one may choose the parameters @xmath32 and @xmath33 in such a way that they further satisfy the conditions of @xmath187 , for all @xmath34s , and all @xmath32s are less than one ( @xmath188 ) . with the settings , if the algorithm reaches its equilibrium at some iteration and the solution of the iteration is also a consensus one , then it must be the global minimal solution ( this global optimality condition can be proved in the exact same way as that of theorem [ sufficient_condition ] ) . the general canonical form can be further generalized to variable propagation matrices , variable forms of lower bound functions , and variable ways of decomposing objective functions . a basic cooperative optimization algorithm ( [ cooperative_optimization ] ) ( or a basic canonical one ( [ cooperative_optimization_canonical ] ) ) is uniquely defined by the objective function decomposition @xmath16 , the cooperation strength series @xmath150 , and the propagation matrix @xmath189 . some general guideline for designing the cooperation strength series has discussed in the previous section . this section focuses on the rest two . a large class of optimization problems have objective functions of the following form , @xmath190 the function @xmath191 is an unary function on variable @xmath12 , for @xmath90 . and the function @xmath192 is a binary function on two variables @xmath193 . to note the collection of all defined binary functions , the set @xmath194 is used which contains non - ordered pairs of variable indices where each pair @xmath195 corresponds to a defined binary function @xmath192 . the above optimization problems are also referred to as binary constraint optimization problems ( binary cop ) in ai . the unary function @xmath196 is called an unary constraint on variable @xmath12 and the binary function @xmath192 is called a binary constraint on variables @xmath193 . binary constraint optimization problems are a very general formulation for many optimization problems arose from widely different fields . examples are the famous traveling salesman problems , weighted maximum satisfiability problems , quadratic variable assignment problems , stereo vision , image segmentation , and many more . solving a binary constraint optimization problem is np - hard in computation . an objective function in form of ( [ binary_cost_function ] ) can be represented with an undirected graph @xmath197 . in the graph , each variable @xmath12 is represented by a node , called a variable node , @xmath198 ; each binary constraint @xmath199 is represented by an undirected edge , connecting the variable nodes @xmath12 and @xmath35 , denoted by a non - ordered pair of variable nodes @xmath200 . by definition , the set @xmath201 of the edges of the graph @xmath202 is @xmath203 . the simple example described in subsection [ sec_simple_example ] is a binary constraint optimization problem . the objective function ( [ obj_simple_example ] ) of the simple example has the form of ( [ binary_cost_function ] ) . it can be represented by an undirected graph as shown in figure [ fig_constraint_graph ] . 2.5 cm if edge @xmath204 , then the variable nodes @xmath193 are called neighbors to each other . in graph theory , they are also called adjacent to each other . each variable node @xmath12 can have a number of neighboring variable nodes . let @xmath36 be the set of the indices of the neighboring variables of @xmath12 . by definition , @xmath205 using the notations , we can rewrite the objective function ( [ binary_cost_function ] ) as @xmath206 the expression ( [ binary_cost_function2 ] ) for an objective function of a binary constraint optimization problem also defines a straightforward way to decompose the energy function . that is , @xmath207 obviously , @xmath208 . this kind of decompositions is so called the straightforward decompositions . the sub - objective functions @xmath15 in the straightforward decompositions can be easily minimized as @xmath209 using the graphical representation of an objective function can help us to visualize the straightforward decomposition of an objective function . for example , the decomposition of the objective function of the simple example presented in subsection [ sec_simple_example ] can be viewed an instance of this kind of decompositions . the original objective function has a graphical representation shown in figure [ fig_constraint_graph ] . each sub - objective function @xmath15 of the decomposition can also be represented by a graph , which must be a subgraph of the original one . the graphical representation of the decomposition is illustrated in figure [ fig_neighborhooddecomposition ] . in the figure we can see that the original loopy graph is decomposed into five loop - free subgraphs . 8.0 cm in general , in a graphical representation , the straightforward decompositions given in ( [ sub_binary_cost_function ] ) can be understood as decomposing a loopy graph into @xmath10 loop - free subgraphs , one subgraph is associated with each variable node in the original graph . in the decomposition , each subgraph is of a star - like structure with its associated variable node as the star center . it consists of the variable node , the neighbors of the node , and the edges connecting the node with its neighbors . the graphical representation of an objective function may contain many loops . that is the major cause of the difficulty at minimizing the objective function . if the graph is loop - free , there exist algorithms with linear complexity ( e.g. , dynamic programming ) that can minimize the objective function efficiently . therefore , if we can decompose an objective function with a loopy graphical representation into a number of sub - objective functions with loop - free graphical representations , a hard optimization problem is , thus , broken into a number of sub - problems of lower complexities . cooperative optimization can then be applied to solve those sub - problems in a cooperative way . this kind of decompositions is called the graph - based decompositions . it is important to note that the modification given in ( [ modified_objective_function ] ) for a sub - objective function does not change its graphical structure . in other words , every modified sub - objective function defined in ( [ modified_objective_function ] ) has the exact same graphical structure as its original one . this is because only unary functions , the assignment constraints @xmath210 , are introduced in the definition . therefore , any optimization algorithm applicable to the original sub - objective functions should also be applicable to the modified ones . in other words , if a sub - objective function is of a tree - like structure , then its modified version defined by ( [ modified_objective_function ] ) must have the exact same tree - like structure . both of them can be minimized efficiently via dynamic programming . in terms of the graph based decompositions , the straightforward decompositions are based on the direct neighbors of each variable node . another possible way of decomposing an objective function is based on the spanning trees of the graph representing the function . a tree is called a spanning tree of an undirected graph @xmath202 if it is a subgraph of @xmath202 and containing all vertices of @xmath202 . every finite connected graph @xmath202 contains at least one spanning tree @xmath211 . given an objective function @xmath11 of @xmath10 variables in form of ( [ binary_cost_function ] ) . let @xmath212 be its graphical representation with @xmath10 variable nodes . without loss of generality , we assume that @xmath202 is connected ( otherwise it implies that the original minimization problem can be broken into several independent sub - problems ) . for each variable node @xmath12 of @xmath202 , we can associate a spanning tree of @xmath202 , denoted as @xmath213 ( @xmath214 shares the same set of nodes as @xmath202 ) . there are @xmath10 such spanning trees in total , @xmath215 ( some trees may possibly be duplicated ) . we also choose those @xmath10 spanning trees in a way such that each edge of @xmath202 is covered at least by one of the @xmath10 trees . figure [ fig_spanningtreedecomposition ] shows an example of decomposing the graph representing the objective function of the simple example into five spanning trees . after decomposing the graph @xmath202 into @xmath10 spanning trees , if we can define a sub - objective function @xmath15 for each spanning tree such that 1 . @xmath216 ; 2 . the graphical representation of @xmath15 is @xmath214 , for @xmath18 . then the set @xmath16 is a legitimate decomposition of the original objective function @xmath11 . this kind of decompositions is so called the spanning - tree based decompositions . given an objective function @xmath11 of a binary constraint optimization problem with a graphical representation @xmath202 and @xmath10 spanning trees , each unary constraint @xmath217 of @xmath11 is associated with a variable node @xmath35 in @xmath202 covered by all the @xmath10 spanning trees . each binary constraint @xmath218 of @xmath11 is associated with an edge @xmath219 in @xmath202 covered at least by one of the @xmath10 spanning trees . assume that the edge @xmath220 is covered by @xmath221 spanning trees , where @xmath221 should be a positive integer . one way for defining sub - objective functions @xmath15 to satisfy the above two conditions is given as follows , @xmath222 for @xmath90 . we can apply algebraic graph theory to reveal some of properties of graphs and their spanning trees . let @xmath197 be a finite simple graph of @xmath10 nodes , with vertices @xmath223 . if @xmath202 is a graphical representation of an objective function , then the vertices @xmath224 is the variable node @xmath12 , i.e. , @xmath225 . the connectivity of a graph @xmath202 can be represented by the adjacency matrix @xmath226 of @xmath202 . the adjacency matrix @xmath226 of @xmath202 is defined as @xmath227 the adjacency matrix of an undirected graph is symmetric . let @xmath228 be the adjacency matrix of @xmath202 where the diagonal entries @xmath229 are replaced by the degrees of vertices @xmath230 . let @xmath231 be the matrix obtained by removing the first row and column of @xmath228 . then the number of spanning trees in g is equal to @xmath232 ( kirchoff s matrix - tree theorem ) . particularly , if @xmath202 is a complete graph , @xmath233 has the determinant @xmath234 . that is , every complete graph with @xmath10 vertices ( @xmath235 ) has exactly @xmath234 spanning trees ( theorem of cayley ) . using factor graph @xcite , any objective function containing @xmath26-ary constraints , where @xmath26 can be any integer number , can be represented as a graph . with the representation , the two aforementioned kinds of decompositions can be easily generalized further to decompose the objective function . special decompositions can also be explored for graphs with special structures or constraints of some special properties to maximize the power of cooperative optimization algorithms . another way to apply the two kinds of decompositions for the @xmath26-ary constraints case is by converting the constraints of orders higher than two into binary constraints via variable clustering technique . as mentioned in the previous subsection , the objective function of a binary constraint optimization problem has a graphical representation @xmath212 . for the straightforward decompositions described in the previous subsection , we can design a propagation matrix based on the adjacency of the graph @xmath202 as follows @xmath236 where @xmath237 is the degree of the variable node @xmath35 . the propagation matrix ( [ p2 ] ) of the simple example is designed in this way . another way to design a propagation matrix is given as follows , @xmath238 such a matrix has all of the diagonal elements of non - zero values . the design of cooperative optimization algorithms is not trivial even with the aforementioned guidelines . in the basic canonical form ( [ cooperative_optimization_canonical ] ) , there are @xmath56 values for the propagation matrix @xmath189 and a series of values for the cooperation strength @xmath32 . to ease the design job for engineers and practitioners , the difference equations ( [ cooperative_optimization_canonical ] ) of the basic canonical form can be simplified to @xmath239 where @xmath240 is the only parameter to be tuned in experiments to maximize performance . it plays the same role as the cooperation strength @xmath32 for controlling the cooperation level among the agents in a system . the above set of simplified difference equations defines the simple form of cooperative optimization . the simple form is derived from the basic canonical form ( [ cooperative_optimization_canonical ] ) by setting @xmath33 be a positive constant @xmath241 , for any @xmath0 and @xmath34 , if @xmath35 is contained in @xmath15 ; and @xmath39 , otherwise . we also let the cooperation strength @xmath32 be of a constant value @xmath86 . let @xmath242 , we have ( [ cooperative_optimization_canonical ] ) simplified to ( [ simple_cooperative_optimization0 ] ) . if the parameter @xmath240 is of a large value , the difference equations ( [ simple_cooperative_optimization0 ] ) of a simple cooperative optimization algorithm may have value overflow problems in computing the assignment constraints @xmath243 . to improve its convergence property , we can offset each @xmath243 by a value at each iteration . one choice is the minimal value of @xmath29 . that is we offset @xmath29 by its minimal value as follows , @xmath244 thus , the offsetting defines an operator on @xmath29 , denoted as @xmath245 . with the notation , the difference equations of a simple cooperative optimization algorithm become @xmath246 just like many other problems in computer vision and image processing , stereo matching can be formulated as a binary constraint optimization problem with an objective function @xmath11 in form of ( [ binary_cost_function ] ) . for detail about the energy function definitions used for stereo matching , please see @xcite . basically , an unary constraint @xmath191 in ( [ binary_cost_function ] ) measures the difference of the intensities between site @xmath0 from one image and its corresponding site in another image given the depth of the site . a binary constraint @xmath247 measures the difference of the depths between site @xmath0 and site @xmath34 . this type of constraints is also referred to as the smoothness constraint in literature . it has also been widely used in solving image segmentation and other vision tasks . in our experiments , we apply the simplified form of cooperative optimization ( [ simple_cooperative_optimization ] ) for stereo matching with the parameter @xmath240 is set to @xmath248 . the maximum number of iterations is set to @xmath249 . the objective function associated with stereo - matching is decomposed based on the spanning - tree based decomposition . the detail of the decomposition and the minimization of the sub - objective functions are offered in the following subsection . often times , the graphical representation of the objective function of an image segmentation problem or a stereo matching problem is of a 2-d grid - like structure . because a 2-d grid - like graph is highly regular in structure , its spanning trees can be easily defined in a systematic way . given an objective function of a 2-d grid - like graphical representation @xmath197 , let @xmath250 be the height of the grid ( the number of rows ) and @xmath251 be the width of the grid ( the number of columns ) . let @xmath252 be the set of all horizontal edges of @xmath202 and @xmath253 be the set of all vertical edges . there are in total @xmath254 nodes , one for each variable . there are in total @xmath255 horizontal edges and @xmath256 vertical edges . with the notations , the objective function can be expressed as @xmath257 or equivalently , @xmath258 the horizontal path @xmath259 through a variable node @xmath12 consists of all the nodes at the same horizontal line as @xmath260 , together with the edges connecting those nodes . the vertical path @xmath261 through a variable node @xmath12 consists of all the nodes at the same vertical line as @xmath260 , together with the edges connecting those nodes . for each variable node @xmath12 , let us define two spanning trees with the node as the root , called the horizontal spanning tree @xmath262 and the vertical spanning tree @xmath263 , respectively . the horizontal spanning tree @xmath262 consists of the horizontal path through the variable node @xmath12 and all the vertical paths through each node in the horizontal path . the vertical spanning tree @xmath263 consists of the vertical path through the variable node @xmath12 and all the horizontal paths through each node in the vertical path ( the illustrations are shown in fig . [ fig_gridgraphhspan ] ) . as their roots.,title="fig:",width=147 ] as their roots.,title="fig:",width=147 ] let the functions @xmath264 be the objective functions associated with the horizontal spanning tree @xmath262 and the vertical spanning tree @xmath263 , respectively . following the general design guideline described in the previous section for the spanning - tree based decompositions ( see eq . ( [ tree_decomposition ] ) ) , we can define @xmath265 and @xmath266 as @xmath267 @xmath268 where @xmath269 , @xmath270 , and @xmath271 . the sub - objective function @xmath15 associated with variable @xmath12 is defined as @xmath272 clearly , we have @xmath273 . as mentioned before , any objective function of a tree - like structure can be minimized efficiently using the dynamic programming technique . it is of a linear computational complexity and is simply based on local message passing from the leave nodes all the way back to the root node . the books @xcite offer a detail explanation about message passing and the dynamic programming technique . when applying the technique for minimizing the objective function of a horizontal or vertical spanning tree , the message flows among the variable nodes are illustrated in figure [ fig_gridgraphmessageflow ] . at their roots.,width=359 ] the middlebury college evaluation framework @xcite for stereo matching is used in the experiments . the script used for evaluation is based on _ exp6_gc.txt _ offered in the framework . the other settings come from the default values in the framework . the results of stereo matching algorithms together with the ground truths for the four test stereo image pairs from the evaluation framework are shown in figure [ fig_cc ] . the quality of solutions of both algorithms are very close to each other from a visual inspection . the following four tables show the performance of the cooperative optimization algorithm ( upper rows in a table ) and the graph cut algorithm ( lower rows in a table ) over the four test image sets . the solution quality is measured in the overall area , no occluded areas , occluded areas , textured areas , texture - less areas , and near discontinuity areas ( see @xcite for the detail description of the evaluation framework ) . both algorithms does not handle occluded areas ( an occluded area is one that is visible in one image , but not the other ) . also , the runtimes of the two algorithms ( co = cooperative optimization algorithm , gc = graph cuts ) are listed . from the tables we can see that the cooperative optimization is very close to graph cuts in terms of solution quality and energy states . however , the cooperative optimization algorithm is around @xmath274 times faster than graph cuts in the software simulation . ccccccc + & + & all & non occl & occl & textrd & textrls & d_discnt + error & 4.04 & 0.85 & 16.18 & 0.85 & 0.36 & 2.84 + bad pixels & 5.35% & 0.18% & 86.78% & 0.18% & 0.00% & 2.46% + error & 3.91 & 1.07 & 15.45 & 1.07 & 0.38 & 3.65 + bad pixels & 5.63% & 0.36% & 88.76% & 0.36% & 0.00% & 4.52% + 0.3 in ccccccc + & + & all & non occl & occl & textrd & textrls & d_discnt + error & 1.46 & 0.61 & 7.92 & 0.63 & 0.33 & 1.56 + bad pixels & 3.93% & 1.35% & 93.06% & 1.48% & 0.14% & 5.96% + error & 1.49 & 0.70 & 7.88 & 0.73 & 0.40 & 1.60 + bad pixels & 3.99% & 1.38% & 94.02% & 1.49% & 0.31% & 6.39% + 0.3 in ccccccc + & + & all & non occl & occl & textrd & textrls & d_discnt + error & 1.30 & 0.99 & 5.41 & 1.00 & 0.97 & 2.01 + bad pixels & 4.77% & 2.59% & 87.38% & 2.57% & 2.61% & 10.63% + error & 1.25 & 0.92 & 5.35 & 1.04 & 0.73 & 2.02 + bad pixels & 4.24% & 2.04% & 87.60% & 2.77% & 1.05% & 10.00% + 0.3 in ccccccc + & + & all & non occl & occl & textrd & textrls & d_discnt + error & 1.58 & 1.11 & 8.29 & 0.93 & 1.42 & 1.49 + bad pixels & 3.29% & 1.65% & 90.72% & 1.38% & 2.20% & 7.28% + error & 1.47 & 0.95 & 8.33 & 0.81 & 1.18 & 1.31 + bad pixels & 3.58% & 1.93% & 91.55% & 1.56% & 2.68% & 6.84% + cooperative optimization offers us a general , distributed optimization method for attacking hard optimization problems . soft decision making , message passing , and solution compromising are three important techniques for achieving cooperation among agents in a cooperative optimization system . the global optimality property of consensus solutions offers an appealing reason for agents in a system to compromise their solutions so that conflicts in their solutions can be resolved . the insights we gained at studying cooperative optimization might help us to apply the cooperation principle to understand or solve more generic decision optimization problems arose from fields like neurosciences , business management , political management , and social sciences . for any numbers @xmath286 in @xmath287 , the following inequality can be proved by the principle of mathematical induction , @xmath288 if @xmath289 converges , there exists @xmath251 such that for all @xmath290 , @xmath291 therefore , defining @xmath292 as @xmath293 we have that for all @xmath290 , @xmath294 therefore , the sequence @xmath295 is a non increasing sequence bounded from below by @xmath296 . it must have a positive limit @xmath297 so that @xmath298 s. geman and d. geman , `` stochastic relaxation , gibbs distributions , and the bayesian restoration of images , '' _ ieee transactions on pattern analysis and machine intelligence _ , vol . pami-6 , no . 6 , pp . 721741 , nov . 1984 . y. boykov and v. kolmogorov , `` an experiemental comparison of min - cut / max - flow algorithms for energy minimization , '' _ ieee transactions on pattern analysis and machine intelligence _ , vol . 26 , no . 9 , december 2004 . r. szeliski and r. zabih , `` an experimental comparison of stereo algorithms , '' in _ vision algorithms : theory and practice _ , ser . lncs , b. triggs , a. zisserman , and r. szeliski , eds . 1883.1em plus 0.5em minus 0.4emcorfu , greece : springer - verlag , sept . 1999 , pp . 119 . m. f. tappen and w. t. freeman , `` comparison of graph cuts with belief propagation for stereo , using identical mrf parameters , '' in _ 9th ieee international conference on computer vision ( iccv 2003 ) , 14 - 17 october 2003 , nice , france_.1em plus 0.5em minus 0.4emieee computer society , 2003 , pp . 900907 . m. j. wainwright , t. s. jaakkola , and a. s. willsky , `` map estimation via agreement on ( hyper)trees : message - passing and linear - programming approaches , '' _ ieee transactions on information theory _ , 51 , no . 11 , pp . 36973717 , march 2005 . x. huang , `` a cooperative search approach to global optimization , '' _ proceeding of the first international conference on optimiztion methods and software _ , vol . december , p. 140 , hangzhou , p.r . china , 2002 . , `` a general framework for constructing cooperative global optimization algorithms , '' in _ frontiers in global optimization _ , ser . nonconvex optimization and its applications.1em plus 0.5em minus 0.4em kluwer academic publishers , 2004 , pp . 179221 . , `` cooperative optimization for solving large scale combinatorial problems , '' in _ theory and algorithms for cooperative systems _ , ser series on computers and operations research.1em plus 0.5em minus 0.4emworld scientific , 2004 , pp . 117156 . , `` cooperative optimization for energy minimization in computer vision : a case study of stereo matching , '' in _ pattern recognition , 26th dagm symposium_.1em plus 0.5em minus 0.4emspringer - verlag , lncs 3175 , 2004 , pp . 302309 . d. j. c. mackay , _ information theory , inference , and learning algorithms_.1em plus 0.5em minus 0.4emcambridge university press , 2003 , available from . [ online ] . available : http://www.cambridge.org/0521642981 j. yedidia , w. freeman , and y. weiss , `` constructing free - energy approximations and generalized belief propagation algorithms , '' _ ieee transactions on information theory _ 51 , no . 7 , pp . 22822312 , july 2005 . t. j. richardson , m. a. shokrollahi , and r. l. urbanke , `` design of capacity - approaching irregular low - density parity - check codes , '' _ ieee transactions on information theory _ , vol . 47 , no . 2 , pp . 619637 , february 2001 . a. blake , `` comparison of the efficiency of deterministic and stochestic algorithms for visual reconstruction , '' _ ieee transactions on pattern analysis and machine intelligence _ , vol . pami-11 , no . 1 , pp . 212 , january 1989 .

given a finite group , a-covering of closed riemannian manifolds , and a so - called-relation , a construction of sunada produces a pair of manifolds and that are strongly isospectral . such manifolds have the same dimension and the same volume , and their rational homology groups are isomorphic . here , we investigate the relationship between their integral homology . the cheeger mller theorem implies that a certain product of orders of torsion homology and of regulators for agrees with that for . we exhibit a connection between the torsion in the integral homology of and on the one hand , and the-module structure of integral homology of the covering manifold on the other , by interpreting the quotients representation theoretically . further , we prove that the-torsion in the homology of is isomorphic to that of for all primes . for , we give examples of pairs of strongly isospectral hyperbolic-manifolds for which the-torsion homology differs , and we conjecture such examples to exist for all primes . -0.8 cm

1601.06821

two closed riemannian manifolds @xmath1 and @xmath2 are said to be _ isospectral _ if the laplace beltrami operators acting on functions on @xmath1 and on @xmath2 have the same spectrum , equivalently if the spectral zeta functions @xmath9 and @xmath10 are equal . the manifolds are said to be _ strongly isospectral _ if the spectra of every natural self - adjoint elliptic differential operator on @xmath1 and @xmath2 agree . in particular , if @xmath1 and @xmath2 are strongly isospectral , then the laplace de rham operators acting on differential @xmath11-forms on the respective manifold have the same spectrum , equivalently the zeta functions @xmath12 and @xmath13 encoding those spectra are equal . following kac s famous question `` can one hear the shape of a drum ? '' @xcite , the following broad questions have received a lot of attention @xcite . [ q:1 ] which isometry invariants of closed riemannian manifolds are isospectral invariants ? which ones are strongly isospectral invariants ? for example , two riemannian manifolds that are strongly isospectral have the same dimension , volume and betti numbers . by analogy with the number theoretic notion of so - called arithmetically equivalent number fields , sunada @xcite proposed a general construction showing that there exist riemannian manifolds that are strongly isospectral , but are not even homeomorphic . let us briefly recall sunada s construction . let @xmath14 be a @xmath0-covering of closed manifolds , where @xmath0 is a finite group . let @xmath15 , @xmath16 be two subgroups of @xmath0 with the property that there is an isomorphism @xmath17\cong \c[g / u_2]$ ] of linear permutation representations of @xmath0 , or equivalently that for every conjugacy class @xmath18 of @xmath0 we have @xmath19 . in this case we say that the formal linear combination @xmath20 is a @xmath0-relation . sunada proves that then the intermediate coverings @xmath21 and @xmath22 are strongly isospectral . in this paper , we want to take the analogy with number fields further . the cheeger - mller theorem @xcite , proving a conjecture of ray and singer @xcite , gives a special value formula for the spectral zeta functions of a riemannian manifold , somewhat analogous to the analytic class number formula for the dedekind zeta function of a number field @xcite . it implies that if @xmath1 and @xmath2 are strongly isospectral riemannian manifolds , then @xmath23 where , for a riemannian manifold @xmath24 , @xmath25 is the covolume of the lattice @xmath26 in the vector space @xmath27 with respect to a certain canonical inner product ( see notation [ not : pairings ] [ not : reg ] ) . the study of torsion homology and regulators of manifolds has attracted a lot of attention recently , see for instance @xcite . in light of equation ( [ eq : cheegermuller ] ) , a natural instance of question [ q:1 ] is whether , for @xmath1 and @xmath2 as above , an integer @xmath28 , and a prime number @xmath8 , we must necessarily have @xmath29=\#h_i(m_2,\z)[p^\infty],\end{aligned}\ ] ] where @xmath30 $ ] denotes the @xmath8-primary torsion subgroup . here , we answer this and similar questions in the context of sunada s construction . more broadly , we study homological properties of covering manifolds in @xmath0-relations , analogous to an old and fruitful line of research in number theory . for example , the following basic result is analogous to results of @xcite on class groups of number fields . [ prop : intro1 ] let @xmath14 be a @xmath0-covering of closed riemannian manifolds , let @xmath31 be a @xmath0-relation , and let @xmath32 be an integer . then for all primes @xmath8 that do not divide @xmath33 , we have @xmath34 \cong h_i(x / u_2,\z)[p^\infty].\ ] ] we actually prove a more general result , see theorem [ thm : nomackeyfunctor ] . in providing counter examples , we have concentrated on closed hyperbolic @xmath7-manifolds , a class of manifolds that plays an important rle in other branches of geometry , as well as in number theory . the following is a result in the opposite direction to that of proposition [ prop : intro1 ] . [ prop : intro2 ] let @xmath6 be a prime number . then there exist strongly isospectral closed hyperbolic @xmath7-manifolds @xmath1 and @xmath2 such that @xmath35\neq \#h_1(m_2,\z)[p^\infty].\ ] ] moreover , if @xmath36 , then there exist @xmath7-manifolds @xmath1 and @xmath2 as above with @xmath35=1,\;\;\text { and } \;\;\#h_1(m_2,\z)[p^\infty]=p.\ ] ] note that hyperbolic @xmath7-manifolds have torsion - free homology in degrees @xmath37 , @xmath38 , and @xmath7 , so in the situation of proposition [ prop : intro2 ] , equation ( [ eq : chmtors ] ) is trivially satisfied in those degrees . it is already known that for every prime @xmath8 , there exist strongly isospectral closed riemannian @xmath39-manifolds @xmath1 and @xmath2 such that @xmath40\neq \#h_1(m_2,\z)[p^\infty]$ ] . this is shown in @xcite , using the following two results . firstly , every finite group can be realised as the fundamental group of a closed smooth 4-manifold @xcite . secondly , for every prime @xmath8 , there exists a finite @xmath8-group @xmath0 with a @xmath0-relation @xmath31 , where the abelianisations of @xmath15 and @xmath16 have different orders ( * ? ? ? * , example 3 ) . at the other extreme , if @xmath24 is a closed oriented 2-manifold , then @xmath41 is torsion - free . riemannian @xmath7-manifolds present an in - between case . it follows from the elliptisation theorem , as proven by perelman @xcite , and from a theorem of ikeda @xcite , that any two isospectral @xmath7-manifolds with finite fundamental groups are homeomorphic . in particular , sunada s construction for 4-manifolds can not possibly generalise to @xmath7-manifolds . instead , we prove proposition [ prop : intro2 ] by a judicious choice of @xmath0-relation @xmath20 , a different one for each prime @xmath8 , and then by performing a computer search among @xmath0-coverings of hyperbolic @xmath7-manifolds and computing the resulting torsion homology of the intermediate coverings corresponding to @xmath15 and @xmath16 . nevertheless , we believe that the evidence of proposition [ prop : intro2 ] justifies the following conjecture in @xcite using techniques introduced in this paper , see in particular proposition [ prop : impliesconj ] ] . [ conj : intro ] for every prime number @xmath8 , there exist strongly isospectral closed hyperbolic @xmath7-manifolds @xmath1 and @xmath2 such that @xmath40\neq \#h_1(m_2,\z)[p^\infty]$ ] . it follows from equation ( [ eq : cheegermuller ] ) that if @xmath1 and @xmath2 are strongly isospectral , then @xmath42 can be expressed in terms of torsion homology of the two manifolds , and in particular is a rational number . moreover , if @xmath1 and @xmath2 arise from sunada s construction , then it follows from proposition [ prop : intro1 ] that the numerator and denominator of this rational number are only divisible by prime divisors of @xmath33 . like in the case of torsion , it would be interesting to understand the quotient of regulators in each degree separately . in this direction we have the following result , which we now state for arbitrary @xmath0-relations ( see definition [ defn : relation ] ) . below , for a group @xmath43 we write @xmath44 $ ] where @xmath45 $ ] denotes the derived subgroup of @xmath43 . [ thm : intro3 ] let @xmath14 be a @xmath0-covering of closed oriented riemannian @xmath46-manifolds , and let @xmath47 , @xmath48 be a @xmath0-relation . then : 1 . for every integer @xmath32 , the product @xmath49 is a rational number that is a product of powers of prime divisors of @xmath33 ; 2 . [ item : regcompq ] for every @xmath50 , the class of @xmath51 in @xmath52 only depends on the isomorphism class of the @xmath0-module @xmath53 ; 3 . [ item : regcompz ] if @xmath8 is a prime number that does not divide @xmath54 for any @xmath55 , then the @xmath8-part of @xmath56 only depends on the isomorphism class of the @xmath0-module @xmath57 . the representation theoretic invariant that appears in parts ( [ item : regcompq ] ) and ( [ item : regcompz ] ) of the theorem was first introduced by dokchitser dokchitser in the context of regulators of elliptic curves ( see e.g. @xcite ) , and is called the regulator constant of the representation with respect to the given @xmath0-relation . we actually relate the quotients of regulators to regulator constants in greater generality , namely without the hypothesis on @xmath8 in part ( [ item : regcompz ] ) of theorem [ thm : intro3 ] . but in general , there is an `` error term '' , which measures the failure of @xmath0-descent for homology in the covering see theorem [ thm : regcomparison ] for the precise statement , but see also remark [ rmrk : noclue ] . vignras @xcite introduced another well - known construction of isospectral manifolds , of a more arithmetic flavour . in @xcite , calegari and venkatesh study torsion homology in pairs of manifolds that are of a similar nature to the examples of vignras , called `` jacquet langlands pairs '' , that are not isospectral but whose spectra are closely related . they raise , in ( * ? ? ? * section 7.10 ) , the question of giving algebraic interpretations to the quotients of torsion homology and regulators separately , analogously to theorem [ thm : intro3 ] . it would also be interesting to investigate the same type of questions in the context of the vignras examples . in section [ sec : examples ] , we give examples of how theorem [ thm : intro3 ] allows one to deduce concrete information about the @xmath58$]-module structure of @xmath59 from the torsion homology of quotients of @xmath60 . it also allows one to formulate a possible representation theoretic line of attack on conjecture [ conj : intro ] , such as the following result , which will be proved as corollary [ cor : impliesconj ] . [ prop : impliesconj ] let @xmath8 be an odd prime , let @xmath61 , and define the subgroups @xmath62 of @xmath0 . suppose that there exists a @xmath0-covering @xmath14 of closed hyperbolic @xmath7-manifolds such that the difference of the betti numbers @xmath63 is odd . then conjecture [ conj : intro ] holds for @xmath8 . to fully appreciate the representation theoretic nature of proposition [ prop : impliesconj ] , see the full statement of corollary [ cor : impliesconj ] . the structure of the paper is as follows . in section [ sec : brauer ] we recall the formalism of burnside groups , representation groups , and @xmath0-relations , and the definition of regulator constants . in section [ sec : manifolds ] we investigate the behaviour of homological invariants of riemannian manifolds in @xmath0-relations , and prove proposition [ prop : intro1 ] and theorem [ thm : intro3 ] . in section [ sec : regconst ] we compute the regulator constants of the rational representations of @xmath64 for odd primes @xmath8 , and of @xmath65 , with respect to certain @xmath0-relations . these computations will allow us to deduce concrete information about the @xmath58$]-module structure of the rational homology of @xmath0-coverings from the torsion homology of intermediate coverings . together with theorem [ thm : regcomparison ] , these calculations will also imply proposition [ prop : impliesconj ] . we prove proposition [ prop : intro2 ] by a direct computer search , and section [ sec : calc ] is devoted to the methods and algorithms used to that end . in section [ sec : examples ] we have collected some interesting examples , illustrating the various phenomena that we investigate here . all our riemannian manifolds will be assumed to be finite - dimensional , oriented , and closed . by an automorphism of a riemannian manifold we mean an orientation preserving diffeomorphism from the manifold to itself that is a local isometry . by a hyperbolic @xmath7-manifold , we mean a quotient of the hyperbolic @xmath7-space by a discrete subgroup of orientation preserving isometries . if @xmath8 is a prime number , we will write @xmath66 for the localisation of @xmath67 at @xmath8 , i.e. the subring @xmath68 of @xmath69 . for a rational number @xmath70 , with @xmath71 and @xmath72 , the @xmath8-adic valuation of @xmath73 is defined to be @xmath74 . when @xmath24 is a @xmath67-module , we write @xmath75 for the torsion submodule of @xmath24 and @xmath76 for the torsion - free quotient of @xmath24 . we would like to thank nicolas bergeron , nathan dunfield , derek holt , emilio lauret , and karen vogtmann for helpful discussions , and the anonymous referee for suggestions that improved the exposition . we recall some standard definitions , for which we refer to ( * ? ? ? 11 ) . by `` module '' we will always mean a finitely generated left module . let @xmath0 be a finite group . the _ burnside group _ of @xmath0 is the free abelian group on isomorphism classes of transitive @xmath0-sets . the set of transitive @xmath0-sets is in bijection with the set of conjugacy classes of subgroups of @xmath0 via the map that assigns to the subgroup @xmath43 the set of cosets @xmath77 . using this identification , we will represent elements of the burnside group as formal sums @xmath78 , where @xmath79 and @xmath80 . let @xmath81 be a domain . representation group _ of @xmath0 over @xmath81 is the free abelian group on isomorphism classes of @xmath81-free indecomposable @xmath82$]-modules , where @xmath82 $ ] is the group ring of @xmath0 over @xmath81 . let @xmath0 be a finite group and @xmath81 be a domain . we write @xmath83 for the free @xmath81-module of rank @xmath84 with trivial @xmath0-action . we have a natural group homomorphism @xmath85}$ ] from the burnside group of @xmath0 to the representation group of @xmath0 over @xmath81 , which sends a @xmath0-set @xmath60 to the permutation module @xmath86 $ ] with an @xmath81-basis indexed by the elements of @xmath60 , and with the @xmath81-linear @xmath0-action given by permutations of the basis . [ defn : relation ] let @xmath81 be a domain . @xmath82$]-relation _ is an element of the kernel of @xmath85}$ ] . a @xmath58$]-relation will be referred to simply as a @xmath0-relation . [ ex : dihedral ] let @xmath8 be a prime , and let @xmath87 be a dihedral group of order @xmath88 . then @xmath89 is a @xmath0-relation . concretely , this means that there is an isomorphism of linear permutation representations of @xmath0 over @xmath69 , @xmath90\oplus \q[g / g]^{\oplus 2}\cong \q[g / c_2]^{\oplus 2 } \oplus \q[g / c_p].\ ] ] moreover , for every prime @xmath91 , the relation @xmath92 is in fact a @xmath93$]-relation , but it is not a @xmath94$]-relation see ( * ? ? ? * proof of proposition 3.9 ) . [ ex:2-gassman ] let @xmath0 be the affine linear group over @xmath95 , i.e. the group of linear transformations @xmath96 of @xmath95 , where @xmath97 and @xmath98 . consider the subgroups @xmath99 and @xmath100 . the group @xmath0 is isomorphic to the semi - direct product @xmath101 , and the subgroups @xmath15 and @xmath16 are both isomorphic to @xmath102 . then @xmath20 is a @xmath94$]-relation for every odd prime @xmath8 , but is not a @xmath103$]-relation , as can be deduced from a direct character computation and lemma [ lem : grelations ] below . [ ex : p - gassman ] let @xmath8 be an odd prime , and let @xmath104 . consider the two subgroups @xmath105 then for every prime @xmath106 , @xmath20 is a @xmath93$]-relation if and only if @xmath91 . observe that @xmath107 contains the subgroup @xmath108 , which is central in @xmath0 . set @xmath109 , and @xmath110 for @xmath111 and @xmath38 . we then get the @xmath112-relation @xmath113 . moreover , the triple @xmath114 minimises the index @xmath115 among all triples @xmath116 , where @xmath117 is a finite group , and @xmath118 is a @xmath117-relation that is not a @xmath119$]-relation ( see @xcite ) . [ lem : grelations ] let @xmath0 be a finite group , let @xmath8 be a prime number not dividing @xmath33 , and let @xmath92 be an element of the burnside group of @xmath0 . then the following are equivalent : 1 . @xmath92 is a @xmath120$]-relation ; 2 . @xmath92 is a @xmath58$]-relation ; 3 . @xmath92 is a @xmath94$]-relation ; 4 . @xmath92 is an @xmath121$]-relation . the equivalence of ( 1 ) and ( 2 ) follows from the fact that if @xmath122 and @xmath123 are @xmath58$]-modules , then @xmath124 ( see @xcite ) . the remaining equivalences are proven in ( * ? ? ? * chapter 5 ) . [ not : burnsidefunctions ] let @xmath125 be any function on the set of conjugacy classes of subgroups of a finite group @xmath0 with values in an abelian group @xmath126 . then @xmath125 extends to a unique group homomorphism from the burnside group of @xmath0 to @xmath126 , defined by @xmath127 , where @xmath128 are subgroups of @xmath0 and @xmath48 . let @xmath129 be a subring of @xmath130 that is a pid . the main examples we are thinking of are @xmath67 , the localisation @xmath66 where @xmath8 is a prime , and @xmath69 . given an @xmath131$]-module @xmath126 , let @xmath132 denote the @xmath131$]-submodule consisting of @xmath129-torsion elements of @xmath126 , and let @xmath133 . since @xmath129 is a pid , for every @xmath131$]-module @xmath126 , the quotient @xmath134 is an @xmath129-free @xmath131$]-module . moreover , @xmath135 is _ self - dual _ , i.e. it is isomorphic to the @xmath136$]-module @xmath137 . thus there exists an @xmath130-valued @xmath129-bilinear @xmath0-invariant non - degenerate pairing on @xmath134 . from now on , we will just say _ pairing on @xmath126 _ when we mean an @xmath130-valued @xmath129-bilinear @xmath0-invariant pairing on @xmath126 that is non - degenerate on @xmath134 . let @xmath0 be a finite group , let @xmath129 a subring of @xmath130 that is a pid , let @xmath81 be a subring of @xmath69 , and let @xmath126 be an @xmath131$]-module . let @xmath138 be a pairing on @xmath126 , and let @xmath139 be an @xmath82$]-relation . the _ regulator constant _ of @xmath126 with respect to @xmath92 is defined as @xmath140 where the @xmath55-th determinant is computed with respect to any @xmath129-basis on @xmath141 . the class in @xmath142 of each determinant is independent of the choice of basis . [ thm : regconst ] the value of @xmath143 is independent of the pairing @xmath138 on @xmath126 . see ( * ? ? ? * theorem 2.17 ) . it follows from theorem [ thm : regconst ] that if @xmath144 , then @xmath145 is well - defined . indeed , @xmath146 , and the pairing can always be chosen to take values in @xmath69 . similarly , if @xmath147 , then @xmath148 is well - defined , while if @xmath149 , then @xmath143 defines a class in @xmath52 . [ cor : multiplicative ] let @xmath150 , @xmath151 be @xmath131$]-modules , and @xmath92 a @xmath0-relation . then @xmath152 . we can choose the pairing on @xmath153 so that the direct summands are orthogonal to each other , making all the matrices block diagonal . [ ex : regconsttriv ] let @xmath0 be a finite group , and let @xmath154 be a @xmath0-relation . then @xmath155 . [ lem : degree0 ] let @xmath0 be a finite group and let @xmath139 be a @xmath0-relation . then we have @xmath156 . by definition , the virtual representation @xmath157^{\oplus n_j}$ ] is zero . the result follows by taking the inner product with the trivial character of @xmath0 . [ prop : orthoreg ] let @xmath0 be a finite group , and let @xmath158 be a @xmath0-relation . let @xmath126 be a @xmath58$]-module that has no simple summand in common with any @xmath159 $ ] . then @xmath160 . see ( * ? ? ? * lemma 2.26 ) . let @xmath0 be a finite group , and @xmath161 a subgroup . the operation on @xmath0-sets of restricting the action of @xmath0 to @xmath161 extends linearly to a restriction map on the burnside groups , and induces a restriction map on relations . with respect to the bases of the burnside groups given by transitive @xmath0-sets , respectively transitive @xmath161-sets , the restriction map is given by mackey s formula : @xmath162 we have the following form of frobenius reciprocity for regulator constants . [ prop : frobenius ] let @xmath0 be a finite group , let @xmath161 be a subgroup , let @xmath126 be an @xmath163$]-module , and let @xmath92 be a @xmath0-relation . then we have @xmath164 see ( * ? ? ? * proposition 2.45 ) . in this section , we investigate the quotients of torsion homology and of regulators of riemannian manifolds arising from @xmath0-relations . we begin by recalling some basic definitions and fixing the notation . the reader is referred to @xcite for the details . if @xmath165 , @xmath166 are two abelian groups equipped with @xmath130-valued bilinear forms @xmath167 , respectively @xmath168 , a _ similitude from @xmath165 to @xmath166 with factor @xmath169 _ is a homomorphism @xmath170 such that @xmath171 for all @xmath172 , @xmath173 . an _ isometry _ is a similitude with factor 1 . [ not : pairings ] let @xmath60 be a @xmath46-dimensional riemannian manifold , and let @xmath32 be an integer . 1 . _ harmonic forms . _ the laplace de rham operator acts on the differential @xmath11-forms on @xmath60 , and we let @xmath174 denote the space of differential @xmath11-forms that lie in the kernel of that operator , the space of harmonic @xmath11-forms . the riemannian metric on @xmath60 induces a canonical inner product on @xmath174 , which induces a canonical isomorphism between @xmath174 and the @xmath130-linear dual @xmath175 . explicitly , the inner product on @xmath174 is given by @xmath176 where @xmath177 , @xmath178 , and where @xmath179 denotes the hodge star operator @xmath180 ( see @xcite ) . we equip @xmath181 with the inner product induced by the one on @xmath174 , i.e. the unique inner product that makes the map @xmath182 an isometry . [ not : reg ] _ regulators . _ we have a homomorphism @xmath183 given by @xmath184 . it follows from the hodge theorem ( ( * ? ? ? * theorem 1.45 ) ) and de rham s theorem that @xmath185 and that the image of @xmath186 spans @xmath181 over @xmath130 . @xmath11-th regulator _ @xmath187 of @xmath60 is defined as the covolume of @xmath188 with respect to the inner product on @xmath181 . + let @xmath189 denote the pullback via @xmath186 of the pairing on @xmath181 , i.e. the unique pairing on @xmath190 that makes @xmath186 an isometry . poincar duality . _ by poincar duality , the map @xmath191 given by cap product with the fundamental class of @xmath60 is an isomorphism . it follows that @xmath192 . [ rmrk : equivariant ] _ actions of automorphism groups . _ let @xmath0 be a group of automorphisms of @xmath60 . then @xmath0 acts on the left by linear transformations on @xmath190 via pushforward , @xmath193 for @xmath194 and @xmath195 . also , @xmath0 acts linearly on the right on @xmath174 via pullback , @xmath196 for @xmath194 and @xmath197 , which induces the dual action on the left on @xmath198 . it follows from the adjointness of pullbacks and pushforwards that the map @xmath186 is a homomorphism of left @xmath0-modules . moreover , recall that our automorphisms are assumed to be orientation preserving , so the poincar duality map @xmath199 is an isomorphism of @xmath0-modules . [ lem : regdet ] let @xmath60 be a riemannian manifold . then we have @xmath200 , where the left hand side is computed with respect to any @xmath67-basis of @xmath201 . in particular , the pairing @xmath202 is non - degenerate on @xmath201 . the claimed equality is the expression of the covolume of a lattice in terms of the determinant of its gram matrix . [ rmrk : poincarereg ] explicitly , let @xmath203 be an orthonormal basis of @xmath174 , and let @xmath204 be a @xmath67-basis of @xmath201 . then we have @xmath205 and for all @xmath206 , @xmath207 , we have @xmath208 . we now prove proposition [ prop : intro1 ] as a consequence of the following result . [ thm : nomackeyfunctor ] let @xmath14 be a finite @xmath0-covering of riemannian manifolds , and let @xmath209 be a @xmath94$]-relation , where @xmath210 and @xmath211 are positive integers . then for every integer @xmath32 , we have @xmath212^{n_j}\cong \bigoplus_k h_i(x / u_k',\z)[p^\infty]^{n_k'}.\ ] ] in particular we have @xmath213 ( see notation [ not : burnsidefunctions ] ) . let @xmath43 be any subgroup of @xmath0 . then the @xmath214-torsion of @xmath215 is isomorphic to the torsion subgroup of @xmath216 . by definition of homology with local coefficients ( see ( * ? ? ? * h ) ) , we have @xmath217)$ ] . since homology with local coefficients is additive in direct sums of modules , the result follows . let @xmath14 and @xmath0 be as in theorem [ thm : nomackeyfunctor ] , let @xmath8 be a prime number that does not divide @xmath33 , and let @xmath92 be a @xmath0-relation . then @xmath212^{n_j}\cong \bigoplus_k h_i(x / u_k',\z)[p^\infty]^{n_k'}.\ ] ] this is a direct consequence of theorem [ thm : nomackeyfunctor ] and lemma [ lem : grelations ] . the rest of the section is devoted to the behaviour of regulators in @xmath0-relations . in particular , we will prove theorem [ thm : intro3 ] . [ lem : h0 ] let @xmath218 be a finite @xmath0-covering of riemannian manifolds of dimension @xmath46 , and let @xmath139 be a @xmath0-relation . then @xmath219 for all @xmath220 , we have @xmath221 . it follows that @xmath222 where the last equality follows from example [ ex : regconsttriv ] and lemma [ lem : degree0 ] . [ lem : simil ] let @xmath223 be a finite @xmath0-covering of riemannian manifolds , and let @xmath32 be an integer . then the following maps are similitudes with factor @xmath33 : 1 . [ item : prepnotisom ] the pullback map @xmath224 ; 2 . [ item : notisom ] its dual @xmath225 ; 3 . [ item : descentpairing ] the pushforward map @xmath226 4 . [ item : ascentpairing ] any map @xmath227 satisfying @xmath228 . 1 . let @xmath229 $ ] and @xmath230 $ ] denote the fundamental class on @xmath60 , respectively on @xmath231 . then @xmath232 = \#g\cdot [ y]$ ] , so for all @xmath233 , we have @xmath234 2 . the assertion immediately follows from part ( [ item : prepnotisom ] ) , from the fact that @xmath235 is an isomorphism , and from the definition of the inner product on the dual space , see notation [ not : pairings ] . 3 . by notation [ not : pairings ] [ rmrk : equivariant ] , the map @xmath186 sends @xmath236 to @xmath237 , and it follows from the adjointness of pullbacks and pushforwards that the diagram @xmath238^{h_x^i } \ar[r]^(.56){f _ * } & h_i(y,\z ) \ar[d]^{h_y^i } \\ ( \ch^i(x)^g)^{\lor } \ar[r]^(.6){(f^*)^{\lor } } & \ch^i(y)^{\lor } } \ ] ] is commutative . note that @xmath237 is canonically isomorphic , as an inner product space , to the @xmath0-coinvariants @xmath239 , which in turn is canonically isomorphic to @xmath240 . this shows that @xmath241 is an isometry . by definition , @xmath242 is also an isometry , and by part ( [ item : notisom ] ) , @xmath243 is a similitude with factor @xmath244 . it follows that @xmath245 is also a similitude with factor @xmath244 . 4 . by part ( [ item : descentpairing ] ) , the map @xmath245 is a similitude with factor @xmath244 , and the claim immediately follows . [ lem : exconvenient ] let @xmath218 be a finite @xmath0-covering of riemannian manifolds , and let @xmath32 be an integer . then the composition @xmath246 satisfies @xmath247 . the pushforward of the fundamental class of @xmath60 is equal to @xmath33 times the fundamental class of @xmath231 . it therefore follows from the naturality of the cap product that @xmath248 , which proves the result . [ lem : index ] let @xmath218 be a finite @xmath0-covering of riemannian manifolds , and let @xmath249 be an integer . let @xmath227 be any map satisfying @xmath228 , and let @xmath250 then @xmath251 is finite and is independent of @xmath252 . the quantity @xmath251 is finite since @xmath253 is an isomorphism . if @xmath254 is another map satisfying the conditions of the lemma , then @xmath255 . it follows from this , and from the fact that @xmath256 is an isomorphism , that @xmath257 . therefore @xmath258 which proves the lemma . given a finite @xmath0-covering @xmath14 and @xmath11 as in lemma [ lem : index ] , we define @xmath259 for any @xmath252 satisfying the hypotheses of the lemma . [ thm : regcomparison ] let @xmath218 be a @xmath0-covering of riemannian manifolds , let @xmath260 be a @xmath0-relation ( see definition [ defn : relation ] ) , and let @xmath32 be an integer . then we have @xmath261 by lemma [ lem : regdet ] , the pairing @xmath262 on @xmath190 is non - degenerate on the quotient modulo torsion . moreover , by lemma [ lem : simil ] ( [ item : descentpairing ] ) applied to @xmath263 for each @xmath194 , this pairing is @xmath0-invariant . for any @xmath80 , let @xmath252 be any map satisfying the conditions of lemma [ lem : index ] for the covering @xmath264 . then we have @xmath265 where the second equality follows from lemma [ lem : simil]([item : ascentpairing ] ) , and the last one from lemma [ lem : regdet ] . this proves the theorem , by taking the product over @xmath55 corresponding to @xmath92 . [ cor : modsquares ] let @xmath218 be a @xmath0-covering of riemannian manifolds , let @xmath260 be a @xmath0-relation , and let @xmath32 be an integer . then we have @xmath266 . let @xmath43 be a group . the abelianisation of @xmath43 is the quotient @xmath267 $ ] , where @xmath45=\langle uvu^{-1}v^{-1}\colon u , v\in u\rangle$ ] , i.e. the maximal abelian quotient of @xmath43 . [ cor : regcomparison ] let @xmath223 be a finite @xmath0-covering of riemannian @xmath46-manifolds , let @xmath139 be a @xmath0-relation , let @xmath8 be a prime , and let @xmath32 be an integer . assume that one of the following holds : 1 . @xmath268 and for each subgroup @xmath80 , the order of @xmath269 is not divisible by @xmath8 ; 2 . for each subgroup @xmath80 , the order of @xmath128 is not divisible by @xmath8 . then @xmath270 . in particular , for all @xmath32 the rational number @xmath271 is a product of the primes dividing the order of @xmath0 . let @xmath43 be any subgroup of @xmath0 . let @xmath272 be the map of lemma [ lem : exconvenient ] . in case ( 1 ) , the 5-term exact sequence coming from the cartan leray spectral sequence in cohomology ( see ( * ? ? ? vii , 7 ) ) is @xmath273 by the universal coefficient theorem , we have @xmath274 , which has order coprime to @xmath8 by assumption . in case ( 2 ) , the group @xmath43 has cohomological dimension @xmath37 over @xmath66 , since @xmath8 does not divide its order , so by the cartan leray spectral sequence we have an isomorphism @xmath275 in both cases , it follows that @xmath276 has cokernel of order coprime to @xmath8 , and since the poincar duality maps @xmath199 and @xmath277 are isomorphisms , the cokernel of @xmath278 is also of order coprime to @xmath8 . the first equality therefore follows from theorem [ thm : regcomparison ] . for the second equality , note that firstly , @xmath279 , and secondly that for an arbitrary @xmath280$]-module @xmath126 , we have @xmath281 . [ rmrk : noclue ] we do not know whether the regulator quotient is a purely representation theoretic invariant in full generality without the restrictive assumptions of corollary [ cor : regcomparison ] , that is whether it only depends on the @xmath280$]-module structure of @xmath190 . this seems to us to be an interesting question . in this section , we compute the regulator constants of rational representations with respect to the relations of examples [ ex:2-gassman ] and [ ex : p - gassman ] . first , let @xmath61 . denote by @xmath282 the subgroup of upper - triangular matrices in @xmath0 , let @xmath283 , and let @xmath15 , respectively @xmath16 be the subgroup of matrices in @xmath282 with square lower right , respectively upper left entry . then @xmath284 is the @xmath0-relation of example [ ex : p - gassman ] . in this section , we compute the regulator constants of the rational irreducible representations of @xmath0 with respect to @xmath92 . we begin by recalling the classification of complex representations of @xmath0 . for any 1-dimensional representations @xmath285 , @xmath286 of @xmath287 , let @xmath288 be the @xmath84-dimensional representation of @xmath282 defined by @xmath289 . [ prop : classif ] the following is a classification of the complex irreducible representations of @xmath0 : * @xmath84-dimensional representations : they are all of the form @xmath290 , where @xmath291 is an irreducible character of @xmath287 . * irreducible principal series , of dimension @xmath292 : they are the inductions @xmath293 , where @xmath294 ; we have @xmath295 if and only if @xmath296 . * special representations , of dimension @xmath8 : they are the @xmath8-dimensional summands @xmath297 of @xmath298 . * cuspidal representations , of dimension @xmath299 : they are the irreducible representations of @xmath0 that are not summands of any @xmath300 . see @xcite . let @xmath301 denote the complex irreducible representation @xmath302 , the steinberg representation of @xmath0 , and let @xmath303 denote the complex irreducible representation @xmath304 , where @xmath305 is the unique character of @xmath287 of order 2 . note that both @xmath301 and @xmath303 are realisable over @xmath69 , so in fact give rise to irreducible @xmath58$]-representations . [ prop : regconstpgassman ] let @xmath306 be an irreducible @xmath58$]-representation . then@xmath307 if @xmath308 , and @xmath309 otherwise . we have @xmath17\cong \c[g / u_2 ] \cong \triv\oplus \st\oplus i$ ] . so if @xmath306 is an irreducible @xmath58$]-representation that is not isomorphic to any of these three direct summands , then @xmath310 by proposition [ prop : orthoreg ] . moreover , we have @xmath311 by example [ ex : regconsttriv ] . to compute @xmath312 , observe that by corollary [ cor : multiplicative ] and by proposition [ prop : frobenius ] , we have @xmath313 let @xmath314 for @xmath111 , @xmath38 . an elementary calculation ( or the bruhat decomposition ) shows that for @xmath111 and @xmath38 , @xmath315 has cardinality 2 , with double coset representatives @xmath84 and @xmath316 . moreover , we have @xmath317 , and @xmath318 . it follows from equation ( [ eq : res ] ) that @xmath319 we conclude from equation ( [ eq : steinberg ] ) and example [ ex : regconsttriv ] that @xmath320 similarly , we have @xmath321 since @xmath322 is 1-dimensional , the contribution from each subgroup @xmath43 in @xmath323 is 1 if @xmath324 is zero , and is @xmath325 otherwise . it follows that @xmath326 . [ cor : impliesconj ] let @xmath8 be an odd prime , let @xmath0 , @xmath282 , @xmath15 , and @xmath16 be as above , and let @xmath14 be a @xmath0-covering of closed hyperbolic @xmath7-manifolds . then the following are congruent modulo 2 : 1 . the @xmath8-adic valuation @xmath327 ; 2 . the multiplicity in @xmath328 of the irreducible @xmath120$]-representation @xmath303 , as defined after proposition [ prop : classif ] ; 3 . the difference of the betti numbers @xmath329 . first , we show the equality between ( 1 ) and ( 2 ) . for closed hyperbolic @xmath7-manifolds , all homology groups with integral coefficients except possibly the first are torsion - free . it therefore follows from equation ( [ eq : cheegermuller ] ) , remark [ rmrk : poincarereg ] , lemma [ lem : h0 ] , and example [ ex : regconsttriv ] , that @xmath330 by corollary [ cor : modsquares ] , the @xmath8-adic valuation of the right hand side is congruent to @xmath331 , where @xmath332 . recall that the right hand side of that last congruence is only well - defined modulo 2 . the equality of ( 1 ) and ( 2 ) therefore follows from corollary [ cor : multiplicative ] and proposition [ prop : regconstpgassman ] . we now prove the equality between ( 2 ) and ( 3 ) . below , we will identify representations of @xmath0 with their characters . let @xmath305 be an arbitrary rational representation of @xmath0 . we have @xmath333 , so the multiplicity of @xmath303 in @xmath305 is equal to @xmath334 where @xmath335 denotes inner products of characters of @xmath43 . the equality between ( 2 ) and ( 3 ) follows from this calculation , with @xmath336 . as in example [ ex : p - gassman ] , let @xmath337 , let @xmath109 , @xmath110 for @xmath111 , @xmath38 , and let @xmath338 , which is a @xmath112-relation . there is a canonical bijection between irreducible @xmath58$]-representations with kernel containing @xmath339 and irreducible @xmath340$]-representations , given by projecting to the quotient . for any irreducible @xmath58$]-representation @xmath306 with kernel containing @xmath339 , let @xmath341 denote its image under this bijection . let @xmath341 be an irreducible @xmath340$]-representation . then@xmath342 if @xmath343 , and @xmath344 otherwise . by ( * proposition 2.45 ) , we have @xmath345 . the result therefore immediately follows from proposition [ prop : regconstpgassman ] . now let @xmath0 be the group of affine linear transformations @xmath96 of @xmath95 , where @xmath97 and @xmath98 , let @xmath99 and @xmath100 , as in example [ ex:2-gassman ] . let @xmath305 be an irreducible faithful representation of @xmath346 . then @xmath347 is a 4-dimensional irreducible @xmath120$]-representation , and is realisable over @xmath69 , so it gives rise to an irreducible @xmath58$]-representation . its isomorphism class does not depend on the choice of the representation @xmath305 . all the other complex irreducible representations of @xmath0 are also realisable over @xmath69 , and are 1- and 2-dimensional . an explicit calculation , which we omit , gives the regulator constants of all the irreducible @xmath58$]-representations with respect to @xmath332 as follows . [ prop:2regcst ] let @xmath306 be an irreducible @xmath58$]-representation . then@xmath348 if @xmath308 , and @xmath349 otherwise . in this section , we present the computational methods used to prove proposition [ prop : intro2 ] . the computations were run on the warwick number theory cluster . we have made the resulting data available at https://arxiv.org/abs/1601.06821 . for each prime @xmath8 , we look for explicit pairs of manifolds @xmath1 and @xmath2 as in proposition [ prop : intro2 ] . the pairs we construct all arise as intermediate coverings of @xmath0-coverings @xmath14 of closed hyperbolic @xmath7-manifolds , for suitable finite groups @xmath0 . we compute many such coverings , and explicitly compute the homology groups of suitable intermediate covers . there are several choices involved , which we now explain in detail . we only compute with arithmetic hyperbolic @xmath7-manifolds . we will recall the definitions , assuming basic facts and terminology on quaternion algebras ; a standard reference is @xcite . let @xmath350 denote the hyperbolic 3-space . we identify the group of orientation preserving isometries of @xmath350 with @xmath351 via the poincar extension ( @xcite ) . for any order @xmath352 in a quaternion algebra , let @xmath353 denote the group of elements in @xmath352 of reduced norm @xmath84 . a number field is called _ almost totally real ( atr ) _ if it has exactly one complex place . a quaternion algebra over an atr number field @xmath354 is called _ kleinian _ if it ramifies at every real place of @xmath354 . for any ring @xmath81 , let @xmath355 denote the algebra of @xmath356 matrices over @xmath81 . an _ arithmetic hyperbolic @xmath7-manifold _ is a manifold commensurable with a quotient @xmath357 , where @xmath358 arises as follows . let @xmath359 be a kleinian quaternion algebra over an atr number field , and let @xmath360 be an isomorphism induced by a complex embedding of @xmath354 . then define @xmath361 an arithmetic hyperbolic @xmath7-manifold has finite volume and is compact if and only if @xmath282 is a division algebra see ( * ? ? ? * theorem 8.2.2 ) . let @xmath359 be a quaternion algebra over a number field , let @xmath362 denote the ring of integers in @xmath354 , and let @xmath363 be a maximal order in @xmath282 . let @xmath364 be an ideal of @xmath362 coprime to the discriminant of @xmath282 , so that @xmath365 , and let @xmath366 be the preimage of the subring of upper - triangular matrices under such an isomorphism . then @xmath367 is an order in @xmath282 . [ lem : torfree ] let @xmath354 be a number field , and let @xmath282 be a quaternion algebra over @xmath354 that is a division algebra . let @xmath368 be the set of primes @xmath106 such that @xmath369 is isomorphic to a quadratic extension of @xmath354 contained in @xmath282 , where @xmath370 denotes a primitive @xmath371-th root of unity in an algebraic closure of @xmath354 . for each @xmath372 , and for each primitive @xmath371-th root of unity @xmath373 , choose a prime ideal @xmath374 of @xmath362 such that the minimal polynomial of @xmath373 is irreducible modulo @xmath374 . let @xmath375 be the set of such @xmath374 . then : 1 . the set @xmath368 is finite ; 2 . for every ideal @xmath364 of @xmath362 divisible by all @xmath376 , the group @xmath377 is torsion - free . if @xmath372 is odd , then @xmath378\le [ f(\zeta_{2q})\colon\q ] = 2[f\colon\q]$ ] , so that @xmath379 + 1 $ ] . this proves ( 1 ) . to prove ( 2 ) , assume for a contradiction that there exists an element @xmath380 such that @xmath73 has prime order @xmath106 modulo @xmath381 . replacing @xmath73 by @xmath382 if necessary , we may assume that @xmath73 has order @xmath383 : if @xmath106 is odd then either @xmath73 or @xmath382 has order @xmath383 ; if @xmath384 , then the only elements of order @xmath38 are @xmath385 , since @xmath282 is a division algebra , so that @xmath73 has order @xmath39 . in particular , @xmath386 is a subfield of @xmath282 isomorphic to @xmath369 . we claim that this is a quadratic extension of @xmath354 , i.e. that @xmath106 belongs to @xmath368 : if @xmath369 is not quadratic , then @xmath387 , but @xmath388 , and @xmath73 does not have order @xmath383 . let @xmath389 be the minimal polynomial of @xmath73 , and let @xmath376 be the corresponding prime ideal . by definition , the image of @xmath73 in @xmath390 upper - triangular . since @xmath391 , this implies that @xmath389 is not irreducible modulo @xmath392 : a contradiction . in our search , we extract atr number fields of degree up to 6 and discriminants with absolute value up to @xmath393 from the pari number fields database @xcite . then , using the algorithms from @xcite implemented in the computer algebra system magma @xcite , we compute 30,000 kleinian quaternion algebras @xmath282 whose maximal orders @xmath394 are such that @xmath395 has covolume at most @xmath396 . using the sufficient condition of lemma [ lem : torfree ] , we find many orders @xmath367 such that @xmath397 is torsion - free , and therefore acts freely on @xmath350 , and hence we produce many arithmetic hyperbolic @xmath7-manifolds @xmath398 . the usual way of proving that arithmetic groups have a torsion - free subgroup of finite index is to consider principal congruence subgroups , which have large index . lemma [ lem : torfree ] allows us to use groups of the form @xmath399 , which have smaller covolume , reducing the cost of the computation . theorem [ thm : nomackeyfunctor ] implies that in order to produce examples as in proposition [ prop : intro2 ] using sunada s method , we need to look for @xmath0-coverings @xmath14 , where @xmath0 is a finite group that admits a @xmath0-relation @xmath332 that is not a @xmath94$]-relation . for @xmath400 , we take @xmath0 and @xmath92 as in example [ ex:2-gassman ] , and for @xmath8 odd , we use the relation of example [ ex : p - gassman ] . in order to find many such @xmath0-coverings of every manifold @xmath401 obtained in section [ sec : y ] , we first compute a finite presentation of @xmath117 , using the algorithms of @xcite . since @xmath402 , @xmath0-coverings of @xmath231 correspond to surjective homomorphisms @xmath403 . using the presentation of @xmath117 , we enumerate surjective homomorphisms @xmath404 up to conjugacy , using the methods described in @xcite . the complexity of the enumeration depends heavily on @xmath244 , so we use the following improvement . let @xmath405 be the free group on @xmath406 generators . recall that for any group @xmath407 there is a canonical bijection @xmath408 . moreover , if @xmath407 is an abelian group , then this bijection is an isomorphism of abelian groups . we will tacitly use this identification in the next result . let @xmath117 be a group with a finite presentation @xmath409 where @xmath81 is the normal closure of the subgroup of @xmath405 generated by @xmath410 elements @xmath411 . let @xmath0 be a group , let @xmath407 be a subgroup of the center of @xmath0 , and set @xmath412 . let @xmath413 be a homomorphism , and let @xmath414 be an arbitrary lift of @xmath415 . let @xmath416 , and let @xmath417 denote the linear map induced by evaluation on the @xmath418 . then : 1 . we have @xmath419 ; and 2 . the lifts @xmath420 of @xmath421 are exactly the homomorphisms @xmath422 , where @xmath125 ranges over @xmath423 . 1 . since @xmath424 is a lift of @xmath413 , we have @xmath425 , and @xmath419 . 2 . since @xmath407 is central , every lift of @xmath415 is of the form @xmath422 , where @xmath426 . such a lift descends to a homomorphism @xmath404 if and only if it vanishes on the generators of @xmath81 . for every such generator @xmath316 , we have @xmath427 if and only if @xmath428 , so that the homomorphism descends if and only if @xmath429 . applying this proposition to @xmath61 , @xmath430 and @xmath431 , we reduce the enumeration of @xmath432 to : * enumerating @xmath433 , * computing inverse images under linear maps @xmath434 with @xmath435 , which can be done efficiently by linear algebra . since the enumeration algorithm is expensive , another simple improvement was useful : if there exists a surjection @xmath403 , then @xmath436 surjects onto @xmath437 . checking this condition before the enumeration is fast and rules out many groups . given a surjective homomorphism @xmath403 , we can compute the homology of the isospectral manifolds as follows . for every @xmath220 , we have @xmath438)\cong h_1(\gamma,\z[g / u])\text{,}\ ] ] and we can compute the first homology group by linear algebra from a presentation of the group @xmath117 . recall that we want to find examples of @xmath0-coverings @xmath218 such that @xmath439 \neq 1 $ ] . since linear algebra over @xmath67 is costly , we used the following strategy : we first compute @xmath440 for @xmath111 , @xmath38 , which only uses linear algebra over @xmath441 , and we only compute the homology over @xmath67 if @xmath442 . that condition by itself already implies that @xmath443\not\cong h_1(x / u_2,\z)[p^\infty]$ ] . in practice , when @xmath444 , this condition often yields examples in which @xmath445 $ ] is actually trivial for one @xmath55 , and isomorphic to @xmath446 for the other one . by computing explicit examples , we obtain proposition [ prop : intro2 ] , which we restate here for convenience . [ intro2bis ] let @xmath6 be a prime number . then there exist strongly isospectral closed hyperbolic @xmath7-manifolds @xmath1 and @xmath2 such that @xmath35\neq \#h_1(m_2,\z)[p^\infty].\ ] ] moreover , if @xmath36 , then there exist @xmath7-manifolds @xmath1 and @xmath2 as above with @xmath35=1,\;\;\text { and } \;\;\#h_1(m_2,\z)[p^\infty]=p.\ ] ] in this section , we give several numerical examples , illustrating some features of the connection between the torsion on the homology of isospectral manifolds , and the @xmath0-module structure of the integral homology of their covering manifold . [ ex:1 ] let @xmath447 , let @xmath15 , @xmath16 be as in example [ ex : p - gassman ] . using the notation of section [ sec : y ] , let @xmath354 be the atr quartic number field generated by a root @xmath448 of @xmath449 . consider the kleinian quaternion algebra @xmath450 over @xmath354 . let @xmath364 be the principal ideal of @xmath362 generated by @xmath451 , which has norm @xmath452 , and let @xmath453 , which is well - defined up to conjugacy in @xmath351 . a fundamental domain for @xmath454 acting on the ball model of @xmath350 is displayed in figure [ pic37 ] . the group @xmath454 has a presentation by generators and relations as follows : @xmath455 let @xmath456 . there is a surjective group homomorphism @xmath457 , given by @xmath458 which gives rise to a @xmath0-covering @xmath14 of arithmetic hyperbolic @xmath7-manifolds . a direct computation , as described in section [ sec : homcomp ] , yields that @xmath459 where @xmath460 denotes the cyclic group of order @xmath406 . here , we have @xmath461 . it follows from corollary [ cor : impliesconj ] , that the multiplicity of @xmath462 ( see section [ sec : regconst ] for the definition ) in the @xmath0-module @xmath59 is odd . in this particular case , we computed directly that this multiplicity is 3 . let @xmath463 . let @xmath354 be the atr number field generated by a root @xmath448 of @xmath464 , and let @xmath282 be the kleinian quaternion algebra @xmath465 let @xmath364 be the principal ideal of @xmath362 generated by @xmath466 , of norm @xmath467 , and let @xmath468 , as in section [ sec : y ] . as in the previous example , @xmath469 is well - defined up to conjugacy in @xmath351 . a fundamental domain for @xmath469 is displayed in figure [ pic19 ] . the hyperbolic manifold @xmath470 has a @xmath0-covering @xmath14 ( which we do not give explicitly this time ) , such that @xmath471\cong c_{19 } \times c_{19},\;\;\;h_1(x / u_2,\z)[19^\infty]\cong c_{19 ^ 2}.\ ] ] here , @xmath472 . corollary [ cor : impliesconj ] implies that the multiplicity of @xmath303 in the @xmath0-module @xmath59 is even , and in fact , in this example it is 0 . moreover , unlike in the previous example , the non - isomorphic torsion can not be detected by the regulator constant of the @xmath0-module @xmath473 . let @xmath474 , and let @xmath332 be as in example [ ex : p - gassman ] . let @xmath354 be the atr number field generated by a root @xmath448 of @xmath475 , and let @xmath282 be the kleinian quaternion algebra @xmath465 let @xmath364 be the principal ideal of @xmath362 generated by @xmath476 , of norm @xmath477 , and let @xmath478 , as in section [ sec : y ] . as in the previous examples , @xmath479 is well - defined up to conjugacy in @xmath351 . a fundamental domain for @xmath479 is displayed in figure [ pic5 ] . the hyperbolic manifold @xmath480 has a @xmath0-covering @xmath14 such that @xmath481\cong c_{5}^3,\;\;\;h_1(x / u_2,\z)[5^\infty]\cong c_{5}.\ ] ] here , @xmath482 . since the quotient @xmath483 is a square , corollary [ cor : modsquares ] together with equation ( [ eq : cheegermuller ] ) imply that the regulator constants @xmath484 are trivial modulo squares , which we confirm by computing that the multiplicity of @xmath303 in @xmath59 is @xmath38 . in particular , the non - triviality of @xmath485 can not be detected in the rational homology of @xmath60 . however , by corollary [ cor : regcomparison ] , the non - triviality of @xmath486 is detected in the @xmath0-module structure of the integral homology via the equality @xmath487 . c. gordon , p. perry and d. schueth , isospectral and isoscattering manifolds : a survey of techniques and examples , geometry , spectral theory , groups , and dynamics , contemp . math . * 387 * , amer . math . ( 2005 ) , 157179 .

in this paper time machines are constructed from anti - de sitter space . one is constructed by identifying points related via boost transformations in the covering space of anti - de sitter space and it is shown that this misner - like anti - de sitter space is just the lorentzian section of the complex space constructed by li , xu , and liu in 1993 . the others are constructed by gluing an anti - de sitter space to a de sitter space , which could describe an anti - de sitter phase bubble living in a de sitter phase universe . self - consistent vacua for a massless conformally coupled scalar field are found for these time machines , whose renormalized stress - energy tensors are finite and solve the semi - classical einstein equations . the extensions to electromagnetic fields and massless neutrinos are discussed . it is argued that , in order to make the results consistent with euclidean quantization , a new renormalization procedure for quantum fields in misner - type spaces ( misner space , misner - like de sitter space , and misner - like anti - de sitter space ) is required . such a self - consistent " renormalization procedure is proposed . with this renormalization procedure , self - consistent vacua exist for massless conformally coupling scalar fields , electromagnetic fields , and massless neutrinos in these misner - type spaces . #

gr-qc9901061

in classical general relativity there are many solutions of einstein equations with closed timelike curves ( ctcs ) @xcite . however , some early calculations of vacuum polarization in spacetimes with ctcs indicated that the renormalized stress - energy tensor diverged at the cauchy horizon or the polarized hypersurfaces @xcite . hawking thus proposed the chronology protection conjecture which stated that physical laws do not allow the appearance of ctcs @xcite . but , many counter - examples to the chronology protection conjecture have been found @xcite . in particular , li and gott @xcite have found a self - consistent vacuum for a massless conformally coupled scalar field in misner space ( see also @xcite ) , which gives an example of a time machine ( i.e. a spacetime with ctcs ) at the semi - classical level ( i.e. the background spacetime is classical but the matter fields are quantized ) . of more interest , recently gott and li have discovered that ctcs could play an important role in the early universe : if we trace backward the history of time , we may enter an early epoch of ctcs , which means that there is no earliest event in time @xcite . according to the theory of quantum foam @xcite , in the early universe ( at the planck epoch ) , quantum fluctuations of spacetime should be very important and the spacetime might have a very complicated topology . very strong fluctuations in the metric of spacetime could cause the lightcones to distribute randomly , which could give rise to a sea of ctcs in the early universe . therefore , we might expect that at very early epochs , the universe could have a tangled network of ctcs . one model of the creation of the universe is the model of tunneling from nothing " @xcite . in this model the universe is supposed to be a lorentzian spacetime [ with signature @xmath1 glued to an early euclidean space [ with signature @xmath2 , thus the universe has a beginning in time ( i.e. the beginning of the lorentzian section ) . this model of tunneling from nothing " has some shortcomings ( see @xcite , and penrose in @xcite ) . contrasting with the model of tunneling from nothing " , in the model of gott and li @xcite , the universe does not need a signature change and has no beginning in time . gott and li s universe is always a lorentzian spacetime but at a very early epoch there is a loop of time . the universe could thus be its own mother and create itself . the model of gott and li has some additional interesting features : the present epoch of the universe is separated from the early ctcs epoch by a past chronology horizon . the only self - consistent solution with this geometry has pure retarded potentials , creating naturally an arrow of time in our current universe , which is consistent with our experience @xcite . thus , ctcs have potentially important applications in the early universe . anti - de sitter space is a spacetime which has ctcs everywhere . it is a solution of the vacuum einstein equation with a negative cosmological constant and has maximum symmetry @xcite . anti - de sitter space plays a very important role in theories of supergravity and superstrings @xcite . if we unfold " anti - de sitter space and go to its covering space , the ctcs disappear . however , if we identify the events related by boost transformations in the covering space of anti - de sitter space , we will get a spacetime with an infinite number of regions with ctcs and an infinite number of regions without ctcs , where the regions with ctcs and the regions without ctcs are separated by chronology horizons . the causal structure is similar to that of misner space , except that misner space has only two regions with ctcs and two regions without ctcs . for misner space , li and gott have found a self - consistent vacuum for a massless conformally coupled scalar field , which is an `` adapted '' rindler vacuum ( i.e. a rindler vacuum with multiple images ) @xcite . in this paper we will show that a self - consistent vacuum also exists for a massless conformally coupled scalar field in the misner - like anti - de sitter space constructed above , which is simply the conformal transformation of li and gott s adapted rindler vacuum . in 1993 , li , xu , and liu constructed a time machine in a space with a complex metric @xcite . in this paper we will show that the lorentzian section of that solution is just the misner - like anti - de sitter space described above . inflation theory proposes that during an early epoch the universe was in a state with an effective positive cosmological constant at the gut ( or planck ) scale @xcite , which is well - described as a de sitter phase . by transition to a zero cosmological constant ( through either quantum tunneling or classical evolution ) , the universe then enters a friedmann big bang stage . but , if there is a transition to a negative cosmological constant ( it does not seem physical theories exclude a negative cosmological constant especially since in supergravity and superstring theories anti - de sitter space is the only known self - consistent solution besides minkowski space @xcite ) , the universe could enter an anti - de sitter phase where ctcs exist . in this paper we will consider some models describing the transition between a de sitter space and an anti - de sitter space , which are obtained by gluing a de sitter space to an anti - de sitter space along a bubble wall , and we will show that self - consistent vacua for these solutions also exist . the generalization to the case of electromagnetic fields and massless neutrinos will also be considered . it is argued that , in order to be consistent with euclidean quantization , a new renormalization procedure for quantum fields in misner - type spaces is required . self - consistent " renormalization procedure is then proposed . with this renormalization procedure , self - consistent vacua exist for massless conformally coupled scalar fields , electromagnetic fields , and massless neutrinos in the misner - type spaces . anti - de sitter space @xmath3 is a hyperbola @xmath4 embedded in a five - dimensional space @xmath5 with metric @xmath6 anti - de sitter space has topology @xmath7 and is a solution of the vacuum einstein equations with a negative cosmological constant @xmath8 , which has maximum symmetry ( i.e. it has ten killing vectors ) . several coordinate systems can be defined in anti - de sitter space : 1 ) _ global static coordinates_. define @xmath9 where @xmath10 , @xmath11 , @xmath12 , and @xmath13 . then the anti - de sitter metric can be written as @xmath14 the global static coordinates ( [ coord1 ] ) cover the whole anti - de sitter space ( except the coordinate singularities at @xmath15 and @xmath16 ) , the time coordinate @xmath17 has a period of @xmath18 . 2 ) _ local static coordinates_. define @xmath19 where @xmath20 , @xmath21 , @xmath22 , and @xmath13 . then the anti - de sitter metric can be written ( in a schwarzschild like form ) as @xmath23 the local static coordinates ( [ coord2 ] ) cover only the region with @xmath24 and @xmath25 in anti - de sitter space . 3 ) _ non - stationary coordinates_. define @xmath26 where @xmath27 , @xmath11 , @xmath12 , and @xmath13 . then the anti - de sitter metric can be written ( in an open cosmological form ) as @xmath28\right\}. \label{metr3}\end{aligned}\ ] ] the non - stationary coordinates ( [ coord3 ] ) cover the region with @xmath29 and @xmath30 , but they can by extended to the region with @xmath31 and @xmath30 by the transformations @xmath32 , @xmath33 , @xmath34 , and @xmath35 . 4 ) the non - stationary coordinates can also be extended to the regions with @xmath36 . define @xmath37 where @xmath20 , @xmath11 , @xmath12 , and @xmath38 . the coordinates @xmath39 cover the region with @xmath40 , where the anti - de sitter metric can be written as @xmath41 . \label{metr3a}\end{aligned}\ ] ] anti - de sitter space is multiply connected and has ctcs everywhere . for example , in the global static coordinates ( [ coord1 ] ) , the world line with @xmath42 , @xmath43 , and @xmath44 ( which is the intersection of the hyperbola given by eq . ( [ ades ] ) with the surface with @xmath45 , @xmath46 , and @xmath47 ) is a ctc with the proper period @xmath48 . if we unfold the anti - de sitter space along the time coordinate @xmath17 in the global static coordinates ( then @xmath17 goes from @xmath49 to @xmath50 ) , we obtain the covering space of the anti - de sitter space , which is simply connected with the topology @xmath51 and does not contain ctcs anymore . ( however , there is no cauchy surface in this covering space @xcite . ) the penrose diagram of the covering space of anti - de sitter space is shown in fig . anti - de sitter space has maximum symmetry , which has one time translation killing vector , three space rotation killing vectors , and six boost killing vectors . in the local static coordinates in ( [ coord2 ] ) , @xmath52 is a boost killing vector . by the continuation @xmath53 where @xmath54 and @xmath55 , the local static coordinates can be extended to the region with @xmath56 on the hyperbola defined by eq . ( [ ades ] ) , where the boost killing vector becomes @xmath57 and the anti - de sitter metric can be written as @xmath58 in the covering space of the anti - de sitter space , if we identify all points related by boost transformations , then we obtain a _ misner - like anti - de sitter space_. with this identification , there are ctcs in the region with @xmath59 but no ctcs in the region with @xmath60 . on the boundary @xmath61 , there are closed null curves . @xmath61 is the chronology horizon . ( see fig . [ fig2 ] for the causal structure of misner - like anti - de sitter space . ) a coordinate system @xmath62 covering the region @xmath63 of anti - de sitter space can also be found , which are given by @xmath64 where @xmath65 . with these coordinates , the anti - de sitter metric can be written as @xmath66 @xmath67 is a boost killing vector . in the covering space of anti - de sitter space , if the points @xmath68 are identified with @xmath69 ( @xmath70 ) , we obtain a misner - like anti - de sitter space . in this space there are ctcs in the regions with @xmath71 , but no ctcs in the region with @xmath72 . in 1993 , li , xu , and liu @xcite constructed a complex space @xmath7 with the metric @xmath73 where @xmath74 , @xmath75 , and @xmath76 are real but @xmath77 is complex , and @xmath74 has a period @xmath78 [ i.e. @xmath79 are identified with @xmath80 where @xmath81 . they showed that this space is a solution of the vacuum complex einstein equation with a negative cosmological constant @xmath82 . here we find that , the lorentzian section of the metric in ( [ metr6 ] ) ( i.e. let @xmath83 ) is just the anti - de sitter metric in ( [ metr5 ] ) with @xmath84 ( i.e. @xmath82 ) . thus the lorentzian section of the complex space of li , xu , and liu is just the misner - like anti - de sitter space obtained from the covering space of the anti - de sitter space by identifying points related by boost transformations . usually there is no well - defined quantum field theory in a spacetime with ctcs . however the problem can be worked using hawking s euclidean quantization procedure @xcite . alternatively , in the case where a covering space exists , we can do it in the covering space with the method of images . in fact , in most cases where the renormalized energy - momentum tensor in spacetimes with ctcs has been calculated , this method has been used ( for the theoretical basis for the method of images see ref . @xcite and references cited therein ) . thus we will begin by using this method to deal with quantum field theory in anti - de sitter space and misner - like anti - de sitter space . anti - de sitter space is conformally flat . with the transformation @xmath85 , the anti - de sitter metric in eq . ( [ metr1 ] ) can be written as @xmath86,\end{aligned}\ ] ] where @xmath87 is the metric of the einstein static universe . with more transformation @xmath88 $ ] and @xmath89 $ ] , the metric of anti - de sitter space can be written as @xmath90 , \label{conf1}\end{aligned}\ ] ] where @xmath91 is just the minkowski metric , and @xmath92 is given by @xmath93 eq . ( [ conf1 ] ) demonstrates that anti - de sitter space is conformally flat . for a massless conformally coupled scalar field in a conformally flat spacetime , there exists a conformal vacuum whose hadamard function @xmath94 is related to the hadamard function @xmath95 of the corresponding vacuum of the massless conformally coupled scalar field in flat minkowski space via @xcite @xmath96 the corresponding renormalized stress - energy tensors are related via @xmath97 , \label{ener}\end{aligned}\ ] ] where @xmath98 and for scalar field we have @xmath99 and @xmath100 @xcite . [ the sign before @xmath101 is positive here because we are using signature @xmath1 . for anti - de sitter space we have @xmath102 , @xmath103 , and thus @xmath104 , @xmath105 . inserting these into eq . ( [ ener ] ) , we have @xmath106 in minkowski spacetime , for the massless conformally coupled scalar field with the minkowski vacuum , the hadamard function is @xmath107 where @xmath108 . the corresponding renormalized stress - energy tensor of the minkowski vacuum is @xmath109 inserting eq . ( [ mg ] ) into eq . ( [ had ] ) , we get the hadamard function for the massless conformally coupled scalar field in the conformal minkowski vacuum in anti - de sitter space @xmath110 \cosh \chi\cosh \chi^\prime-1-\sinh \chi\sinh \chi^\prime \cos\theta_2 } , \label{adg}\end{aligned}\ ] ] where @xmath39 are the global static coordinates of anti - de sitter space . clearly @xmath111 satisfies the periodic boundary condition @xmath112 thus it is a suitable hadamard function in anti - de sitter space which has ctcs everywhere . inserting eq . ( [ me ] ) into eq . ( [ ener1 ] ) , we get the renormalized stress - energy tensor for the massless and conformally coupled scalar field in the conformal minkowski vacuum in anti - de sitter space @xmath113 which is the same as that for de sitter space with radius @xmath114 . if we insert the energy - momentum tensor in eq . ( [ ener2 ] ) into the semiclassical einstein equations @xmath115 and recall that for anti - de sitter space we have @xmath116 , we find that the semiclassical einstein equations are satisfied if and only if @xmath117 if @xmath118 , the two solutions to eq . ( [ eina ] ) are @xmath119 , which corresponds to minkowski space ; and @xmath120 , which corresponds to a de sitter space with radius @xmath121 . thus , if the bare cosmological constant @xmath118 , there is no self - consistent anti - de sitter space ( though there are a self - consistent de sitter space with radius @xmath122 @xcite and a self - consistent minkowski space ) . if @xmath123 , the two solutions to eq . ( [ eina ] ) are @xmath124 which corresponds to an anti - de sitter space with radius @xmath114 , and @xmath125 which corresponds to a de sitter space with radius @xmath126 . ( if @xmath127 , the two solutions both correspond to de sitter spaces @xcite . ) it is interesting to note that if @xmath123 there are two self - consistent spaces , one of which is an anti - de sitter space , and the other is a de sitter space . ( [ eina ] ) tells us that , if @xmath123 and @xmath128 , we have @xmath129 , which implies that the self - consistent de sitter space supported by a negative cosmological constant has a sub - planckian radius . ( see sec . v for further discussion . ) for the misner - like anti - de sitter space which is obtained from the covering space of anti - de sitter space by identifying points related by boost transformations , as in the case of misner - like de sitter space @xcite , it is easily to show that the adapted conformal minkowski vacuum is not a self - consistent quantum state for the massless conformally coupled scalar field . the renormalized stress - energy tensor of the adapted conformal minkowski vacuum diverges as the chronology horizon is approached . but , as in the cases of misner space @xcite and misner - like de sitter space @xcite , we can show that an adapted conformal rindler vacuum is a self - consistent vacuum state for a massless conformally coupled scalar field in misner - like anti - de sitter space . to do so , it will be more convenient to write the anti - de sitter metric in the local static coordinates [ eqs . ( [ coord2 ] ) and ( [ metr2 ] ) ] and the minkowski metric in rindler coordinates @xmath130 where the rindler coordinates @xmath131 are defined by @xmath132 where @xmath62 are the cartesian coordinates in minkowski spacetime . with the transformation @xmath133 where @xmath134 are the local static coordinates in eq . ( [ coord2 ] ) , the anti - de sitter metric ( [ metr2 ] ) can be written as @xmath135 here @xmath92 is @xmath136 ^ 2.\end{aligned}\ ] ] with the conformal relation between anti - de sitter space and rindler space given by eqs . ( [ rin3 ] ) and ( [ adsr ] ) , the time coordinate @xmath17 of the anti - de sitter space in local static coordinates is mapped to the rindler time coordinate @xmath137 . construct a misner space with all @xmath138 ( @xmath139 ) identified , then the map given by eq . ( [ rin3 ] ) gives rise to a misner - like anti - de sitter space with all @xmath140 ( @xmath139 ) identified , where @xmath141 eqs . ( [ rin3])-([t0 ] ) give a natural conformal map between misner space and misner - like anti - de sitter space . for a massless conformally coupled scalar field in misner space , the hadamard function for the adapted rindler vacuum is @xcite @xmath142 } , \label{ring}\end{aligned}\ ] ] where @xmath143 is defined by @xmath144 the corresponding renormalized stress - energy tensor of the adapted rindler vacuum is @xcite @xmath145 \left(\begin{array}{cccc } -3&0&0&0\\ 0&1&0&0\\ 0&0&1&0\\ 0&0&0&1 \end{array } \right ) , \label{e69}\end{aligned}\ ] ] which is expressed in the rindler coordinates . we see that @xmath146 ( also @xmath147 ) diverges at the chronology horizon where @xmath148 unless @xmath149 . if @xmath149 , however , we have @xmath150 throughout the misner space ( though rindler coordinates cover only a quadrant in misner space , the results can be analytically extended to the whole misner space @xcite where @xmath151 is also zero , see @xcite for further discussion ) . inserting eq . ( [ ring ] ) into eq . ( [ had ] ) , we obtain the hadamard function for the adapted conformal rindler vacuum of the massless and conformally coupled scalar field in misner - like anti - de sitter space @xmath152 } , \label{e79}\end{aligned}\ ] ] where @xmath143 is written in @xmath134 as @xmath153\right\}. \label{e80}\end{aligned}\ ] ] clearly the hadamard function in eq . ( [ e79 ] ) satisfies the periodic boundary condition @xmath154 where @xmath139 , thus it defines a reasonable quantum state in the misner - like anti - de sitter space . inserting eq . ( [ e69 ] ) into eq . ( [ ener1 ] ) , we obtain the renormalized stress - energy tensor of the adapted conformal rindler vacuum of the conformally coupled scalar field in misner - like anti - de sitter space @xmath155 \left(\begin{array}{cccc } -3&0&0&0\\ 0&1&0&0\\ 0&0&1&0\\ 0&0&0&1 \end{array } \right)-{1\over960\pi^2 \alpha^4}\delta_{\mu}^{~\nu } , \label{e73}\end{aligned}\ ] ] where the coordinate system is the local static coordinate system @xmath134 . again , @xmath156 ( also @xmath157 ) diverges at the chronology horizon @xmath158 , unless @xmath159 . but , if @xmath160 we get a renormalized stress - energy tensor which is regular throughout the misner - like anti - de sitter space @xmath161 which is exactly the same as that of de sitter space and anti - de sitter space . [ though the local coordinates @xmath134 cover only the region with @xmath24 in anti - de sitter space , the results can be easily extended to the whole misner - like anti - de sitter space ( where @xmath162 is finite and given by eq . ( [ e75 ] ) ) , as in the cases of misner space and misner - like de sitter space @xcite . ] similarly , the misner - like anti - de sitter space solves the semiclassical einstein equation with a negative cosmological constant @xmath163 and the energy - momentum tensor in eq . ( [ e75 ] ) ( and thus is self - consistent ) if @xmath164 . phase transitions play important roles in the evolution of the early universe . with phase transitions various bubbles could form ; inside and outside the bubbles the spacetimes have different stress - energy tensors and thus are described by different spacetime metrics @xcite . the inside and outside of a bubble are separated by a wall a spacetime structure which can be approximately treated as a three - dimensional hypersurface . usually it is assumed that the outside of the bubble is in a state dominated by a positive cosmological constant at gut ( or planckian ) scale . thus the spacetime metric outside the bubble is well approximated with a de sitter metric . inside the bubble , the cosmological constant could be zero and the stress - energy tensor could be zero thus the spacetime inside would be minkowskian which is a model of inflation decaying through a first order phase transition in the old inflation theory @xcite ; or , inside the bubble the cosmological constant could also be positive and at gut ( or planckian ) scale so the spacetime inside the bubble is still inflating , but after a while the cosmological constant falls off a plateau and evolves classically to zero and the universe inside the bubble enters a hot big bang phase which is a model of transition from inflation to an open big bang cosmology through a second order phase transition @xcite . but , either via the first order phase transition or the second order phase transition , as another alternative , the inside of the bubble could become dominated by a negative cosmological constant instead and thus the spacetime inside the bubble would be described with an anti - de sitter metric . in this paper we are interested in this case since anti - de sitter space has ctcs . in this paper we will discuss bubbles that are pre - existing rather than ones that form by quantum tunneling . we thus only consider how to glue a de sitter space and an anti - de sitter space together at a boundary ( i.e. at a wall ) , and we investigate the causal structure of the spacetime so obtained . the conditions for two spacetimes to be glued together along a wall ( so that the einstein equations are satisfied at the wall ) are @xcite : 1 ) the metrics on the wall induced from the spacetimes at the two sides agree ; 2 ) the surface stress - energy tensor of the wall defined by @xmath165 should satisfy ( in the gaussian normal coordinates near the wall ) @xmath166 where @xmath167 is the normal vector of the hypersurface @xmath168 of the wall ( @xmath169 if the @xmath168 is timelike ; @xmath170 if the wall is spacelike . regarding @xmath168 as a hypersurface embedded in either the spacetime inside it , or the spacetime outside it , by the first condition , @xmath168 should be either timelike or spacelike in both . here we do not consider the case of a null wall ) , and @xmath171\equiv k_{ab}^+ -k_{ab}^- \label{join1}\end{aligned}\ ] ] is the difference of the extrinsic curvature of @xmath168 embedded in the spacetimes at the two sides of the wall @xcite , and @xmath172 . the definition for the extrinsic curvature @xmath173 is @xmath174 where we have treated @xmath167 as an vector field in the neighborhood of @xmath168 extended from the normal vector defined only on @xmath168 . in the gaussian normal coordinates of @xmath168 , the components of @xmath173 can be written as @xmath175 ( here we use the definition of the extrinsic curvature @xmath173 with an opposite sign as that used in @xcite . ) the evolution of the wall is governed by @xmath176=0 , \label{evol}\end{aligned}\ ] ] where `` @xmath177 '' denotes the covariant derivative associated with the three - metric @xmath178 on @xmath168 . ( here the greek letters @xmath179 label vectors and tensors in the four - dimensional spacetime , the latin letters @xmath180 label vectors and tensors in the three - dimensional space @xmath168 . ) de sitter space is a hyperbola @xmath181 embedded in a five - dimensional space @xmath5 with the metric @xmath182 de sitter space is a solution of the vacuum einstein equations with a positive cosmological constant @xmath183 , which has maximal symmetry ( it has six space rotation killing vectors and four boost killing vectors ) . there are various coordinate systems for de sitter space @xcite . for our purpose here , the following two coordinate systems are convenient : 1 ) . define @xmath184 where @xmath185 , @xmath11 , @xmath12 , and @xmath13 . the coordinates @xmath39 in ( [ dco2 ] ) cover the region on the hyperbola ( [ des ] ) with @xmath29 and @xmath186 , where the de sitter metric can be written as @xmath187\right\}. \label{dmetr2}\end{aligned}\ ] ] the section @xmath188 in de sitter space is an open ( i.e. negatively curved ) , homogeneous , and isotropic space . the metric in ( [ dmetr2 ] ) could describe an open inflation @xcite . the coordinates @xmath39 in ( [ dco2 ] ) can be extended to the region with @xmath31 and @xmath186 by the reflection @xmath189 and @xmath33 , @xmath190 , and @xmath35 . 2 ) the coordinates in ( [ dco2 ] ) could be extended to the region with @xmath191 . define @xmath192 where @xmath20 , @xmath193 , @xmath12 , and @xmath13 , then the de sitter metric can be written as @xmath194 . \label{dmetr1}\end{aligned}\ ] ] the coordinates @xmath39 cover the region with @xmath191 in the de sitter space as a hyperbola given by eq . ( [ des ] ) . the penrose diagram of de sitter space is shown in fig . [ fig3 ] . by gluing anti - de sitter space onto de sitter space we can obtain various spacetimes with ctcs . these spacetimes with ctcs differ from the anti - de sitter space with ctcs by the fact that in these glued spacetimes there are regions without ctcs which are separated from the regions with ctcs by chronology horizons , while in the usual anti - de sitter space ctcs exist everywhere . here we only show one typical example , which is obtained by gluing an anti - de sitter space onto a de sitter along a timelike hypersurface ( @xmath195 ) . this spacetime could describe a bubble of anti - de sitter space existing for all time in an eternal de sitter space . consider a de sitter space as a hyperbola described by eq . ( [ des ] ) in the embedding space with the metric in eq . ( [ des1 ] ) . cut this de sitter space along the hypersurface @xmath196 with @xmath197 and throw away the part with @xmath198 . denote the part of the de sitter space with @xmath199 as @xmath200 . then we have a de sitter space with a boundary @xmath196 at @xmath201 . suppose @xmath202 , then @xmath196 is timelike . in such a case , the hypersurface @xmath196 is a three - dimensional timelike hyperbola with @xmath203 . with the coordinates in ( [ dco1 ] ) , @xmath196 is at @xmath204 . the metric @xmath205 on @xmath196 induced from the de sitter metric is @xmath206 . \label{hab1}\end{aligned}\ ] ] the normal vector of @xmath168 is @xmath207 ( @xmath169 ) . the extrinsic curvature @xmath208 of @xmath196 is @xmath209 consider an anti - de sitter space as the hyperbola given by eq . ( [ ades ] ) in the embedding space with the metric in eq . ( [ metr0 ] ) . cut the anti - de sitter space along the hypersurface @xmath210 with @xmath211 , throw away the part with @xmath212 . we denote the anti - de sitter space with @xmath213 as @xmath214 . then we have an anti - de sitter space with @xmath213 and a boundary @xmath210 at @xmath215 . suppose @xmath216 , then @xmath210 is timelike . in such a case , @xmath210 is a three - dimensional timelike hyperbola with @xmath217 . with the coordinates @xmath39 in ( [ coord3a ] ) , the hypersurface @xmath210 is at @xmath218 . the metric @xmath205 on @xmath210 induced from the anti - de sitter metric is @xmath219 . \label{hab3}\end{aligned}\ ] ] the normal vector of @xmath210 is @xmath220 ( @xmath169 ) . the extrinsic curvature @xmath221 of @xmath210 is @xmath222 to glue the anti - de sitter space and the de sitter space together , let us identify @xmath196 with @xmath210 by identifying their coordinates @xmath223 . the spacetime so obtained is schematically shown in fig . [ fig4 ] . in the sections with @xmath224 in de sitter space and anti - de sitter space , an @xmath225 is glued together with an @xmath226 at a cross - section @xmath227 . from eqs . ( [ hab1 ] ) and ( [ hab3 ] ) , the metric @xmath205 on @xmath228 induced from de sitter space and anti - de sitter space agree if and only if @xmath229 by gluing two spacetimes along a hypersurface , usually a surface stress - energy tensor is induced . the surface stress - energy tensor induced on the hypersurface @xmath168 is determined by the difference in the extrinsic curvature of @xmath196 which is embedded in the de sitter space and the extrinsic curvature of @xmath210 which is embedded in the anti - de sitter space through eq . ( [ join ] ) with @xmath169 . by inserting the extrinsic curvatures derived above into eqs . ( [ join1 ] ) and ( [ join ] ) , we obtain the surface stress - energy tensor for @xmath168 @xmath230 where @xmath231 , the metric @xmath205 on the timelike @xmath168 is given by @xmath232 . \label{sab11}\end{aligned}\ ] ] the surface stress - energy tensor given by eq . ( [ sab1 ] ) is like a positive three - dimensional cosmological constant . the timelike hypersurface @xmath168 with the metric ( [ sab11 ] ) is a three - dimensional de sitter space . the penrose diagram of the spacetime obtained by gluing @xmath200 with @xmath214 along the timelike @xmath168 is shown in fig . there are ctcs in the region of @xmath200 with @xmath233 and the whole @xmath214 . but there are no ctcs in the region of @xmath200 with @xmath234 . the null hypersurface @xmath235 is the chronology horizon which separates the region with ctcs from that without ctcs . ( see fig . [ fig5 ] . ) mathematically , a de sitter space can also be glued to an anti - de sitter space along a spacelike hypersurface , the resultant spacetime has similar properties as the example described above . the surface stress - energy tensor induced on the spacelike hypersurface is like a negative cosmological constant in a three - dimensional euclidean space . the spacelike hypersurface is a three - dimensional hyperbola @xmath226 . the causal structure of the spacetime obtained by gluing a de sitter space to an anti - de sitter space along a spacelike hypersurface is the same as that of the spacetime obtained by gluing a de sitter space to an anti - de sitter space along a timelike hypersurface . there are ctcs in the region of @xmath200 with @xmath233 and the whole @xmath214 . but there are no ctcs in the region of @xmath200 with @xmath234 . the null hypersurface @xmath235 is the chronology horizon which separates the region with ctcs from that without ctcs . for both cases , we have @xmath236 and @xmath237 , thus the evolution equation ( [ evol ] ) is satisfied automatically . from eqs . ( [ sab1 ] ) we see that if @xmath238 , i.e. if @xmath168 becomes null , we have @xmath239 . though the coordinates @xmath39 are singular at the null @xmath168 with @xmath240 and @xmath241 , we can show that as @xmath242 , the scalars @xmath243 and @xmath244 also vanish as @xmath242 . ( these conclusions also hold if @xmath168 is spacelike . ) however , since the metric @xmath205 on a null hypersurface is degenerate , the formalism discussed above can not be used to the junction at null hypersurfaces . thus the case of null @xmath168 requires more discussion . here we do not discuss this complicated but interesting issue . ( for detail discussions on the junction conditions at null hypersurfaces , see references @xcite ) . as discussed in sec . iii , an anti - de sitter space with radius @xmath114 is self - consistent ( i.e. the semi - classical einstein equations ( [ ein ] ) are satisfied with @xmath245 being the renormalized stress - energy tensor of vacuum polarization ) if the negative cosmological constant is @xmath246 while a de sitter space with radius @xmath247 is self - consistent if the cosmological constant is @xcite @xmath248 @xmath249 is always negative . @xmath250 could be either positive or negative , depending on the value of @xmath247 . from eq . ( [ solb ] ) , if @xmath251 , @xmath250 is positive ; if @xmath252 , @xmath250 is negative . thus , interestingly , _ a de sitter space with sub - planckian radius could be self - consistent only if the bare cosmological constant is negative_. from eqs . ( [ sola ] ) and ( [ solb ] ) , we see that if @xmath252 , @xmath253 if and only if @xmath254 this together with eq . ( [ ab ] ) gives a self - consistent spacetime which is obtained by gluing a de sitter space to an anti - de sitter space with _ a unique negative bare cosmological constant throughout_. this could be realized since for a negative cosmological constant there are two self - consistent solutions of the semi - classical einstein equations , one is anti - de sitter space , the other is de sitter space , and these two could be glued together as discussed in sec . the spacetime obtained by gluing a de sitter space with an anti - de sitter space as discussed in the last section , is a self - consistent solution of the semi - classical einstein equations if eqs . ( [ sola ] ) and ( [ solb ] ) are satisfied and on the wall separating the de sitter region and the anti - de sitter region there is a surface stress - energy tensor given by eqs . ( [ sab1 ] ) . since the hadamard function is continuous across the wall ( the hadamard function does not contain any derivatives of the metric ) , the wall does not introduce any additional vacuum polarization effects . the above discussions of self - consistent solutions are for a massless conformally coupled scalar field in de sitter / anti - de sitter spaces . the results can be easily extended to other matter fields . if their vacua are invariant under de sitter / anti - de sitter transformations , it could be expected that their renormalized stress - energy tensor should have the form of @xmath255 . if there are many matter fields with their vacua being invariant under de sitter / anti - de sitter transformations , the renormalized stress - energy tensor could be written as @xmath256 where @xmath257 is a dimensionless number determined by the number and spins of matter fields existing in the de sitter / anti - de sitter space with radius @xmath258 ( @xmath259 for anti - de sitter space , @xmath260 for de sitter space , in practice @xmath261 ) . correspondingly , with the appearance of many matter fields with de - sitter / anti - de sitter invariant vacua , eq . ( [ sola ] ) and ( [ solb ] ) should be replaced by @xmath262 and @xmath263 @xmath249 is always negative . @xmath250 is negative if @xmath264 , positive if @xmath265 . and , @xmath253 if @xmath266 if eqs . ( [ ab ] ) and ( [ solc1 ] ) are satisfied simultaneously , the spacetime obtained by gluing the de sitter space to the anti - de sitter space is self - consistent and has a unique negative bare cosmological constant through both de sitter and anti - de sitter regions . such a solution is very interesting . the semi - classical einstein equations with a negative bare cosmological constant can have two self - consistent solutions , one being an anti - de sitter space , the other being a de sitter space . these two spacetimes could transit from one to the other across a bubble wall ( through whatever quantum processes ) without changing the cosmological constant . but , for misner - type spaces ( misner space , misner - like de sitter space , or misner - like anti - de sitter space ) , the situation for matter fields other than the massless and conformally coupled scalar field is more complicated since the self - consistent vacua are not lorentzian or de sitter / anti - de sitter invariant . to see this , let us consider the electromagnetic field in misner space . as an alternative to the method of images , euclidean quantization is another more powerful tool for dealing with quantum field theory in an acausal space @xcite . the euclidean method provides a convenient bridge between the conical space around a cosmic string @xcite and misner space , which could conveniently translate the results of quantum field theory in a conical space to that in misner space ( since a conical space and misner space have the same euclidean section the euclidean conical space ) . in @xcite , from the well - known renormalized stress - energy tensor of the conformally coupled scalar field in the conical space around a cosmic string , using the euclidean method ( first translate the results in the conical space around the string to that in the euclidean conical space , then translate these results to that in misner space ) , cassidy has successfully predicted that there should be a quantum state with vanishing renormalized stress - energy tensor in misner space when the boost parameter @xmath267 is @xmath78 , which corresponds to no cosmic string . with li and gott s independent discovery of the self - consistent vacuum state ( an adapted rindler vacuum ) for a conformally coupled scalar field in misner space @xcite , cassidy s prediction has been confirmed . and , the euclidean quantization procedure gives a beautiful geometrical explanation for the self - consistent vacua in misner - type spaces namely that when @xmath149 the corresponding euclidean section is flat with no conical singularity and thus has @xmath268 throughout @xcite . recently , with the method of euclidean quantization , li and gott have found a self - consistent vacuum for a model of inflation in the kaluza - klein theory , and found a relation between the fine structure constant and the inflationary energy scale which is consistent with the energy scale usually talked about in inflation and gut theory @xcite . thus , we adopt the method of euclidean quantization . the euclidean section of misner space is a euclidean conical space with metric @xmath269 where @xmath270 is identified with @xmath271 ( @xmath272 ) . if we make the continuation @xmath273 and @xmath274 , we obtain misner space @xcite . ( on the other hand , if we make the continuation @xmath275 , we obtain the spacetime of a cosmic string . ) the quantum field effects in the conical space have been investigated by many people ( see @xcite and references therein ) . thus , for the electromagnetic field in the euclidean conical space , the renormalized stress - energy tensor of vacuum polarization is @xmath276 \left[\left(2\pi\over\phi_0\right)^2 + 11\right ] \left(\begin{array}{cccc } 1 & 0 & 0 & 0\\ 0&-3 & 0 & 0\\ 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 1 \end{array } \right ) , \label{tab1}\end{aligned}\ ] ] where the cylindrical coordinates @xmath270 are used . clearly , if @xmath277 , the renormalized stress - energy tensor of the electromagnetic field is zero and thus the semi - classical vacuum einstein equations ( i.e. the einstein equations with the renormalized stress - energy tensor being that of vacuum polarization ) are satisfied . this is what we expect since for @xmath277 the euclidean conical space becomes the regular flat euclidean space @xmath51 . but , when the euclidean conical space is continued to misner space by @xmath273 and @xmath274 in eq . ( [ tab1 ] ) , the renormalized stress - energy tensor becomes @xmath278 \left[-\left(2\pi\over\eta_0\right)^2 + 11\right ] \left(\begin{array}{cccc } -3 & 0 & 0 & 0\\ 0 & 1 & 0 & 0\\ 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 1 \end{array } \right ) , \label{tab2}\end{aligned}\ ] ] where the rindler coordinates @xmath131 are used . we see that , unlike the case for a massless conformally coupled scalar field , even if @xmath149 , the renormalized stress - energy tensor given by eq . ( [ tab2 ] ) is nonzero . in fact , @xmath279 in eq . ( [ tab2 ] ) diverges at @xmath148 unless @xmath280 . if @xmath280 , the @xmath279 given by eq . ( [ tab2 ] ) is zero and thus it would solve the semi - classical einstein equations for this locally flat space . but , if this were a self - consistent solution , it would be surprising that the self - consistent vacua of different matter fields have different values of @xmath267 . recall that , for the massless conformally coupled scalar field the self - consistent vacuum has @xmath149 ( @xcite , and sec . iii of this paper ) . and , for the case with @xmath149 , there is an excellent geometric explanation : the corresponding euclidean section with @xmath277 is the regular euclidean space @xmath51 without conical singularity @xcite ; while for @xmath280 , we can not find any simple geometric explanation . another surprise would be that , for the case of the electromagnetic field , the euclidean section with @xmath277 is a self - consistent solution of the semi - classical euclidean einstein equations ; but , by adopting the continuation @xmath281 and @xmath274 , the resultant misner space with @xmath149 is _ not _ a self - consistent solution of the semi - classical einstein equations . this implies that the semi - classical einstein equations are broken during this continuation : that is , with this particular continuation the solutions of the semi - classical einstein equations can not be translated between the lorentzian and euclidean sections . this raises a question as to the procedure of euclidean quantization : is the euclidean quantization still valid for the electromagnetic field in misner space ? a fundamental requirement for euclidean quantization should be that _ during the continuation between the lorentzian section and the euclidean section , the einstein equations should be preserved _ ( otherwise the euclidean quantization loses its significance ) . note that if we first take @xmath277 in the euclidean section of misner space , @xmath279 will be zero in the euclidean section according to eq . ( [ tab1 ] ) . then , if we use the continuation @xmath273 and @xmath282 ( in this case @xmath283 ) @xmath284 ( in this case @xmath285 ) when going from the euclidean conical space to misner space , naturally we expect `` zero '' should be continued to `` zero '' in the renormalized stress - energy tensor . then we should expect that the @xmath279 in the misner space should also be zero , which would be conflict with the results obtained above by going to the lorentzian section first and then setting @xmath149 . what causes this non - self - consistency ? if we check the continuation procedure carefully , we find that the problem is at renormalization . in the euclidean section , the renormalization is equivalent to subtracting from the original non - renormalized hadamard function @xmath94 a reference hadamard function @xmath286 with @xmath277 , i.e. the regularized hadamard function is @xmath287 where @xmath288 if we make the continuation with @xmath273 and @xmath274 but keep @xmath277 unchanged in @xmath289 , @xmath289 would be continued as @xmath290 then the obtained @xmath289 in the lorentzian section is the usual hadamard function for the minkowski vacuum ( see @xcite ) , and the corresponding renormalized stress - energy tensor ( which is obtained by operating on @xmath291 with a differential operator @xcite ) is just given by eq . ( [ tab2 ] ) . but , in this continuation procedure , it would be surprising why in both @xmath291 and @xmath292 the parameter @xmath282 was changed to @xmath293 but in @xmath289 the parameter @xmath277 was unchanged , thus this procedure is not self - consistent . the result of this non - self - consistent procedure is that the semi - classical einstein equations are broken , as mentioned above . for a flat euclidean space @xmath51 with cartesian coordinates @xmath294 and metric @xmath295 , if we go to the lorentzian section by the continuation @xmath296 , the euclidean space @xmath51 is naturally continued to the usual simply connected minkowski space with cartesian coordinates @xmath62 and metric @xmath297 . since in the euclidean section @xmath298 goes from @xmath49 to @xmath50 , naturally in the lorentzian section @xmath17 also goes from @xmath49 to @xmath50 . with this continuation , the hadamard function for the euclidean vacuum in the euclidean @xmath51 space which is the reference hadamard function for renormalization in the euclidean section is continued to the hadamard function for the lorentzian vacuum in minkowski space which is the reference hadamard function for renormalization in minkowski space . but in our case , we start with a flat euclidean space @xmath51 with cylindrical coordinates @xmath270 and metric @xmath299 where @xmath300 has a period of @xmath78 . when we go to the lorentzian section by the continuation @xmath273 , very naturally @xmath137 has also a period of @xmath78 . then the obtained lorentzian section is a misner space with the boost parameter @xmath149 . with this continuation , the hadamard function for the euclidean vacuum in the euclidean section which is the reference hadamard function for renormalization in the euclidean section is continued to the hadamard function for li and gott s adapted rindler vacuum with @xmath149 very naturally which should also be the reference hadamard function for renormalization in misner space . we start with the reference space for a euclidean conical space which is the flat euclidean space without conical singularity , and end with the corresponding the reference space for misner space which is the misner space with @xmath149 . this procedure is very natural . therefore , here we propose the following _ self - consistent renormalization procedure _ for quantum fields in misner - type spaces : when we make the continuation from the euclidean section to the lorentzian section by @xmath273 and @xmath301 , we should also make the continuation @xmath302 ( see fig . [ fig6 ] ) , then the reference hadamard function is continued to be the hadamard function in the lorentzian section ( misner space ) with @xmath149 . with this self - consistent renormalization procedure , instead of eq . ( [ continue1 ] ) , the reference hadamard function is continued to be @xmath303 making all the problems mentioned above go away . with this self - consistent renormalization procedure , instead of eq . ( [ tab2 ] ) , the renormalized stress - energy tensor of electromagnetic field in misner space becomes ( substituting @xmath293 for @xmath282 and @xmath304 for @xmath78 in eq . ( [ tab1 ] ) ) @xmath305 \left[\left(2\pi\over\eta_0\right)^2 + 11\right ] \left(\begin{array}{cccc } -3 & 0 & 0 & 0\\ 0 & 1 & 0 & 0\\ 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 1 \end{array } \right ) , \label{tab3}\end{aligned}\ ] ] in the rindler coordinates @xmath131 . clearly , if @xmath149 , we have @xmath268 and thus the semi - classical einstein equations are satisfied . the situation is similar for massless neutrinos in misner space . if we make the continuation @xmath306 , @xmath274 , @xmath302 , using the self - consistent renormalization procedure , we obtain the renormalized stress - energy tensor of vacuum polarization for massless neutrinos in misner space @xmath307 \left[\left(2\pi\over\eta_0\right)^2 + 17\right ] \left(\begin{array}{cccc } -3 & 0 & 0 & 0\\ 0 & 1 & 0 & 0\\ 0 & 0 & 1 & 0\\ 0 & 0 & 0 & 1 \end{array } \right ) , \label{tab4}\end{aligned}\ ] ] which is also zero for @xmath149 . thus , with the new renormalization procedure , we obtain self - consistent vacua for electromagnetic fields and massless neutrinos in misner space . it is very easy to check that this new renormalization procedure does not change the results for a massless conformally coupled scalar field in misner space . the euclidean result for a conical space is @xmath308 \left(\begin{array}{cccc } -3&0&0&0\\ 0&1&0&0\\ 0&0&1&0\\ 0&0&0&1 \end{array } \right ) , \label{e69a}\end{aligned}\ ] ] substituting @xmath293 for @xmath282 and @xmath304 for @xmath78 into eq . ( [ e69a ] ) gives eq . ( [ e69 ] ) just as before . thus , li and gott s adapted rindler vacuum @xcite is still a self - consistent vacuum for the massless conformally coupled scalar field in misner space with @xmath149 . these results can be easily transplanted to misner - like de sitter space and misner - like anti - de sitter space , since these fields ( massless and conformally coupled scalar fields , electromagnetic fields , and massless neutrinos ) are conformally invariant and de sitter space and anti - de sitter space are conformally flat . the results are : if the boost period is @xmath309 ( where @xmath258 is the radius of the de sitter space or anti - de sitter space ) , the renormalized stress - energy tensors are finite and given by eq . ( [ star ] ) with @xmath310 for one neutrino field and @xmath311 for the electromagnetic field . with eqs . ( [ sold ] ) and ( [ sole ] ) , self - consistent solutions of the semiclassical einstein equations with cosmological constant can be found . thus , with our new renormalization method , various self - consistent vacuum states in misner - type spaces with ctcs can be found . the chronology protection conjecture @xcite was originally based on the fact that for a massless and conformally coupled scalar field with the adapted minkowski vacuum in the misner space the renormalized stress - energy tensor of vacuum polarization diverges at the chronology horizon @xcite . after the appearance of many counter - examples @xcite , cassidy and hawking @xcite demonstrated that the back - reaction of vacuum polarization does not enforce chronology protection . the results in this paper ( and @xcite ) support this demonstration . however , cassidy and hawking @xcite turned to the proposition that the effective action of matter fields in spacetimes with ctcs always diverges at the chronology horizon and that it is this that enforces chronology protection . the effective action plays an important role in euclidean quantum gravity , which gives the probability for the creation of spacetime through quantum tunneling . consider a simple example of a massless conformally coupled scalar field in misner space . the euclidean effective lagrangian density is @xcite @xmath312\left[\left(2\pi\over\phi_0\right)^2 + 11\right ] . \label{lag}\end{aligned}\ ] ] the euclidean effective action is obtained by integrating the euclidean effective lagrangian density over a suitable volume of the euclidean section . clearly , if @xmath313 , the effective euclidean lagrangian diverges at @xmath148 ( thus the euclidean effective action diverges if the domain for integration includes the conical singularity at @xmath148 ) . but , if @xmath277 , we have @xmath314 , which is not surprising since when @xmath277 the conical singularity at @xmath148 disappears and the euclidean section becomes the usual regular flat @xmath51 space . the euclidean effective action is also zero since the euclidean effective lagrangian is zero everywhere . by the continuation @xmath281 , @xmath274 , and with the self - consistent renormalization procedure outlined above , the reference hadamard function is continued to be that with @xmath149 in misner space ( which means that the ratio @xmath315 in eq . ( [ lag ] ) is changed to @xmath316 since @xmath274 and @xmath302 ) , then the effective lagrangian for a massless conformally coupled scalar field in misner space is @xmath317\left[\left(2\pi\over\eta_0\right)^2 + 11\right ] , \label{lag1}\end{aligned}\ ] ] which is zero everywhere if @xmath149 . thus the lorentzian effective action is also zero for @xmath149 . this shows that , for the self - consistent vacuum in misner space , the effective action is zero . it can be expected that this result can be extended to other quantum fields or other spacetimes with ctcs , where the ( euclidean or lorentzian ) effective lagrangian would be finite throughout the space [ and thus the ( euclidean or lorentzian ) effective action would also be finite ] for the self - consistent vacua . thus cassidy and hawking s argument that the effective action ( or equivalently the entropy ) enforce chronology protection is questionable . from the covering space of anti - de sitter space , a misner - like anti - de sitter space can be constructed . this misner - like anti - de sitter space has ctcs but the regions with ctcs are separated from the regions without ctcs by chronology horizons . in the appropriate coordinates , this misner - like anti - de sitter space is just the lorentzian section of the complex space with ctcs constructed by li , xu , and liu in 1993 @xcite . for a massless conformally coupled scalar field in this space , a self - consistent vacuum is found , whose renormalized stress - energy tensor is like that of a positive cosmological constant which when added to an appropriate negative bare cosmological constant can self - consistently solve the semi - classical einstein equations . by gluing a de sitter space to an anti - de sitter space along a bubble wall , another new spacetime with ctcs is obtained . this spacetime could describe the transition between de sitter space and anti - de sitter space . in this spacetime , the region with ctcs and the region without ctcs are separated via chronology horizons . for the de sitter / anti - de sitter invariant vacua in these spacetimes , the renormalized stress - energy tensors are like positive cosmological constants . a self - consistent solution can be obtained if there is a single negative bare cosmological constant in the two regions with the renormalized stress - energy tensor of vacuum polarization adding different positive cosmological constants to the two sides of the bubble wall so that the effective cosmological constant ( bare @xmath318 renormalized ) is positive on one side and negative on the other . on the hypersurface separating de sitter space from anti - de sitter space , in order that the einstein equations are satisfied , a surface stress - energy tensor should be induced . if the hypersurface is timelike , the surface stress - energy tensor is like that of a three - dimensional positive cosmological constant . the self - consistent solutions of the semi - classical einstein equations with cosmological constant and the renormalized stress - energy tensor of vacuum polarization in de sitter / anti - de sitter space are investigated . if the bare cosmological constant is positive , there are two self - consistent solutions , both of them are de sitter spaces . if the bare cosmological constant is zero then there are two self - consistent solutions one is minkowski space and the other is a planckian scale de sitter space . if the bare cosmological constant is negative , there are also two self - consistent solutions , one of them is an anti - de sitter space , but the other is a sub - planckian scale de sitter space . and , at the sub - planckian scale , self - consistent solutions ( either de sitter space or anti - de sitter space ) exist only for a bare negative cosmological constant . the generalization to electromagnetic fields and massless neutrinos in spacetimes with ctcs are discussed . it is argued that , for misner - type spacetimes , in order that the semi - classical einstein equations are preserved under continuation between the euclidean and lorentzian sections , a new renormalization procedure should be introduced . we have proposed such a self - consistent renormalization procedure , with which self - consistent vacua for electromagnetic fields and neutrinos are found . k. gdel , rev . mod * 21 * , 447 ( 1949 ) . a. h. taub , ann . math . * 53 * , 472 ( 1951 ) . e. t. newman , l. tamburino , and t. j. unti , j. math . * 4 * , 915 ( 1963 ) . c. w. misner , in _ relativity theory and astrophysics i : relativity and cosmology _ , edited by j. ehlers , lectures in applied mathematics , vol . 8 ( american mathematical society , providence , 1967 ) , p. 160 . m. s. morris , k. s. thorne , and u. yurtsever , phys . lett . * 61 * , 1446 ( 1988 ) . j. r. gott , phys . lett . * 66 * , 1126 ( 1991 ) . w. a. hiscock and d. a. konkowski , phys . d * 26 * , 1225 ( 1982 ) . v. p. frolov , phys . d * 43 * , 3878 ( 1991 ) . kim and k. s. thorne , phys . d * 43 * , 3929 ( 1991 ) . j. d. e. grant , phys . rev . d * 47 * , 2388 ( 1993 ) . s. w. hawking , phys . d * 46 * , 603 ( 1992 ) . d. g. boulware , phys . d * 46 * , 4421 ( 1992 ) . li , j. -m . xu , and l. liu , phys . rev . d * 48 * , 4735 ( 1993 ) . li , phys . d * 50 * , r6037 ( 1994 ) . t. tanaka and w. a. hiscock , phys . d * 52 * , 4503 ( 1995 ) . li , class . quantum grav . * 13 * , 2563 ( 1996 ) . s. v. krasnikov , phys . d * 54 * , 7322 ( 1996 ) . s. v. sushkov , class . quantum grav . * 14 * , 523 ( 1997 ) . m. visser , phys . d * 55 * , 5212 ( 1997 ) . m. j. cassidy , class . quantum grav . * 14 * , 3031 ( 1997 ) . li and j. r. gott , phys . lett * 80 * , 2980 ( 1998 ) . j. r. gott and l. -x . li , phys . d * 58 * , 023501 ( 1998 ) . j. a. wheeler , phys . rev . * 97 * , 511 ( 1955 ) . a. vilenkin , phys . b * 117 * , 25 ( 1982 ) . j. b. hartle and s. w. hawking , phys . d * 28 * , 2960 ( 1983 ) . s. hawking and r. penrose , _ the nature of space and time _ ( princeton university press , princeton , 1996 ) . s. w. hawking and g. f. r. ellis , _ the large scale structure of space - time _ ( cambridge university press , cambridge , 1973 ) . l. castellani , r. dauria , and p. fr , _ supergravity and superstrings : a geometric perspective _ ( world scientific , singapore , 1991 ) . e. witten , preprint hep - th/9805112 ( 1998 ) . a. h. guth , phys . d * 23 * , 347 ( 1981 ) . a. d. linde , phys . 108b * , 389 ( 1982 ) . a. albrecht and p. j. steinhardt , phys . lett * 48 * , 1220 ( 1982 ) . s. w. hawking , in _ general relativity : an einstein centenary survey _ , edited by s. w. hawking and w. israel ( cambridge university press , cambridge , 1979 ) , p. 746 . s. w. hawking , phys . d * 52 * , 5681 ( 1995 ) . n. d. birrell and p. c. w. davies , _ quantum fields in curved space _ ( cambridge university press , cambridge , 1982 ) . j. r. gott , nature * 295 * , 304 ( 1982 ) . j. r. gott , in _ inner space /outer space _ , edited by e. w. kolb et al ( university of chicago press , chicago , 1986 ) . s. coleman , phys . d * 15 * , 2929 ( 1977 ) . s. coleman and f. de luccia , phys . rev . d * 21 * , 3305 ( 1980 ) . c. w. misner , k. thorne , and j. a. wheeler , _ gravitation _ ( w. h. freeman and company , san francisco , 1973 ) . e. schrdinger , _ expanding universes _ ( cambridge university press , london , 1956 ) . c. j. s. clarke and t. dray , class . quantum grav . * 4 * , 265 ( 1987 ) . c. barrabes and w. israel , phys . d * 43 * , 1129 ( 1991 ) . j. r. gott , ap . j. 288 , 422 ( 1985 ) . w. a. hiscock , phys . d * 31 * , 3288 ( 1985 ) . li and j. r. gott , phys . d * 58 * , 103513 ( 1998 ) . j. s. dowker , quantum field theory around conical defects , in _ the formation and evolution of cosmic strings _ , edited by g. gibbons , s. hawking , and t. uachaspati ( cambridge university press , 1990 ) . d. iellici , preprint gr - qc/9805058 , ph.d . m. j. cassidy and s. w. hawking , phys . d * 57 * , 2372 ( 1998 ) .

project 1640 is a high contrast near - infrared instrument probing the vicinities of nearby stars through the unique combination of an integral field spectrograph with a lyot coronagraph and a high - order adaptive optics system . the extraordinary data reduction demands , similar those which several new exoplanet imaging instruments will face in the near future , have been met by the novel software algorithms described herein . the project 1640 data cube extraction pipeline ( pcxp ) automates the translation of closely packed , coarsely sampled spectra to a data cube . we implement a robust empirical model of the spectrograph focal plane geometry to register the detector image at sub - pixel precision , and map the cube extraction . we demonstrate our ability to accurately retrieve source spectra based on an observation of saturn s moon titan .

1104.5233

in recent years an assortment of new astronomical techniques have evolved to address the challenges of imaging faint objects and disk structure at close angular separations to nearby stars . a major scientific motivation for these efforts is the direct detection and characterization of low - mass companion bodies orbiting at separations between @xmath1 5 and 100 au . these objects are beyond the reach of conventional optical imaging due to the extreme contrast in brightness with respect to the primary star . in such cases , even under ideal observing conditions the diffracted light of the primary star overwhelms the neighboring source of interest . the various methods of manipulating a star s light to enable investigation of its immediate environment are collectively referred to as high - contrast imaging . for a recent review of this field , see @xcite . the acquisition of spectra of young , sub - stellar mass objects in this newly opened parameter space will ultimately lead to a breakthrough in our understanding of exoplanet populations @xcite . project 1640 ( p1640 ) is the first of several instruments to approach the high contrast imaging problem through a combination of high - order adaptive optics , a lyot coronagraph , and an integral field spectrograph @xcite . forthcoming instruments using a similar design include the gemini planet imager ( gpi ) @xcite , the very large telescope spectro - polarimetric high - contrast exoplanet research ( vlt - sphere ) project @xcite , and the subaru telescope planetary origins imaging spectrograph ( poise ) @xcite . while previous efforts have used integral field spectrographs for high contrast imaging ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and lyot coronagraphs have also been employed for surveys of nearby stars ( e.g. * ? ? ? * ; * ? ? ? * ) , p1640 is the first instrument to combine these two technologies . the coronagraph component , based on the fourier optics concept described in @xcite , rejects the core of the target star s point spread function ( psf ) and attenuates the surrounding diffraction rings . provided that the adaptive optics ( ao ) system upstream of the coronagraph has corrected the star s psf to near the diffraction limit , then the dominant source of noise in the image exiting the coronagraph takes the form of a halo of speckles surrounding the occulted star , as in figure [ fig : examplecube ] @xcite . these relatively long - lived , point source - like artifacts are caused by uncorrected wave front aberrations , and limit the dynamic range of the data unless further processing is carried out @xcite . the integral field spectrograph , also referred to as the integral field unit ( ifu ) , is situated after the coronagraph and provides spatially resolved spectra for a grid of points across the field of view @xcite . the reduced form of data acquired with an ifu is a stack of simultaneous narrowband images spanning the instrument s wavelength range , often called a data cube . an example of part of a p1640 data cube is shown in figure [ fig : examplecube ] . one benefit the ifu provides is enabling the observer to measure the spectrum of any source at any position in the field of view . this is not possible with a conventional spectrograph , which can only use one spatial dimension at a time to discriminate against other sources in the field of view . the second purpose of the ifu is to exploit the chromatic behavior of the speckles @xcite . speckles are optical in origin and their separation from the target star is linearly proportional to wavelength , at least to a first - order approximation ( see figure [ fig : examplecube ] ) . by comparing the images in different channels of the data cube , the observer can discriminate the speckles from a true point source , whose position should remain constant with wavelength . furthermore , automated post - processing software can use the chromaticity of the speckles to subtract a large component of them from the data , taking a reduced data cube as input and generating a speckle - suppressed version @xcite . the unusual properties of data generated by the p1640 ifu necessitate novel reduction techniques to reach the point where inspection , spectrophotometry , astrometry , and advanced post - processing techniques like speckle suppression can begin . in the scope of this article , we describe the software created to translate rapidly the raw data from the ifu camera to a set of data cubes ready for further analysis . during operation at palomar observatory , p1640 receives a wave front - corrected beam of the target star s light from the 200 " hale telescope ao system . the current ao system , the 241-actuator palao @xcite , will soon be upgraded to the 3388-actuator palm-3000 @xcite . upon entering the instrument , the light passes through an apodized lyot coronagraph , followed by an integral field spectrograph , which contains a near - infrared camera . in addition to the focal plane mask and lyot stop of a traditional lyot coronagraph , p1640 uses a pupil plane apodization mask to optimize the starlight suppression based on the telescope pupil shape @xcite . the beam exiting the coronagraph comes to a focus on a @xmath2 square microlens array at the entrance of the spectrograph . immediately after the microlens array , the light is collimated to form a pupil on a wedge - shaped prism , which disperses the light over the 1.1 - 1.8 @xmath3 m wavelength range of operation spanning the @xmath4 and @xmath5 bands . additional optics focus the @xmath6 resulting spectra onto a teledyne hawaii-2 @xmath7 pixel , hgcdte , near - infrared detector . the field of view of the final image , designed to match the control radius of the palm-3000 ao system , is @xmath8 . for further details on the optical and mechanical design , see @xcite . for each exposure , the camera controller performs a sequence of non - destructive reads on the detector array . in other words , the digitized value of each pixel is periodically sampled while its voltage escalates . this technique , known as up - the - ramp sampling , can result in the read noise being reduced by a factor of @xmath9 in a reduced image when the counts versus read slope is fit for the @xmath10 samples of each pixel @xcite . up - the - ramp sampling also adds an advantageous temporal dimension to our data . speckle suppression algorithms work best when the positions of the speckles are well - defined . for bright stars with high signal - to - noise in individual read differences , it may be helpful to `` freeze '' the speckle pattern with the higher time resolution enabled in a read - by - read data reduction . our pipeline reduces the detector data with both approaches : the non - destructive read ( ndr ) slope fit and consecutive read differences . the read sample interval is fixed at 7.7 seconds by the camera controller . the sequence of reads are stored in a binary file containing the arrays of 16-bit unsigned integer samples , which we refer to as a _ dat _ file . the camera controller also generates a separate fits file with a header containing the information about the telescope and instrument status , the target ( coordinates , magnitude , parallax , etc . ) , and the name of the _ dat _ file corresponding to the exposure . a typical observation of a project 1640 science target is made up of 15 exposures , each containing 20 reads , giving a cumulative exposure time of @xmath11 minutes . the resulting volume of raw data is 160 mbytes for each exposure s _ dat _ file and 2.4 gbytes in total . the structure of the p1640 ifu focal plane , illuminated by moonlight , is depicted in figure [ fig : focpldiagram ] . the microlens array , represented by the dotted grid superimposed on the left panel , is rotated with respect to the detector . this configuration interleaves the adjacent rows of microlens spectra , thereby maximizing the efficiency of focal plane area usage . along a given column of microlenses , the mean interval between neighboring spectra is 3.3 pixels in the horizontal direction and 10.0 pixels in the vertical direction . each spectrum takes up a length of approximately 27 pixels in the dispersion direction . rather than relying purely on design predictions , we have written procedures to empirically determine the ifu response , capturing the minute optical distortions and alignment changes unique to each observing run . two forms of calibration data are used as input for the focal plane model . the first kind are the spectrograph images formed by illuminating the instrument pupil with a tunable laser source . these allow us to characterize the response at fixed wavelengths across the passband . secondly , during each observing run we observe a broadband source of nearly uniform brightness across our field of view either the moon or the twilight sky . in the case of the moon calibration images , the telescope ao correction loop was turned off , and several exposures with different pointings were averaged . these images constrain the geometry of the focal plane , including the positions and shapes of the individual spectra formed on the detector , as well as large - scale variations in sensitivity across the field of view due to vignetting . an accurate model of the monochromatic spectrograph point spread function ( psf ) is at the core of the ifu focal plane model . we emphasize the distinction here from the coronagraph psf , which is formed on the mircolens array at the entrance of the ifu . the spectrograph psf , on the contrary , is the signal formed on the ifu focal plane from monochromatic light incident on an individual microlens . in a laboratory environment before the first scientific observing run , we illuminated the ifu with a tunable laser source and recorded narrowband emission ( bandwidth @xmath12 4 nm ) images at wavelengths in 0.01 @xmath3 m increments spanning the 0.7 @xmath3 m operating band of the instrument . any given laser image shows a grid of @xmath0 point spread functions , each corresponding to a microlens illuminated by the beam entering the spectrograph . from these images , we derived an analytic model of the spectrograph psf specific to the recorded wavelength , as follows : first , a script looped through all of the psfs in the spectrograph image , forming a 9@xmath139 pixel _ mean _ psf based on the subset having centroids within 0.05 pixels of a detector pixel center . next , we experimented with a variety of two - dimensional functional forms to represent the psf , progressively adding parameters until finding one with a good match to the data . since in our case the detector pixel width is comparable to the psf full - width half maximum value , it was necessary to take into account not only the effect of the finite detector pixel area in sampling the function , but also intra - pixel sensitivity variations . charge diffusion is the largest contribution to non - uniform sensitivity within any given pixel . during the technology development phase of a space mission to survey extragalactic supernovae , @xcite measured the effect of charge diffusion on the intra - pixel sensitivity of a hawaii-2 detector . he found a good empirical fit to a typical pixel s response by convolving a tophat function ( with width equal to that of the detector pixel ) with a hyperbolic secant diffusion term , @xmath14 , where r is radius from origin and @xmath15 is the diffusion length . with the established diffusion length of 1.9 @xmath3 m ( compared to the 18 @xmath3 m full pixel width ) , the response falls to about 50% of the peak at the middle of each pixel edge . for lack of similar measurements of our own hawaii-2 detector , we assumed the same charge diffusion behavior . we discretized the two - dimensional functions representing the psf , @xmath16 , and intra - pixel response , @xmath17 , at a resolution of @xmath18 times that of the detector , where @xmath18 is an odd number @xmath19 3 . in other words , the image model has @xmath20 samples contained within each detector pixel , one always aligned with the center of a pixel . the intra - pixel response function , @xmath21 , is only defined over an area of one pixel , so that @xmath22 , whereas @xmath23 is defined over the entire @xmath24 area of the mean psf cutout , corresponding to @xmath25 . in this notation , the detector - downsampled psf , @xmath26 is determined by @xmath27 where @xmath28 are independent variables representing detector samples over the @xmath29 pixel mean psf cutout . we found a satisfactory functional form to match the monochromatic psf by taking the sum of two piecewise , two - dimensional gaussian profiles , defined as follows : @xmath30 where @xmath31 nine parameters fully describe the psf in this formulation : two amplitudes , six characteristic widths , and one rotation . the piecewise definition allows freedom from reflective symmetry across the @xmath32 axis . hence there is a pair of characteristic widths for each side of the @xmath33 axis one for @xmath34 ( @xmath35 and @xmath36 ) the other for @xmath37 ( @xmath38 and @xmath39 ) . also note the coordinate transformation built into the definition ( equation [ eqn : coordxform ] ) . the operations in the right hand column vector serve two purposes . first , they shift the effective origin from the lower left corner of the @xmath29 cutout array its original position for the purpose of simplified indexing in equation [ eqn : psfbinning]to its center . at the same time , the @xmath40 factor scales both coordinates to units of detector pixel width . finally , the @xmath41 matrix facilitates a rotation of the overall surface by angle @xmath42 in the counterclockwise sense . using mpfit , the non - linear least squares fitting program written by @xcite , we determined the function parameters for the mean psf cutouts at wavelengths 1.25 @xmath3 m and 1.58 @xmath3 m . the results , based on a model spatial sampling rate of @xmath43 times that of the detector , are listed in table [ tab : psfparams ] . in each case , the amplitudes were scaled so as to give unity peak intensity in the detector - downsampled psf . as in equation [ eqn : psfdef ] , the characteristic widths are in units of detector pixel widths . in figure [ fig : psf_cross_sections ] we have plotted orthogonal cross sections of the best - fit psf functions . in the same figure we drew bars to represent the corresponding detector - downsampled psf cross sections . note that the peak of each model function is significantly higher than that of the detector - downsampled version , due to the sensitivity roll - off away from the pixel center . at both wavelengths , the mean residual disparity between the downsampled best - fit model and the original mean laser psf cutout ( not shown in the plot ) is the less than @xmath44 of the peak intensity . cccccccccc 1.25 & 1.31 & 0.33 & 0.52 & 0.85 & 0.38 & 1.22 & 0.91 & 1.85 & @xmath45 + 1.58 & 1.32 & 0.27 & 0.55 & 0.81 & 0.44 & 1.14 & 0.90 & 1.87 & @xmath46 + we built upon knowledge of the spectrograph psf to characterize the coarse near - infrared spectra distributed across the ifu focal plane . here we turned to our moon and twilight sky calibration exposures , during which each microlens was illuminated with a strong , uniform , broad spectrum of light . an example image of this kind was illustrated in figure [ fig : focpldiagram ] . by isolating the small detector area containing an individual microlens signal , we can fit a set of parameters encoding the spectrum geometry . ccccc position @xmath47 & @xmath48 & n / a & ( 0.02047.0 , 0.02047.0 ) & n / a + height ( @xmath49 ) & @xmath50 & 23.9 pixels & 23.524.7 pixels & 0.2 pixels + tilt ( @xmath51 ) & @xmath52 & 0.044 & 0.00900.080 & 0.021 + in table [ tab : specparams ] we have listed the parameters needed to describe the spectrum image of an individual microlens . the @xmath47 position coordinates are the most fundamental of these . they are referenced to the @xmath53 = 1.37 @xmath3 m point , which coincides with the sharp ( blue ) edge of the telluric water absorption trough between the @xmath4 and @xmath5 bands . the @xmath54 coordinates index the full detector array from an origin at the pixel in the lower left ( sw ) corner of the image ; all integral values align with a pixel center . using similar notation , we defined the spectrum height and tilt based on the relative positions of the @xmath53 = 1.10 @xmath3 m and 1.76 @xmath3 m points , which roughly correspond to the edges of our passband . from the position , height , and tilt , the coordinates of an arbitrary wavelength in the spectrum can be calculated by the following parametrized equations : @xmath55 where @xmath49 is height , @xmath51 is tilt , and @xmath56 m . by this definition , each integral step in @xmath57 corresponds to 0.03 @xmath3 m , so there is a total of 22 such increments from 1.10 @xmath3 m and 1.76 @xmath3 m . using the same wavelength parameter , we can specify the intrinsic spectrum incident on the focal plane by some function @xmath58 for @xmath59 . we introduce the concept of spectrum _ trace _ to model the layout of the spectrum by an ideal `` skeleton '' image formed by a train of impulse functions , unencumbered by diffraction and focus effects . the trace is discretized in the same manner as the psf model , at a resolution @xmath18 times higher than that of the detector . in this case , however , we use a lower spatial sampling rate factor of @xmath60 to balance reasonable execution speed and performance . by convolving the trace with the reverse of the psf ( rotated 180@xmath61 ) , and downsampling the result , we can synthesize a spectrum image as measured by the detector ( figure [ fig : moddescrip ] ) . the position , height , and tilt parameters , along with the intrinsic spectrum can then be adjusted by a least squares fitting algorithm until the downsampled result matches the data cutout . a cutout spanning an area of @xmath62 detector pixels is sufficient to enclose an individual spectrum as well as major portions of the two nearest neighbors . at the start of the fitting procedure , the cutout is aligned such that the ( @xmath63 ) reference point of the spectrum nearly corresponds to pixel position ( 4 , 16 ) in the detector - downsampled trace array . two free parameters in the model , @xmath64 and @xmath65 , allow the algorithm to refine the initial position guess alongside the other geometrical properties . for a given microlens spectrum , the following equations define the conversion between @xmath66 detector indices and @xmath67 trace array indices : @xmath68 from which it follows that the trace indices of the 1.37 @xmath3 m reference point are @xmath69 . based on the typical interval between spectra along a microlens , we set the 1.37 @xmath3 m reference points of the neighboring spectra by @xmath70 and @xmath71 . the trace signal is dispersed over the same line segment defined in equation set [ eqn : specxyparam ] , with intrinsic spectrum function @xmath58 . the neighboring spectra are parametrized by the same shape with respect to their own reference points , ( @xmath72 ) and ( @xmath73 ) . we form separate trace arrays for the @xmath4- and @xmath5-band halves of the spectra , designated @xmath74 and @xmath75 . we do this in anticipation of separate convolution operations with the @xmath4- and @xmath5-band psfs ( figure [ fig : psf_cross_sections ] ) . the two respective trace arrays are defined as follows : @xmath76 where @xmath77 , @xmath78 , and @xmath79 . to test a given set of spectrum model parameters , we convolve the @xmath4 and @xmath5 trace arrays with the reverse of the psf model ( such that @xmath80 ) , giving a high - resolution model of the light distribution on the focal plane , @xmath81 : @xmath82 implicitly , we have zero - padded the trace arrays before the convolution , and trimmed the result to the original @xmath83 array size . the detector - downsampling operation is similar to that used earlier for the psf model : @xmath84 however , one new variable has been introduced in equation [ eqn : binnedspec ] : @xmath85 , a constant offset added to each pixel in the downsampled image model . this is one more parameter open to adjustment by the fitting procedure , which takes into account any background level of scattered light present in the data cutout . for a point source , in some focal plane locations this background reaches up to 3% of the 99.5 percentile - level count rate , considering all detector pixels . therefore , it becomes especially significant for an exposure of a source as bright as the moon . the resulting synthetic spectrum image @xmath86 can be directly compared with the data cutout ( figure [ fig : moddescrip ] ) . in principle , the least squares fitting algorithm ( the mpfit program in our case ) converges on the data cutout over many loops , switching between revising the trace model parameters and comparing the downsampled result to the data . the combination of unknown position , shape , and intrinsic spectrum @xmath58 presents too many free parameters for a fitting algorithm to accurately solve for at once . in practice , we need to iteratively build up constraints , starting from as few assumptions as possible . one aspect of the moon / sky calibration exposure we can take advantage of is the fact that the intrinsic spectrum , @xmath58 , is identical across the image , apart from scale factors due to large - scale variations in sensitivity over the field of view . in addition , by referring to the laser calibration images , we can make very good initial guesses of the height and tilt for a given region of the focal plane . still , we found these constraints alone were insufficient to reach consistent solutions . the exact vertical position of the spectrum ( encoded by @xmath65 in equation set [ eqn : uvdef ] ) proved especially difficult to determine with only limited information about the light source and the instrument response . to get over this barrier , we chose to use prior knowledge of the atmosphere s transmission function in particular , the shape imposed on the spectrum by the deep water absorption trough in the middle of the p1640 passband . figure [ fig : watertrough ] shows the expected transmission function of the atmosphere from 1.28 @xmath3 m to 1.52 @xmath3 m . the data points are based on the measurements made by @xcite from kitt peak ( at altitude 6875 ft , comparable to the 5618 ft altitude of the palomar observatory hale telescope ) , here averaged over 0.01 @xmath3 m bins . instead of allowing the points inside the water trough ( @xmath87 ) to vary freely , we impose the condition @xmath88 where @xmath89 is the peak - normalized atmospheric transmission function plotted in figure [ fig : watertrough ] . inside the water trough , @xmath58 is discretized in 0.01 @xmath3 m bins ; outside , in 0.03 @xmath3 m bins ( integral values of @xmath57 ) . once the above assertion is in place , the fitting algorithm could at last reliably determine both the position and shape of the spectrum , as achieved in the example shown in figure [ fig : moddescrip ] . [ eqn : watercond ] we repeated the spectrum fitting procedure across the entire spectrograph image to form a global solution unique to the specific epoch of the moon / sky calibration exposure . to minimize the number of free parameters before executing this , we first determined a mean intrinsic spectrum @xmath90 based on average of the @xmath58 fit results from a sub - area of about 100 spectra near the center of the field of view . with the spectrum shape fixed , however , there needs to be a parameter that captures variations in overall signal strength across the focal plane . we designated an amplitude parameter @xmath91 to acts as a multiplicative constant , applied to @xmath90 , and freely adjusted alongside @xmath92 , @xmath93 , @xmath49 , @xmath51 , and @xmath94 . in figure [ fig : param_maps ] we have displayed maps of the height , tilt , and amplitude parameters for one epoch . these maps proved essential to the challenging process of debugging the fitting routines . they also enable easy visual comparisons between focal plane properties at different times , and can serve as diagnostic tools during periods of modifications and upgrades to instrument optics . the maps in figure [ fig : param_maps ] appear as rotated squares because the microlens array is by design rotated with respect to the detector ( as shown previously in figure [ fig : focpldiagram ] ) . we index the microlenses using cartesian coordinates @xmath95 and @xmath96 relative to an origin at the lower left corner . with these coordinates , a range of @xmath97 is sufficient to enclose the @xmath0 microlens spectra on the detector . the ability to analyze the _ spatial _ distribution of the ifu spectra is also of great interest . a vector field , like those depicted in figure [ fig : distortion ] , is an effective way to illustrate the evolution of the global spectrograph focal plane geometry . to make these plots , we first partitioned the calibration image into an array of @xmath98 boxes , each of width 256 detector pixels . for any two comparison epochs , each with spectrum position arrays @xmath99 and @xmath100 , we calculated the median of the differences @xmath101 and @xmath102 inside each box , resulting in an @xmath98 array of vectors . before plotting those vectors , we subtracted the median difference vector at the image center . in this way , we removed the effect of a trivial bulk shift between the focal plane patterns . depending on the duration of time between the pair of calibration images under consideration , the quivers representing the vectors need to be scaled up by different factors to reveal the subtle evolution . once this is done , it becomes clear that the overall scale and orientation of the spectrograph focal plane pattern vary with time . more complicated , non - uniform distortions also play a role . in all cases , the magnitude of these changes are small enough that they would never be obvious from a mere `` blinking '' comparison of the source images . for example , the transformation from march 2009 to june 2009 can mostly be attributed to a rotation of the microlens array with respect to the detector ( or vice versa ) by an angle of merely @xmath103 . likewise , between 28 june 2009 and 29 june 2009 , the focal plane was magnified by about 0.003% . it is a fair guess that the changes we observe in global focal plane geometry are due to minute variations of environmental conditions inside the ifu dewar . however , it remains unclear what the relative contributions are from the various optics and the mechanical support structures . to an extent , the origin of these changes is not an issue , so long as each science data set can be attached to a solution that accurately reflects its particular geometry . another important product of our global fitting procedure is a synthetic image of the entire spectrograph focal plane . using the established geometric parameters , we can inject an arbitrary source spectrum @xmath104 to simulate the distribution of light incident on the detector . the synthetic focal plane image is useful for inspecting the results of the global fit , and is also an essential ingredient in the algorithm used by the cube extraction pipeline to register the spectrograph image of an individual science exposure at sub - pixel precision ( described in [ sec : specreg ] ) . the formalism is analogous to that described for the individual spectrum cutout model in equations [ eqn : uvdef][eqn : binnedspec ] . we designate @xmath105 and @xmath106 to represent the @xmath4- and @xmath5-band model trace arrays of the full spectrograph focal plane image , discretized at a spatial sampling rate @xmath18 times that of the detector . since the hawaii-2 detector array size is @xmath7 , the trace arrays are defined over @xmath107 . here we again settled on a sampling factor of @xmath18 = 7 . since as before , we require @xmath18 to be an odd integer @xmath108 , there is always a pair of trace array indices @xmath109 aligned with the center of a given detector pixel @xmath66 : @xmath110 the trace array is determined by the position , height , and tilt solutions , now indexed by microlens coordinates @xmath111 : @xmath112 @xmath113 from the trace arrays , we obtain the spectrograph focal plane image model @xmath114 in the same manner as in [ sec : spectrummodel ] , by convolving them with their corresponding reversed psf models : @xmath115 we implement these convolution operations in the fourier domain to save computational time , which is otherwise a nuisance for the large dimensions of our arrays @xcite . with a spatial sampling factor of @xmath116 , we obtain a factor of @xmath1 50 using fft - based convolution over direct convolution . in our experience , this cuts the execution time needed to form the synthetic image down from a few hours to a few minutes ( assuming the global solution is already done ) . finally , we can obtain the detector - downsampled synthetic focal plane image , @xmath117 , by binning @xmath118 to the detector resolution using the assumed intra - pixel response : @xmath119 note that this synthetic detector image is idealized in the sense that we have left out the amplitude modulations across the image ( @xmath120 ) as well as background light parameters ( @xmath121 ) . for the purpose of registering a science spectrograph image , this is preferred , since we are only concerned with matching the shapes and positions of the spectra . the project 1640 data cube extraction pipeline ( pcxp ) , written in the gnu c programming language , automates the processing of raw p1640 detector images and their translation to reduced data cubes . a block diagram summarizing the steps applied to each image is shown in figure [ fig : blockdiagram ] . by design , the program is fast enough to use while observing , so that newly acquired images can be inspected in real time to monitor instrument performance and check for unknown objects . the data pipeline can also be used to process an arbitrarily large set of raw data at a later date . for post - processing , we feed the output of the pcxp into the project 1640 speckle suppression pipeline ( pssp ) , described by @xcite . two outer loops comprise the pcxp execution . first , the program steps through the detector data in the input directory specified by the user at the start time , processing each non - destructive read ( ndr ) sequence to form a reduced , registered spectrograph image . the second stage of the pipeline loops through the finished spectrograph images and extracts data cubes from each of them . throughout these steps , the pipeline relies on the empirical model of the spectrograph focal plane described in [ sec : focplmodel ] . the pipeline identifies pixels contaminated by cosmic rays by checking for anomalous jumps in digitized count values within the ndr sequence . for each detector pixel , our algorithm determines the median increase in counts between successive reads over the course of the exposure . a count increment greater than five times the median is flagged as a cosmic ray event . at each pixel meeting this criterion , the count contribution of the cosmic ray event is subtracted from the read corresponding to the event as well as all the following reads , canceling out its influence . we chose our threshold based on inspections of images of faint occulted stars , with relatively noisy slopes . we blinked `` before '' and after images to check that all apparent cosmic ray events , and no starlight - dominated pixels were erroneously flagged . this method of cosmic ray removal only works for exposures consisting of more than two reads . for a shorter exposure there is no way to take advantage of the ndr detector mode to identify cosmic rays . in this case the pipeline passes the detector image through the iraf noao cosmic ray cleaning algorithm . during each observing run , a set of `` dark '' ndr sequences are obtained by taking calibration exposures with the ifu in a cryogenic state identical to the scientific data acquisition mode , except that the coronagraph beam entrance window is capped to obstruct external light . these dark exposures record the bias , thermal , dark current , and badly - behaved `` hot '' pixel count values of the detector array at each read interval . the median of 11 dark ndr sequences for each exposure time is added to a permanent library directory of dark exposures , marked by date and exposure time . see figure [ fig : dark ] for an example of a dark exposure . after loading the _ dat _ file of a science target , the first processing step of the pipeline is to find the most appropriate dark ndr sequence and perform a readwise subtraction . after subtracting the bias / dark component and removing the cosmic rays , the pipeline fits a slope to the adu count versus time values recorded in the ndr sequence . this reduces the detector data for a given exposure to a single @xmath7 pixel representation of the spectrograph image . we employ an ordinary least - squares linear regression to determine the count rate for each pixel , eventually storing the floating point values in a fits file in units of counts / second . the slope fitting is complicated by pixel saturation caused by bright sources . however , the ndr detector mode is advantageous for handling this . in the case where a pixel reaches saturation at some point after the first two reads , the affected reads are simply excluded from that pixel s linear regression . this approach , recommended after tests described by @xcite , extends the effective dynamic range of a long exposure by a factor of @xmath122 . for pixels that reach saturation before the second read , a slope computation is not possible . if this saturation occurred between the first and second reads , then the slope can at least be approximated based on the difference between the first read value and an assumed zero level from the dark ndr sequence . however , in the case of a pixels saturating before the first read , this result will not be physically meaningful . to prevent erroneous measurements from being made by investigators analyzing images affected by saturation , the pipeline sets an appropriate header variable in the reduced fits files . this header keyword indicates whether any of the detector pixels saturated , and if so , whether that occurred at the first , the second , or a subsequent read . a externally controlled lamp inside the dewar of the p1640 ifu can fully illuminate the detector . to counteract pixel - to - pixel variations in detector sensitivity , we constructed a detector flat field map based on the mean of 12 dewar lamp exposures . since the lamp intensity is not uniform across the detector , we used iraf to fit a cubic spline surface to the normalized , mean dewar lamp image . we divided by the resulting spline surface to form the final detector flat , with large - scale variations removed ( see the next section for an explantation of how large - scale variations in sensitivity are corrected for during cube extraction ) . the standard deviation of pixel values in the detector flat field map is 0.13 . after the ndr slope - fitting step , the pipeline divides the spectrograph image by the flat field map to compensate for pixel - to - pixel variations . for locations in the flat field map with exceptionally low values ( @xmath123 ) , no division is carried out since doing so would tend to enhance the noise induced by weak , problematic pixels . due to flexure varying mechanical stress on the instrument while the telescope slews the projection of the microlens array onto the detector changes over the course of an observing period . the plot in figure [ fig : offsets ] illustrates the magnitude of this effect based on measurements from three observing runs . between targets , the positions of the spectra on the detector can uniformly shift by up to 2 pixels in each direction . between observing runs there is a more pronounced , systematic shift in the spectrograph - detector alignment . to accurately extract the data , the pipeline needs to register the precise offset between each spectrograph image and the focal plane model of the corresponding epoch . we accomplish this through two stages : first a crude estimate based on a cross - correlation with the downsampled spectrograph image model ( array @xmath124 in equation [ eqn : binnedimagemodel ] ) , followed by a more elaborate approach to refine the offset to sub - pixel precision . for efficiency , the initial cross - correlation is restricted to a @xmath125 square pixel section of the science image @xmath126 . we denote this cutout box with a tilde accent on top of the original array symbol : @xmath127 likewise we use @xmath128 to represent the same section from the downsampled focal plane image model . the center of the box , @xmath129 , is chosen based on the average count rate computed within @xmath130 partitions across the image , so as to enclose spectra with relatively high signal strength . in a typical science image with the star occulted by the coronagraph , this is near the center of the image , where the residual starlight is brightest . we cross - correlate @xmath128 and @xmath131 to determine the _ crude _ offset . a given science focal plane pattern is not expected to stray more than two pixels away from the calibration exposure of the matching epoch . furthermore , the periodicity of the spectrum pattern ensures that large lags will merely introduce degenerate solutions . therefore , we do not compute the full two - dimensional cross - correlation array , but merely a small region bounded by horizontal and vertical lags up to 4 pixels in each direction : @xmath132 the summation limits take the lag range into account in order to avoid the influence of non - overlapping array edges . the lag combination that maximizes @xmath133 , which we denote @xmath134 , is our initial guess for the horizontal and vertical displacement of @xmath135 with respect to @xmath124 . the second stage of the registration routine determines separately the _ fine _ @xmath136 and @xmath137 offsets . as apparent in figure [ fig : focpldiagram ] , the spectral dispersion is almost completely aligned with the @xmath137 axis of our detector image coordinate system . as a consequence , the shape of a spectrum s horizontal cross - section at a given wavelength is determined much more by the spectrograph psf shape than by the intrinsic spectrum of the light source . it is effectively the cross - dispersion profile commonly referred to in literature on more conventional spectroscopic observations ( e.g , * ? ? ? therefore , to measure the effect of a slight horizontal offset on the detector - sampled image , we can start with the high - resolution model of the light distribution @xmath118 ( equation [ eqn : imagemodel ] from [ sec : globalsol ] ) , even though its intrinsic spectrum does not necessarily match the data . to simulate how the detector would `` see '' the image model for a range of small fractional - pixel offsets from the initial alignment , we downsample @xmath118 as in equation [ eqn : binnedimagemodel ] , but with the detector sampling array shifted by a range of horizontal sub - pixel offsets indexed by the variable integer @xmath138 : @xmath139 now we reevaluate the cross - correlation peak for the @xmath18 fine horizontal offsets values , with @xmath32 and @xmath33 fixed at @xmath140 and @xmath141 : @xmath142 the fractional offset index @xmath138 that maximizes @xmath143 gives the fine horizontal displacement of the data , @xmath138 , with respect to the crude initial guess , @xmath140 . the full horizontal offset is @xmath144 detector pixel widths . we originally intended to use the same approach to determine the fine vertical offset . unfortunately , in this case the disparity between the intrinsic spectrum of the data and the image model strongly biases the cross - correlation result . in a typical science image , the cutout box @xmath131 contains @xmath110 speckles . their chromatic position dependence ( as illustrated in figure [ fig : examplecube ] ) causes steep brightness gradients in the spectra formed on the spectrograph focal plane , since a given microlens will collect light from a speckle over only a fraction of the passband . whatever intrinsic spectrum is built into the image model , @xmath118 , will significantly differ from that of most of the sample . we found that the effects of these disparities do not average out over an ensemble . instead , they systematically push the cross - correlation result up or down by a degree that does not reflect the actual relative wavelength alignment . instead of using the image model as an alignment template , we return to the fitting approach described in [ sec : spectrummodel ] . this time , however , rather than fitting the full spectrum parameter set , we concentrate on the region with the most information about the vertical position : the telluric water absorption trough . therefore , we confine the least - squares fit region to a @xmath145 box , aligned such that the 1.37 @xmath3 m reference point is near the middle pixel on the 8th row . we further simplify the spectrum fit by describing the local light source with merely two parameters : an amplitude and color . the other free parameters are the background light offset and the vertical position . the height and tilt are already known from the calibration image solution , and the horizontal position is fixed based on the previous step in the registration algorithm . as in equation [ eqn : troughcond ] , the spectrum trace points with wavelengths @xmath146 m are again tied to the transmission function plotted in figure [ fig : watertrough ] . the anchor points at @xmath1471.28 @xmath3 m and 1.58 @xmath3 m are set based on the amplitude and color parameters . to get a diverse set of spectrum shapes spanning a wide region of the speckle halo , during the fine vertical offset fitting procedure we sample 121 spectra over a @xmath148 pixel box ( as compared to the @xmath2 pixel box used for the horizontal registration ) . of the 121 fits , the median vertical offset is taken as the final value , and rounded to the nearest 1/7th of a pixel to match the quantization of the fine horizontal offset . in trial runs , we found the vertical offsets determined from the full set of sample spectra follow a gaussian distribution , with standard deviation 0.20.4 pixel widths , depending on the source image . we accept this as the uncertainty in the vertical registration . in order to form a data cube , the pipeline must `` know '' where individual spectra are positioned on the focal plane , and furthermore , which points of those spectra correspond to a given wavelength . we rely on the global spectrograph focal plane solution described in [ sec : globalsol ] to establish the image geometry for each epoch under consideration . one of the products of the calibration image fitting procedure is a text file tabulating the positions of all @xmath0 spectra alongside their corresponding microlens indices . this table , combined with the results of the registration algorithm ( [ sec : specreg ] ) and the maps of height and tilt parameters give all the information needed to organize the detector data for a given science image . the amplitude map produced during the global fit ( see figure [ fig : param_maps ] ) also has an important role . it compliments the dewar lamp flat described in [ sec : dewarflat ] by capturing the larger - scale variations in sensitivity across the field of view . by looking up the amplitude parameter associated with a given microlens , we can appropriately scale any detector samples from that spectrum to compensate for optical effects such as vignetting . despite the numerous stages in the detector image processing , some minor extraneous background structure persists into the processed focal plane image . this component , superimposed on the real signal , is caused by a combination of residual bias counts , scattered light within the instrument , and thermal contamination from outside the dewar ( unaccounted for in the dark subtraction ) . in the cube extraction routine , after loading an individual focal plane image , the pipeline forms a map of background count rates based on measurements between spectra . figure [ fig : extraction ] shows the regions used to estimate the background count rate associated with a given microlens . the upper left box is situated so that its bottom row is matched with the @xmath149 m point ( rounded to the nearest row ) , and the bottom row of the lower right box is on level with @xmath150 m . for both background boxes , the near side is spaced three columns from the rounded center of the spectrum . the pipeline takes the median of the sample of the pixels in both @xmath151 dark regions and stores this in a residual background map . after forming the background estimates for all microlenses , the resulting map is smoothed with a box median filter and stored for use in the inner extraction loop . our cube extraction method is summarized in figure [ fig : extraction ] . after forming the background map , the extraction routine loops through microlens indices @xmath95 and @xmath96 . for each microlens , we retrieve the parameters from the global spectrograph focal plane solution : position , height , tilt , and amplitude ( represented by the variables @xmath152 , @xmath153 , @xmath154 , and @xmath120 , respectively ) . an inner loop steps through 23 wavelength channels in 0.03 @xmath3 m increments between @xmath155 m and 1.76 @xmath3 m . we index these channels by integer values of @xmath57 ( @xmath59 ) , and determine the extraction target point for each cube element , or _ spaxel _ , as follows ( cf . equation set [ eqn : specxyparam ] ) : @xmath156 where @xmath140 and @xmath141 are the crude horizontal and vertical offsets , and @xmath138 and @xmath157 are the fine horizontal and vertical offset indices determined by the registration algorithm ( [ sec : specreg ] ) for the current reduced spectrograph image , @xmath126 . we use a hat symbol to designate the same coordinates rounded to the nearest pixel center : @xmath158 . the spaxel for each microlens and wavelength combination is based on the weighted sum over a @xmath159 detector pixel square centered on @xmath158 : @xmath160 the weights @xmath161 applied to the detector samples are based on the psf model , downsampled and truncated to the @xmath159 pixel extraction box as follows : @xmath162 defined for @xmath163 . the formulas for the @xmath4- and @xmath5-band psfs , @xmath164 and @xmath165 , as well as the intra - pixel response , @xmath21 , can be found in [ sec : psfmodel ] . the integers @xmath166 and @xmath167 encode the offsets of the extraction target point from the extraction box center : @xmath168 the resulting indices take on integer values in the range @xmath169 ( corresponding to offsets up to @xmath170 of a pixel width in each direction when @xmath60 ) . lastly , the @xmath171 factor in front of each weight formula compensates for the effect that the offset between the psf center and the extraction box center has on the sum of products in equation [ eqn : weightedsum ] . the need for this can be qualitatively understood by the fact that the overall flux in a @xmath159 pixel sample of the psf decreases when the peak is offset from the center . the @xmath171 correction factor varies from unity at perfect alignment up to 1.09 in the worst case for extreme offsets . for a given microlens , the separation between the extraction target points of the first and last channels corresponding to @xmath172m is typically about 24 detector pixels . since we use @xmath159 pixel boxes to extract a signal for each of 23 channels spanning that length , the footprints of adjacent channels necessarily overlap . we have examined the effect of this by extracting data cubes directly from the laser calibration images ( discussed in [ sec : psfmodel ] ) . from these cubes , we compared the mean flux in neighboring cube channels . the results , plotted in figure [ fig : laserresponse ] for 1.25 @xmath3 m and 1.58 @xmath3 m emission , reveal the effective filter shape of an individual data cube channel . the cube channel filters exhibit a full width half - maximum value of @xmath1 70 nm at both @xmath4 and @xmath5 band . knowledge of this profile is essential for comparisons between p1640 data and existing astronomical spectra . we can not simply bin a reference spectrum to the channel spacing ; we must also convolve it with the cube channel filter before comparing it to data cube measurements . suppose an object appears in a data cube , and we carry out channel - wise photometry to find a spectrum @xmath173 , @xmath59 . to compare this meaningfully to an established spectrum , @xmath174 , acquired by some other instrument with wavelength bin width @xmath175 , requires two steps . first , we re - bin @xmath174 to the cube channel interval , 0.03 @xmath3 m , to form an intermediate - resolution spectrum @xmath176 : @xmath177 we then convolve the intermediate - resolution spectrum @xmath178 with the cube channel filters @xmath179 and @xmath180 . the filter functions are defined to follow the profiles shown in figure [ fig : laserresponse ] for @xmath181 ( so that @xmath182 corresponds to the central peak of the filter ) , and are zero - valued outside that range . @xmath183 the ratios in front of each convolution sum compensate for the effect of the spectrograph passband edges on the filtering ( they are unity when @xmath57 is at least three channels from both passband edges ) . the resulting spectrum , @xmath184 , is smoothed to the same resolution as the cube - derived spectrum @xmath173 . if @xmath185 has been corrected for the p1640 spectral response , then the two spectra can be directly compared , apart from some scale factor . if on the other hand , @xmath173 is a `` raw '' cube spectrum , with undetermined spectral calibration , and @xmath184 refers to the same source , then it can be used to correct @xmath173and , in general , any p1640 data cube , as described next . we characterize the wavelength - dependent sensitivity of p1640 , or spectral response function , by comparing the `` raw '' data cube count values of an unocculted reference star ( observed off - axis from the coronagraph focal plane mask ) with its established near - infrared spectrum . in practice , we have chosen stars with spectra archived in the nasa infrared telescope facility ( irtf ) spectral library to calibrate our response @xcite . the spectral response function is determined by dividing the re - binned , smoothed reference spectrum ( @xmath184 , in the notation above ) by the spectrum of the same source derived from a p1640 data cube . we measure the signal of the observed reference star by carrying out aperture photometry on each channel image making up the data cube , enclosing the third airy ring . to capture the wavelength - dependent scaling of the coronagraph psf , we linearly increased the photometric aperture radius from 13 to 20 spaxels across the passband . the resulting response curve for one calibration star , hd 75555 ( @xmath186 ; spectral type f5.5iii - iv ) , is shown in figure [ fig : response ] . the response curve shows the expected roll - off at the edges of the operating range due to telluric water absorption features . the valley centered near @xmath187 m is likewise due to water absorption between @xmath4 and @xmath5 band . the overall climb in the response towards longer wavelengths is caused by the wavelength - dependence of three effects in combination : the energy per photon as dictated by the plank relation , @xmath188 ; detector quantum efficiency ; and the transmission of the blocking filter at the ifu entrace . we normalize the spectral response function to its mean value before storing it for general application to data cubes . the cube extraction pipeline loads one of these mean - normalized spectral response functions into memory before beginning the cube extraction routine . if the appropriate switch is set by the user , the pipeline will divide the spaxel values in each channel image by the corresponding response function value . uncertainty in the global spectrograph focal plane solution , the image registration , and the weighting function combine to contribute a pseudo - random error to each cube point , at a level of @xmath1 5% of the spaxel value . we extracted cubes from the moon calibration images to estimate the magnitude of this error . since spatial amplitude variations are compensated for during extraction , ideally any given spatial cross - section of a cube extracted from a calibration image would appear flat . instead , we observe a standard deviation of 3% , with some variation between channels and areas of the image . there are also systematic errors caused by light from adjacent microlenses overlapping on the focal plane , which we refer to as cross - talk . the horizontal space between spectra on the focal plane is 3.3 detector pixels , and yet we know from the spectrograph psf model ( figure [ fig : psf_cross_sections ] ) that about 10% of the downsampled psf flux falls outside the central three columns . as a result , a small fraction of light from one microlens is inevitably counted during the extraction of a neighboring spectrum . consider the spectrograph image cutout shown in figure [ fig : extraction ] . for each channel of a given spectrum , you can attribute the dominant contamination to a different channel belonging to the neighboring spectrum positioned either above or below along the microlens column . we know from examining the laser calibration data cubes that the upper limit of the flux incorrectly extracted into a cube point is @xmath1 5% of the cube value in a neighboring spectrum . in other words , suppose @xmath189 is the cube point whose contamination we are trying to assess . based on our spectrum model , we can determine that some channel @xmath190 of microlens @xmath191 is the dominant contamination source for channel @xmath192 of microlens @xmath193 . therefore , we estimate the cross - talk error @xmath194 , added in quadrature with the uncertainty described earlier . using a table of established cross - talk channel pairs , we can repeat this error estimate for each channel of a spectrum of interest . the spectral response function plotted in figure [ fig : response ] and the source spectrum plotted in figure [ fig : titan ] reflect this analysis . the pcxp stores the reduced data output in fits files , and organizes them in a directory tree by object and date . three channels from an example cube based on an occulted star observation are displayed in figure [ fig : examplecube ] . in addition to the normal cube extraction described in [ sec : cubeextraction ] , there are several other products the pipeline derives from the raw data . from the brightest stars , there is enough signal recorded in a single 7.7 second read to form a cube without using the full exposure time . one option of the pipeline takes advantage of this , checking if the @xmath195-band magnitude is less than 2.0 , and if so then making cubes from each pair of consecutive reads in the ndr sequence . the resulting `` read - wise '' cubes have a speckle pattern resolved to a higher time resolution , which may eventually be exploited to improve speckle suppression . at the opposite time scale , the pipeline can form cubes from the mean of all spectrograph images acquired on the same data of a given target . the pipeline also forms `` collapsed '' images by summing all the channel images of a cube , as well as the subsets of channels corresponding to the @xmath4 and @xmath5 bands . to demonstrate the efficacy of our data extraction and calibration procedures , we apply them here to an observation of saturn s moon titan acquired on 2009 march 15 . after locking the ao system on titan , its image was positioned off - axis from the focal plane mask so that no part of the @xmath196 disk was occulted by the coronagraph . we used the pipeline to generate a data cube from a single 138 s exposure , calibrating the relative channel fluxes with the response function shown in figure [ fig : response ] . next , we averaged 1900 spaxels inside the resolved disk of titan . after normalizing the disk - averaged spectrum to the mean channel flux , we compared it to unpublished data obtained by emily schaller two days earlier using the spex near - infrared spectrograph @xcite at the nasa infrared telescope facility ( irtf ) . following the procedure in [ sec : speccal ] , we re - binned and smoothed the irtf spectrum to match the resolution of the data cube ( equations [ eqn : rebinspec][eqn : convspec ] ) . the resulting spectra are plotted in figure [ fig : titan ] . the near - infrared spectrum of the moon is marked by a series of broad ch@xmath197 absorption troughs @xcite . at the two albedo peaks in our passband , 1.3@xmath198 m and 1.6@xmath198 m , titan s atmospheric opacity is low enough for the direct reflection of sunlight off the water ice surface to constitute the observed flux , rather than diffuse scattering in its stratospheric haze @xcite . we expect the disk - averaged spectra acquired on these two dates to be similar . as titan rotates over a 16-day period , in synchronicity with its orbit around saturn , the near - infrared albedo observed from earth ( through the 1.3@xmath198 m and 1.6@xmath198 m methane `` windows '' ) varies in a cycle with an amplitude on the order of 10% . this variation is caused by a change in surface features between the leading and trailing hemispheres @xcite . however , despite the 45 degree rotation of titan with respect to earth between the irtf and p1640 observations , previous monitoring by several investigators indicates no significant albedo change between our specific pair of planetographic longitudes ( @xmath199 and @xmath200 ) @xcite . furthermore , long - term monitoring of titan s albedo only occasionally reveals deviations from predicted reflectivity due to transient cloud features ( e.g. * ? ? ? * ; * ? ? ? . the anticipated resemblance of the two spectra is confirmed in figure [ fig : titan ] : the average absolute difference between the irtf and p1640 titan data over the 19 channels used in figure [ fig : titan ] is 7% of the mean flux , and most flux points agree within the error bars of the p1640 data . the only channel flux showing significant disparity with the irtf data , centered at 1.13 @xmath3 m , is located near the edge of a telluric water absorption trough , and therefore is more susceptible to calibration errors than most channels on the plot . for examples of m dwarf stellar spectra that have been measured from p1640 data cubes , see @xcite and @xcite . starting from the economic constraint of limited detector area , the design of any integral field spectrograph must reach compromises between the competing parameters of spatial resolution , field of view , spectral resolution , and spectral range . for project 1640 , the need to nyquist - sample the starlight speckle pattern inside the angular extent of the adaptive optics system `` control radius '' largely determined the balance of these tradeoffs . the other major factor was the overarching science goal of distinguishing astrophysically interesting features in the spectral energy distributions of young giant exoplanet atmospheres . based on these considerations , the p1640 collaboration concluded on an ifu design that strongly favored a high density of spatial elements over spectral resolution , to a greater extreme than previous microlens - based integral field spectrographs . for example , the broadband mode of the tiger ifu had 572 spatial elements dispersed at spectral resolution @xmath201 , and the broadband mode of the osiris ifu has 1024 spatial elements dispersed at @xmath202 @xcite . p1640 , by comparison , has 38,000 spatial elements with @xmath203 , corresponding to over a factor of 30 increase in spatial elements , and a similarly substantial reduction in spectral resolution . only two other ifus will join p1640 in this operating regime in the next year : gpi and vlt - sphere both designed for exoplanet imaging . since the properties of p1640 data are unusual even in the context of preceding ifus , the instrument commissioning has required development of novel extraction and calibration approaches . consider that each detector pixel width spans approximately 27.5 nm in spectral dispersion4% of the instrument passband and that the physical length of an individual microlens spectrum s footprint is merely 0.5 mm at the focal plane . as such , great care has been required to accurately map the spectrograph focal plane data , in a manner that is resilient to subtle long - term changes in optical alignment . we described an answer to this problem in [ sec : focplmodel ] , using a hierarchical fitting procedure to build a comprehensive , epoch - specific model of the full spectrograph focal plane . the last obstacle to securing the layout of the data , instrument flexure , must be dealt with individually for each exposure . therefore , unlike the case of building the spectrograph focal plane model , we register each science spectrograph image `` on the fly '' inside the data pipeline , as explained in [ sec : specreg ] . the relatively long execution time required to build the spectrograph focal plane solution ( @xmath1 12 hours on a single high - performance workstation ) means that it is sometimes necessary to rely on an outdated solution to extract data . such will be the case , for example , during the first night of an observing run when the calibration moon / sky image has not yet been acquired and fit . we know from the geometric evolution illustrated in figure [ fig : distortion ] that consequent errors in the positions of the spectra , along with other properties , will inevitably degrade the quality of the cube . however , for a preliminary inspection of data , and to check on instrument performance , the result will usually be acceptable . recall that the spectrograph image registration uses a region of the image with the strongest signal to align the model . therefore , under circumstances where the solution is old , the extraction will still tend to be fairly accurate near the brightest region of the image , but will progressively worsen towards the outskirts of the focal plane . the main advantages of our weighted sum approach to ifu spectrum extraction , described in [ sec : weightedsum ] , are simplicity and speed . our reasoning behind using the spectrograph psf itself to shape the weighting function is that it mimics the cross - dispersion profile of the spectrum . this way , along the middle horizontal row of the @xmath159 extraction box , we weight the detector samples by an estimate of their relative intensity ( see figure [ fig : extraction ] for an example ) . @xcite originally devised this strategy for extracting coarsely - sampled ccd spectra . he demonstrated that matching the weights to the expected cross - dispersion profile optimizes the fidelity of the extraction in the case where read noise is prevalent . this makes sense intuitively because we want detector samples with high count rates to have more influence than weak , noisy ones . unlike the case of horne s spectrograph , however , we are dealing with many closely packed spectra , and so we are forced to truncate our weighting area to a region smaller than the actual extent of the psf . we attempt to account for this in the weighting formula in equation [ eqn : weights ] . since this correction depends not just on the shape of the psf but also the target spectrum , it is necessarily only an estimate , and can contribute errors on the order of a few percent to the spaxel value . as compared to the cross - dispersion axis , for the dispersion axis there is more freedom in the choice of extraction weights . the tradeoffs here are signal - to - noise ratio per channel versus spectral resolution . ultimately , for any weighted sum approach , the spectral resolution is limited by the width of the monochromatic ifu response along the dispersion axis . the spectrograph psf fitting result plotted in the right - hand panel of figure [ fig : psf_cross_sections ] shows this is about 2 detector pixels for p1640 , which translates to @xmath1 50 nm in the dispersion direction . for simplicity , we chose to remain with the psf again to set the weights ; the resulting cube spectra have a resolution of about 70 nm ( @xmath204 ) . in the future , it would be worthwhile to experiment with a hybrid weighting function that combines the cross - dispersion profile of the psf with a different vertical profile , to investigate how much the spectral resolution can be improved . as one example of an alternative , @xcite propose extracting spectra from the gpi ifu by summing strictly along a single row / column of pixels in the cross - dispersion direction . another possible improvement to our data pipeline is a completely different extraction approach based on deconvolution or fitting . a well - designed fitting algorithm might disentangle the flux contributions of wavelengths with overlapping footprints . we have done a few experiments in this direction for example , we applied the same mpfit - based algorithm we used to model the microlens spectra in the moon / sky calibration image ( [ sec : spectrummodel ] ) to a science image . the results were of significantly poorer quality than our normal data cubes , partly as a consequence of not having the luxury to average the fit spectra over many microlenses , as we do with calibration images . in another program , two of our co - authors created a program that fits each spectrum cutout as a train of scaled psfs , each one representing a different channel . while its implementation is not complete , this method shows promise of forming data cubes with slightly higher spectral resolution than the existing extraction . since any fitting approach is inherently much slower than a weighted sum translation , one can imagine two cube extraction algorithms co - existing for different purposes : one as an offline procedure reserved for images of the greatest interest , and the other program applied to all p1640 data . we have developed a collection of algorithms to reduce the data acquired by p1640 , a coronagraphic integral field spectrograph designed for high contrast imaging . our aim has been to describe our data pipeline software in enough detail that upcoming microlens - based imaging spectrograph projects can take advantage of our experience in treating closely packed , coarsely sampled spectra . an essential element of our approach is an empirical model of the spectrograph focal plane image , based on calibration exposures in which the entire microlens array is illuminated , in turn , by broadband and monochromatic light . to derive a solution specific to each observation epoch , we fit a set of parameters describing each microlens spectrum ( position , tilt , height , and overall signal amplitude ) . we use the resulting table of solved parameters to determine the extraction location on the focal plane for any given combination of microlens and wavelength , and hence build the data cube . we implement the cube extraction with a weighted sum that optimizes the signal - to - noise ratio by mimicking the expected cross - dispersion profile , as constrained by the sub - pixel spectrograph image registration . sources of error in the final data cube are cross - talk between adjacent microlens spectra , uncertainty in the spectrograph focal plane model , uncertainty in sub - pixel registration , and uncertainty in the determination of extraction weights . nevertheless , based on an observation of saturn s moon titan , we have demonstrated our ability to retrieve strong - featured near - infrared spectra to @xmath1 5% accuracy . as our methods for handling this new form of data evolve , we expect p1640 to continue its pioneering role in high contrast astronomy . thanks are due to anand sivaramakrishnan , emily rice , michael mcelwain , james gunn , david zurek , and charles beichman for helpful discussions . we also thank emily schaller for providing us with unpublished spex / irtf titan data to serve as a reference spectrum in figure [ fig : titan ] . project 1640 is funded by national science foundation grants ast-0520822 , ast-0804417 , and ast-0908484 . part of this work was performed under a contract with the california institute of technology ( caltech ) funded by nasa through the sagan fellowship program . the members of the project 1640 team are also grateful for support from the cordelia corporation , hilary and ethel lipsitz , the vincent astor fund , judy vale , andrew goodwin , and an anonymous donor .

we study the effect that the temperature structure has on the determination of the primordial helium abundance , . we provide an equation linking(o ) , the temperature derived from the [ o ] lines , and(he ) , the temperature of the he i lines , both for h regions with o only and for h regions where a fraction of o is present . by means of(he ) , which is always smaller than(o ) , we derive the helium abundances of 5 objects with low and very low metallicity ( ngc 346 , ngc 2363 , haro 29 , sbs 0335052 , and i zw 18 ) ; these objects were selected from the literature because they include the 3 low metallicity objects with the best line determinations and the 2 objects with the lowest metallicity . from these abundances we obtain that(nhc ) , a value 0.0088 lower than that derived by using(o ) . we call this determination(nhc ) because the collisional contribution to the balmer line intensities has not been taken into account . all the recent determinations in the literature have not taken into account the collisional contribution to the balmer line intensities . by considering the collisional contribution to the balmer line intensities of these five objects we derive that(+hc) .

astro-ph0107189

the determination of the pregalactic , or primordial , helium abundance by mass @xmath0 is paramount for the study of cosmology , the physics of elementary particles , and the chemical evolution of galaxies . in this paper we present a new determination of @xmath0 based on observations of the metal - poor extragalactic h regions ngc 346 , i zw 18 , ngc 2363 , haro 29 , and sbs 0335052 . this determination is compared with those carried out by other authors . a preliminary account on the results for i zw 18 and ngc 2363 is presented elsewhere @xcite . there are three problems affecting the @xmath0 determination that need to be further analyzed : @xmath6 ) the temperature structure , @xmath7 ) the ionization structure , and @xmath8 ) the collisional excitation of the hydrogen lines . this paper will be mainly concerned with the effect of the temperature structure of the nebulae on the helium line intensities . in a future paper we will discuss in more detail the problem of the collisional excitation of the hydrogen lines @xcite . the most accurate @xmath0 values in the literature have been derived under the assumption of no contribution to the hydrogen balmer lines due to collisional excitation , we will refer to these determinations in this paper as @xmath0(nhc ) and to those that include the hydrogen collisional effect as @xmath0(+hc ) . due to the collisional contribution to the balmer line intensities the @xmath0(nhc ) determinations are lower limits to the @xmath0(+hc ) value . there are several pieces of evidence indicating that @xmath1(he ) is smaller than @xmath1(o ): @xmath6 ) observations of ngc 346 [ hereinafter paper i ] pei00z , @xmath7 ) photoionization models of giant extragalactic h regions @xcite , @xmath8 ) in photoionization models @xmath1(o ) is smaller than @xmath1(o ) for objects with @xmath1(o ) higher than 12,360 k @xcite , and a considerable fraction of the he lines originates in the o@xmath3 zone . this small difference is significative because it lowers the he@xmath3/h@xmath3 determination . by combining the observed he lines it is found that @xmath1(he ) and @xmath9(he ) are coupled in the sense that a lower @xmath1 implies a higher @xmath9 reducing the he@xmath3/h@xmath3 determination ( see paper i ) . another problem that has to be considered to determine very accurate he / h values of a given h region is its ionization structure . the total he / h value is given by : @xmath10 for objects of low degree of ionization it is necessary to consider the presence of he@xmath11 inside the h@xmath3 zone , while for objects of high degree of ionization it is necessary to consider the possible presence of h@xmath11 inside the he@xmath3 zone . for objects of low degree of ionization @xmath12 might be larger than 1.00 , while for objects of high degree of ionization @xmath13 might be smaller than 1.00 . this problem has been discussed by many authors ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? each h region is different and a good photoionization model is needed to estimate an @xmath12 of very high accuracy . in this paper we will assume that the h regions are chemically homogeneous . in section 2 we will adapt the @xmath14 and @xmath15 formalism introduced by @xcite to relate @xmath1(o ) and @xmath1(o ) with @xmath1(he ) , the full mathematical treatment is presented in the appendix . in section 3 we will rediscuss the @xmath16 determination of ngc 346 , the most luminous h region in the smc , carried out in paper i. in sections 4 to 7 we redetermine the he@xmath3/h@xmath3 values for i zw 18 , ngc 2363 , haro 29 , and sbs 0335052 , based on the observations of @xcite and @xcite and on photoionization models for these objects . these four objects were selected from their sample based on the following criteria : the two objects with the lowest heavy element abundances and the two objects with the smallest observational errors @xcite . we derive he@xmath3/h@xmath3 , @xmath9(he ) , @xmath1(he ) , and @xmath17(3889 ) self - consistently based on all the observed he line intensities , with the exception of those strongly affected by underlying absorption . by combining the he@xmath3/h@xmath3 values of these four objects with that of ngc 346 we derive a new @xmath0(nhc ) in section 8 . the collisional contribution to the balmer lines is estimated in section 9 . the discussion and conclusions are presented in sections 10 and 11 . we first consider an h region where the he , the o , and the h volumes coincide perfectly . if we further assume that the temperature is constant we have a very simple model which gives us a first approximation to a real nebula . if we now assume that the h region has an average temperature given by @xmath18 and a mean square temperature variation given by : @xmath19 we find that the observed @xmath1(4363/5007 ) , derived from the @xmath20(4363)/@xmath20(5007 ) ratio , is given by @xcite : @xmath21 \frac{t^2}{2 } \right ) \label{eto3-t0 } , \ ] ] and that the temperature associated to the he recombination lines , that originate in the he region , is ( see paper i ) : @xmath22 this means that the temperature that should be used to measure the helium abundance is given by : @xmath23 \frac{t^2}{2}\right ) \label{ethe2-to3 } .\ ] ] in figure [ fto3-the ] , and based on equation [ ethe2-to3 ] , we present @xmath1(he ) /@xmath1(o ) versus @xmath14 for different values of @xmath1(o ) . we consider now another h region where the he volume coincides with the h volume , but now oxygen is present both in the form of o@xmath2 and o@xmath3 , with the o@xmath3 weight described by the fraction @xmath24 : @xmath25 if the average temperatures in the o and o zones are given by : @xmath26 and @xmath27 respectively , then the average temperature for the whole ionized region @xmath15 is equal to @xmath28 the fractional difference between @xmath29 and @xmath15 can thus be characterized by the parameter @xmath30 so that @xmath31 and the ratio between @xmath32 and @xmath15 is : @xmath33 in the general case when the o and o zones are not of uniform temperature , but include temperature fluctuations , represented by : @xmath34 and @xmath35 respectively , the mean square temperature fluctuation for the entire h region , @xmath14 , is given by : @xmath36 were the first two terms represent the deviation from uniform temperature given by the difference in temperature between each zone and the average for the whole h region , while the other two represent the additional contribution due to inhomogeneities in each zone . substituting the first 2 terms with equations [ ebeta_2 ] and [ ebeta_1 ] and by keeping the last 2 terms to second order we obtain : @xmath37 @xmath29 and @xmath38 are related to @xmath1(o ) by ( see equation [ eto3-t0 ] ) : @xmath39 \frac{t_3 ^ 2}{2 } \right ) \label{eto3-t03 } ; \ ] ] while @xmath32 and @xmath40 are related to an o temperature , determined from the @xmath20(3727)/@xmath20(7325 ) ratio , that in the low density limit is given by @xcite : @xmath41 \frac{t_2 ^ 2}{2 } \right ) \label{eto2-t02 } .\ ] ] it is convenient to obtain an equation , with the same physical meaning than that of equations [ eto3-t03 ] and [ eto2-t02 ] , that relates @xmath15 and @xmath14 with a temperature that can be derived from observations , particularly with those of the o and o forbidden lines . intuitively we can define the following equation ( the formal derivation of equation [ eto2 + 3_t0 m ] is given in the appendix ) : @xmath42 \frac{t^2}{2 } \right ) \label{eto2 + 3_t0 m } , \ ] ] where @xmath15 and @xmath14 correspond to the values for the entire volume , @xmath43 represents an effective average over the excitation energy needed to produce the oxygen lines , @xmath44 and @xmath45 is an average between @xmath1(o ) and @xmath1(o ) . however , it can be shown ( see appendix ) that , the simple average of the temperatures weighted by @xmath24 and @xmath46 does not give an equation accurate to second order and that @xmath47 needs a factor @xmath48 of the form : @xmath49\gamma \label{eto2 + 3def1 } , \ ] ] @xmath48 is very close to unity for objects where @xmath50 or where @xmath1(o ) is similar to @xmath1(o ) ( @xmath51 ) ; but for h regions where there are appreciable fractions of both o@xmath3 and o@xmath2 and where there is an appreciable difference between @xmath1(o ) and @xmath1(o ) , @xmath48 must be calculated as described in the appendix . ( all computations done in this paper were done using the formalism described in the appendix . ) finally , solving equation [ eto2 + 3_t0 m ] for @xmath15 and substituting it in equation [ ethe2-t0 ] it follows that @xmath52 \frac{t^2}{2}\right ) \label{ethe2-to2 + 3 } , \ ] ] which can now be used to determine the he abundances . to derive @xmath47 from the observed @xmath1(o ) and @xmath1(o ) values we need to use equations [ ealpha ] , [ etheta ] , and [ eto2 + 3def1 ] for @xmath53 ; for @xmath48 larger than one we also need to use equations [ egamma ] , [ edelta_1 ] , and [ egamma ] . if only @xmath1(o ) is available we need a photoionization model to obtain a relationship between @xmath1(he ) and @xmath1(o ) . in figure [ fto2 + 3 ] we present the relationship between @xmath1(he ) and @xmath1(o ) as a function of @xmath14 for objects with different o@xmath3 fractions . figure [ fto2 + 3 ] is based on equations [ ealpha ] , [ etheta ] , [ eto2 + 3def1 ] , [ egamma ] , [ edelta_1 ] , and [ egamma ] and the relationship @xmath54 derived from the photoionization models by @xcite . from this equation it is obtained that : for @xmath29 = 12,360 k then @xmath55 , for higher temperatures @xmath56 , and for lower temperatures @xmath57 . from figure [ fto2 + 3 ] and equation [ estasinska ] we note the following cases for objects with @xmath1(o ) @xmath58 12,360 k : @xmath6 ) for @xmath59 and @xmath60 , then @xmath1(he ) = @xmath1(o ) and @xmath61 . @xmath7 ) for @xmath24 @xmath62 and @xmath63 , then @xmath1(he ) is _ always _ smaller than @xmath1(o ) and @xmath14 is _ always _ larger than 0.000 . @xmath8 ) for @xmath59 , and @xmath64 , then @xmath1(he ) @xmath65 @xmath66(o ) and @xmath67 . @xmath68 ) for @xmath24 @xmath62 and @xmath69 , then @xmath1(he ) is even smaller than in case @xmath7 and @xmath14 is even larger than in case @xmath7 . very metal - poor extragalactic h regions , those that have been used to derive the primordial helium abundance , have @xmath1(o ) larger than 12,360 k and are best represented by case @xmath68 . figure [ fto2 + 3 ] is also related to the averaged quantities block of cloudy s output @xcite . in that section of cloudy s output there is a line labeled `` peimbert '' with the following entries : `` t(o iiir ) '' that corresponds to @xmath1(o ) in this paper ; `` t(bac ) '' is the hydrogen temperature resulting from the predicted balmer jump and h@xmath30 for the optically thick case , corresponds to the outer surface of the object and does not represent typical h regions ; `` t(hth ) '' is the hydrogen temperature resulting from the predicted balmer jump and h@xmath30 for the optically thin case , corresponds to @xmath1(bac ) defined by @xcite and used by many authors , and applies to typical h regions ; `` t@xmath70(hstr ) '' is the structural value computed across the h region , and it is called @xmath14 in this paper . `` t(o3-bac ) '' and `` t@xmath70(o3-bc ) '' are the average temperature and the mean square temperature fluctuation values derived from the predicted @xmath1(o ) and @xmath1(bac ) values , they correspond to @xmath15 and t@xmath70(hstr ) only when all the o is in the o@xmath2 stage . in the presence of o@xmath3 and according to figure [ fto2 + 3 ] and equation [ estasinska ] we can distinguish two cases : when @xmath1(o ) is higher than 12,360 k then `` t@xmath70(o3-bc ) '' is higher than @xmath14 , but when @xmath1(o ) is smaller than 12,360 k then `` t@xmath70(o3-bc ) '' is smaller than @xmath14 . from photoionization models of h regions it has been found that @xmath14 is in the 0.002 to 0.03 range , with typical values around 0.005 ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) ; from observations it has been found that @xmath14 is in the 0.01 to 0.04 range , with typical values around 0.03 ( e.g. * ? ? ? * ; * ? ? ? * and references therein ) . as a first approximation due to equation [ estasinska ] we expect minimum @xmath14 values from photoionized models with @xmath71 12,360 k and an increase in @xmath14 for hotter and cooler models . most of the observational determinations of @xmath14 come from h regions with @xmath1(o ) @xmath72 8000 k which are biased toward one of the higher ends of the @xmath14 distribution . very metal - poor h regions are biased to the other high end of the @xmath14 distribution , particularly those with wr stars and a large fraction of o@xmath3 . the @xmath14 values derived from photoionization models correspond to @xmath1(he ) values 3% to 11% smaller than @xmath1(o ) . in the presence of additional sources of energy , to those provided by photoionization , the difference between @xmath1(he ) and @xmath1(o ) might be larger . in paper i we have analyzed ngc 346 , the most luminous h region in the small magellanic cloud . due to the relatively small distance to ngc 346 it is possible to place the observing slit avoiding the brightest stars , thus minimizing the effect of the underlying absorption in the he@xmath3/h@xmath3 determination . the determination of @xmath0 based on ngc 346 can have at least four significant advantages and one disadvantage with respect to those based on distant h region complexes : @xmath6 ) no underlying absorption correction for the helium lines is needed because the ionizing stars can be excluded from the observing slit , @xmath7 ) the determination of the helium ionization correction factor can be estimated by observing different lines of sight , @xmath8 ) the accuracy of the determination can be estimated by comparing the results derived from different lines of sight , @xmath68 ) the electron temperature is generally smaller than those of metal poorer h regions reducing the effect of collisional excitation from the metastable 2@xmath73 s level of he and from the ground level of h , and @xmath74 ) the disadvantage is that the correction due to the chemical evolution of the smc is in general larger than for the other systems . from the observations of the he lines @xmath75 3889 , 4026 , 4387 , 4471 , 4922 , 5876 , 6678 , 7065 , and 7281 ( see table [ tlines ] ) and from the temperatures associated with the o lines and the balmer discontinuity in region a we have used a maximum likelihood method to determine simultaneously and self - consistently several parameters . our results are : @xmath9(he ) @xmath76 , @xmath1(he ) @xmath77 , @xmath78 , and @xmath17(3889 ) = 0.0 . for this temperature range we obtain that o / h@xmath79 . in table [ tchi ] we present @xmath80 and @xmath81 as a function of several @xmath9(he ) and @xmath1(he ) values , which include a few representative temperatures and the densities favored by those temperatures . from this table it can be seen that the minimum @xmath81 values imply a strong correlation between the temperature and the density , in the sense that the lower the temperature , the higher the density , and consequently the lower the @xmath80 value . in paper i it is also found that @xmath66(he ) is @xmath82 smaller than @xmath66(o ) . the parameters that represent this model are presented in table [ tparameters ] . in table [ t346 ] we present the @xmath16(nhc ) values for different temperatures and densities , where we have included the he@xmath2/h@xmath3 ratio ( see paper i ) and an @xmath83 . @xcite have computed photoionization models for ngc 346 with cloudy , they have found : @xmath6 ) that there is a negligible amount of h@xmath11 inside the he@xmath3 zone , in agreement with paper i , @xmath7 ) that @xmath1(he ) is 5% smaller than @xmath1(o ) in their best model , and @xmath8 ) that the @xmath1(o ) computed by the models is 9% smaller than the observed value , indicating the presence of additional heating sources not included in the photoionization models and probably implying that the real difference between @xmath1(he ) and @xmath1(o ) is higher than predicted by the models . up to now , with the exception of ngc 346 , it has not been possible to derive the @xmath12 , @xmath9(he ) , @xmath1(he ) , @xmath17(3889 ) , and the he@xmath3/h@xmath3 ratio based only on the helium lines . usually the observed @xmath1(o ) value has been used to complement the information provided by the he lines , moreover it has also been assumed that the observed @xmath1(o ) is equal to @xmath1(he ) . for i zw 18 , sbs 0335052 , ngc 2363 , and haro 29 we will make use of the observed @xmath1(o ) , but we will not assume that it is equal to @xmath1(he ) , instead we will make use of photoionization models computed with cloudy to relate both quantities . in general photoionization models predict @xmath1(o ) values smaller than observed @xcite indicating the possible presence of an additional heating source not considered by the models ; this result also implies that the @xmath1(he ) values predicted by the models will be smaller than those @xmath1(he ) values derived from the observed @xmath1(o ) values . since the lower the temperature , the lower the derived he / h value ( see for example table [ tzw ] ) , the use of the temperatures predicted by the models yields spuriously low he / h values . on the other hand the models can give us an estimate of @xmath14(o ) , probably a lower limit . the formalism presented in section [ stemp ] , together with @xmath14(o ) , provides a relationship between the observed @xmath1(o ) and @xmath1(he ) . from photoionization models of i zw 18 based on cloudy @xcite we find that : @xmath83 , @xmath84 , and @xmath14(o ) @xmath85 . we consider @xmath14(o ) as a lower limit to @xmath14 ( see equations [ et2 ] and [ et22t32 ] ) . from the formalism of section [ stemp ] together with @xmath86 and @xmath1(o ) @xmath87 k @xcite we obtain @xmath88 and consequently that @xmath1(he ) is 10.2% smaller than @xmath1(o ) . to derive the he@xmath3/h@xmath3 ratio we have used the maximum likelihood method , mlm ( see paper i ) . the inputs are : @xmath6 ) @xmath84 , @xmath7 ) @xmath88 , @xmath8 ) @xmath1(o ) @xmath87 k and @xmath68 ) the observations of @xmath75 3889 , 4026 , 4471 , 5876 , 6678 , and 7065 by @xcite , where we have adopted the line intensity ratios presented in table [ tlines ] , corrected by underlying absorptions of 3.3 , 0.8 , and 0.8 for @xmath75 3889 , 4026 and 4471 respectively , that correspond to a burst model of 4 myr @xcite . the underlying absorptions predicted by the burst model have been reduced by about 10% to take into account the presence of the nebular continuum emission ; a similar reduction has been applied to the underlying absorptions predicted by the starburst models for ngc 2363 , haro 29 , and sbs 0335052 . to obtain he@xmath3/h@xmath3 values we need a set of effective recombination coefficients for the he and h lines , the contribution due to collisional excitation to the helium line intensities , and an estimate of the optical depth effects for the helium lines . the recombination coefficients that we used were those by @xcite for h , and @xcite for he . the he collisional contribution was estimated from @xcite and @xcite . the optical depth effects in the he triplet lines were estimated from the computations by @xcite , and based on paper i we will assume that the he singlet lines are produced under case b. the maximum likelihood solution amounts to @xmath6 ) @xmath9(he ) @xmath89 , @xmath7 ) he@xmath3/h@xmath90 , @xmath8 ) @xmath91 , and @xmath68 ) @xmath66(he ) @xmath92 , 10.2% smaller than @xmath1(o ) ( see table [ tparameters ] ) . for this temperature range we obtain that o / h @xmath93 . there is not enough information to independently derive the @xmath14 and @xmath17(3889 ) values from the observed lines in this object , therefore their output values are equal to the input values . table [ tzw ] presents the @xmath16(nhc ) values for four @xmath14 values : 0.024 ( the optimum value ) , 0.030 ( + 1@xmath94 ) , 0.018 ( - 1@xmath94 ) , and 0.006 ( the minimum possible value given by the size and temperature of the o@xmath3 zone ) ; and 6 different densities that include those provided by the mlm for each @xmath14 . the @xmath16(nhc ) values include the contribution due to he@xmath2 and an @xmath83 . the @xmath1(o ) values for the four objects observed by @xcite and @xcite were derived by us from the observed line intensities , we obtain the same results for i zw 18 and sbs 0335052 than @xcite , but we obtain values 650 k higher for ngc 2363 and haro 29 than @xcite . @xcite produced detailed photoionization models of ngc 2363 , they find also that the @xmath1(o ) predicted by the models is considerably smaller than 15,750 k , the observed value ; from their models they find also that @xmath14 is in the 0.007 to 0.029 range . this amounts to a @xmath1(he ) 6% @xmath95 2% smaller than @xmath1(o ) . from the models of ngc 2363 and the slit used by @xcite we also find that : @xmath96 , indicating the presence of neutral hydrogen inside the ionized helium region ( see equation [ eicf ] ) , and @xmath97 . support for the relative high density comes from the @xmath9[s ] derived by @xcite and by @xcite that amount to @xmath98 and @xmath99 respectively and the @xmath9 [ ar ] values in the 400 to 800 @xmath100 range derived by @xcite and the @xmath101 value derived by @xcite ; the [ s ] lines trace the outer parts while the [ ar ] lines trace the inner parts of the h region . for the mlm we use the following inputs : @xmath6 ) @xmath102 , @xmath7 ) @xmath103 , @xmath8 ) @xmath1(o ) @xmath104 k , and @xmath68 ) the observations of @xmath75 3820 , 3889 , 4026 , 4387 , 4471 , 5876 , 6678 , 7065 , and 7281 by @xcite slightly modified to take into account underlying absorptions of 0.3 , 1.8 , 0.4 , 0.2 , and 0.4 for @xmath75 3820 , 3889 , 4026 , 4387 , and 4471 respectively ( see table [ tlines ] ) , that correspond to a burst model of 3 myr @xcite . we did not consider @xmath1054922 because its intensity was many @xmath94 away from the solution increasing @xmath81 to unacceptable values ; we also adopted an error for @xmath1053889 three times larger than the one given by @xcite due to our estimate of the combined errors produced by the reddening correction and the underlying absorption correction . with these inputs we obtain @xmath6 ) @xmath9(he ) @xmath106 , @xmath7 ) @xmath107 , @xmath8 ) @xmath108 , @xmath68 ) he@xmath3/h@xmath109 , @xmath74 ) @xmath110 , and @xmath111 ) @xmath1(he ) @xmath112 , a value 6.5% smaller than @xmath1(o ) ( see table [ tparameters ] ) . note that the @xmath14 and @xmath113 output values are not the same as the input values . this is because the quality of the measurements for ngc 2363 is better than that of i zw 18 , and thus the values of @xmath14 and @xmath113 can be directly estimated from the lines , however these values are not as accurate as those of ngc 346 and the error bars would be very large , therefore the observed line intensities get mixed with the @xmath14 and @xmath113 input values in the mlm , giving us a different output . for this temperature distribution we obtain o / h@xmath114 . table [ t2363 ] presents the @xmath16(nhc ) values for four @xmath14 values : 0.021 ( the optimum value ) , 0.031 ( + 1@xmath94 ) , 0.011 ( - 1@xmath94 ) , and 0.001 ( the minimum possible value given by the size and temperature of the o@xmath3 zone ) ; and 6 different densities that include those provided by the mlm for each @xmath14 . the @xmath16(nhc ) values include the contribution due to he@xmath2 and an @xmath96 . haro 29 , also known as i zw 36 or markarian 209 , is one of the two blue compact galaxies with the smallest observational errors in the list of @xcite , the other being ngc 2363 . from photoionization models of haro 29 based on cloudy by @xcite and the slit used by @xcite it is again found that the @xmath1(o ) values predicted by the models are in the 14,550 to 15,100 k range , values considerably smaller than the observed value of 16,050 k ; from these models it is also found that @xmath115 0.995 - 0.996 , @xmath116 , and @xmath117 0.002 -0.004 . for the mlm we use the following inputs : @xmath6 ) @xmath118 , @xmath7 ) @xmath119 , and @xmath8 ) the observations of @xmath75 3820 , 3889 , 4026 , 4387 , 4471 , 5876 , 6678 , 7065 , and 7281 by @xcite slightly modified to take into account underlying absorptions of 0.4 , 2.2 , 0.6 , 0.3 , and 0.6 for @xmath75 3820 , 3889 , 4026 , 4387 , and 4471 respectively ( see table [ tlines ] ) , that correspond to a burst model of 3.2 myr @xcite . we did not consider @xmath105 4922 because its intensity was many @xmath94 away from the solution increasing @xmath81 to unacceptable values ; we also adopted errors for @xmath753889 and 7065 three and two times larger than the one given by @xcite due to our estimate of the combined errors produced by the reddening correction and the underlying absorption correction . from these inputs we obtain : @xmath9(he ) @xmath120 , @xmath121 , @xmath122 , and he@xmath3/h@xmath123 . the density result is of great concern , not only is the lower limit unphysical , but the derived value is similar to the root mean square density , @xmath124 , that amounts to @xmath125 . giant h regions usually show @xmath126 values considerably smaller than those densities determined through a forbidden - line ratio or from the he lines based on the maximum likelihood method , @xmath127 ; to quantify the difference the filling factor @xmath128 is used and is defined by means of the relation : @xmath129 typical values of @xmath128 for giant h regions are in the 0.001 to 0.1 range ( e.g. paper i ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) that would imply a @xmath9(he ) in the 149 to 1490 @xmath100 range . since the density derived from the mlm implies an @xmath130 , an unacceptable value , we decided to make another estimate of @xmath9(he ) . from the cloudy models by @xcite we found that a good fit to the observed line intensities is obtained with @xmath131 , which corresponds to an @xmath9(he ) @xmath132 . fixing this density as our preferred value we obtain @xmath133 , @xmath134 ( see table [ tparameters ] ) , and he@xmath3/h@xmath135 , other values can be obtained or interpolated from table [ th29 ] . the @xmath16(nhc ) values include the contribution due to he@xmath2 and an @xmath136 . for this temperature distribution we obtain o / h@xmath137 . after i zw 18 , sbs 0335052 is the extragalactic h region with the second lowest metallicity known ; and being brighter than i zw 18 makes it critical in determining @xmath0 . we have computed a series of cloudy models for sbs 0335052 @xcite . we find that the @xmath1(o ) predicted by the models is @xmath1(o ) = 17,500 - 18,500 k , again considerably smaller than the observed value of 20,500 k. for the three central values of the observing slit used by @xcite ( center , 0.``6sw , 0.''6ne ) we have extracted from the models the following predictions : @xmath138 , @xmath139 , and @xmath14 in the 0.010 to 0.015 range . to correct for underlying absorption the blue helium lines and the balmer lines observed by @xcite we adopted the values predicted by @xcite for a 3 myr old instantaneous burst with @xmath140 . the corrections for @xmath75 4026 , 4471 , 4922 , 3889 and h@xmath30 amount to 0.4 , 0.4 0.3 1.8 and 2.0 respectively . note that @xcite corrected the balmer lines for an underlying absorption of 0.2 and that they did not correct the helium lines . we adopted the @xmath141(h@xmath30 ) value determined by @xcite that is mainly based on the @xmath20(h@xmath24)/@xmath20(h@xmath30 ) ratio , this ratio is not affected by underlying absorption ( but it might be affected by collisional excitation of the balmer lines , see below ) . the line ratios are presented in table [ tlines ] . from the mlm and the following inputs : @xmath6 @xmath142 , @xmath7 ) @xmath119 , and @xmath8 ) the modified observations of @xmath75 3889 , 4026 , 4471 , 5876 , 6678 , and 7065 ( @xmath105 4922 was not considered because its intensity was many @xmath94 away from the solution increasing the @xmath81 to unacceptable values ) ; we obtain @xmath9(he ) @xmath143 , he@xmath3/h@xmath144 , @xmath145 , and @xmath146 , this @xmath14 corresponds to a @xmath1(he ) @xmath147 ( see table [ tparameters ] ) . for this temperature distribution we obtain o / h@xmath148 . table [ t0335 ] presents the @xmath16(nhc ) values for four @xmath14 values : 0.021 ( the optimum value ) , 0.028 ( + 1@xmath94 ) , 0.014 ( - 1@xmath94 ) , and 0.004 ( the minimum possible value given by the size and temperature of the o@xmath3 zone ) ; and 6 different densities that include those provided by the mlm for each @xmath14 . the @xmath16(nhc ) values include the contribution due to he@xmath2 and an @xmath138 . in table [ tcomp ] we compare the @xmath16(nhc ) maximum likelihood values computed in this paper for @xmath149 ( see tables [ t346]-[t0335 ] ) with the @xmath16(nhc ) values for @xmath150 computed by @xcite . the differences in the @xmath16 values between both types of determinations are of the order of 0.0070 , and are manly due to the treatment of the temperature . the errors that we quote are larger than those by izotov and colaborators , both groups include the errors in the observed line intensity ratios , however we also include the uncertanties on the @xmath9(he ) , @xmath14 , and @xmath113 values while they do not . none of the @xmath16(nhc ) errors include the uncertanties due to the collisional excitation of the balmer lines ( which will be discussed in the next section ) nor the prescence of other possible systematic effects . in figure [ fy - oh ] we also present the @xmath16(nhc ) versus o / h diagram for all the objects . to determine the @xmath0(nhc ) value for all the objects it is necessary to estimate the fraction of helium present in the interstellar medium produced by galactic chemical evolution . we will assume that @xmath151 where @xmath152 is the oxygen abundance by mass . the @xmath153 baseline given by the five objects in the sample is very small and consequently produces large errors in the @xmath154 determination , therefore we decided to adopt @xmath155 , the slope derived in paper i. from this slope and the five @xmath16(nhc ) values for @xmath149 presented in table [ tcomp ] we derive @xmath0(nhc)@xmath156 . in table [ tcomp ] we present the @xmath0(nhc ) determination derived from the @xmath16(nhc ) values for @xmath150 obtained by @xcite and @xcite for haro 29 , ngc 2363 , i zw 18 , and sbs 0335052 , and find that our result is 0.0088 smaller than theirs . the main difference is due to our use of @xmath1(he ) instead of @xmath1(o ) used by them . @xcite were the first to estimate the collisional contribution to the balmer lines and its effect on the determination of @xmath0 ; they made a crude estimate of this effect for i zw 18 and concluded that the contribution to @xmath20(h@xmath24 ) _ may _ be roughly 2% . all the subsequent determinations of @xmath0 in the literature have been derived under the assumption of no contribution to the hydrogen balmer lines due to collisional excitation , we have referred to these determinations in this paper as @xmath0(nhc ) and as @xmath0(+hc ) to those primordial helium abundance determinations that include the collisional contribution effect . further discussion of the relevance of the collisional contribution to the balmer lines is presented elsewhere @xcite . from observations it is not easy to estimate this effect for two reasons : small changes are expected in the balmer line ratios , and the increase in @xmath20(h@xmath24)/@xmath20(h@xmath30 ) due to collisions can be ascribed to a spuriously higher @xmath141(h@xmath30 ) . the he / h line ratios are affected by the increase of the balmer line intensities due to collisions and by the decrease of @xmath141(h@xmath30 ) due to the increase of @xmath20(h@xmath24)/@xmath20(h@xmath30 ) . from a series of cloudy models it is found that the collisional contribution to @xmath20(h@xmath30 ) for i zw 18 and sbs 0335052 is in the 2% to 6% range , for haro 29 and ngc 2363 in the 1% to 3% range and for ngc 346 in the 0.6% to 1.2% range . the effect of collisions on the balmer line intensities together with the cloudy models for i zw 18 , sbs 0335052 , and haro 29 will be discussed extensively elsewhere @xcite . the cloudy models for ngc 2363 and ngc 346 are those by @xcite and @xcite respectively . to estimate @xmath16(+hc ) for ngc 346 , ngc 2363 , haro 29 , and i zw 18 we used the published balmer line intensities without modifying the @xmath141(h@xmath30 ) determination , the reason is that the published @xmath141(h@xmath30 ) values are already very low and do not seem to indicate that they have been overestimated . for ngc 346 the published @xmath141(h@xmath30 ) value is 0.15@xmath157 and the value derived from the embedded stellar cluster amounts to @xmath158 ( see paper i ) , for ngc 2363 and haro 29 @xmath141(h@xmath30 ) amounts to 0.11 and 0.00 respectively @xcite , and for i zw 18 it amounts to @xmath159 @xcite . the @xmath16(+hc ) estimates are presented in table [ tcomp ] , where the collisional contribution was obtained from the models mentioned in the previous paragraph . the errors in the @xmath16(+hc ) determinations include the same errors as the @xmath16(nhc ) determinations plus the error produced by the uncertainty in the collisional contribution effect , but they do not include the errors due to other possible systematic effects . alternatively @xcite derived for the three central positions of sbs 0335052 a @xmath141(h@xmath30 ) value of @xmath160 , but the value derived from @xmath20(h@xmath24)/@xmath20(h@xmath30 ) is higher than those derived from @xmath20(h@xmath161)/@xmath20(h@xmath30 ) and @xmath20(h@xmath162)/@xmath20(h@xmath30 ) ( after correcting the balmer lines for underlying absorption as mentioned in section 7 ) ; this difference could be due to collisional excitation of the balmer lines , therefore from the corrected balmer lines for underlying absorption and from the effect of collisions , as predicted by one of our cloudy models , we obtained a @xmath141(h@xmath30 ) of 0.15 ; for this value of the reddening we corrected all the line intensities and from them we derived the @xmath16(+hc ) value presented in table [ tcomp ] . our preliminary result of @xmath0(+hc ) , is about 0.0028 larger than @xmath0(nhc ) ( see table [ tcomp ] ) . from table [ tcomp ] it can be seen that the @xmath16(+hc ) values for ngc 346 , ngc 2363 and haro 29 are similar to those for i zw 18 and sbs 0335052 . this result _ may _ indicate that the corrections adopted for collisional excitation of the balmer lines have been overestimated because the production of helium due to galactic chemical evolution is expected to be higher for the first three objects than for the last two ( see equation [ edeltao ] ) . the brightest extragalactic h regions with temperatures in the 14,000 to 16,000 k range might be the best objects to determine @xmath0 . the reasons are the following : @xmath6 ) there are many of these objects available and the brightest of them are brighter than the metal - poorest objects known which implies that the errors in the line intensity ratios will be smaller , and @xmath7 ) the effect of collisional excitation of the balmer lines is considerably smaller than for the metal - poorer objects ( those with temperatures in the 18,000 to 22,000 k range ) reducing the error in the @xmath16 determination due to this effect . on the other hand the correction to the @xmath16 determination to obtain @xmath0 based on objects with temperatures in the 14,000 to 16,000 k range increases the error in the @xmath0 determination , but in general the size of this error is expected to be smaller than those introduced in the previous paragraph because the correction is small and can be obtained with high accuracy under the assumption that @xmath163 ( see paper i ) . from chemical evolution models it is found that changes in @xmath154 are small even in the presence of a very strong burst of star formation @xcite ; moreover these objects seem to have been forming stars over periods of at least 1 to 2 gyr implying that most of their oxygen formed before the present burst , as in the case of haro 29 ( e.g. * ? ? ? * ) , and consequently that the expected changes in @xmath154 due to bursts are very small . the @xmath0(nhc ) value derived by us is significantly smaller than the value derived by @xcite , from the @xmath16 o / h linear regression for a sample of 45 bcgs , and by @xcite , from the average for the two most metal deficient galaxies known ( i zw 18 and sbs 0335052 ) , that amount to @xmath164 and @xmath165 respectively . our @xmath66(he ) determinations , that we prefer , produce @xmath16 values that are about 0.007 smaller than those derived by @xcite and @xcite , a small but significative difference , as we will see below . it should be noted that the abundances for ngc 2363 , sbs 0335052 , haro 29 , and i zw 18 obtained by both groups are based on the same observations . from photoionization models computed with cloudy we estimate that @xmath1(he ) , the temperature that should be used to determine the helium abundance , should be at least 5% smaller than @xmath1(o ) . moreover , if in addition to photoionization there is additional energy injected to the h region the difference between @xmath1(o ) and @xmath1(he ) might be even larger . figure [ feta ] shows the helium , deuterium , and lithium abundances predicted by standard big bang nucleosynthesis computations with three light neutrino species for different values of @xmath166 , the baryon to photon ratio @xcite , also in this figure we present observational abundance determinations of these elements . the implications of this figure for the determination of @xmath167 , the baryon content of the universe , are presented in table [ tbar ] . based on the best observations of extragalactic h regions of low metallicity available in the literature we redetermine the primordial helium abundance , @xmath0(nhc ) , taking into account all the observed he line intensities with the exception of those strongly affected by underlying absorption . we derive he / h , @xmath9(he ) , @xmath1(he ) , and @xmath17(3889 ) self - consistently . from the same data for ngc 2363 , sbs 0335052 , haro 29 , and i zw 18 we obtain @xmath16(nhc ) values about 0.007 smaller than those derived by izotov , thuan and collaborators , the differences are small but significant and are mainly due to the lower @xmath1(he ) values adopted by us . in the self - consistent solutions the lower @xmath1(he ) values imply higher densities ; the higher the density , the higher the collisional contribution to the he line intensities and consequently the lower the helium abundances ( see tables [ tchi ] and [ t346]-[t0335 ] ) . the @xmath0(nhc ) value derived by us from ngc 346 , ngc 2363 , sbs 0335052 , haro 29 , and i zw 18 is in good agreement with that derived from ngc 346 in paper i. the @xmath16(nhc ) values are lower limits to the real @xmath16 values because they do not consider the excitation of the balmer lines due to collisions . a preliminary estimate of @xmath16(+hc ) is presented in table [ tcomp ] . in a future paper we will give a full discussion of this problem @xcite based on detailed models for each object . the @xmath0(nhc ) of @xmath168 combined with standard big bang nucleosynthesis computations @xcite implies that , at the @xmath169 confidence level , @xmath167 is in the 0.011 to 0.020 range for @xmath170 = 0.7 ( see table [ tbar ] ) . also in table [ tbar ] we present seven other determinations of @xmath167 not based on @xmath0 , our result is in good agreement with five of them , those based on : the primordial lithium determination , li@xmath171 , the cosmic background imager , the baryon budget in the galactic vicinity ( @xmath172 = 0 ) , the baryon budget at @xmath173 3.00 and that derived under the assumption of a flat universe , the snia magnitude redshift data to derive @xmath174 , and a baryon fraction estimate based on x - ray observations of rich clusters of galaxies ( this determination is labeled snia in table [ tbar ] ) ; on the other hand our result is in disagreement with two of them , those based on : the primordial deuterium determination , d@xmath171 , and boomerang . our @xmath0(+hc ) value of @xmath175 is also in good agreement with the five determinations of @xmath167 mentioned above and is closer than the @xmath0(nhc ) value to the other two determinations but still in disagreement with them . the errors in all determinations need to be reduced to constrain even further the domain available to non standard bbn models . it is a pleasure to acknowledge several fruitful discussions with : g. ferland , b. e. j. pagel , e. skillman , g. stasiska , g. steigman , s. torres - peimbert , and s. viegas . in this appendix we prove the validity of equation [ eto2 + 3_t0 m ] . to do this , we need to give a physical meaning to @xmath45 and consequently to @xmath48 , presented in equation [ eto2 + 3def1 ] . therefore we need to derive the functional form of @xmath48 that will make equation [ eto2 + 3_t0 m ] accurate to second order . equation [ eto3-t0 ] is designed ( and by analogy [ eto3-t03 ] and [ eto2-t02 ] ) to take into account the increase in emissivity of the forbidden o lines with increasing temperature , therefore giving higher @xmath176 differences to regions with larger temperature fluctuations . we want to give the same physical meaning to equation [ eto2 + 3_t0 m ] , thus we want the @xmath177 difference to increase with @xmath14 . this can be physically explained by examining equation [ et2 ] where the three terms are positive and thus increase @xmath14 . the first term is related to the temperature difference between the o and o zones and the last two terms are related to the temperature fluctuations in each zone . the fluctuations associated with the last two terms correspond to the @xmath178 and @xmath179 gw3o3h)w differences , therefore by adopting a @xmath45 of the form @xmath180 the fluctuations will produce a @xmath181 difference . the first term should also contribute to the @xmath182 difference , but since this term does not represent a change in @xmath40 or @xmath38 , @xmath183 and @xmath184 will not change either , consequently another term is needed in the definition of @xmath47 this effect is introduced by multiplying the direct average by a factor @xmath48 , which represents the second order correction due to the difference in temperature between the o and o zones , thus @xmath45 will be of the form : @xmath185\gamma \label{eto2 + 3def2 } , \ ] ] where @xmath186 here @xmath162 is the relative difference between @xmath183 and @xmath184 and @xmath161 represents a ( free ) parameter which will permit to recover the physical meaning of equations [ eto3-t03 ] and [ eto2-t02 ] . thus @xmath162 is given by @xmath187 to have a second order correction in @xmath30 it is enough to determine @xmath162 to first order ; to first order @xmath188 and @xmath189 , therefore @xmath190 substituting equations [ eto3-t03 ] and [ eto2-t02 ] into equation [ eto2 + 3def2 ] we obtain @xmath191 \nonumber \\ & = & \gamma\left[1 - 3 \left(\left(1-\alpha\right)\frac{t_3 ^ 2}{2}+ \alpha\frac{t_2 ^ 2}{2}\right ) \right . \nonumber \\ & & + \left.\frac{1}{t_0}\left(\left(1-\alpha\right)90800\frac{t_3 ^ 2}{2 } + \alpha97300\frac{t_2 ^ 2}{2}\right)\right ] \label{eto2 + 3_1 } ; \end{aligned}\ ] ] substituting @xmath48 and keeping this equation accurate to second order in the temperature variations ( @xmath30 , @xmath192 , @xmath193 , and @xmath194 ) we obtain @xmath195 to simplify this equation and recover the physical meaning of equation [ eto3-t0 ] it is necessary for @xmath161 to be @xmath196 here it is enough to derive @xmath161 to zeroth order , so dividing @xmath43 by any one of the available temperatures ( @xmath15 , @xmath32 , @xmath29 , @xmath183 or @xmath184 ) would be approximately right , thus we obtain @xmath197 under the assumption that @xmath198 equation [ et2 ] becomes @xmath199 and finally equation [ eto2 + 3_3 ] becomes @xmath42 \frac{t^2}{2 } \right ) \label{eto2 + 3_t0d } , \ ] ] or @xmath200 \frac{t^2}{2}\right ) \label{et0-to2 + 3 } .\ ] ] equation [ et0-to2 + 3 ] can now be combined with those equations that relate @xmath15 and @xmath14 , to some observable temperature like @xmath1(bac ) or @xmath201 . to adopt a value of @xmath53 in equation [ eto2 + 3def1 ] is equivalent to assume that the first term in equation [ et2 ] is equal to zero . we have already mentioned the parameter space where it is a good approximation to assume @xmath53 . to obtain very accurate he / h values the exact value of @xmath48 , that includes the second order effects implied by @xmath14 , should be used . lr@@xmath95lr@@xmath95lr@@xmath95lr@@xmath95lr@@xmath95l 3820 & & & 0.012&0.001 & 0.010&0.001 & + 3889 & 0.0940&0.0017 & 0.0898&0.0071 & 0.087&0.001 & 0.095&0.001 & 0.0924&0.0028 + 4026 & 0.0185&0.0007 & 0.0202&0.0036 & 0.018&0.001 & 0.020&0.001 & 0.0173&0.0006 + 4387 & 0.0047&0.0002 & & 0.005&0.001 & 0.005&0.001 & + 4471 & 0.0384&0.0005 & 0.0387&0.0025 & 0.040&0.001 & 0.039&0.001 & 0.0382&0.0007 + 4922 & 0.0100&0.0002 & & 0.013&0.001 & & 0.0093&0.0004 + 5876 & 0.1064&0.0012 & 0.0936&0.0028 & 0.106&0.001 & 0.102&0.001 & 0.1027&0.0017 + 6678 & 0.0296&0.0002 & 0.0263&0.0018 & 0.029&0.001 & 0.029&0.001 & 0.0263&0.0006 + 7065 & 0.0211&0.0002 & 0.0239&0.0016 & 0.029&0.001 & 0.024&0.001 & 0.0358&0.0007 + 7281 & 0.0063&0.0003 & & 0.006&0.001 & 0.005&0.001 & + lccccccccc 0.001 & & 12,800 & & 8052 & 8014 & 7980 & 7924 & 7865 & 7810 + & & & & ( 12.4 ) & ( 14.0)&(18.4)&(31.7)&(54.2)&(83.3 ) + + 0.014&&12,240 & & 8036 & 8001 & 7969 & 7917 & 7861 & 7809 + & & & & ( 13.0)&(8.81)&(7.53 ) & ( 10.8)&(21.7)&(39.0 ) + + 0.022&&11,860 & & 8026 & 7992 & 7961 & 7911 & 7858 & 7808 + & & & & ( 22.3)&(14.5)&(9.58)&(6.53)&(10.0)&(19.8 ) + + 0.030&&11,480 & & 8015 & 7983 & 7953 & 7906 & 7855 & 7807 + & & & & ( 39.0)&(27.9)&(19.7)&(10.8)&(7.53 ) & ( 10.3 ) + lccccc @xmath1(o ) & @xmath202 & @xmath203 & @xmath204 & @xmath205 & @xmath206 + @xmath1(he ) & @xmath207 & @xmath208 & @xmath209 & @xmath210 & @xmath211 + @xmath14 & @xmath212 & @xmath213 & @xmath214 & @xmath215 & @xmath216 + @xmath217(he ) & @xmath218 & @xmath219 & @xmath220 & @xmath221 & @xmath222 + @xmath113 & @xmath223 & @xmath224 & @xmath225 & @xmath226 & @xmath227 + @xmath228(he ) & 1.000 & 1.000 & 0.993 & 0.995 & 0.9985 + lccccccccc 0.001 & & 12,800&&0.2436 & 0.2428&0.2420&0.2407&0.2393&0.2381 + 0.014&&12,240&&0.2433&0.2425&0.2417 & 0.2405&0.2392&0.2380 + 0.022&&11,860&&0.2430&0.2423&0.2415&0.2404&0.2392&0.2380 + 0.030&&11,480&&0.2428&0.2420&0.2414&0.2403&0.2391 & 0.2380 + lccccccccc 0.006 & & 17,920&&0.2404&0.2375 & 0.2362&0.2354&0.2346&0.2307 + 0.018&&17,390&&0.2400&0.2373&0.2359 & 0.2351&0.2344&0.2307 + 0.024&&17,120&&0.2398&0.2371&0.2358&0.2350&0.2343&0.2307 + 0.030&&16,860&&0.2396&0.2370&0.2357&0.2349&0.2342 & 0.2307 + lccccccccc 0.001 & & 15,620&&0.2500&0.2459 & 0.2445&0.2430&0.2414&0.2395 + 0.011&&15,160&&0.2493&0.2455&0.2442 & 0.2427&0.2412&0.2394 + 0.021&&14,710&&0.2487&0.2451&0.2438&0.2424&0.2410&0.2393 + 0.031&&14,260&&0.2480&0.2446&0.2434&0.2421&0.2407 & 0.2391 + lccccccccc 0.001 & & 15,860&&0.2477&0.2451&0.2427&0.2390 & 0.2356&0.2296 + 0.016&&15,190&&0.2471&0.2446&0.2424&0.2389 & 0.2356&0.2200 + 0.023&&14,880&&0.2468&0.2444&0.2422&0.2388&0.2356&0.2302 + 0.030&&14,570&&0.2465&0.2441&0.2421&0.2388 & 0.2356&0.2304 + lccccccccc 0.004 & & 20,020&&0.2439&0.2425 & 0.2409&0.2396&0.2382&0.2351 + 0.014&&19,590&&0.2442&0.2430&0.2413 & 0.2400&0.2387&0.2356 + 0.021&&19,290&&0.2445&0.2433&0.2416&0.2403&0.2390&0.2360 + 0.028&&18,990&&0.2447&0.2435&0.2419&0.2406&0.2393 & 0.2363 + lcccc ngc 346 & @xmath229 & ..... & @xmath230 & @xmath231 + ngc 2363 & @xmath232 & @xmath233 & @xmath234 & @xmath235 + haro 29 & @xmath236 & @xmath237 & @xmath238 & @xmath239 + sbs 0335052 & @xmath240 & @xmath241 & @xmath242 & @xmath243 + i zw 18 & @xmath244 & @xmath245 & @xmath246 & @xmath247 + @xmath0(sample ) & & @xmath248 & @xmath249 & @xmath250 + lcc@l d@xmath171 & 0.035 - 0.049 & ( @xmath169 ) & @xcite + li@xmath171 & 0.013 - 0.026 & ( @xmath169 ) & @xcite + boomerang & 0.031 - 0.059 & ( @xmath169 ) & @xcite + cbi & 0.006 - 0.060 & ( min - max ) & @xcite + ( @xmath251 ) & 0.007 - 0.041 & ( min - max ) & @xcite + ( @xmath252 ) & 0.010 - 0.060 & ( min - max ) & @xcite + snia & 0.018 - 0.066 & ( @xmath169 ) & @xcite + @xmath0(nhc),@xmath150 & 0.025 - 0.037 & ( @xmath169 ) & this paper + @xmath0(nhc),@xmath149 & 0.011 - 0.020 & ( @xmath169 ) & this paper + @xmath0(+hc),@xmath149 & 0.012 - 0.027 & ( @xmath169 ) & this paper +

using relativistic , steady , axisymmetric , ideal magnetohydrodynamics ( mhd ) we analyze the super - alfvnic regime of a pulsar wind by means of solving the momentum equation along the flow as well as in the transfield direction . employing a self - similar model , we demonstrate that ideal mhd can account for the full acceleration from high ( ) to low ( ) values of , the poynting - to - matter energy flux ratio . the solutions also show a transition from a current - carrying to a return - current regime , partly satisfying the current - closure condition . we discuss the kind of the boundary conditions near the base of the ideal mhd regime that are necessary in order to have the required transition from high to low in realistic distances , and argue that this is a likely case for an equatorial wind . examining the mhd asymptotics in general , we extend the analysis of heyvaerts & norman and chiueh , li , & begelman by including two new elements : classes of quasi - conical and parabolic field line shapes that do not preclude an efficient and much faster than logarithmic acceleration , and the transition after which the centrifugal forces ( poloidal and azimuthal ) are the dominant terms in the transfield force - balance equation .

astro-ph0309292

the spin down luminosity of a rotating pulsar powers the emission of an associated synchrotron nebula by driving an initially poynting flux - dominated wind ( the poynting - to - matter energy flux ratio is @xmath4 ) that flows along the open magnetic field lines . the density of the outflowing matter is enhanced by pair - creating cascades near the source and the resulting number density is sufficiently large to screen the electric field along the flow and provide the necessary charge and current densities for ideal magnetohydrodynamics ( mhd ) to hold . at a distance @xmath5 cm specifically for the crab the ram pressure of the wind equals the total nebula pressure and the wind terminates in a standing shock @xcite . the `` wisps '' ( which appear at distance @xmath6 from the center ) are an observational signature of the coupling between the fast pulsar wind and the slowly expanding nebula @xcite . by applying ideal mhd in the nebula , the matching shock conditions require that the pulsar wind has become completely matter - dominated at the position of the shock ; the @xmath2 function should be as low as @xmath7 just upstream of the termination shock ( kennel & coroniti 1984a , b ) , and the question arises as to how this transition from high to low values of @xmath2 can happen , the so - called @xmath2 problem . we note , however , that the very existence of the shock has been questioned by @xcite , who argues that no shock is necessary and that the observed emission is created by current dissipation in a poynting flux - dominated wind . furthermore , even if a shock indeed terminates the pulsar wind , abandoning the axisymmetry or the ideal mhd in the nebula results in undetermined @xmath8 @xcite . even so , @xmath8 could in general be small , and the interpretation of the wisps as ion driven compressions @xcite , as well as the fitting of the observed nebula emission @xcite , provide additional support for the value @xmath9 . the crab - pulsar / nebula is the first and most analyzed object of this kind . recent x - ray observations have shown that there exist other objects with common characteristics [ e.g. , the vela pulsar @xcite , the nebulae associated with psr b1509 - 58 @xcite , and psr b1957 + 20 @xcite ] . the study of approximately monopole magnetic field geometries that are extremely inefficient accelerators gave the impression to the community that ideal mhd is in general unable to solve the @xmath2 problem , and many authors refer to the `` @xmath2 parameter '' and not to the `` @xmath2 _ function _ '' . moreover , several alternative ( non - ideal mhd ) attempts ( see , e.g. , @xcite for a review ) gave unsatisfactory results , and the puzzle remains unsolved for almost three decades . however , there are in the literature @xmath10 self - similar , ideal mhd solutions that start with @xmath4 near the origin of the flow and reach @xmath11 ( @xcite ; @xcite ) , or even more efficient accelerators with @xmath12 @xcite , or @xmath13 @xcite . ( actually , @xmath10 self - similar are the only for the present known _ exact _ solutions of relativistic mhd in the sense that they satisfy the momentum equation in both directions : along the flow _ and _ in the transfield direction . ) we note in this connection that the main acceleration mechanism is not the centrifugal , but the _ magnetic_. the former may be important in the sub - alfvnic regime of a disk - wind and results in poloidal velocity of the order of the initial keplerian speed , while the latter caused by the magnetic pressure - gradient force results in much higher velocity that depends on the initial poynting - to - mass flux ratio @xmath14 . employing the @xmath2 function we may write @xmath15 , thus the maximum final lorentz factor is attained in solutions with @xmath16 . in this paper we show that ideal mhd can account for the full acceleration from @xmath17 to @xmath18 , _ if _ the boundary condition @xmath19 at the base of the ideal mhd accelerating regime is satisfied ( here @xmath20 is the poloidal current , @xmath21 the poloidal magnetic flux function , and @xmath22 the field angular velocity ) . the transfield force - balance equation is the most important in determining the acceleration by means of controlling the field line shape and consequently how fast the quantity @xmath23 decreases ( here @xmath24 is the cylindrical distance and @xmath25 the poloidal magnetic field ) . in the main part of the acceleration the transfield component of the electromagnetic force equals the poloidal centrifugal force that is proportional to the curvature of the poloidal field lines @xmath26 . if this is still the case after the end of the acceleration in the far - asymptotic regime where @xmath27 the electromagnetic force exactly vanishes yielding the solvability condition at infinity ( @xcite ; @xcite ) . however , if during the acceleration phase sufficiently small @xmath2 values are reached , the azimuthal centrifugal force the other part of the inertial force becomes important . in this regime it is the difference between poloidal and azimuthal centrifugal forces that equals the much smaller electromagnetic force in the transfield direction , and no solvability condition can be derived . we organize the paper as follows . in [ mhd ] we present the ideal mhd formalism focusing on the super - alfvnic asymptotic regime where the puzzling transition from @xmath28 to @xmath29 happens . in [ zss ] we derive a novel @xmath30 self - similar class of solutions describing the polar region of the wind and present the results of the integration . in [ genanalysis ] we analyze the asymptotic shape of the flow in relation to the acceleration , demonstrating that the same results are expected in non - self - similar cases as well . in [ equatorialwind ] we discuss the problem of the pulsar magnetosphere , pointing out the problems that the force - free picture faces , and suggesting qualitatively a scenario that seems plausible for creating an equatorial wind with the appropriate conditions for efficient acceleration from high to low-@xmath2 values satisfied . a summary follows in [ conclusions ] . the system of equations of special relativistic , steady , ideal mhd , consists of the maxwell equations , ohm s law , and the continuity , entropy and momentum equations : @xmath31 @xmath32 @xmath33 @xmath34 here @xmath35 is the velocity of the outflow , @xmath36 the associated lorentz factor , @xmath37 the electric and magnetic fields as measured in the central object s frame , @xmath38 the charge and current densities , @xmath39 the gas rest - mass density and pressure in the comoving frame , while for the polytropic equation of state ( with index @xmath40 ) the enthalpy - to - rest - mass ratio is @xmath41 assuming axisymmetry [ @xmath42 , in spherical @xmath43 and cylindrical @xmath44 coordinates with @xmath45 along the rotation axis and the central object at ( @xmath46 , five conserved quantities along the flow exist . if @xmath21 is the poloidal magnetic flux function , they are ( e.g. , @xcite ) : @xmath47 where subscripts @xmath48/@xmath49 denote poloidal / azimuthal components . an important combination of the above field line constants is the `` michel s magnetization parameter '' @xmath50 the physical quantities can be expressed as functions of @xmath21 and the `` alfvnic '' mach number @xmath51 : [ quantities ] @xmath52 @xmath53 @xmath54 @xmath55 here @xmath56 is the cylindrical distance in units of the light cylinder s lever arm on each field - streamline @xmath57const , and @xmath58 is its value at the alfvn point . , although in the general case where @xmath59 it is not a cylinder . ] alternatively , using the equivalent to equation ( [ gamma ] ) , @xmath60 and the poynting - to - matter energy flux ratio @xmath61 we may rewrite all quantities as functions of ( @xmath62 ) or ( @xmath63 ) . the two remaining equations are the bernoulli and the transfield force - balance relations . the identity @xmath64 gives the bernoulli equation @xmath65 ^ 2 + \left [ \frac{\xi \gamma - \mu \left(1- x_{\rm a}^2 \right ) } { \xi x } \right]^2\end{aligned}\ ] ] where @xmath66 the projection of equation ( [ momentum ] ) along @xmath67 gives the transfield force - balance equation @xmath68 where : * the azimuthal centrifugal term @xmath69 * the rest of the inertial force density along @xmath70 @xmath71 -{f}_{c \bot } = -\frac{b_p^2}{4 \pi { \cal r } } m^2 \nonumber\end{aligned}\ ] ] is the poloidal centrifugal term . here @xmath72 is the curvature radius of the poloidal field lines . * the pressure gradient force density along @xmath70 @xmath73 * the `` electric field '' force density @xmath74 * the `` magnetic field '' force density along @xmath70 @xmath75 the total electromagnetic force density in the transfield direction @xmath76 can be decomposed as @xmath77 } _ { f_{em1 } } + \underbrace { \frac{{b_p^2(1-x^2)}}{{4 \pi { \cal r } } } } _ { f_{em2 } } + \underbrace { \frac{b_p^2}{4 \pi \varpi } \hat{n } \cdot \hat{\varpi}}_{f_{em3}}\ ] ] altogether , they give the following form of the transfield force - balance equation : @xmath78 + \nonumber \\ + \frac{b_p^2}{4 \pi \varpi } \left[1 + \left(\frac{m v_{\phi } } { v_p}\right)^2 \right ] \hat{n } \cdot \hat{\varpi } - \hat{n } \cdot { \boldsymbol{\nabla } } p \,.\end{aligned}\ ] ] ( using equations [ [ quantities ] ] , the latter becomes a second - order partial differential equation ( pde ) for @xmath21 . ) at the origin of the flow ( subscript @xmath79 ) , @xmath80 , @xmath81 , and ( using eq . [ [ gamma ] ] ) @xmath82 . as the flow moves downstream , the functions @xmath83 and @xmath84 increase , and when the alfvn surface is reached they become @xmath85 , @xmath86 . the pulsar wind is still strongly magnetized in the neighborhood of the alfvn surface , and the field is close to force - free there : @xmath87 , or , @xmath88 , implying @xmath89 . could be in general larger than unity . ] the light cylinder @xmath90 is located slightly after the alfvn surface . next , the flow enters the super - alfvnic regime . in this regime @xmath91 , @xmath92 . also thermal effects play no longer any role and the fluid is cold , @xmath93 . it is reasonable to assume that the above inequalities hold from the classical fast - magnetosound point downstream . at this point @xmath94 , and for cold flows , @xmath95 ( e.g. , @xcite ) . , or , ( using eq . [ [ gamma ] ] with @xmath96 , @xmath81 ) , @xmath97 , implying @xmath89 . ] + employing the comoving magnetic field @xmath98 , the bernoulli equation ( [ bernoulli ] ) gives the exact result @xmath99}{\left[x^2 - 1+\mu ( 1-x_{\rm a}^2 ) / \xi \gamma \right]^2 } \frac{b_\phi^2}{\gamma^2 } \,.\ ] ] in the asymptotic regime where @xmath100 the latter equation implies @xmath101 thus , the magnetic field is mainly azimuthal and the electric field approximately equals the magnetic field _ exactly _ equals @xmath102 , eq . ( [ eapprox ] ) implies a `` linear accelerator '' @xmath103 @xcite . however , in the general case @xmath104 does not _ exactly _ equals @xmath102 , and eq . ( [ eapprox ] ) gives just the small difference between them . note that the @xmath105 is not exactly satisfied in the numerical solution of @xcite either . ] @xmath106 the approximate form of the bernoulli equation ( [ bernoulli ] ) is @xmath107 we are now in a position to examine the highly nonlinear transfield force - balance equation ( [ transfield ] ) . in that equation , the @xmath108 and @xmath109 are the dominant terms that almost cancel each other ( e.g. @xcite ) . the algebraic sum of their main contributions is the @xmath110 term , which can be written as ( using eq . [ [ bpbphi ] ] ) @xmath111 . this term must be balanced by the other terms of equation ( [ transfield ] ) . ignoring @xmath112 , we are left with the `` poloidal curvature '' term @xmath113 and the term @xmath114 \hat{n } \cdot \hat{\varpi } = -f_{c \bot } -f_{em3}$ ] . note that the last term , which henceforth we call `` centrifugal '' , has contributions from the poloidal magnetic field ( the part @xmath115 which is the dominant one in the @xmath4 regime ) , and from the azimuthal centrifugal force @xmath116 . ( the ratio of the two `` centrifugal '' parts is @xmath117 , and the term @xmath118 becomes important in the matter - dominated regime . ) thus , the asymptotic form of the transfield force - balance equation ( [ transfield ] ) is @xmath119 ^ 2 + \left [ 1 + \frac{m^2 \gamma^2 } { x^2 ( \gamma^2 - 1 ) } \right ] \hat{n } \cdot \hat{\varpi } \,.\ ] ] this important equation , with the `` centrifugal '' term omitted , was derived by @xcite , while recently @xcite derived the same equation with the `` centrifugal '' term included , but the poloidal curvature term omitted . it is convenient to rewrite the bernoulli and transfield force - balance equations in terms of the unknown functions ( @xmath63 ) ( also assuming @xmath120 ) . the bernoulli equation ( [ bernoulliapprox ] ) becomes ) yields @xmath121 . the last term on the right - hand side is nonnegligible near the classical fast - magnetosound surface in the sense that the derivative of the latter equation ( including the last term ) along the flow gives @xmath122 , implying @xmath123 , @xmath124 . however , @xmath125 and we neglect this term in our analysis . ] @xmath126 , or ( using eqs . [ [ sigma ] ] , and [ [ sigma ] ] ) , [ systemfinal ] @xmath127 the transfield force - balance equation ( [ transfieldapprox ] ) becomes ( using eqs . [ [ b - e ] ] , [ [ m2 ] ] , [ [ sigma ] ] , and [ [ bernoullifinal ] ] ) @xmath128 equations ( [ bernoullifinal ] ) , and ( [ transfieldfinal ] ) together with the equation for the curvature radius , @xmath129 form a closed system for the functions @xmath2 and @xmath21 . a possible solution of the system will describe the outflow downstream from the classical fast - magnetosound surface ( where the super - alfvnic asymptotic conditions surely hold ) . our task is to find a field geometry @xmath130 , such that the flow accelerates smoothly from @xmath4 ( at the classical fast - magnetosound surface ) to @xmath29 ( in the matter - dominated regime ) . equation ( [ bernoullifinal ] ) shows that the key quantity @xmath131 should be a decreasing function of the distance from the source , along any field line . according to the model of kennel & coroniti ( 1984a , b ) for the crab nebula , the solution should give @xmath132 , @xmath133 , and @xmath134 g , at the distance @xmath5 cm of the termination shock . at the classical fast - magnetosound surface @xmath135 and , from equation ( [ sigma ] ) , @xmath136 . + the number density of the particles in the pulsar magnetosphere highly exceeds the goldreich - julian value , and a tiny deviation from neutrality is enough to support the electromagnetic field . assuming that in the upper ( lower ) hemisphere @xmath137 , the poynting - flux is outflowing when the electric field points toward the rotation axis , corresponding to a negative value of the charge density in the polar regions ( @xmath138 ) . for a highly relativistic poloidal motion , @xmath139 ( where a suffix `` @xmath140 '' denotes the component of a vector along the poloidal field ) , and the flow that originates from the negatively - charged polar region is current - carrying ( @xmath141 ) . depends on the tiny relative speed between the positively and negatively charged species . ] the current - closure condition demands that each field line reaches a point where @xmath142 , and after that enters the distributed `` return - current '' regime @xmath143 . the poloidal current density is @xmath144 , with @xmath145 , and the meridional current lines represent the loci of constant total poloidal current ( @xmath146 ) . as a flow element crosses successive current lines , the absolute value of @xmath20 decreases and asymptotically vanishes ( @xmath147 ) . + an alternative scenario is that the current closes in a thin sheet at the end of the ideal mhd asymptotic regime , in which case @xmath148 . + the field is force - free near the pulsar and the energy is poynting flux - dominated : @xmath4 . this remains the case near the classical fast - magnetosound surface , where @xmath149 . as the lorentz force accelerates the matter meaning that the flow crosses current lines with decreasing @xmath150 the function @xmath2 also decreases . the relation between @xmath2 and @xmath20 is ( by combining equations [ [ mu ] ] , [ [ sigma ] ] ) @xmath151 at @xmath152 a few the force - free approximation brakes down and the back - reaction of inertial forces on the magnetic field becomes important . the final value of @xmath2 could be in general very small ( or exactly zero ) , depending on the small ( or zero ) value of @xmath153 . + the `` centrifugal '' term in the transfield equation ( the last term in eq . [ [ transfieldfinal ] ] ) is negligible in the force - free regime . since the flow velocity is mainly poloidal , it is expected that this term is also negligible inside the matter - dominated regime , in which case the poloidal curvature is controlled by the electromagnetic force @xmath110 . nevertheless , as the flow becomes more and more matter - dominated , the electromagnetic force @xmath110 decreases ( it vanishes at @xmath154 ) . as a result , a point @xmath155 is reached where the two terms on the right - hand side of equation ( [ transfieldfinal ] ) become comparable , and after that point the azimuthal centrifugal term takes over . + for @xmath156 when the centrifugal term is negligible equation ( [ transfieldfinal ] ) yields ( using eqs . [ [ sigmam ] ] , [ [ bernoullifinal ] ] ) @xmath157 here @xmath158 is the opening half - angle of the flow ( the angle between the poloidal magnetic field and the rotation axis ) . + if the flow reaches a constant value ( in principle different from field line to field line ) @xmath159 such that @xmath160 , then the right - hand side of equation ( [ sintheta ] ) is a function of @xmath21 alone and @xmath161 would diverge logarithmically with @xmath24 @xcite . the only way to avoid this divergence is to have @xmath162const . by using equations ( [ b - e ] ) , and ( [ sigma ] ) the solvability condition at infinity @xmath163const ( heyvaerts & norman 1989 ; chiueh et al . 1991 ) is recovered . + at sufficiently small values of @xmath2 ( @xmath164 ) , the curvature of the poloidal field - streamlines is controlled by the azimuthal centrifugal force , equals @xmath165 and can be omitted in the matter - dominated regime . ] ; the electromagnetic force is much smaller than its inertial counterparts , and equation ( [ transfieldfinal ] ) gives @xmath166 in other words , in the @xmath167 regime we can not neglect terms of order @xmath168 , @xmath169 in the transfield equation , simply because the electromagnetic term is even smaller . + in this regime the solution of the momentum equation is simply @xmath170const and the motion is ballistic ( in the poloidal plane the azimuthal and poloidal centrifugal forces cancel each other and the total inertial force vanishes ) . by decomposing a straight in three dimensions , constant velocity streamline in cylindrical coordinates , we get : @xmath171 , @xmath172 ( from angular momentum conservation ) , and an increasing @xmath173 . it can be shown that the poloidal streamline shape ( the projection of the straight in three dimensions streamline onto the poloidal plane ) is hyperbolic , @xmath174 with some constant @xmath175 . + we emphasize that for flows with @xmath176 , equation ( [ sintheta ] ) should be replaced by equation ( [ hyperregime ] ) and there is no problem related to the divergence of @xmath161 . in the absence of a solvability condition at infinity , any value @xmath177 ( and the corresponding current distribution , [ [ currdistr ] ] ) is a possible solution . even so , the most plausible case is @xmath16 corresponding to a distributed return - current regime with @xmath147 @xcite . + the classical fast - magnetosound surface is not singular when we solve simultaneously the bernoulli _ and _ the transfield force - balance equations ( it just separates the elliptic and hyperbolic regimes of the pde problem ) . the singular surface is the so - called modified fast - magnetosound surface ( or fast - magnetosound separatrix surface ) that lies inside the hyperbolic regime , coincides with a limiting characteristic , and plays the role of the event horizon for the propagation of fast waves @xcite . the causality principle is satisfied ( disturbances in the asymptotic regime can not influence the flow near the origin ) only if the solution is trans - modified fast . completely matter - dominated flows will have super modified fast - magnetosound velocities ( simply because the fast - magnetosound speed vanishes as @xmath178 ) . assuming that the poloidal field line shape is approximately cylindrical , i.e. , @xmath179 ) to [ system_cyl ] @xmath180 to construct classes of analytical solutions , we shall make two assumptions : @xmath30 is assumed a product of a function of @xmath21 times a function of @xmath24 ( the @xmath30 self - similar ansatz ) @xmath181 ( where @xmath182const ) , and @xmath183 following the algorithm described in @xcite , it is possible to separate the variables @xmath21 and @xmath24 in the system ( [ system_cyl ] ) only if the following relations hold : [ flux1 ] @xmath184 here @xmath185 and @xmath186 are constants . the system ( [ system_cyl ] ) becomes @xmath187 using the expression for the curvature radius @xmath188 we get finally a system of ordinary differential equations ( odes ) [ ode_cyl ] @xmath189 \nonumber \\ & & \quad \ \ \times \left [ \frac{\mu_0 ^ 2 \tau^2 \varpi^2 } { \varpi_0 ^ 2 \zeta^2 \sigma^2 } - \sigma \right]^{-1 } \ , . \label{ode_cyl2}\end{aligned}\ ] ] the latter system can be easily integrated ; the only difficulty is that the solution should cross a singular point that appears when the denominator in equation ( [ ode_cyl2 ] ) vanishes . by rewriting this denominator as @xmath190 it is evident that the singular point corresponds to the modified fast - magnetosound point , where the phase speed of the fast - magnetosound waves propagating along @xmath191 is zero ( see appendix c in @xcite ) . using equation ( [ z_cyl ] ) we may find @xmath192 and hence the magnetic field , @xmath193 whereas the current density component @xmath194 is given by @xmath195 the magnetic flux function is @xmath196 it is instructive to derive simple analytical scalings from equations ( [ ode_cyl ] ) : they turn out to be in very good agreement with the results of the numerical integration presented below . before analyzing separately each regime , we rewrite schematically equation ( [ ode_cyl2 ] ) as @xmath197 all the terms appearing on the right - hand side are positive ( for @xmath198 and @xmath199 ) . it is important to note that the terms @xmath200 , @xmath201 come from the poloidal curvature term of the transfield equation , the @xmath202 and @xmath203 have electromagnetic origin , while @xmath204 comes from the `` centrifugal '' term . + in this regime the denominator in equation ( [ schemasigma ] ) is negative , @xmath205 , i.e. , the term @xmath203 dominates . since @xmath206 , the numerator should be positive . thus , the dominant term is the @xmath202 ( the only positive term ) . as we expected , in the sub - modified fast regime the magnetic field dominates over the inertial terms . one may write @xmath207 . the integration gives @xmath208 and we finally have ( by employing eq . [ [ ode_cyl1 ] ] ) @xmath209 equation ( [ jparallel ] ) gives @xmath210 . thus , for @xmath211 the flow is current - carrying ( @xmath212 ) near the origin . + in this regime , @xmath213 . since @xmath206 , the inertial terms dominate in the numerator as well , resulting in @xmath214 . as long as @xmath156 we get that not only the poloidal curvature terms are important and should not be neglected , but also that the transfield force - balance equation reduces to @xmath215 . in other words , it is not correct to substitute @xmath216 ( the `` pseudo - force - free '' condition according to @xcite ) in the transfield equation and keep only the other terms . on the contrary , _ the @xmath216 is the transfield equation itself . _ by including the `` centrifugal '' term we simply modify the above conclusions and the transfield equation reduces to equation ( [ hyperregime ] ) . + thus , for @xmath156 the solution of the transfield equation is simply a straight line @xmath217 $ ] . by employing equation ( [ ode_cyl1 ] ) we finally get @xmath218 ( in the most general , i.e. , non - self - similar case , the solution is the conical ia shape that we discuss in [ genanalysis ] . ) + equation ( [ jparallel ] ) gives @xmath219 . thus , for @xmath220 the current density component @xmath194 becomes positive at large distances . the combination @xmath221 corresponds to a transition from current - carrying ( @xmath141 , at @xmath4 ) to a return - current regime ( @xmath222 at @xmath29 ) , required by the current - closure condition . + the azimuthal centrifugal force becomes important after the point where the two terms on the right - hand side of equation ( [ trans1 ] ) become comparable . using the scalings of equation ( [ regime2 ] ) , this happens at @xmath223 , so @xmath224 . + if we continue the integration for sufficiently large @xmath30 , the fast decrease of @xmath225 stops at the point where @xmath226 . at this point the product @xmath227 is of the order of @xmath228 ( see eq . [ [ flux2 ] ] ) . as we discuss in [ genanalysis ] the flow could continue to be accelerated after that point , but with a much lower ( logarithmic ) rate . numerically we may integrate the equations ( [ ode_cyl ] ) as follows . suppose that we examine the flow along the field - streamline @xmath21 such that @xmath229 , and @xmath230 , @xmath231 rad s@xmath232 ( for which eq . [ [ varpi_0 ] ] gives the value of @xmath233 @xmath234 cm , and the independent variable of the odes is @xmath235 ) . we give the model parameters ( @xmath186 , @xmath236 ) , and a trial value for the @xmath2 function at the modified fast - magnetosound singular point , @xmath237 . at this point , both the numerator and denominator of equation ( [ ode_cyl2 ] ) vanish , and we are able to find the values @xmath238 and @xmath239 . using lh@xmath240pital s rule we find the slope @xmath241 and start the integration from the singular point downstream . at some value of the independent variable @xmath242 we find @xmath243 @xmath244 . at this point the distance should be @xmath245 ( @xmath246 cm ) , and according to equation ( [ z_cyl ] ) the equality @xmath247 @xmath248 should also hold . if the latter equality is not satisfied , we change the trial value @xmath249 and repeat the whole procedure until we find @xmath250 . for the correct @xmath249 we are able to integrate downstream until @xmath251 ( at the distance @xmath252 , just before the termination shock ) , as well as upstream until @xmath253 , i.e. , until the solution encounters the classical fast - magnetosound surface . ( using the value of the azimuthal magnetic field @xmath254 g and eq . [ [ bifield ] ] , we find the constant @xmath255 . ) + the representative solution @xmath256 corresponds to the set of parameters ( @xmath257 ) . for @xmath258 the integration gives @xmath9 at @xmath259 cm . figure [ fig1 ] shows the field lines as well as the current lines in the poloidal plane . the line shape is quasi - conical with small opening angle consistent with the model assumption @xmath260 . the current density component @xmath261 changes sign from negative close to the classical fast - magnetosound surface to positive at larger distances , satisfying ( at least partly ) the current - closure condition ( according to the analysis of [ scalings ] , this is expected for @xmath221 ) . + figure [ fig2 ] shows the function @xmath225 and the transition from poynting- ( @xmath262 ) to matter- ( @xmath263 ) dominated regime . the @xmath2 function scales as @xmath264 at @xmath265 a few ( force - free regime ) and as @xmath266 when the flow becomes matter - dominated . + the components of the magnetic field are shown in figure [ fig3 ] . the main component is the azimuthal one , and at the distance @xmath252 it becomes @xmath267 g . it is evident from the figure that the self - consistency condition @xmath260 is everywhere satisfied . + figure [ fig4 ] shows the transfield components of the various force densities as functions of @xmath268 along the reference field line @xmath269 . as expected , the dominant forces are @xmath270 and @xmath271 , and they almost cancel each other . for @xmath272 , @xmath273 and @xmath274 , corresponding to @xmath141 and @xmath275 . for larger @xmath268 both @xmath108 and @xmath109 change sign and the flow enters the return - current regime @xmath276 , where the charge density is positive ( @xmath277 ) . + the forces @xmath278 , @xmath279 , and @xmath280 are proportional to the first , second , and third terms in equation ( [ transfieldfinal ] ) , respectively . at small heights @xmath281 and the curvature radius is @xmath282 : its exact value is controlled by the electromagnetic forces @xmath283 . however , at @xmath284 , we encounter the point @xmath285 , where the `` centrifugal '' force @xmath286 ( that mainly consist of the azimuthal centrifugal part @xmath287 ) becomes important . after that point @xmath288 and the transfield equation becomes @xmath289 . the small difference between these two terms is the electromagnetic term @xmath290 . + figure [ fig5 ] shows another solution similar to solution @xmath256 . the only difference is that here @xmath291 . equations ( [ ode_cyl ] ) show that for different @xmath186 , the value @xmath249 remains roughly the same , while @xmath239 scales as @xmath292 . thus , the cylindrical distances in solution @xmath293 are ten times larger compared to the ones for solution @xmath256 . + figure [ fig6 ] shows a solution corresponding to @xmath294 . in this case the constrain on @xmath2 at @xmath252 implies @xmath295 . because @xmath296 , the current density component @xmath194 remains negative all the way from the source to infinity . if we continue the integration to sufficiently large distances ( not shown in fig . [ fig6 ] ) the poloidal current becomes parallel to the flow and the acceleration stops ; the function @xmath225 then approaches a finite asymptotic value @xmath297 . the outcome of the magnetic acceleration is either @xmath298 or @xmath16 . although the second possibility incorporates the current - closure condition and seems more natural , the first one is also possible ( it requires that the current closes in a thin sheet at the end of the ideal mhd asymptotic regime , but it can not be ruled out , simply because all mhd winds practically terminate at some finite distance ) . besides the value of @xmath299 , another important issue is the acceleration rate . in this section we focus on the field line asymptotic shape in the regime @xmath300 and its relation to the acceleration . ( the analysis holds for nonrelativistic flows as well . ) note that the discussion of the present section as well as the characteristics of the flow described in [ character ] is general and not restricted to the self - similar solutions presented in [ zss ] . it applies in the asymptotic regime @xmath301 where the motion is mainly poloidal , @xmath302 . equation ( [ omega ] ) yields @xmath303 , or , ( using eq . [ [ currdistr ] ] ) , @xmath304 ( see also eq . [ [ bernoullifinal ] ] ) . approaching its maximum value @xmath305 , a similar relation , namely @xmath306 , holds . ] thus , an accelerating flow corresponds to a decreasing @xmath23 , a function that depends on the line shape . in other words , how fast the acceleration takes place depends on the magnetic flux distribution , by the solution of the transfield equation . we already know that a field line shape of the form @xmath307 gives only logarithmic acceleration ( chiueh et al . the deviations from monopole magnetic field also result in inefficient @xcite , or logarithmic @xcite acceleration . ( the monopole geometry itself which however does not satisfy the transfield equation gives extremely inefficient acceleration : @xmath308 ; @xcite . ) the logarithmic acceleration gives asymptotically @xmath16 , but over completely unrealistic ( exponentially large ) distances . the numerical simulations by @xcite also give inefficient acceleration , possibly because the initial configuration is a split monopole field and does not change much during the simulation . the @xmath10 self - similar solutions of li et al . ( 1992 ) and @xcite give final equipartition @xmath309 , but the same model could give smaller @xmath310 as well @xcite . however , because it can not capture the transition from positive to negative curvature radius , it is impossible to obtain @xmath16 . the asymptotics of the @xmath10 self - similar model are straight poloidal field lines ( @xmath311 ) ; as a result , the transfield component of the electromagnetic force asymptotically equals the `` centrifugal '' force ( which could be very small , but definitely not exactly zero ) , and @xmath312 . even so , the value @xmath313 could be very small , and the model should be explored more carefully in that respect . in the @xmath30 self - similar model presented in [ zss ] , we kept the line shape free : @xmath314 , and the solution itself finds the shape through the function @xmath315 . the result is a quasi - conical shape , and the distribution of the magnetic flux is different from the monopole field ( see eq . [ [ flux ] ] ) . ccccc & + & @xmath178 & @xmath316 & @xmath312 & @xmath317 + & @xmath318 & @xmath319 & @xmath319 & @xmath320 + & @xmath321 & @xmath322 & @xmath323 & @xmath324 + & parabolic ii or iia & hyperbolic & conical i or ia & conical i or ia + the @xmath325 acceleration phase & & & or parabolic i or ia & or parabolic i or ia + @xmath326 & & & or cylindrical & or cylindrical + conical ia : @xmath327 & & & & + @xmath328 & @xmath329 & @xmath329 & @xmath329 & @xmath329 + parabolic ia : @xmath330 , @xmath331const & & & & + @xmath332 & @xmath329 & @xmath329 & @xmath329 & @xmath329 + parabolic ii : @xmath333 & & & & + @xmath334 & @xmath329 & @xmath329 & & + parabolic iia : @xmath335 & & & & + @xmath336 & @xmath329 & @xmath329 & & + + + collimation of relativistic outflows is in general much more difficult than in the nonrelativistic cases , firstly because the electric force almost cancel the transfield component of the magnetic force , and secondly because the effective matter inertia is larger ( e.g. , @xcite ) . the result is negligible curvature , @xmath337 ( e.g. , @xcite ) , and only close to the origin where the lorentz factor is smaller than a few tens is collimation efficient @xcite . this result , together with the envisioned close to @xmath338 field line shape we employ the term `` type i conical '' for this shape led to the erroneous conclusion that acceleration is impossible when @xmath337 . however , negligible curvature does not necessarily mean @xmath339 ; the most general case is a `` type ia conical '' shape given by @xmath340 differentiating equation ( [ straightshape ] ) we get for the poloidal magnetic field @xmath341 ( here primes denote derivative with respect to @xmath21 . ) it is evident from equation ( [ straightfield ] ) that for @xmath342 , @xmath343 the quantity @xmath23 is a decreasing function of @xmath24 , meaning that the flow is accelerated . magnetic flux conservation implies @xmath344 , where @xmath345 is the cross - section area between two neighboring magnetic flux surfaces @xmath21 , @xmath346 ( @xmath347 , see fig . [ fig7 ] ) . since @xmath348 , it is clear that acceleration is possible when as the flow moves @xmath345 increases faster than @xmath349 . this is the case in type ia conical flow , while in type i conical it is exactly @xmath350 and the quantity @xmath227 remains constant . the previous analysis shows that the argument of chiueh et al . ( 1998 ) related to the global inefficiency of the magnetic acceleration in relativistic flows ( based on the fact that the collimation is inefficient ) , can be circumvented . the conical ia field lines are perfectly straight , satisfying the @xmath351 , nevertheless , they expand in the sense that @xmath345 increases faster than @xmath349 . in other words , we may have expansion ( increasing @xmath352 ) , without bending ( @xmath353 ) . + by combining equations ( [ straightfield ] ) and ( [ bernoullifinal ] ) , we find that the @xmath225 function could change from a @xmath354 value at @xmath355 ( the conical ia shape starts roughly at this distance ) , and becomes @xmath356 at distances @xmath357 , with @xmath358 since the factor @xmath359 is not @xmath1 ( this would require the opening of the field lines asymptotically to change rapidly from line - to - line which is unlikely to happen ) the only case that gives @xmath360 is the @xmath361 . this is an important requirement for the above mechanism to work , and it is equivalent to @xmath362 in the super - alfvnic regime and as long as the @xmath2 is @xmath0 [ in this regime @xmath363 . hence , as @xcite state , the poloidal field lines need to be bunched within a small solid angle before expansion takes place . contrary to their claim , however , the expansion is not necessarily sudden near the light cylinder , but extended to @xmath364 . in [ equatorialwind ] we discuss how such a field configuration may be created ( see also fig . [ fig7 ] ) . the general asymptotic expansion of the field line configuration in both , relativistic and nonrelativistic cases is @xmath365 for which we get @xmath366 chiueh et al . ( 1991 ) examined two types of parabolic cases : type i ( @xmath367 ) when @xmath368 is a field line constant , and type ii ( @xmath369 ) when the @xmath370 slowly ( logarithmically ) decreases and asymptotically vanishes . they also examined the conical type i case ( @xmath371 ) . besides the conical type ia case , new types of parabolic cases corresponding to @xmath372 should be added in the analysis . we henceforth call them type ia , for @xmath331const@xmath373 , and type iia , for @xmath374 . these new parabolic types together with the type ia conical , are much faster than logarithmic accelerators in relativistic as well as in nonrelativistic flows . table [ table1 ] summarizes the various possible geometries in the acceleration phase @xmath375 ( first column ) combined with the possible final results of the acceleration including possible acceleration after the @xmath3 transition and the far - asymptotic final geometries . the final current distribution and the dominant forces in the transfield direction are also shown . compatible combinations of the geometry in the acceleration phase @xmath375 and the far - asymptotic final state are marked with a `` @xmath329 '' . for example , if the flow has a type ia conical shape ( this is the most plausible case for relativistic mhd in which the field lines have negligible curvature @xmath376 ) , the @xmath2 function decreases according to equation ( [ bernoullifinal ] ) and the shown in the table [ table1 ] function @xmath23 , and reaches a minimum value @xmath377 at distances @xmath378 ( see eq . [ [ sigmamin ] ] ) . if the azimuthal centrifugal force is still negligible at these distances ( i.e. , @xmath379 ) the acceleration stops and the far - asymptotic regime is the one in the fifth column ( most likely with unchanged conical ia shape ) . if @xmath380 , then the azimuthal centrifugal force starts to play a role in determining the poloidal curvature . one possibility is that the acceleration does not continue even with slightly different line shape , resulting in the fourth column case . the remaining case is that the @xmath2 continues to decrease , and the line shape becomes hyperbolic . since the asymptotic form of a hyperbolic flow is again conical , the @xmath225 function reaches a new minimum given from an equation similar to ( [ sigmamin ] ) . if the acceleration stops at this point , we get a final situation corresponding to the third column of table [ table1 ] . however , even tiny deviations from an _ exactly _ radial flow @xmath215 could give a slow ( logarithmic ) acceleration to the completely matter - dominated regime @xmath16 , as was argued by @xcite . this situation corresponds to a continuous acceleration under a parabolic ii or iia shape ( second column ) . we note , however , that for realistic applications , the interaction of an outflow with its environment brakes - down the ideal mhd conditions and this probably happens before the realization of this last part of the acceleration . chiueh et al . ( 1991 ) examined the far - asymptotics of relativistic mhd outflows , generalizing the work on nonrelativistic flows by @xcite . they assumed that the various quantities become functions of @xmath21 alone ( the characteristic of the far - asymptotic regime ) , neglected the poloidal centrifugal term @xmath381 as well as terms of order @xmath382 in the transfield equation , and found that conical field lines satisfying @xmath383 should enclose a finite current @xmath384 ( the solvability condition at infinity , see [ character ] ) . as already discussed in [ character ] , this is the case when @xmath160 , but for smaller values of @xmath299 the asymptotic transfield equation reduces to the balance between the poloidal and azimuthal centrifugal forces [ i.e. , they are exactly the @xmath385 terms that remain in the transfield ] ; the electromagnetic term is much smaller , but nevertheless not exactly zero and no solvability condition can be derived . our self - similar solutions fall in this category : the condition @xmath386 is not satisfied , because @xmath387 . as we see in table [ table1 ] the solvability condition at infinity is mandatory only for the fifth column . as a result we may have , e.g. , nonrelativistic conical or type i / ia parabolic flows that carry poloidal current @xmath388 ( with negative or positive @xmath389 when they are in the current - carrying or return - current regime respectively ) , provided that @xmath390 . as we already stated , the analysis of this section applies to nonrelativistic flows as well . a first step towards finding nonrelativistic mhd solutions with @xmath391 is the super modified fast - magnetosound solution of @xcite , which ends slightly after the modified fast singular point with @xmath392 . the full acceleration ( @xmath16 ) should be examined using a model that allows a transition from positive to negative curvature radius ( the asymptotic transfield force - balance equation for nonrelativistic flows implies that the curvature radius has the sign of the @xmath393 , thus changing sign when the flow moves from the current - carrying to the return - current regime , e.g. , @xcite ) . the most efficient acceleration mechanism that we described is the one based on the conical ia field line shape . the @xmath30 self - similar solutions are just a numerical example confirming that the mechanism indeed works . although the presented solutions describe an outflow very close to the polar axis , the same exactly physical mechanism is expected to work for the equatorial wind as well . only practical difficulties do not allow us to derive semianalytical solutions in that regime . as we discuss in [ genanalysis ] , an important requirement for the above mechanism to work is that @xmath361 . in fact , as we argue below , such a configuration is most likely to happen near the equator . the pulsar magnetosphere has been traditionally considered as being force - free , and inertial effects are usually neglected . on the other hand , it is generally accepted that it is because of the inertia that the magnetic field line topology becomes open and an azimuthal magnetic field component is created . moreover , the pair creation and their acceleration to high lorentz factors ( e.g. , @xcite ) involve non - ideal mhd effects that may be important in the final form of the magnetosphere . a question arises : is the force - free a good approximation ? @xcite have solved the force - free problem ( although with the current - closure not fully satisfied ) and @xcite have repeated their numerical work with higher resolution , deriving similar results . the usual problem with the force - free dynamics is that the condition @xmath394 is not always satisfied , since the bernoulli equation that relates the electromagnetic field with the lorentz factor of the system of reference where the electric field vanishes , is omitted ( see eq . [ [ bernoulli - comb ] ] ) . in regimes where @xmath395 the drift velocity is larger than the light speed this is indeed the case in the aforementioned force - free solution . according to @xcite the drift velocity becomes larger than the light speed at distance a few times the light cylinder ( their figure 5 shows that typically this happens at cylindrical distance @xmath396 ) thus , the force - free assumption brakes - down much before the fast - magnetosound point ( @xmath397 ) . another problem of the force - free solution is that it implies negative azimuthal velocities near the source . @xcite demonstrated how we can find the flow speed in cases where the electromagnetic field is known . we need to solve the system of equations ( coming from eq . [ [ omega ] ] , and the second of eqs . [ [ v ] ] ) @xmath398 together with the identity @xmath399 . the solution for @xmath36 is actually equation ( [ bernoulli - comb ] ) .. this happens exactly in the positions were the drift velocity equals the light speed , i.e. , where @xmath400 . ] the problem related to negative azimuthal speeds comes from equation ( [ forcefree1 ] ) . the second part of the right - hand side is much smaller than unity very close to the source in a possible corotating regime , and should increase up to an asymptotic value equal to unity as we move to larger @xmath83 ( in order to have a negligible asymptotic azimuthal speed ) . practically speaking , the ratio @xmath401 should be @xmath402 at distances @xmath403 a few . after that point @xmath404 , and this continues to be the case even in the superfast ( non - force - free ) regime where both the @xmath150 and the @xmath227 decrease . but , what happens relatively close to the light cylinder ? the poloidal field lines follow the dipolar shape up to the point where they approach the equator , but then they remain parallel to it for some distance . we know that a relativistic flow is difficult to bend and this is a characteristic of a force - free regime as well . the curvature @xmath26 is of the order of @xmath405 , meaning that at distances @xmath403 a few , the poloidal field lines are practically straight . holds if we assume that @xmath406 in eq . ( [ transfield ] ) . very close to the midplane , the field may change in a small scale . in addition , the pressure gradient term may be important . it is because of these two effects that the field lines do not cross the equator . ] as a result , the quantity @xmath23 increases , because the surface between two neighboring field lines increases slower that @xmath407 . for a force - free case where @xmath150 remains constant , we get a decreasing ratio @xmath408 . since its asymptotic value should be unity , near the source is larger than unity , implying @xmath409 . in fact @xcite found negative azimuthal velocities . the only way to avoid this effect in the framework of force - free field , is to have very small initial poloidal velocity ( @xmath410 ) , meaning that our `` initial '' surface is very close to the pulsar . however , in that case we should take into account non - ideal effects related to the pair creation region . a solution to this problem could be that the flow in the regime where the lines remain approximately parallel to the equator is _ decelerating _ , meaning that kinetic ( or enthalpy ) energy flux is transfered to poynting , and the @xmath2 function increases . in that case , the @xmath411 would increase and cancel the increase of @xmath227 in the expression of @xmath412 , resulting in @xmath413 . during the phase where the flow is approximately parallel to the equator , the @xmath414 increases almost linearly with @xmath24 ( @xmath415 with @xmath416 ) . since the @xmath417 is forced to follow this increase ( in order to be consistent with a negligible azimuthal motion ) , the matter energy part is decreasing linearly with @xmath418 , and an important ( or even the most significant ) part of the poynting flux may be created there . as the flow decelerates , at some point it reaches @xmath419 . only after that point the curvature of the poloidal field lines @xmath420 may be important and the lines start to expand to larger heights above the equator . the expansion in principle allows for acceleration ( if @xmath421 increases faster than @xmath349 ) , and this time the electromagnetic energy is transfered to the matter . we consider this point as the origin of the ideal mhd regime that we examine in this paper . another problem comes when we consider the `` last '' current loop ( @xmath422 ) that closes on the equator . along this equatorial current line the azimuthal magnetic field vanishes . since @xmath423 , the inequality @xmath394 holds only in the unlikely case where the poloidal field also vanishes on the equator . the remaining case is to include non - ideal mhd effects ( then the electric field is not equal to @xmath424 and @xmath425 could be positive ) . although the details of the full magnetosphere remain to be explored , and this requires to solve the full problem ( not only the non - force - free , but also the non - ideal mhd especially near the `` last '' current line that closes on the equator ; not to mention the inner / outer gaps and the non - axisymmetry ) , the picture described above looks appealing and consistent with the observed crab - pulsar equatorial wind . the scenario also explains why the quantities @xmath426 , @xmath427 , and @xmath428 are likely @xmath0 just after the end of this `` parallel to the equator '' phase . these are the `` initial '' conditions for the ideal mhd phase . it may not be just a coincidence that the efficient and faster than logarithmic acceleration to low-@xmath2 values during the ideal mhd phase was possible for these `` initial '' conditions only . note also that the part of the total open magnetic flux that contributes to the `` parallel to the equator '' phase , is @xmath429 ( for exactly dipolar field , @xmath430 of the open magnetic flux correspond to field lines that have @xmath431 when cross the light cylinder ) . the rest @xmath429 of the open field lines likely follows a dipolar shape till @xmath432 a few , and then becomes conical ia ( again @xmath433 is negligible ) . since there is no obvious reason for an increasing @xmath227 , the acceleration gives asymptotically @xmath434 or @xmath435 , if the initial @xmath225 value is @xmath436 or @xmath437 , respectively . associated with the parallel electric field inside the non - ideal mhd regime , which points toward the axis for @xmath438 where @xmath439 . ] the `` initial '' conditions @xmath362 are in principle possible in other astrophysical settings as well , such as gamma - ray burst sources or agns . for example , a corona or a thermal / radiation pressure field may keep the field lines that emanate from a disk parallel to each other ( at least for some distance ) , in which case the product @xmath440 again increases and a matter energy - to - poynting flux transformation is realized ( e.g. , @xcite ) . in the first part of the paper ( [ governing ] , [ superalfvenic ] ) we derived the ideal mhd equations that describe the super - alfvnic asymptotic regime . the resulting system of equations remains intractable enough , and we could only find a @xmath30 self - similar model ( [ zss ] ) describing the regime near the rotation axis . we regard the solution @xmath256 as the best representative result of that model . it includes all the expected characteristics of an efficient mhd accelerator , in particular : * the transition from @xmath441 to @xmath18 , showing that ideal mhd can account for the full acceleration , resulting in a matter - dominated flow . * the transition from a current - carrying to a return - current regime , showing how the poloidal current lines close . * the transition from positive to negative poloidal curvature radius . when the flow becomes sufficiently matter - dominated ( @xmath442 ; see [ character ] ) it has a negative curvature radius and the line shape becomes hyperbolic . this is simply the result of angular momentum conservation : the azimuthal component of the velocity decreases , and the constancy of the total velocity implies an increasing cylindrical component . ( this small deviation from a conical line shape could be enhanced in the termination shock . ) the most important implication of this effect is not the hyperbolic line shape itself ( which still remains practically conical ) , but the new status quo in the transfield force - balance . it is now the poloidal and azimuthal centrifugal inertial forces that dominate and their difference is the much smaller transfield electromagnetic force component . * the transition from sub to super modified fast - magnetosound flow . this surface should be crossed since in the matter - dominated regime the fast - magnetosound speed becomes progressively smaller ( it vanishes at @xmath154 ) . in this sense it is a signature of an efficient acceleration . it is also related to the causality principle : only in the super modified fast - magnetosound regime none of the mhd waves can propagate upstream and reach the origin of the flow . we tried to present our numerical solutions in accord with the model of kennel & coroniti for the crab nebula . in solution @xmath256 the transition from @xmath443 to @xmath9 happens over less than five decades in distance from the pulsar , and the value @xmath8 is reached at distance of @xmath444 cm . as we discussed in [ scalings ] , @xmath2 decreases as @xmath445 in the non - force - free regime @xmath446 a few ( and even faster when @xmath265 a few ) . the assumption @xmath260 required in order to derive the @xmath30 self - similar model does not allow us to examine the wind near the equator . nevertheless , the presented numerical solutions serve as the first examples of low-@xmath2 asymptotic flows , and , more importantly , they guided us to think which is the situation in the most general ( non - self - similar ) case . our main conclusions ( described and justified in [ character ] and [ genanalysis ] ) are in summary : * type ia conical flows ( eq . [ [ straightshape ] ] ) are efficient accelerators . the @xmath2 decreases much faster than logarithmically . * the same for the parabolic ia , iia ( they likely apply to nonrelativistic flows ) . * for sufficiently small asymptotic values of the @xmath2 function , a transition @xmath3 exist , where the azimuthal centrifugal force is comparable to the electromagnetic force in the transfield direction . the acceleration may or may not continue after this transition . in both cases , however , the solvability condition at infinity ( heyvaerts & norman 1989 ; chiueh et al . 1991 ) is not mandatory . the latter should be satisfied only in cases where @xmath160 . note that for the equatorial pulsar wind the ratio @xmath447 is @xmath0 at the position of the shock ( for @xmath448 , @xmath449 , and @xmath450 ) . however , the solvability condition may not be satisfied , if the value of the @xmath2 required at the shock is not the asymptotic @xmath299 , but the @xmath2 function still decreases at the position of the shock . moreover , in applications to jets related to other astrophysical settings , the transition @xmath3 may be important . * in flows where the acceleration continues to @xmath451 , the line shape is hyperbolic and the curvature radius of the poloidal field - streamlines becomes negative . a required condition for the above acceleration mechanism to work is @xmath452 . as we discuss in [ genanalysis ] , this condition means that at the inner super - alfvnic regime @xmath227 should be @xmath453 , and the current that flows along the field lines should be much larger than the typical values obtained in monopole solutions , @xmath454 . we argue in [ equatorialwind ] that this condition is likely satisfied for the equatorial wind . the scenario presented in that section is outlined below with the help of the figure [ fig7 ] . the various transitions / regimes as we move along the reference ( thicker ) field line , marked with the numbered vertical dotted lines are : + origin @xmath455 1 : this is the most uncertain regime , and very important problem for future research . we can only speculate what may happen there . the poloidal field possibly starts with a dipolar shape , however , inertial effects ( associated with the azimuthal velocity ) create an open magnetosphere . ( note that even under force - free conditions the last closed field line does not in principle intersect the equator at the light cylinder ; see @xcite ) . non - ideal mhd effects accelerate created pairs to @xmath456 . one has to solve the dynamics by including the back - reaction from the various emission mechanisms to yield the exact value of @xmath457 . in principle , this value could correspond to matter - dominated flow @xmath458 ( although the poloidal field is strong , the poynting flux associated with the @xmath459 product may not be larger than the matter energy flux ) . the initial conditions after the pair acceleration in the non - ideal mhd regime may correspond even to super - alfvnic conditions in which case the light cylinder has not a particular physical meaning . is no longer related to the matter rotation . it is just the ratio @xmath460 ( e.g. , * ? ? ? * ; * ? ? ? ] + 1 : here @xmath461 a few . this may be the base of the ideal mhd regime ( although there should be non - ideal effects near the midplane even after that point , see [ equatorialwind ] ) . downstream from this point the super - alfvnic asymptotic analysis holds . according to equation ( [ bpbphi ] ) , @xmath462 , and @xmath463 also the poloidal curvature is @xmath464 and the poloidal field lines are approximately straight and roughly parallel to the equator . + 1 @xmath455 2 : the product @xmath227 increases as @xmath465 , where @xmath466 is the distance between two neighboring field lines @xmath21 , @xmath346 ( see fig . [ fig7 ] ) . since the @xmath467 increases , @xmath468 should also increase in order to keep the inequality @xmath469 true . thus , matter kinetic energy ( and possibly thermal energy ) is transfered to poynting . figure [ fig7 ] shows a current loop that crosses the reference field line between points 1 and 2 . here @xmath470 and the associated lorentz force @xmath471 _ decelerates _ the matter , increasing the @xmath2 function . also @xmath141 , @xmath138 , but the transfield component of the lorentz force [ @xmath472 remains small , resulting in @xmath473 , ( e.g. * ? ? ? + 2 : here the lorentz factor is sufficiently small ( @xmath474 ) for the field lines to start slightly bending . this is the base for the accelerating ideal mhd regime described in the previous sections . this point corresponds to an `` initially '' poynting flux - dominated wind . the `` initial '' conditions satisfy @xmath475 and the @xmath2 function is @xmath0 . + 2 @xmath455 4 : the field lines have a parabolic shape : @xmath476 slightly increases , resulting in a tiny decrease in @xmath227 . the @xmath2 function decreases according to the equation ( [ bernoullifinal ] ) . for the poloidal current density , @xmath141 , @xmath477 . + 3 : classical fast - magnetosound point , @xmath478 , @xmath479 . + 4 : the @xmath2 function is unity . equation ( [ bernoullifinal ] ) implies that @xmath480 . + 4 @xmath455 6 : this is the conical ia phase ( the discussion of [ genanalysis ] refers to this regime ) . the field lines become @xmath481 ( tangent to the dot - dashed lines in fig . [ fig7 ] ) . it is seen in figure [ fig7 ] how the conditions @xmath482 , @xmath343 are realized . the @xmath227 decreases according to equation ( [ straightfield ] ) . as a result , the @xmath2 function decreases from @xmath483 to @xmath484 given by equation ( [ sigmamin ] ) . + 5 : the current loop that the reference field line crosses at this point has @xmath142 , i.e. , here the current - carrying regime ( @xmath212 ) ends and the return - current regime begins ( @xmath485 ) . + 6 : here the @xmath486 is of the order of unity . the fast acceleration can not continue after that point . however , due to a tiny deviation from a @xmath487 it may continue logarithmically ( see @xcite and references therein ) . if at point 6 the value of @xmath488 is @xmath9 while at point 2 the flow is poynting dominated ( @xmath489 ) , then the ratio @xmath490 and the required value for the @xmath426 at point 2 is @xmath491 . assuming further that ideal mhd holds during the 1 @xmath455 2 phase and that the flow started at 1 with @xmath492 , we get @xmath493 . thus , @xmath494 . for a dipolar field before the point 1 , @xmath495 , and we get a required value for @xmath496 . however , it is not clear that all the above assumptions hold , so we can not be sure for the derived value @xmath497 . in general , the fast ( faster than logarithmic ) acceleration in magnetized outflows gives asymptotically @xmath498 , where @xmath227 is the `` initial '' value at the poynting dominated regime ( point 2 in fig . [ fig7 ] ) . besides the application to pulsar winds on which we focussed in this paper , the formulation presented is quite general and agn or grb relativistic as well as yso nonrelativistic jet asymptotics can also be considered ( the asymptotic analysis presented in [ mhd ] applies in all these cases , no matter if asymptotically @xmath29 or @xmath354 , or even @xmath499 ) . in addition , the @xmath30 self - similar model of [ zss ] could be used for examining agn or grb jets in their super - alfvnic regime . the author is grateful for useful discussions with arieh knigl and ruben krasnopolsky . this work was supported in part by nasa grant nag5 - 12635 and by the u.s . department of energy under grant no . b341495 to the center for astrophysical thermonuclear flashes at the university of chicago . support from a mccormick fellowship at the enrico fermi institute is also acknowledged .

emission in coincidence following absorption at rest in nuclei is studied using quantum mechanical scattering theory and nuclear wave functions . absorption is assumed to occur on two protons in the nucleus . in the formalism , emphasis is put on the study of the final state interaction ( fsi ) effects of and with the recoiling nucleus . we include elastic scattering and single nucleon knock - out ( ko ) channels in the fsi . calculations are presented for thec nucleus , using shell model wave functions , and without any extra mass modification of the system in the nucleus . calculated results are presented for the angular correlation distribution between and , their invariant mass distribution and the momentum spectra of and . these results are compared with the corresponding experimental measurements . with only elastic scattering fsi included , the angular correlation distribution and the momentum spectra are found to be in good accord with the corresponding measurements . with full fsi the calculated invariant mass distribution is found to have two peaks , one corresponding to the elastic scattering fsi and another to single nucleon ko fsi . the ko peak agrees fully , in position and shape , with the peak observed in ref . . the peak corresponding to elastic scattering fsi does not seem to exist in the measured distribution . considering that such a two peak structure is always seen in the inclusive ( , ) and ( , ) reactions in nuclei at intermediate energies , absence of the elastic scattering peak in the reaction is intriguing .

1003.1615

the @xmath9 interaction is attractive in s - wave and isospin , t = 0 state . because of this there is much interest in its study and the study of its implications in the possible existence of the @xmath1nucleus quasi - bound states in nuclei . experimentally , the existence of such bound states is indicated in the finuda measurements @xcite of the stopped @xmath1 absorption on li , c and other target nuclei . these experiments using the finuda spectrometer installed at the da@xmath10ne collider detect a @xmath3 hyperon and a proton pair in coincidence following @xmath1 absorption at rest on several nuclei . the emitted @xmath3-@xmath2 pair is found to emerge , predominantly back to back in all target nuclei , and have their invariant mass distributions peaking significantly below the sum of a kaon and two proton mass in free state ( 2.370 gev ) . if it is assumed that the @xmath3-@xmath2 pair is emitted from a @xmath5 " system in the nucleus , this mass shift implies a bound @xmath5 system in nuclei with the binding energy above 100 mev . in a more elaborate second run of these experiments carried out recently @xcite it is further reported that these mass shifts occur only for the @xmath5 module , and not for the @xmath11 cluster . the absorption on an @xmath12 pair gives @xmath3-@xmath13 and @xmath14-@xmath2 pairs in the final state . recent analysis of the old disto data from the saturne accelerator on @xmath15 reaction too suggests the existence of a @xmath5 cluster with the binding energy around 100 mev @xcite . theoretically , following the extraordinary success of the su(3 ) chiral perturbation theories in describing the @xmath16-@xmath17 and @xmath18-@xmath17 systems , the @xmath1-@xmath2 system has also been studied under this framework , though , unlike pion and @xmath18 cases , the basic interaction here is relatively strong . this , however , is incorporated in these studies by including terms up to order @xmath19 in the chiral lagrangian expansion . then , in combination with non - perturbative coupled - channel techniques this framework has been found quite appropriate for the study of antikaon - nucleon interaction in the literature . it was first developed in ref . @xcite , and subsequently expanded in ref . various channels involved for s = @xmath201 meson - baryon scattering are @xmath21 , @xmath22 , @xmath23 , @xmath24 , @xmath9 , @xmath25 . with the proper choice of parameters entering in these calculations , all available low - energy scattering data in these channels are reproduced well . the @xmath9 scattering amplitude resulting from these calculations have a two pole structure between @xmath26 and @xmath27 thresholds . the pole which is located close to the real axis couples strongly to the @xmath27 channel , while the one coupling strongly to @xmath28 channel lies away from the real axis . empirically , only available information pertaining to @xmath9 scattering below threshold is the @xmath26 invariant mass distribution . this mass spectrum has its maximum at 1405 mev , and has a width of about 50 mev . in the particle data group table @xcite this is identified as t = 0 , spin half , s = @xmath201 @xmath29 @xmath9 bound state with a binding energy of 27 mev . however , it is noticed in the literature that there is some subtlety involved in assigning a mass to the @xmath29 . the observed @xmath26 spectrum has t = 0 . hence , in principle , it can have a generic s - wave t = 0 source . this means that one needs to fit a superposition of the contributions from both , the @xmath27 and @xmath30 poles mentioned above , to reproduce the observed @xmath30 spectrum and assign a mass to @xmath29 . within this scenario it turns out that both the poles contribute roughly in equal measure to reproduce the measured @xmath26 spectrum , with a tendency towards higher @xmath27 share . a more thorough investigation of this issue has been carried out recently in ref . they conclude that for the study of @xmath27 scattering , amongst various channels involved in the coupled channel calculations , the @xmath27 and @xmath26 channels dominate and they couple strongly . they also conclude that the mass of the @xmath29 state is in fact 1420 mev , and not 1405 mev , thus making it only 12 mev below @xmath27 threshold . recent measurements on the @xmath31 reaction at cosy @xcite , however , seem to present experimental evidence which do not support the above two pole model for the @xmath29 . the shape and position of the @xmath29 distribution in these measurements is reconstructed cleanly in the @xmath32 channel using invariant- and missing - mass techniques . the mass of the @xmath29 is found to be @xmath33 1400 mev and width @xmath33 60 mev . theoretical search for the antikaon - nucleus bound state has been carried out in the literature following the variational approach @xcite and the faddeev method @xcite for @xmath1 plus 2 - 3 nucleons and heavier nuclei . all these calculations need , as input , realistic choice for the @xmath34 and the @xmath1-@xmath17 potentials . for the nucleon - nucleon potential , following extensive work over the years on this subject , it is always possible to make a correct choice . for the @xmath1-@xmath17 potential , however , the situation is uncertain . some calculations generated a pseudo - potential for it by reproducing the @xmath1-@xmath2 bound state of 27 mev binding energy , while others used a leading order chiral interaction . they all found @xmath5 bound states with about 50 mev or more binding energies . latest calculation in ref . @xcite , which , following the two pole model , uses the @xmath1-@xmath17 effective potential corresponding to 1420 mev mass of @xmath29 , finds a @xmath5 system bound by around 19 mev only . this state has a width between 40 and 70 mev . this suggests that the @xmath5 module might not be sufficiently bound to produce any experimentally observable signal corresponding to the antikaon - nucleus mesic bound state . the situation is further confused by the suggestion in ref . @xcite , that the down - shift observed in the invariant mass of the @xmath2 and @xmath3 in the finuda experiment could be the result of the final state interaction of these particles with the recoiling nucleus . this seems quite plausible because , due to q - value of the @xmath35 reaction being around 317 mev , the kinetic energies of outgoing @xmath2 and @xmath3 are 160 mev or so . at these energies , it is well known that in the nucleon - nucleus scattering the reactive cross section mainly consists of the single nucleon knock - out channel @xcite . thus , the knock - out of one nucleon in the nucleus by the out going @xmath2 or @xmath3 can shift their energies considerably . the calculations in ref . @xcite indeed reproduce the observed mass shifts in the finuda experiments . however , as mentioned earlier , the similar effects not observed in run ( 2 ) of finuda in @xmath1-@xmath36 absorption can not be reconciled with this explanation @xcite . therefore , the situation on the ( @xmath1 , @xmath0 ) reaction in the nucleus seems very confusing . it calls for more studies on the description of the reaction dynamics , as well as the @xmath1-nuclear binding . in the present paper we reexamine the hypothesis of magas _ et al . _ @xcite of the origin of the observed @xmath3-@xmath2 peak in the finuda measurements to the single nucleon knock - out events in the final state . the calculations reported in ref . @xcite are the computer simulations of the internuclear cascade model for the nucleon - nucleus scattering . this approach describes the sequence of nucleon - nucleon collisions of the outgoing nucleon while passing through the residual nucleus in the final state in the framework of classical physics . the trajectory of each nucleon is followed . after a mean free path a @xmath17-@xmath17 collision takes place and its results are computed by monte carlo or some similar method . apart from the pauli principle there are no quantum mechanical effects in this approach of describing the fsi . the nucleus too is described by the fermi gas model , thus being totally devoid of any nuclear structure effect . in view of the crucial role played by the fsi in interpreting the finuda @xmath3-@xmath2 measurements for @xmath1-nuclear bound states , it is absolutely necessary that the fsi in this reaction is described using quantum mechanical scattering theory and nuclear wave functions . this is the purpose of the present paper . the paper is organized as follows . in section 2 , first , based on physical reasoning , we present an overall description of the @xmath37 reaction following @xmath1 absorption in the nucleus along with an appropriate final state interaction . this is followed by the formalism utilized for the evaluation of the cross section . calculated results along with a discussion around them are presented for @xmath4c target nucleus . this is followed by the conclusions . to remain specific in our discussion and presentation , in the following we consider @xmath4c target nucleus all along . the @xmath1 meson after being captured in a high atomic orbit reaches the @xmath38 orbit through electromagnetic transitions . from there onward it comes under the influence of the strong nuclear interaction and gets absorbed . the x - ray transition width for the @xmath38 @xmath39 @xmath40 transition and the nuclear capture from the @xmath38 orbit are reported in @xcite to be 0.0749 ev and 0.98 @xmath41 0.19 ev respectively . this gives the relative population of kaons in @xmath40 and @xmath38 orbits , using @xmath42 around 7@xmath43 . such a small population of kaons in the @xmath40 orbit also makes nuclear capture of kaons from the @xmath44 orbit as insignificant . we , thus , consider in our calculations the capture of the kaon from both the @xmath38 and @xmath40 atomic orbits . the @xmath1 absorption yield for the @xmath45 branch is written as the weighted average yield from these orbits as @xmath46 however , as we will see later ( fig . 1 ) , due to larger centrifugal barrier , the overlap of the @xmath38 orbit with the nuclear wave functions is about 2 orders of magnitude smaller than that of the @xmath40 orbit . furthermore , as the calculation of the absorption yield , @xmath47 involves the square of this overlap , despite the factor of 0.07 in the above for the @xmath40 orbit , the contribution from the @xmath38 orbit to the @xmath1 nuclear capture remains about two orders of magnitude smaller than that from the @xmath40 orbit . therefore , in the following we present calculations considering the capture from the @xmath40 orbit only . in the absorption process , the @xmath3 hyperons in the final state are produced either in the quasi - free process @xmath48 or the two - nucleon absorption process @xmath49 . for the measurements which involve @xmath2-@xmath3 detection in coincidence , obviously only the two - nucleon absorption contributes . the absorption on more than two nucleons is expected to be weak because the probability of finding three or more nucleons together in the nucleus is small . the q - value of the @xmath50 process ( ignoring nuclear binding of the absorbing protons ) is 317 mev . this energy is mainly shared by the emerging nucleon and the @xmath3 hyperon . furthermore , since the fermi momentum of the nucleons in the nucleus is not large , the outgoing nucleon and the @xmath3 hyperon following @xmath1-absorption at rest emerge back to back and have their momenta centred around 570 mev / c . of course , due to fermi motion , in an actual situation the back to back correlation is smeared into two narrow cones , and momenta of the nucleon and the @xmath3 are spread around 570 mev / c by the fermi motion . additionally , the target nucleus , after absorption , is left into a two hole shell model state . structurally , two aspects of nucleon motion in the nucleus appear in the @xmath51 absorption process in the nucleus . because of the predominantly back to back emission of @xmath2 and @xmath3 hyperon and their momenta being centred around 570 mev / c , the absorbing pair of nucleons in the nucleus needs to be as close as around 0.2 fm to each other at the time of @xmath1 absorption . since the @xmath17-@xmath17 potential at these distances is very strong , the relative wave function of these protons has strong short range correlations . these correlations in the nucleus , however , heals very fast @xcite and the wave function goes over to the shell model mean field wave function . the centre of mass of these two protons , however , has no such constraints on it , hence , it always moves in the most probable trajectory given by the nuclear mean field . the appropriate 2-proton wave function in the nucleus for the kaon absorption , therefore , has the form , @xmath52 where @xmath53 s are the shell model wave functions and @xmath54 is a jastrow - type correlation function @xcite . the outgoing @xmath2 and @xmath3 will also be correlated similarly by a correlation function , say @xmath55 . the healed @xmath2-@xmath3 wave function here , however , will be a phase shifted wave function . consequent to the above completely two different space scales involved , the absorption probability for @xmath1 in the nucleus for the @xmath2-@xmath3 branch ( for absorption on a @xmath56 pair , say ) factors into two parts ( shown in the next section ) , @xmath57 where @xmath58 and @xmath59 are the centre - of - mass and the total momenta of the @xmath2 and @xmath3 respectively . @xmath60 and @xmath61 are respectively the absorption strength for @xmath62 process in their centre of mass and the momentum probability distribution of nuclear wave functions corresponding to the total momentum , @xmath63 . because of the back - to - back emission of @xmath2 and @xmath3 , obviously the magnitude of @xmath64 is very large and that of @xmath63 is small . due to these vastly different momentum scales for @xmath64 and @xmath63 , over most of the variables measured in the @xmath51 reaction , while @xmath65 can go through a large variation , the factor @xmath66 does not change much . showing the localization of the @xmath1 absorption for the @xmath40 and @xmath38 orbits . the vertical line shows the @xmath67 radius of the @xmath4c nucleus . ( b ) distribution of the @xmath40 and the @xmath38 @xmath1 wave functions.,width=604,height=264 ] before we talk about the final state interaction ( fsi ) , let us mention that the @xmath1 absorption in the nucleus occurs on its surface . quantitatively , this region is determined by the overlap of the @xmath1 @xmath40 and @xmath38 atomic orbits with the spatial distribution of the absorbing protons . 1a shows these overlaps , where @xmath68 for the @xmath4c nucleus we have taken the two protons moving in the @xmath69 shell model orbital . their radial distribution is described by the harmonic oscillator wave function . the kaonic atomic orbits are given by the hydrogenic wave function . to show their relative localization clearly , in fig . 1a we plot @xmath70 s for the @xmath40 and the @xmath38 @xmath1 orbits , which have different magnitudes , on the same scale with arbitrary units . we see that both the overlap functions peak around r = 2.5 fm , with @xmath38 overlap function about half a fermi ahead . the @xmath4c rms radius is known to be 2.4 fm . the magnitude of the @xmath38 overlap function , we note is two orders of magnitude smaller than that of the @xmath40 orbit . this happens because the @xmath38 wave function rises slower than the @xmath40 @xmath1 wave function in the region of the overlap , as we see in fig . above localization of the @xmath70 beyond the @xmath4c rms radius coupled with the back to back emission of @xmath2 and @xmath3 following absorption , suggests that only one of the two emitted particles , @xmath2 and @xmath3 , goes into the nucleus at a time . the other particle moves outward . hence , the fsi with the recoiling nucleus is mainly suffered by only one particle , @xmath2 or @xmath3 at a time . the channels which dominate in contribution to this interaction are the elastic channel and the single nucleon knock - out ( ko ) channel . the latter is known to constitute about 80@xmath43 of the total reactive cross section in the proton - nucleus inelastic scattering in nuclei around 160 mev proton energy @xcite , which is the relevant proton energy in the present study . out of these , the effect of the elastic channel at intermediate energies is mostly absorptive , while that of the ko channel is dispersive as well as absorptive . in our calculations , we include both the channels . the inclusive probability for a process like @xmath71 is therefore written as @xmath72 the sum of two terms in the above is incoherent because in principle ( by making exclusive measurements ) we can distinguish between elastic scattering and single scattering with one target nucleon being knocked out . the elastic " process for the @xmath73 reaction is shown in fig . the @xmath1 gets absorbed on a pair of protons in the target nucleus and produces a @xmath2-@xmath3 pair . this pair is detected in coincidence in the final state . no measurements are made on the recoiling nucleus , hence the measurements are inclusive in that sense . the recoiling nucleus is left in a _ two hole _ state centred around the excitation corresponding to the summed binding energy ( @xmath74 ) of nucleons 1 and 2 in the nucleus a. let us denote different states of b around this excitation by @xmath13 . the inclusive absorption probability for protons in shells ( @xmath75 ; @xmath76 ) in the nucleus is then given by @xmath77 where @xmath78 and @xmath79 is the excitation energy of the state @xmath13 in @xmath80 . @xmath81 denotes the kinetic energy of the particle @xmath82 . bar on the sum in the above expression denotes the average and sum over the spins in the initial and final states respectively . the transition matrix element @xmath83 is given by @xmath84^{1/2}\,\int d\xi\,\,dx_1\,\,dx_2\,\ , \psi_{b\,,\,n}^*(\xi)\,\,\chi^{-*}(x_1,x_2)\,\,h_{12}^{k^-}\,\ , \psi_a(\xi , x_1,x_2)\,\,\phi_k,\ ] ] where @xmath85 are the active number of absorbing proton pairs in shell ( @xmath75 ; @xmath76 ) in the target nucleus , @xmath86 . @xmath87 is the nuclear wave function , and @xmath88 is the elastically scattered @xmath2 and @xmath3 wave function . @xmath89 is the absorption vertex , and it depends only on the proton coordinates , @xmath90 and @xmath91 in the nucleus a. @xmath92 represents collectively the coordinates of the @xmath93 nucleons . to proceed further we note that since the excited states ` @xmath13 ' in the nucleus @xmath80 are the hole states corresponding to two nucleons , they are not likely to have much energy spread . hence , in the energy delta function in eq . ( 8) we replace @xmath79 by @xmath94 . with this we obtain @xmath95\,\sum_n \bar{\sum _ \sigma}\,|m_{fi}|^2,\ ] ] with @xmath96 $ ] , the phase space factor , given by @xmath97\,=\,\frac{1}{(2\pi ) ^5}\,\,\delta(m_i\,-\,t_\lambda\,-\,t_p\ , -\,t_b\,-\,b_{12})\,\,d\vec{p_\lambda}\,\,d\vec { p_p},\ ] ] and @xmath98 . sum over ` @xmath13 ' is now performed using the closure relation " , yielding @xmath99\,\,\bar{\sum _ \sigma}\,\int d\xi\,\,\left|\,\int dx_1\,\,dx_2\,\,\chi^{-*}(x_1,x_2)\,\,h_{12}^{k^-}\,\ , \psi_a(\xi , x_1,x_2)\,\,\phi_k\,\right|^2.\end{aligned}\ ] ] since we are interested only in the inclusive absorption strength we take a simple description of the target nucleus , where the absorbing protons move in shell model orbitals @xmath100 and @xmath101 and the core of @xmath93 nucleons is a spectator . with this description , above expression reduces to @xmath102\,\,\bar{\sum _ \sigma}\,\,\left|\ , \int dx_1\,\,dx_2\,\,\chi^{-*}(x_1,x_2)\,\,h_{12}^{k^-}\,\ , \psi_{n_1l_1m_1;n_2l_2m_2}(x_1,x_2)\,\,\phi _ k\,\right|^2,\ ] ] where @xmath103 is the properly anti - symmetrized two proton wave function in the nucleus . in the @xmath104 representation it is written as @xmath105 for two protons in the same shell antisymmetry requires that @xmath106 . further on , performing sum over @xmath107 and @xmath108 contained in @xmath109 in eq . ( 13 ) , we get @xmath110 the correlation functions @xmath54 and @xmath55 mentioned in eq . ( 3 ) and after it are absorbed here in the absorption vertex @xmath111 . to proceed further , let us now utilize the fact that the momentum @xmath113 appearing in the absorption vertex @xmath111 has large magnitude and a short range . because of this , in the expression for @xmath114 we can factorize the expectation value of @xmath111 from the rest , and write , @xmath115\,\,\times \\ \nonumber & & \left[\,\frac{n_{l_1\,l_2}}{(2l_1 + 1)(2l_2 + 1)}\,\,\sum_{lm}\ , \left|\,\int d\vec{r_1}\,\,d\vec{r_2}\,\,\chi ^{- * } ( \vec{r_1},\vec{r_2})\,\,\phi_k(\vec{r_1})\,\,\phi_{lm } ( \vec{r_1},\vec{r_2})\,\,\delta(\vec{r_1}-\vec { r_2})\,\right|^2 \,\right].\end{aligned}\ ] ] two expressions in the square brackets in above can be identified with two terms of eq . ( 4 ) , i.e. @xmath116,\ ] ] and @xmath117\,\,n_{l_1\,l_2}\frac { 1}{(2l_1 + 1)(2l_2 + 1)}\,\times \\ & & \nonumber \sum_{lm}\,\,\left|\,\int d\vec{r_1}\,\,d\vec{r_2}\,\ , \chi^{-*}(\vec{r_1},\vec{r_2})\,\,\phi_k(\vec{r_1})\,\ , \phi_{lm}(\vec{r_1},\vec{r_2})\,\,\delta(\vec{r_1}-\vec{r_2})\ , \right|^2.\end{aligned}\ ] ] as the energies of the proton and the lambda following @xmath1 absorption is around 160 mev or so , we describe the scattering of these particles by the recoiling nucleus using eikonal approximation . the basic assumption in this description is that the propagating particle is mainly scattered in the forward direction . taking z - axis parallel to the proton momentum , @xmath119 , the proton distorted wave , @xmath120 is written in eikonal approximation as @xmath121 where the distortion function @xmath122 is given in terms of an optical potential , @xmath123 by @xmath124,\ ] ] where @xmath125 . for writing the @xmath3 distorted wave we recall that the @xmath3 moves opposite to the proton . therefore , the momentum vector @xmath126 is anti - parallel to the chosen z - axis . the distortion factor , @xmath127 for @xmath3 therefore becomes @xcite @xmath128.\ ] ] combining @xmath129 for the proton and the lambda we then get @xmath130.\ ] ] if we make the @xmath131 approximation for @xmath132 and assume forward scattering for @xmath133 , we get @xmath134,\ ] ] where @xmath135 and @xmath136 are respectively the total cross section and the ratio of the real to imaginary part of the scattering amplitude for the @xmath137 system . now , if we ignore the difference between the proton and the lambda elementary scattering parameters and take them as that for the better studied @xmath138 system at some mean value of the @xmath2 and @xmath3 energies , above expression simplifies to @xmath139 \\ & = & exp\,\left[\,\frac{i}{2}\,\,\sigma_t^{pn}(i+\beta _ { pn})\,\ , t(\vec{b})\,\right],\end{aligned}\ ] ] where @xmath140 is the nuclear density and @xmath141 is the total nuclear material seen by the proton and the lambda _ together _ at an impact parameter , @xmath142 . it is given by @xmath143 with @xmath144 . now if we observe eq . ( 24 ) for @xmath145 a little closely , we realize that this , in fact , is the total distortion factor for the passage of a proton from one end of the nucleus to another . this , thus , is the mathematical description for the statement made in an earlier section that , because of the peripheral localization of the @xmath1 absorption and the back - to - back emission of @xmath2 and @xmath3 , total scattering of the @xmath2 and the @xmath3 can be included in the final state by considering the passage of only one particle ( @xmath2 or @xmath3 ) through the whole nucleus . furthermore , since the effect of distortion at the energies of @xmath2 and @xmath3 in the studies here ( @xmath33 160 mev ) is mainly absorptive , and most of the measurements on @xmath146 following @xmath1absorption are of inclusive type , it will be reasonably correct to include the overall effect of the distortion in @xmath65 by multiplying it by an attenuation factor , @xmath147 , given by @xmath148 and removing the distortion factor from the integral in eq . factor @xmath149 in above has been introduced before @xmath150 to include the effect of pauli - blocking of the nucleons in the nucleus after scattering . a nuclear matter estimate for its value above twice the fermi energy is given by @xcite @xmath151 where @xmath152 , with @xmath153 denoting the fermi energy . with above treatment of distortion , eq . ( 18 ) for @xmath65 simplifies to @xmath154\,\,|\eta_a(t_p)|^2\,\,\sum_{lm}|\,f_{lm}^{l_1l_2}(q)\ , \nonumber \sum_{lm}|\,f_{lm}^{l_1\,l_2}(q)\,|^2&=&\sum_{lm}\left|\ , \sqrt{\frac{n_{l_1\,l_2}}{(2l_1 + 1)(2l_2 + 1)}}\,\,\int d\vec{r_1}\ , \,e^{-i\,\vec{q}\,\cdot\,\vec{r_1}}\,\,\phi_k(\vec{r_1})\,\ , \phi_{lm}(\vec{r_1},\vec{r_1})\,\right|^2 \\ & = & n_{l_1\,l_2}\,\,\sum_l ( l_1\,l_200\,/\,l\,0)^2\,\,|\ , g_{l_1\,l_2}^l(q)\,|^2,\end{aligned}\ ] ] with @xmath155 where @xmath156 are the nucleon radial wave functions in the nucleus . ( 26 ) ) for @xmath157li , @xmath4c and @xmath158ca nuclei.,width=415,height=340 ] the value of @xmath147 depends upon the proton energy , @xmath159 through @xmath160 and upon the nucleus through the thickness function , @xmath161 . to get an idea about the value and its variation , in fig . 3 we plot @xmath147 in the proton kinetic energy range 50 - 250 mev for three nuclei , @xmath157li , @xmath4c and @xmath158ca . for calculating the pauli blocking factor , @xmath149 , fermi momentum is taken equal to 200 mev / c . the nuclear densities are taken from ref . we find that in the energy range ( 100 - 200 mev ) of interest in the present calculations the attenuation factor approximately remains constant . the values of this factor for the three nuclei are around 0.9 , 0.7 and 0.4 respectively . prescription to describe the absorption vertex , @xmath162 is not clear and also not simple . one thing , which is definite about it is that , it involves large momentum transfer , hence , spatially it can not be localized over any extended volume . dynamically , the one - nucleon absorption mechanism , @xmath163 is understood to involve the @xmath164 , which decays in to a pion and a hyperon . the range of this vertex is determined by the @xmath164 propagator . a study in ref . @xcite suggests that the absorption probability depends upon this range , and in their estimate this range could be around 0.75 fm or so . the two nucleon mechanism , due to strong attractive @xmath1-@xmath2 interaction in the t = 0 state , involves dynamically a strongly correlated system of @xmath165 , where the @xmath1 is continuously exchanged between two protons . detailed dynamical composition of this system is determined in the @xmath166pts by the non - perturbative coupling amongst various s = @xmath201 , t = 0 channels , @xmath21 , @xmath22 , @xmath167 , @xmath24 , @xmath9 , @xmath25 . this system eventually decays in to @xmath146 . exact mechanism of this decay is not immediately obvious . however , in line with the one - nucleon @xmath1 absorption , one mechanism could be that at some stage in the @xmath165 system a @xmath164 is produced , which , as suggested in ref . @xcite , interacts with another proton through an exchange of pions or a pair of @xmath16-@xmath168 and goes over to the @xmath2-@xmath3 final state . in ref . @xcite , using the range parameter 0.2- 1.2 fm for the absorption vertex the authors estimate the decay width of 2 - 8 mev for the @xmath165 system , with the maximum width occurring for the range around 0.7 fm . thus , it appears that , even if the details of the absorption vertex is not known very clearly , two things are clear : ( i ) the magnitude of the two proton @xmath1 capture depends upon the spatial extension of the absorption vertex , and ( ii ) the probable range of the vertex is such that the variation of the capture probability with momentum , @xmath64 could not be very rapid . the ( @xmath1 , @xmath0 ) reaction measurements have been done on the distributions of the @xmath169 ( and proton ) momentum , invariant @xmath0 mass , and their angular correlations . in all these distributions , as we will discuss in the next section , each point involves a folding of @xmath66 and @xmath65 . however , the values of @xmath64 for all the measurements are large ( around 500 mev / c ) and do not have much spread ( only up to 10@xmath43 ) . because of this , the factor @xmath162 ( eq . ( 17 ) ) in the formalism enters in determining the absolute magnitude of the ( @xmath1 , @xmath0 ) process only . the shapes of different distributions are determined by the factor @xmath65 . furthermore , as the available data from the experiments exist only in arbitrary units , we have taken @xmath162 as a constant factor , denoted by @xmath170 , in our calculations . before we present the results let us make some points about the finuda experiment . the @xmath3 s are detected in this experiment by reconstructing the invariant mass of the @xmath3 decay products , @xmath2 and @xmath171 . however , the restriction on the low momentum threshold for @xmath171 in the finuda spectrometer is such that it cuts out the @xmath169 hyperons with a momentum lower than 300 mev / c . therefore , the @xmath169 from the quasifree process ( @xmath172 ) in this experiment is hardly observed . above around 400 mev / c , since the main contribution comes from two - nucleon absorptions ( @xmath173 ) , the finuda measurements have major contribution from this capture process . furthermore , the measurements are done with the @xmath2 and @xmath169 in coincidence , with the cosine of the angle between them restricted as @xmath174 cos @xmath175 . the latter constraint is put because the cross section beyond these limits is insignificant ( see fig . 4 ) . for calculations , putting all factors together from the above section , the differential absorption strength , @xmath176 for @xmath1-absorption on two protons in shell model orbitals , ( @xmath75 ; @xmath76 ) is finally written as @xmath95\,c\,|\eta_a(t_p)\,|^2\,n_{l_1\,l_2 } \,\,\sum_l(l_1\,l_2\,0\,0\,/\,l\,0)^2\,\,|\,g_{l_1\,l_2}^l(q)\,|^2,\ ] ] where all the terms are as defined in above sections . we present here calculated results using this expression for the @xmath4c target nucleus with an appropriate phase space factor , @xmath96 $ ] . the proton wave functions in @xmath4c are generated in an oscillator potential , whose length parameter , @xmath177 is taken equal to 1.67 fm . this parameter fits the elastic electron scattering data on the @xmath4c nucleus @xcite . the binding energies of protons in @xmath69 and @xmath44 shells are taken as given by the ( @xmath2 , @xmath40 ) and ( @xmath7 , @xmath178 ) reactions @xcite . for @xmath4c they are 15.96 and 34.0 mev respectively . the @xmath1 absorption is considered to occur on proton pairs in @xmath69 , @xmath44 and @xmath179 shells . the value of @xmath85 for these absorptions is taken equal to the number of possible proton pairs in these shells , as an upper limit . they are in the ratio of @xmath180 . the phase space factor , @xmath181 for calculating the angular correlation between the @xmath2 and @xmath169 and their momentum distribution is written as latexmath:[\[ps\,=\,\frac{4\,\pi\,m_p\,m_{\lambda}}{(2\pi)^5}\,\ , \frac{|\vec{p_p}|^2|\,\vec{p_\lambda}|^2}{(m_{\lambda}\,+\,m_b)\ , ( \theta_{\lambda\,p})}\,d\,cos\,(\theta_{\lambda\,p})\,d\ , in fig . 4 we present the calculated angular correlation between the @xmath3 and the proton along with the finuda measurements . we find them to agree very well with each other . the steep rise in the distribution towards @xmath183 = 180@xmath184 is the strong confirmation of the two - proton absorption mechanism . both the results are given in relative units . and @xmath2 along with the finuda measurements @xcite.,width=415,height=340 ] measurements have also been done in the finuda experiment on the momentum distributions of the proton and @xmath3 . the measured @xmath3 distribution from ref . @xcite is shown in fig . 5 . if one believes that the observed @xmath2 and @xmath3 come from the two - proton absorption vertex and do not undergo any further fsi except the elastic scattering , this distribution should be similar to the one calculated using eq . we show the calculated @xmath3 and proton momentum distributions in fig . 5 , and find that the peak position of the @xmath3 distribution nearly agrees with the corresponding measured distribution . the shape of the calculated distributions is however found to be less broad compared to the observed one . larger magnitude of measured events below 400 mev / c , which makes it broader can not be understood easily . they can not be attributed to the quasifree process either because the finuda spectrometer cuts off @xmath169 s below 300 mev / c . another source of this deficiency can be that , in our calculations we have taken @xmath162 ( denoted by @xmath170 in eq . ( 31 ) ) as constant . we examined it . as we see in the phase space expression ( eq . ( 32 ) ) , calculation of the absorption probability for each value of the @xmath185 involves an integral over @xmath186 , which in the finuda measurements lies between @xmath201.0 and @xmath200.8 . this implies in the calculation for each @xmath187 an integral over a certain range of @xmath64 and @xmath63 corresponding to this range of @xmath146 angle . for 400 , 500 and 600 mev / c values of @xmath187 ( which more or less covers fig . 5 ) this range of @xmath64 is about 510 - 495 , 520 - 505 and 515 - 500 mev / c respectively . these values , as we see are large and have about same range for all the @xmath3 momenta . therefore our assumption about the constancy of @xmath162 in the calculations should not be the cause of concern . the shape of the @xmath187 distribution is , in fact , mainly determined by @xmath65 through the @xmath63 dependence of the nuclear wave functions . the range of @xmath63 for the above three @xmath188 s in the @xmath189 are 200 - 260 , 35 - 220 , and 160 - 270 mev / c respectively . this range shows why the @xmath187 momentum distribution peaks around 500 mev / c . it also suggests that it can probably be made broader by enriching the nucleon shell model wave functions in high momentum components . however , to reproduce the width of the observed momentum distribution , we believe that it will require a considerable modification of the nuclear wave function . and @xmath2 inclusive momentum distributions with @xmath174 cos @xmath190 along with the measured @xmath169 momentum distribution ( represented by histogram).,width=415,height=340 ] next we calculate the @xmath2-@xmath169 invariant mass distribution . the phase space for this is written as @xmath191 where @xmath192 denotes the reduced mass of the @xmath193 system . the magnitudes of @xmath194 and @xmath113 are determined for a given invariant mass @xmath195 through @xmath196 and @xmath197 the angle @xmath198 between @xmath194 and @xmath113 is constrained such that @xmath199 . calculated invariant mass distribution along with the corresponding measured finuda distribution @xcite are shown in fig . we observe that , compared to the mass of @xmath200 in the free state ( 2370 mev ) the calculated distribution is down - shifted in mass by about 50 mev due to proton binding in @xmath4c . the measured mass distribution , however , is still below this by an additional 70 mev or so . this , incidentally , is around the reported calculated binding energy in the literature of the @xmath5 " module taking @xmath164 mass 27 mev below @xmath1-@xmath17 threshold . the range of the values of @xmath64 and @xmath63 for @xmath201 mev , ( which covers the calculated distribution ) are 520 - 490 and 27 - 320 mev / c respectively . this says that the values of @xmath64 are large and do not vary much , hence @xmath66 is not likely to influence the shape of the mass distribution . its shape would be mainly determined by the nuclear wave function through @xmath65 and the factor @xmath63 in the phase space . and @xmath2 along with finuda measurements . @xmath174 cos @xmath175.,width=453,height=377 ] if we include in the kinematics of our calculations an additional binding energy of 70 mev in the initial system , then , we find that the peak position in the calculated mass distribution ( not plotted here ) , of course , comes near to the peak position of the measured distribution , but the shape of the calculated distribution turns out much sharper than the experimental one . the calculated @xmath3 momentum distribution with the additional binding also gets shifted towards lower momenta . this then spoils the agreement of the calculated results with the experiments shown above in fig . 5 without any @xmath200 " binding . the angular correlation between @xmath3 and proton , however , remains unchanged . in the two - nucleon @xmath1 absorption model , with only elastic scattering of @xmath2 and @xmath3 included in the final state , the measured @xmath3-@xmath2 angular correlation distribution is reproduced well , and the momentum spectrum of @xmath3 s to a reasonable extent without introducing any additional binding of the @xmath5 " module . the calculated invariant @xmath3-@xmath2 mass distribution , however , peaks around 70 mev higher than the measured one . attributing this shift to the additional binding of the @xmath5 " module , the calculations including this binding in kinematics , obviously , reproduce the peak position of the measured mass distribution , but the shape of the distribution remains very narrow . it also spoils the agreement with the @xmath3 momentum distribution achieved without including any @xmath200 " binding . thus , in an overall conclusion it seems that the observed invariant mass distribution in the ( @xmath1 , @xmath0 ) reaction does not correspond to the peak predicted in this section with only elastic scattering included in the fsi . the knock - out scattering means that the proton or @xmath169 , emanating from vertex i , encounters a hard collision with a nucleon in the residual nucleus and knocks it out ( fig . this collision alters significantly the momentum distribution of the striking particle . to calculate the knock - out contribution we recollect that , as discussed in the earlier section , @xmath2 and @xmath169 move back to back at vertex i and the vertex itself is localized on the surface of the target nucleus . because of this only one particle goes into the nucleus , the other one moves out . for such trajectories , as discussed earlier in section a(2 ) for distorted waves in elastic scattering , here too the contribution of @xmath2 and @xmath169 to ko fsi can be included by considering the passage of only one of the two particles , but through the whole nucleus . this can be reasonably correct if the elastic scattering parameters for @xmath138 and @xmath3-@xmath17 systems at intermediate energies are not much different . considering that it is so , in the following we calculate the ko contribution induced by the proton . let @xmath202 describe the momentum distribution of the proton when it incidents on the vertex ii , where @xmath203 is the momentum of the proton at the time of leaving the vertex i. the altered momentum of this proton after the ko scattering is denoted by @xmath119 . the proton with this momentum along with the @xmath3 are detected in the experiment . the knocked - out nucleon from the nucleus is not seen . the probability for this inclusive " process is determined by the product of @xmath202 and the proton induced inclusive single nucleon knock - out cross section as @xmath204 where @xmath205 is the total proton - nucleus cross section at the momentum @xmath206 . momentum @xmath207 is given in terms of the variables at the first vertex . in terms of @xmath126 it is given by @xmath208^{1/2}\right],\ ] ] + where all the notations are self explanatory . @xmath81 represents the kinetic energy of the particle @xmath82 . function @xmath210 is obtained from eq . ( 31 ) , by first writing the phase space @xmath96 $ ] for @xmath126 as , @xmath97\,=\,\frac{m_\lambda\,m_b}{(2\pi)^5}\,\,\frac{|\vec{p_p^\prime}| \vec { p_p^\prime}|m_\lambda cos\,(\theta_{\lambda p^\prime})}\,\ , d\vec { p_\lambda},\ ] ] and then substituting this phase space expression in eq . ( 31 ) , and identifying @xmath202 with ( @xmath211 ) . this gives @xmath212 \\ & & \times\,\,\left[c\,|\eta_a(t_p)|^2\,n_{l_1\,l_2}\ , \sum_l(l_1l_200/l0)^2 |g_{l_1\,l_2}^l(p_b)|^2\right]\end{aligned}\ ] ] however , before we proceed further let us mention that eq . ( 36 ) for the knock - out contribution , which has a factorization of the vertex i and ii , holds under a certain approximation . more correctly , as shown in diagram 2(b ) these two vertices should have been correlated in space through the proton ( or lambda ) propagator between them . this propagator would be a proton scattered wave between the two vertices with an outgoing boundary condition . that is , we would have between two vertices a factor like @xmath213 where @xmath214 collectively represents all other factors . @xmath215 represents the proton interaction with the medium . this integral has two parts , one originating from the principal value and another from the energy conserving @xmath216-function part of the propagator . physically these two parts represent the off - shell and the on - shell scattering in the intermediate state . the on - shell part can be shown to be roughly proportional to the proton momentum , hence dominates at higher energies . the off - shell part dominates at lower energies ( see for example @xcite ) . in our case , since the energies of the proton ( or lambda ) are in the intermediate energy range , we have restricted ourselves to the energy conserving on - shell contribution only . this , essentially is the assumption implicit in writing eq . ( 36 ) with the proton being described at both vertices by distorted waves . proton induced single nucleon knock - out reaction at intermediate energies is a well studied subject experimentally as well as theoretically @xcite . using the notations given in diagram ( 2b ) , the expression for the ( @xmath2 , @xmath2@xmath17 ) knock - out reaction is given by @xmath217 where @xmath218 denotes the average and sum over the spins in the initial and final states respectively . @xmath219 is the momentum of the recoiling nucleus . @xmath220 and @xmath79 denote its intrinsic excitation and the excitation energy respectively . to evaluate this expression , first we integrate over @xmath221 and utilize the momentum conserving delta function , giving @xmath222 with @xmath223 , where @xmath224 is the momentum transfer from the incident proton . the transition matrix @xmath225 in above describes the t - matrix for the knock - out of a nucleon from the nucleus b and leaving the residual nucleus in a one - hole excited state denoted by ` @xmath13 ' . it is given by @xmath226 where @xmath80 and @xmath220 denote the nuclear wave functions in the initial and final states respectively . @xmath227 is the @xmath17-@xmath17 scattering amplitude . this amplitude is half off - shell if the distortion of the continuum nucleons is ignored and becomes fully off - shell if the distortions are included . however , at the energies of concern to us , the off - shell effects are known not to be significant . hence , normally the @xmath17-@xmath17 t - matrix here is taken on - shell and the energy , @xmath228 at which it is evaluated is taken corresponding to the incident momentum @xmath203 . the @xmath229 itself is related to the @xmath17-@xmath17 cross section in the centre of mass through @xmath230 where @xmath231 is the @xmath17-@xmath17 momentum in its centre of mass , and @xmath232 is the nucleon mass . the sum over ` @xmath13 ' in eq . ( 42 ) means the sum over the excited states in the recoiling nucleus consistent with the momentum conservation . since the reaction mechanism involves only a nucleon in the nucleus b , ` @xmath13 ' would have only a small spread . therefore , in the energy delta function in eq . ( 42 ) we can replace , to a reasonable approximation , @xmath79 by the binding energy , @xmath233 of the knocked - out nucleon in b. this simplifies the summation over @xmath13 in eq . ( 42 ) . using closure , we can then write @xmath234 where @xmath92 collectively denotes coordinates of all the spectator core nucleons in the nucleus @xmath80 . to proceed further we now use a simple description of the nucleus . we write the target nucleus b wave function as a product of a single nucleon ( ` n ' ) wave function in a shell with quantum numbers @xmath235 " and the core nucleus wave function @xmath236 . with this eq . ( 45 ) reduces to @xmath237 with @xmath238 where @xmath239 is the nucleon spectroscopic factor . in the present simplified description of the nuclear wave function , it is equal to the number of nucleons in the shell @xmath235 " . for the bound nucleon we use the momentum space representation , @xmath240 , where @xmath241 from the momentum conservation at the @xmath242 vertex and that following eq . ( 42 ) equals to , @xmath243 . with this identification eq . ( 46 ) then factorizes as @xmath244 \\ & & \times\,\,\left[\,\frac{1}{4}\,\,\frac{s(l)}{2l+1}\,\,\sum_{m_l } \,|\,\phi_{nlm_l}(-\,\vec{q_r})\,|^2\,\right].\end{aligned}\ ] ] the single nucleon knock - out cross section expression ( eq . ( 42 ) ) subsequently becomes @xmath245 \\ & & \times\,\,\left[\,\frac{1}{4}\,\,\frac{s(l)}{2l+1}\,\,\sum_{m_l } \,|\,\phi_{nlm_l}(-\vec{q_r})\,|^2\,\right].\end{aligned}\ ] ] to obtain the expression for the inclusive cross section we still need to integrate this expression over @xmath246 . following ref . @xcite we use the energy delta function to remove angle integration , and with some algebraic manipulations obtain @xmath247 \\ & & \hspace{4cm}=\,s(l)\,\,\frac{m}{2q}\,\ , \int_{q_r^{min}}^\infty q_r\,\,dq_r\,\,|\,\phi _ { nl}(q_r)\,|^2.\end{aligned}\ ] ] substituting this in eq . ( 49 ) above , and also writing @xmath248 in terms of the elementary @xmath17-@xmath17 differential cross section ( eq . ( 44 ) ) we get @xmath249 \,\,\left[\,\frac{d\sigma}{d\omega}(\epsilon)\,\right]_{cm}^{pn}\,\ , \left[\,s(l)\,\int_{q^{min}_r}^\infty q_r\,\,dq_r\,\,\left|\ , \frac{1}{(2\pi)^{3/2}}\,\,\phi _ { nl}(q_r)\,\right|^2\,\right].\ ] ] here @xmath250 is that minimum momentum which a nucleon of binding energy @xmath233 must have in the nucleus for the scattered proton to be observed at a scattering angle @xmath251 with a momentum @xmath252 . its value is given by @xmath253}{[\,p_{p^\prime}^2\,-\,p_p^\prime\,p_p\,cos\,(\theta)\,+\ , m\,b_n\,]^{1/2}}\,\,.\ ] ] eq . ( 51 ) for the knock - out cross section assumes that the incoming proton and the outgoing nucleons do not suffer any additional scattering except the hard knock - out collision . this additional scattering , however , can be incorporated in the formalism by replacing the plane wave description of the continuum nucleons by the distorted waves " , which would be solutions of the wave equation with appropriate optical potentials in it . several studies of the distortion effect in the knock - out reaction have been carried out in the literature and it has been found that in the energy range of nucleons of interest to us here , the effect of distortion is mainly absorptive . the dispersive effect is very small . the absorption factor for @xmath4c ( @xmath2 , @xmath40 ) @xmath254 reaction at 160 mev beam energy , for example , in ref . @xcite is found to be around 0.5 . finally , before closing this section we determine the extent of accuracy to which the expression in eq . ( 51 ) describes the measured inclusive proton induced single nucleon knock - out cross section . we calculate the inclusive cross sections at 160 mev beam energy for @xmath4c and compare them with the measured ( @xmath2,@xmath6 ) ones at the same beam energy @xcite . the calculated results are summed over the knocked - out nucleon ( including both neutron and proton ) from @xmath44 and @xmath69 shells . the single nucleon binding energies for them are taken from ref . the nucleon - nucleon differential cross section in the centre - of - mass required in the calculations are taken from the analytic form given in ref . @xcite , i.e. @xmath255\,(\,1\,+\,0.1\,cos^2\,\theta\,),\ ] ] where the cross section is in mb and energy ( kinetic ) , t in mev . this form is in good agreement with the energy dependence of the observed cross sections for the range 20 @xmath256 t @xmath256 200 mev . the bound nucleon wave function is described by the oscillator potential form with the length parameter , b = 1.67 fm . the spectroscopic factor , @xmath239 is taken equal to the number of nucleons ( neutrons+protons ) in the orbital @xmath235 . c after a single nucleon knock - out scattering from the nucleus along with the experimental points at the same energy @xcite . , width=415,height=340 ] with these inputs and the distortion factor equal to 0.5 ( as discussed above ) , the calculated energy spectrum for the proton going at 30@xmath257 , for example , in the lab . frame along with the experimental cross sections is shown in fig . 7 . as we see , the agreement between them , both in shape and magnitude , is very good . this validates the accuracy of eq . ( 51 ) for the description of the inclusive knock - out channel and gives confidence for its use in calculations of the ( @xmath1 , @xmath0 ) reaction . the final expression for calculating the knock - out contribution to the @xmath1 absorption probability is obtained by substituting eq . ( 51 ) for @xmath258 in eq . we get @xmath259\,\ , f(\vec { p_p^\prime})\,\times\,\frac{1}{\sigma _ t^{pb}(t_{p^\prime } ) } \,\,\left[\,\frac { 2}{qp_p^\prime}\,\right]\,\,\left[\,\frac{d\sigma } { d\omega}(\epsilon)\,\right]_{cm}^{pn}\times \\ & & \left[\,s(l)\,\int_{q^{min}_r}^\infty q_r\,\,dq_r\,\left|\ , \frac{1}{(2\pi ) ^{3/2}}\,\,\phi _ { nl}(q_r)\,\right|^2\,\right],\end{aligned}\ ] ] where @xmath260 is given by eq . ( 39 ) . as in the case of elastic scattering " we present here calculated results for the target nucleus @xmath4c . the results are the probabilities summed over two proton hole states in various pairs of ( @xmath75 ; @xmath76 ) shell model orbitals at vertex i and , for each of these pairs , summed over various one nucleon @xmath235 orbitals in nucleus @xmath80 at the knock - out vertex ii . since in the final state after knock - out we do not detect the knocked out nucleon , we use the spectroscopic factor , @xmath239 summed over both , the neutrons and protons . the radial part of the bound state wave functions , as discussed in the last section , is taken for the oscillator potential . the differential scattering cross section for @xmath2-@xmath17 is described by eq . ( 53 ) with energy taken corresponding to momentum @xmath261 of the proton incident at the vertex ii ( fig . required @xmath262 for proton on nucleus @xmath80 at energies corresponding to different proton momentum , @xmath207 is taken from ref . this cross section , of course , does not vary significantly over the @xmath263 energy range of interest here . the physical effect of knock - out scattering at vertex ii is to reduce the energy of the proton @xmath263 and deflect it from its direction of incidence . the amount of these changes , as can be seen from the experimental results on inclusive ( @xmath2 , @xmath263 ) reaction at 160 mev in ref . @xcite , are about 30 mev and above for the energy reduction and about 30@xmath257 and above for the deflection . immediate consequence of these numbers would be that the angular correlation between @xmath2 and @xmath3 shown in fig . 4 , coming from vertex i , will be widened significantly and the energy spectrum of @xmath2 and @xmath3 , shown in fig . 5 , will be shifted towards lower energies . both these effects will , therefore , spoil the agreement shown in these figures ( figs . 4 - 5 ) between the calculated results from vertex i and the corresponding finuda measurements . to get a quantitative idea about the extent of deflection the vertex ii may introduce in the proton , @xmath263 , in fig . 8 we show its angular distribution relative to the @xmath169 motion after being scattered from the vertex ii . initially the @xmath264 is taken to move at 180@xmath257 w.r.t . @xmath3 with energy as fixed at the vertex i , including the spread due to fermi motion . without the knock - out scattering , this distribution will be just a point at cos(@xmath265)=@xmath201 in this figure . due to scattering this point gets a significant spread , as we see in this figure . calculated results use eq . ( 54 ) and are integrated over the energy spread of @xmath263 . mass distribution from both the vertices along with the experimental points @xcite.,width=415,height=377 ] the consequence of above is expected to be a significant modification in the invariant mass distribution of the @xmath2 and @xmath169 . however , before we show these results we may mention that , because of various sums , integrals and constrain checks on kinematic variables , the calculations are very involved and tedious . therefore , to keep the calculations somewhat simpler and physically transparent we have put some constraints in the calculations without , of course , losing any essence of the physics of the results . we have seen earlier in the calculations on vertex i that the @xmath2 and @xmath169 from it emerge mainly back - to - back with a very small cone angle . we have , therefore , done the vertex ii calculations with @xmath265 = 180@xmath184 only . the energy variation of the @xmath2 and @xmath3 due to fermi motion of the absorbing protons in the nucleus a , however , has been kept intact . with this , in fig . 9 we show the calculated invariant mass distribution of @xmath2-@xmath3 after knock - out scattering along with the finuda measurements @xcite . it is extraordinary to see that the calculated mass distribution totally agrees with the measured ones . they agree in mass shift as well as in the shape of the distribution . this is the same observation as reported in ref . @xcite . however , to get the complete @xmath2-@xmath169 invariant mass distribution we need to add to above the contribution from the elastic scattering " fsi corresponding to vertex i ( fig . therefore , in fig . 9 we also show the @xmath2-@xmath3 invariant mass distribution due to vertex i , and the sum of the elastic " and knock - out " contributions . the summed distribution obviously has two peaks , one corresponding to elastic scattering " fsi and another to the single nucleon knock - out " fsi . this is similar to the inclusive inelastic spectrum normally seen experimentally in ( @xmath2 , @xmath263 ) or ( @xmath7 , @xmath266 ) scattering on any nucleus at intermediate energies " @xcite . unusual thing following @xmath1 absorption seen here is that the peak corresponding to elastic scattering " seems to be missing in the finuda data . this is intriguing . we summarize our observations for the vertex ii as , 1 . the calculated invariant @xmath2-@xmath3 mass distribution totally reproduces the experimentally observed distribution . this is in line with the finding in ref . though we have not shown the full calculations for the inclusive energy spectra for the proton or lambda and angular correlation between them , we believe that due to large deflection and energy shift by the single - nucleon knock - out scattering the agreement seen in figs . 4 - 5 between the observed and calculated angular correlation between the @xmath2 and @xmath169 and the inclusive @xmath2 and @xmath169 energy spectra using only the vertex i will be spoiled considerably . we have not calculated these spectra fully because , due to several integrations over various kinematic variables , they are very long and involved . we have calculated the inclusive differential absorption probability for @xmath1 at rest in @xmath4c nucleus for the @xmath0 exit channel . the @xmath1 is assumed to be absorbed on a pair of protons in the nucleus . the final state interaction in the reaction includes the elastic scattering of @xmath2 and @xmath169 in the final state and the single nucleon knock - out from the recoiling nucleus . the calculated invariant @xmath2-@xmath3 mass distribution shows two peaks , one due to elastic scattering and another due to knock - out channel . the latter peak overlaps in position and width with the peak observed in the finuda measurements . the peak corresponding to elastic scattering is not seen in the experiments . measured angular correlation between @xmath2 and @xmath169 and their inclusive energy distribution agree with the corresponding calculated results including only the elastic scattering in the final state . inclusion of the knock - out channel is likely to spoil this distribution . thus , finally , we may conclude that , seen in isolation , the experimentally observed shift in the invariant @xmath2-@xmath3 mass distribution could be interpreted as due to single nucleon knock - out final state interaction . but , if we include other results , like the absence of elastic scattering peak in experiments , full agreement of the calculated @xmath2-@xmath169 angular correlation and their inclusive spectra using only vertex i with the corresponding measurements , the situation becomes quite a bit confusing . as a final comment in the present work on the @xmath200 " cluster interpretation of the observed downshift of about 100 mev in the finuda measurements of the @xmath267 invariant mass compared to its free value , the knock - out reaction in the final state seems to be a definitive alternative for this shift . only discomfiture in this conclusion comes from the absence of the elastic scattering " peak ( fig . 9 ) in the observed invariant mass distribution in the finuda measurements . this absence can not be attributed to the cut off of the @xmath3 hyperons below 300 mev / c momentum in the finuda spectrometer as these momenta in the elastic scattering " peak are above 400 mev / c . availability of absolute measurements may help to understand the ( @xmath1 , @xmath0 ) reaction better . the authors are extremely grateful to dr . neelima kelkar and dr . kanchan khemchandani for many useful suggestions and their comments on the presentation of the work . this work was initiated sometime back when bkj visited the institute of particle and nuclear studies , kek , japan . he thanks prof . shunzo kumano and other members of the institute for many useful academic discussions and their hospitality during the stay . the work was done under the financial grant from the department of science and technology , govt . of india . t. yamazaki _ et al . _ , proceedings of the international conference on exotic atoms and related topics and international conference on low energy antiproton physics ( exa / leap 2008 ) , vienna , austria , 2008 ; arxiv:0810.5182v1 . b. krippa , phys . c * 58 * , 1333 ( 1998 ) ; e. oset and a. ramos , nucl . phys . * a635 * , 99 ( 1998 ) ; j. a. oller and u. g. mei@xmath268ner , phys . lett . * b500 * , 263 ( 2001 ) ; m. f. m. lutz and e. e. kolomeitsev , nucl . phys . * a700 * , 193 ( 2002 ) . g. jacob and th . j. maris , nucl . phys . * 31 * , 139 ( 1962 ) ; rev . * 38 * , 121 ( 1966 ) ; t. berggren and h. tyren , ann . * 16 * , 153 ( 1966 ) ; daphne f. jackson , adv . phys . * 4 * , 1 ( 1971 ) ; r. shanta and b. k. jain , nucl . phys . * b175 * , 417 ( 1971 ) ; r.e . c * 21 * , 1014 ( 1980 ) ; j. s. oconnell _ et al . _ , _ ibid _ * 35 * , 1063 ( 1987 ) .

we present moderate and high - dispersion 12.5 spectra of the-radius nebula around p cygni , dominated by bright emission lines of [ fe ii ] . observed [ fe ii ] line ratios disagree with theoretical transition rates in the literature , so we use the spectrum of p cygni s nebula to constrain the atomic data for low - lying levels of [ fe ii ] . of particular interest is the ratio [ fe ii ]/ , often used as a reddening indicator , for which we empirically derive an intrinsic value of 1.49 , which is 1040% higher than previous estimates . high - dispersion spectra of [ fe ii ] constrain the geometry , detailed structure , and kinematics of p cygni s nebula , which is the major product of p cygni s outburst in 1600 a.d . we use the [ n ii]/[n i ] line ratio to conclude that the nebula is mostly ionized with a total mass of.1 m ; more than the mass lost by the stellar wind since the eruption . for this mass , we would expect a larger infrared excess than observed . we propose that the dust which obscured the star after the outburst has since been largely destroyed , releasing fe into the gas phase to produce the bright [ fe ii ] emission . the kinetic energy of this shell is ergs , far less than the kinetic energy released during the giant eruption of car in the 1840s , but close to the value for car s smaller 1890 outburst . in this respect , it is interesting that the infrared spectrum of p cygni s nebula resembles that of the `` little homunculus '' around car , ejected in that star s 1890 eruption . the mass and kinetic energy in the nebulae of car and p cygni give insight to the range of parameters expected for extragalactic car - like eruptions .

astro-ph0510836

p cygni is a key object for understanding mass loss in the most luminous evolved stars . at a distance of 1.7 kpc ( adopted throughout this paper ; najarro et al . 1997 ) it is the nearest luminous blue variable ( lbv ) , and it is one of only two stars in our galaxy to have been observed during a giant lbv eruption ( p cyg s eruption in 1600 and @xmath5 car s giant eruption in the 1840s ; see humphreys et al . giant eruptions of lbvs where the bolometric luminosity increases by a few magnitudes and the progenitor star survives the event have been recognized in other galaxies as well ; examples include the `` type v '' supernovae sn1961v in ngc 1058 ( van dyk et al . 2002 ; filippenko et al . 1995 ; goodrich et al . 1989 ) and sn1954j ( v12 ) in ngc 2403 ( tamann & sandage 1968 ; smith et al.2001b ; van dyk et al . 2005 ) , and several more recent @xmath5 car analogs classified as type iin supernovae ( e.g. , filippenko 2005 ; van dyk et al . 2000 ; wagner et al . 2004 ; drissen et al . the lightcurves of @xmath5 car and p cyg showed dips of several magnitudes after their eruptions ( humphreys et al . 1999 ; de groot 1988 ; lamers & de groot 1992 ) , usually attributed to dust formation . nebulae around these stars provide useful clues to the physics of their outbursts , such as the explosion geometry , chemical composition , ejection speed , mass , and kinetic energy . since the extragalactic @xmath5 car analogs are too distant to spatially resolve their nebulae , and since @xmath5 car itself is such an extreme example , p cyg s nebula provides an important benchmark for understanding the evolution of intermediate - luminosity lbvs ( see smith et al . 2004 ) . p cyg s nebula is harder to observe than the homunculus of @xmath5 car because of its weaker contrast with the central star at all wavelengths . this difference is partially due to the longer time elapsed since p cyg s eruption ( 400 yr , as opposed to only 160 for @xmath5 car ) , allowing the ejecta more time to disperse and fade . excess infrared ( ir ) emission from dust in p cyg s nebula is much less pronounced than in @xmath5 car dust emission at 220 @xmath0 has not been detected above the thermal - bremsstrahlung emission from p cyg s stellar wind ( smith et al . 2001a ; waters & wesselius 1986 ) . the cool dust excess at far - ir wavelengths ( waters & wesselius 1986 ) is also meager compared to other lbv nebulae like the homunculus of @xmath5 car ( morris et al . 1999 ; smith et al . 2003 ) or the ring nebula around ag car ( mcgregor et al . 1988 ) . while the most basic physical quantities ( mass , kinetic energy ) of the ejecta from @xmath5 car s giant eruption are now fairly well constrained ( smith et al . 2003 ) , those same quantities for p cyg s nebula are not . p cyg is surrounded by complex emission - line nebulosity , including an inner shell with a radius of @xmath11011 , an outer shell with a radius of @xmath11@xmath66 , and some more extended infundibular debris in a `` giant lobe '' several arcminutes across ( barlow et al.1994 ; oconnor et al . 1998 ; meaburn 2001 ; meaburn et al . 1996 , 1999 , 2004 ) . in this paper we focus on the near - ir spectrum of the bright `` inner shell '' , which is also observed in radio free - free emission ( skinner et al . johnson et al . ( 1992 ) found the inner nebula to be nitrogen rich , indicating cno - processed ejecta like in @xmath5 car ( e.g. , smith & morse 2004 ) . a valuable spectral diagnostic of p cyg s nebula is its bright emission from ir lines of [ fe ii ] , such as @xmath212567 , @xmath216435 , and related transitions between the @xmath7 term to the @xmath8 and @xmath9 terms . these are collisionally - excited lines with an upper level @xmath11 ev above the ground state . they are bright in the nebula of p cyg and in the nebulae of several other lbvs ( smith 2002a , 2002b , 2005 ; hamann et al . 1994 ; allen et al . 1985 ) . while bright ir emission from these [ fe ii ] transitions is seen in shock - excited sources like herbig - haro objects and supernova remnants ( e.g. , hartigan et al . 2004b ; graham et al . 1987 ; oliva et al.1990 ; mckee et al . 1984 ) , and so is often taken as a signpost of shocks , the excitation mechanism in all lbvs is not uniform . in some lbv nebulae the [ fe ii ] emission is due to shocks , while in others uv excitation is responsible ( see smith 2002a ) . analysis of these [ fe ii ] lines is complicated by disagreement in the literature on their relative transition rates ( nussbaumer & storey 1988 ; quinet et al . 1996 ; bautista & pradhan 1998 ) . preliminary near - ir spectra of p cyg in the h band showed bright [ fe ii ] @xmath216435 out to 10 from the star ( smith 2001 , 2002a ) , consistent with the inner shell seen in images ( barlow et al . 1994 ) . here we investigate the ir emission from [ fe ii ] in more detail with spectral observations from nasa s infrared telscope facility ( irtf ) , sampling many [ fe ii ] transitions across the 12.5 @xmath0 wavelength range at moderate resolution using spex , and at high spectral resolution for the 16435 line alone using cshell . we obtained long - slit 1 - 2.5 @xmath0 spectra of p cygni s nebula on 2004 august 19 with the near - ir spectrograph spex ( rayner et al.2003 ) mounted on the irtf . spex uses a 1024@xmath101024 aladdin3 insb array , with a spatial pixel scale of 0@xmath1115 . we used spex in long - slit mode with the 0@xmath113-wide slit ( providing a spectral resolution of @xmath122000 ) oriented along p.a.=270@xmath13 , and positioned 4 north of the star to sample emission from the nebula while avoiding the bright central star ( see figure 1 ) . sky subtraction was accomplished by nodding 30 along the slit so that the nebula was always in part of the slit ; the total on - source exposure time was 16 minutes in each of four filters ( k , h , and two filters in the j band ) . wavelength calibration was measured using an internal emission lamp , while flux calibration and correction for telluric absorption were accomplished using observations of spectroscopic standard stars . on the night these observations were obtained , the conditions were photometric and the seeing was @xmath10@xmath118 . figure 2 shows the spectrum extracted from a 2-wide segment of the long - slit data , corresponding to a bright knot in the nebula located 4 north and 8 east of the star ( see fig . 1 ) . however , this is not a true sample of the nebular emission , since there is significant contamination from instrumental scattered starlight , and starlight reflected by dust ( continuum emission and stellar wind lines are seen ) . therefore , figure 3 shows the same spectrum , but with the reflected starlight subtracted . the reflected starlight was sampled from a different region along the slit closer to the star where the continuum emission was much stronger relative to the nebular emission ( it was simple to disentangle the contributions of nebular emission and scattered light , since [ fe ii ] lines like @xmath216435 are absent in the spectrum of the central star ) . intensities of nebular lines measured in this spectrum are listed in table 1 , where noise in fitting the continuum level dominates the uncertainties . additional uncertainties in the flux calibration between various wavelength regions depend on the seeing and the exact position of the slit aperture during observations of the standard star , potentially causing changes in the fraction of the total point spread function included in the slit ; these uncertainties are harder to quantify . however , it is reassuring that the scattered continuum light from the star in figure 2 follows a smooth trend from one band to the next there is no strong discontinuity in the continuum level between the @xmath14 and @xmath15 bands , for example , indicating no severe difference in slit position between exposures of the standard star . figure 4 shows tracings of the @xmath212567 and @xmath216435 intensities and ratios along the slit , as well as the @xmath215335/@xmath216435 flux ratio , which is a diagnostic of variations in the electron density . on 2003 july 31 we observed p cygni with the high - resolution spectrograph cshell ( greene et al . 1993 ) mounted on the irtf . these observations were conducted remotely from the university of colorado . cshell has a 256@xmath10256 sbrc insb array with a spatial pixel scale of 0@xmath112 , although only 160 pixels are illuminated in the spatial direction , yielding an effective slit length of roughly 30 . we used a slit width of 0@xmath115 , providing a spectral resolving power of @xmath1243,000 or 7 km s@xmath16 . the circular variable filter ( cvf ) wheel isolates a single order , and we chose a central wavelength corresponding to the bright [ fe ii ] @xmath216435 emission line ( vacuum wavelength 16439.98 ) . we observed three separate pointings with the slit oriented at p.a.=0 and centered on the central star , 3 east , and 4 west ( see fig . 1 ) . at each position , we used total on - source integration times of 2030 minutes , and sky - subtraction was accomplished with identical observations of an off - source position . the resulting 2-d spectra at these three pointings are shown in figure 5 . wavelengths were calibrated using telluric absorption lines , adopting vacuum wavelengths in the telluric spectrum available from noao . velocities in figure 5 are heliocentric , and have been corrected for the motion of the earth ( to convert to lsr velocities for p cygni , add 18.1 km s@xmath16 ) . to supplement these ir spectra ( specifically for the purpose of investigating the electron density and ionization fraction ) , we also obtained visual - wavelength spectra of the nebula . we observed p cygni on 2005 may 5 using the goldcam spectrograph on the kitt peak national observatory ( kpno ) 2.1 m telescope . we used two different settings to cover the full optical spectrum from about 3400 7200 . for blue wavelengths , we used grating 58 in second order , and for red wavelengths we used grating 47 in first order . both observations used a 1@xmath112@xmath105 slit aperture offset 4 north of the star ( identical to the spex slit position shown in fig . total exposure times were 30 minutes in the blue , and 40 minutes for the red spectrum . flux calibration was performed with reference to spectra of the central star , p cygni itself , for which the visual spectral energy distribution has been studied extensively ( e.g. , lamers et al . some relative line intensities are discussed further in 5.1 and 5.2 to aid our interpretation of the ir spectra . after subtracting the scattered starlight , the near - ir spectrum of p cygni s nebula in figure 3 is dominated by lines of [ fe ii ] . the strength of the [ fe ii ] lines is unusual compared to circumstellar nebulae around other types of evolved stars , such as planetary nebulae and wolf - rayet ring nebulae , and is a testament to the high density of the gas , and the high luminosity and relatively low effective temperature of the central star . the bright [ fe ii ] spectrum of p cygni s nebula resembles that of the homunculus nebula around @xmath5 car ( smith 2002@xmath17 ) , although p cygni shows no emission lines of molecular hydrogen . the lack of h@xmath18 is probably due to the much lower mass of p cygni s nebula ( see 5.3 ) and to the extreme youth of the homunculus . the lack of h@xmath18 is also consistent with the fact that we detect no emission from the 2.3 @xmath0 co emission features . these bright near - ir lines of [ fe ii ] are common in the circumstellar nebulae of lbvs in general ( smith 2002@xmath19 ) . the long - slit aperture allows us to investigate spatial variations across the nebula . for example , figure 4 shows that across the spex slit placed 4 north of the star , there is little variation in either the reddening or electron density , traced by the [ fe ii ] ratios @xmath212567/@xmath216435 and @xmath215335/@xmath216435 , respectively . this verifies that in general , the physical conditions we derive for one particular region of the nebula probably apply to most of the bright [ fe ii ] shell as well . figure 3 and table 1 also show several emission lines of [ ni ii ] . of particlar interest are [ ni ii ] @xmath223079 and @xmath223688 ( both are @xmath20 ) , which are rarely seen in circumstellar nebulae . @xmath219388 ( @xmath20 ) , is expected to be strong but was not within the wavelength range of our observations . ] these lines share the same upper term as [ ni ii ] @xmath27379 and @xmath27412 ( @xmath21 ) , which are extremely bright in the nebula of p cygni ( johnson et al . 1992 ; barlow et al . 1994 ) , and whose strength is thought to be a consequence of fluorescence ( lucy 1995 ) . these same lines are also strong in the crab nebula ( dennefeld & pequignot 1983 ; henry et al . 1984 ) , and are seen in herbig - haro jets very close to their sources ( hartigan et al . it is not known if a similar continuum - fluorescence process affects the [ fe ii ] lines . the numerous [ fe ii ] lines in our p cyg spectra make it possible to constrain einstein a coefficients . these calculations are important because the a values for these transitions computed in two previous studies ( quinet et al . 1996 , hereafter q96 ; nussbaumer & storey 1988 , hereafter ns88 ) differ by as much as 30% . a 30% error in the [ fe ii ] @xmath212567/@xmath216435 flux ratio , used to estimate reddening because these lines originate from the same upper level , translates into a much larger error when reddening corrections are extrapolated to optical wavelengths . for example , if the intrinsic ratio of [ fe ii ] @xmath212567/@xmath216435 increases by 30% , then an object which would have had a@xmath22=0 using the lower ratio now has a@xmath22=2.85 . the spectra in fig . 2 recorded ten [ fe ii ] emission lines between the second - excited term @xmath7 and the first - excited term @xmath23f , and an additional seven [ fe ii ] lines from @xmath7 to the ground term @xmath9 . the lowest 13 energy levels of fe ii , are respectively , @xmath24 , @xmath25 , and @xmath26 , and in what follows we simply refer to these levels as 1 through 13 . for example , in this notation the 12567 and 16435 transitions are 10 - 1 , and 10 - 6 , respectively . table 2 compiles the 17 observed transitions into 30 line ratios in four groups , where each group originates from a common upper state . within each group we adopted the a value for the brightest transition from ns88 , whose compilation agrees somewhat better with our data than do the values of q96 . other a values within the group then follow from the observed emission line ratio and known reddening ( a@xmath22=1.89 , lamers et al . 1983 ) . table 2 shows the predicted emission line ratios for a@xmath22=0 and a@xmath22=1.89 for q96 and ns88 , together with the line ratios predicted from the new list of a - values in table 3 , reddened with a@xmath22=1.89 . by construction , the new dereddened a coefficients match the observations in table 2 . the last column of table 2 is the uncertainty in the observed ratio . the first and third lines of table 3 are of particular interest to all reddening estimates using near - ir [ fe ii ] lines . the new ratio of a values for the [ fe ii ] lines at 12567 and 16435 is 1.13 , @xmath110% higher than the value of 1.04 of ns88 and over 40% higher than the q96 value of 0.79 . this corresponds to an intrinsic [ fe ii ] @xmath212567/@xmath216435 line intensity ratio of 1.49 , higher than q96 ( 1.04 ) or ns88 ( 1.36 ) . as table 3 shows , when ns88 and q96 disagree most we typically find a values closer to ns88 , but often the new values do not lie between the estimates of ns88 and q96 . thus , our results indicate that previous studies that adopted the intrinsic @xmath212567/@xmath216435 line ratios from q96 or ns88 have underestimated the true reddening . the @xmath212567/@xmath216435 line ratio is often used to estimate reddening in agn as well ; rodriguez - ardila et al . ( 2004 ) have also argued that the a - values of q96 and ns88 are incorrect , but they favor an even _ lower _ value for the intrinsic line ratio of only 0.98 . our study of p cygni reveals that this can not be correct , since even our raw _ observed _ value for this ratio is 1.25 . we did not attempt to constrain collision strengths between the various levels or compare a values between different groups because the densities within the p cygni nebula are not known well , and the flow is clumpy ( see 4 ) . however , such a calculation could be done with a better understanding of the electron densities in the shell . a deeper optical search for faint emission lines would help in this regard . figure 5 shows high - dispersion spectra of [ fe ii ] @xmath216435 at three different positions in the nebula , with the slit oriented in the north / south direction and centered 3 east of the star , on the star , and 4 west . the morphology seen in figure 5 is consistent with three slices through a roughly spherical expanding shell . however , it is obviously not a uniformly smooth sphere , as prominent knots and corrugations exist . the detailed structure of the shell is reminiscent of vishniac instabilities when a thin shell with dense knots is swept up by a faster low - density wind ( e.g. , garcia - segura et al . this is indeed the case , as the shell expansion is slower than the terminal velocity of the present - day stellar wind ( see 4.2 ) . aside from these perturbations with typical sizes of 12 , the only striking departure from spherical symmetry is a possible blow - out at 80 to 0 km s@xmath16 ( heliocentric ) at the southern edge of the shell , most apparent in figure 5@xmath17 . although the overall shape of the nebula is basically spherical , it does show some large - scale brightness asymmetry . for example , in all three slit positions in figure 5 , the brightest emission is north of the nebula s center and on the blueshifted side of the shell . there are corresponding local brightness maxima on the opposite side ( to the south and redshifted ) of the shell in figures 5@xmath19 and @xmath17 . this hints at a sort of reflected symmetry consistent with cross sections through an inclined ring or torus of brighter emission embedded in the shell , with its northern polar axis tilted away from us by @xmath145 . the [ fe ii]-emitting knots in figure 5 have resolved linewidths of typically fwhm@xmath271520 km s@xmath16 . this is obviously too broad for an iron line to be due to thermal broadening , so it represents dynamic expansion of the gas or perhaps broadening due to shocks as the stellar wind interacts with slower nebular material . the high - resolution spectra in figure 5 have clearly resolved the spatial and kinematic structure of p cyg s nebula , allowing us to quantitatively constrain several of its basic physical parameters . the slit position centered on the star is the most useful as it crosses through the middle diameter of the nebula , sampling the largest spatial extent and the highest velocities . figure 6 shows the same data as in figure 5@xmath17 for the slit passing through the star , but with two ellipses drawn to represent the idealized inner and outer boundaries of a single spherical shell ( with radii of @xmath28 and @xmath29 , respectively ) . because of the knots and corrugations in the nebula , these ellipses are obviously not perfect boundaries for the shell , but they are useful for guiding our interpretation of the nebular structure and for eventually estimating its mass and kinetic energy in 5 below . derived parameters for these two surfaces are discussed below and are summarized in table 4 . _ systemic velocity _ : the centroid velocity of each of the two ellipses gives the center - of - mass velocity of the nebula . the dashed vertical line in each panel of figures 5 and 6 shows the approximate centroid velocity of the shell , taken to represent the systemic velocity of p cygni at @xmath30=38@xmath315 km s@xmath16 ( the uncertainty here is dominated by the irregular structure in the shell , rather than statistical measurement uncertainty ) . relative to the local standard of rest , this systemic velocity would be about @xmath32=20@xmath315 km s@xmath16 , which is in reasonable agreement with 22.6@xmath311.5 km s@xmath16 estimated by barlow et al . ( 1994 ) from optical lines . _ expansion velocity _ : the inner boundary of the shell ( @xmath28 ) has an expansion velocity of @xmath31120 km s@xmath16 with respect to the systemic velocity , and the outer edge ( @xmath29 ) expands at @xmath31152 km s@xmath16 . these are slower than the present - day terminal speed of p cygni s wind , which is roughly 185 km s@xmath16 ( najarro et al.1997 ; lamers et al . thus , it is reasonable to expect a 65 km s@xmath16 shock at the inside surface of the nebula where the faster stellar wind catches up . however , this shock will not profoundly affect the gas temperature or the relative line intensities in the nebular spectrum , since radiative energy input from the star exceeds that of the kinetic energy from the wind by a factor of @xmath110@xmath23 . the mean expansion speed of @xmath31136 km s@xmath16 that we measure from [ fe ii ] @xmath216435 is close to the value of 140 km s@xmath16 measured from optical [ n ii ] lines by barlow et al . ( 1994 ) , but significantly higher than the 110 km s@xmath16 expansion they measured from [ ni ii ] lines . barlow et al . explained the discrepancy between the apparent expansion speeds of [ n ii ] and [ ni ii ] by proposing that the [ ni ii ] emission arises in slower dense knots that have been overtaken by the stellar wind , producing more extended bow shocks around the knots that emit [ n ii ] and other nebular lines . our new results contradict this conjecture , since the [ fe ii ] and [ ni ii ] transitions both share the same low ionization and high critical densities that motivated the explanation . the densest knots should emit [ fe ii ] as well . _ shell radius and thickness _ : the inner and outer radii of the idealized spherical shell in figure 6 are @xmath28=7@xmath118 or 2.0@xmath1010@xmath33 cm and @xmath29=9@xmath117 or 2.5@xmath1010@xmath33 cm . some emission is seen to extend beyond these boundaries in the form of dense knots ( interior ) or diffuse emission ( exterior ) . the overall thickness of the shell is @xmath34=1@xmath119 or 5@xmath1010@xmath35 cm . aside from a few redshifted knots located 56 south of the star , the shell is hollow ( i.e. there is no younger shell expanding inside this one ) . _ dynamical age _ : if the ellipses in figure 6 trace spherical surfaces of the shell , then at a distance of 1.7 kpc , r/@xmath36 gives the dynamical age of each feature assuming linear motion . for @xmath28 the dynamical age is 531 yr , and for @xmath29 it is 524 yr ; these are in good agreement with one another considering that the uncertainty is of order @xmath315% . of course , the dynamical age reflects linear motion and does not necessarily signify the true age of the ejecta , since the shell may have been accelerated by the stellar wind or decelerated by slower ambient material . the present - day nebular expansion speed ( 136 km s@xmath16 ) is in fact lower than the stellar wind speed ( 185 km s@xmath16 ) , suggesting that the shell has decelerated to @xmath13/4 of its initial speed . coincidentally , the dynamical age we measure assuming linear motion is 4/3 of the time since the 1600 outburst . in other words , by assuming linear motion we overestimate the age by precisely the same factor that the shell s velocity has apparently slowed . this agreement would seem too fortuitous if the bright shell around p cygni were not ejected in the 1600 a.d . outburst . given the lack of any comparably bright ejecta closer to the star , _ we conclude that the bright _ [ fe ii ] _ shell in figures 5 and 6 was indeed ejected during the 1600 outburst and has been decelerated by interaction with the ambient medium_. _ filling factor _ : while the overall thickness of the shell derived above is roughly 1@xmath119 , it is clear from figure 5 that the shell is significantly thinner and corrugated with several knots and loops ; i.e. it has complex substructure . the [ fe ii ] @xmath216435 emission is dominated by the densest material in clumps and filaments , so for estimating the mass in the next section , a representative density filling factor @xmath37 is necessary . assigning a value for the filling factor based on the observed structure of the nebula is subjective , but we can nevertheless make an educated guess , since the filling factor of the thick shell drawn in figure 6 is clearly not 1 . a value closer to 0.2 seems more appropriate , but an even lower value might apply if our spatial resolution is inadequate to resolve the sizes of features in the nebula . we provisionally take @xmath37=0.2@xmath310.1 . in the previous section , we concluded that the bright [ fe ii ] shell seen in our spectra is indeed the major product of the 1600 outburst of p cygni . thus , the mass and kinetic energy of this shell trace the total mass ejected and the mechanical luminosity of the outburst . if p cygni s nebula can be approximated as a spherical hollow shell , then a rough estimate of the total gas mass of the nebula can be expressed as @xmath38 where @xmath39=2.2 is the mean molecular weight for a neutral ( see 5.2 ) he mass fraction of @xmath40=0.55 ( najarro et al . 1997 ) ; f@xmath41 is the hydrogen ionization fraction , and @xmath37 is the filling factor in the shell . the geometry is constrained ( see 4.2 ) from our high - resolution spectra of [ fe ii ] @xmath216435 , while the filling factor is only marginally constrained . this leaves the electron density and the h ionization fraction as the unknown quantities that need to be addressed before we can estimate the mass from equation 1 . johnson et al . ( 1992 ) measured an electron density of @xmath42=600 @xmath43 using the [ s ii ] @xmath446717,6731 doublet , although their spectra had low sensitivity . from the same [ s ii ] doublet in higher quality data , barlow et al . ( 1994 ) measured higher densities of 7002000 @xmath43 at various positions in the nebula . in our new optical spectra , with the slit offset 4north of the star , the [ s ii ] ratio indicates a range of electron densities from about 2,500 to 11,000 @xmath43 . in figure 7 we show tracings of the spectrum at three positions along the slit with the reflected starlight subtracted . however , much of the mass may potentially reside in dense clumps or filaments that emit [ fe ii ] lines , but where the [ s ii ] lines are collisionally de - excited . therefore , when deducing the mass from geometric parameters that apply to [ fe ii ] emission regions , it is more appropriate to estimate the average electron density in the [ fe ii ] emitting gas itself , which tends to favor the highest densities . the ratio @xmath215335/@xmath216435 is a diagnostic of electron density , and the bottom panel in figure 4 suggests that this ratio has a fairly uniform value of @xmath10.15 across the nebula . adopting the transition probabilities of nussbaumer & storey ( 1988 ) , our new transition rates for these two lines do not differ from those of nussbaumer & storey by more than 40% . ] , the @xmath215335/@xmath216435 ratio of 0.15 would suggest an electron density of roughly 60008000 @xmath43 . this is in agreement with some of the denser material traced by the [ s ii ] ratio in our spectra . we adopt @xmath42=6000 @xmath43 as representative for the dense [ fe ii]-emitting gas , with the caveat that it could be in error by as much as a factor of 2 in either direction . to estimate the true mass of the nebula , we must first deduce the hydrogen ionization fraction f@xmath41 = n(h ii)/(n(h i)+n(hii ) ) . we experimented with the spectral synthesis code cloudy ( ferland 1996 ) to simulate a thick shell with the geometry described above , illuminated by a central star with t@xmath45=1.9@xmath1010@xmath23 k and l=6@xmath1010@xmath46 l@xmath3 ( najarro et al . 1997 ; lamers et al . 1983 , 1996 ) . we found that he was neutral and h was almost fully ionized in such a situation . however , it is not necessarily safe to assume that p cyg s nebula is fully ionized , since the dense stellar wind extinguishes much of the star s lyman continuum radiation if h recombines along the flow . for example , najarro et al . ( 1997 ) find that in the outermost parts of the stellar wind , the ionization fraction is @xmath47 0.5 . returning to the cloudy simulations , by varying the amount of escaping lyman continuum radiation that is extinguished by the wind we could produce a wide range of values for the ionization fraction in the nebula ( 0.05@xmath48f@xmath491 ) in any of these cloudy calculations . ] . furthermore , if much of the lyman continuum is indeed extinguished and the nebula would otherwise be mostly neutral , then we may need to include the effects of shock heating by the stellar wind that overtakes the shell . barlow et al.(1994 ) considered shocks to be important in the observed spectrum of p cygni s nebula . in other words , we need more information to constrain this problem . we could , in principle , use the observed ratios of [ o i ] and [ o ii ] lines in the nebula to trace the hydrogen ionization fraction . in fact , this was the goal in obtaining our optical spectra . unfortunately , we did not detect [ o i ] @xmath26300 or [ o ii ] @xmath23727 in our spectra of p cygni s nebula , perhaps owing to a depleted oxygen abundance or very bright scattered light from the star at blue wavelengths . however , [ n ii ] @xmath26583 is a bright line in the nebula , and our optical spectra also show weak [ n i]@xmath25200 at about the 2@xmath50 level ( in this discussion , [ n i]@xmath25200 refers to the sum of the individual doublet lines at 5199 and 5201 ) . the large flux ratio of the [ n ii ] to [ n i ] lines suggests that the gas is mostly ionized . the relative flux of [ n ii]@xmath26583 and [ n i]@xmath25200 is determined by three parameters : the electron temperature , electron density , and the ratio of the densities of [ n ii ] and [ n i ] . the average electron density is @xmath16000 @xmath43 , as mentioned above , and the electron temperature for photoionized gas should be @xmath110@xmath23k . the case of a radiative shock is somewhat more complicated , as the temperature in the [ n ii]-emitting region is higher than that for [ n i ] . in this case we adopt values typical for the cooling zones of a low ionization shock , where t(n ii)@xmath5110@xmath23k and t(n i)@xmath516500 k. to obtain the density ratio n(n ii)/n(n i ) , we performed a full 5-level atom calculation for n i and n ii using the latest available atomic data for the einstein - a values and collision strengths ( hudson & bell 2005 ; storey & zeippen 2000 ; bell et al.1995 ; mclaughlin & bell 1993 ; baluja & zeippen 1988 ; zeippen 1987 ; froese fischer & saha 1985 ; butler & zeippen 1984 ; berrington & burke 1981 ; le dourneuf & nesbet 1976 ; pradhan 1976 ; dopita et al.1976 ) . collisions and radiative decay determine the relative populations of the energy levels and therefore allow us to infer the density ratio of n ii / n i from the observed emission line ratios . physically , the value of the density ratio n(n ii)/n(n i ) is set by collisional ionization , photoionization , and charge exchange between n i , n ii , h i , and h ii . using our 3@xmath50 upper limit for the strength of [ n i ] , we measure a value for the line ratio [ n ii ] @xmath26583/[n i ] @xmath25200 greater than 127 in the shell of p cygni , with the slit offset 4 north of the star ( if our 2@xmath50 detection is to be believed , the same line ratio would be 189 ) . using the 2@xmath50 result for the constant temperature ( photoionization ) case , we found nitrogen ionization fractions n(n ii)/(n(n i)+n(n ii ) ) of 0.900.98 for t@xmath52 of 5000 - 20000 k ; the corresponding 3@xmath50 limits over this temperature range are @xmath530.86 to @xmath530.98 . for the shock - like case , the observed [ n ii]/[n i ] ratio would imply nitrogen ionization fractions of @xmath53 0.73 and 0.80 for the 3@xmath50 and 2@xmath50 measurements , respectively . so in all cases , we find that nitrogen is mostly ionized . the next question is how to convert the nitrogen ionization fraction into the hydrogen ionization fraction . if , as should occur in the cooling zone of a shock , charge exchange is the dominant process that ties n ii / n i to h ii / h i , then @xmath54 the exponential factor arises from the different ionization potentials of n ( 14.53 ev ) and h ( 13.59 ev ) , while the factor of 4.5 reflects the relative statistical weights of the n ii+h i and n i+h ii outlet channels . at @xmath55k , n(n ii)/n(n i ) = 1.5 n(h ii)/n(h i ) , so that f@xmath41 = 2f@xmath56/(3-f@xmath56 ) , where f@xmath41 is the ionization fraction of h and f@xmath56 is the ionization fraction of n. at @xmath55k , when f@xmath56 = 0.9 , we find f@xmath41 = 0.86 . hence , whenever charge exchange dominates the ionization states and n is mostly ionized , so is h. as the ionization parameter ( ratio of density of ionizing photons to total h density ) increases , photoionization becomes increasingly more important relative to charge exchange in determining the ionization fractions of h and n. in p cygni s nebular shell at the adopted density , the ionization parameter is roughly 10@xmath57 . the value for the hydrogen ionization fraction derived from the observed [ n ii]/[n i ] ratio ( a few percent neutral ) agrees with what we would expect _ a priori _ from this ionization parameter based on the lyman continuum flux escaping the thick wind of p cygni using an upper limit of @xmath5810@xmath59 s@xmath16 from recent modeling work that is in preparation by najarro & hillier ( najarro 2005 , private comm . ; see also najarro et al . 1997 ) . in any case , the observed [ n ii]/[n i ] ratio implies that p cygni s nebula is mostly ionized . this result removes a potentially large source of uncertainty in estimating the mass of the outburst . following the previous subsections , if we take @xmath60 to be @xmath16000 @xmath43 and 0.8@xmath48f@xmath490.95 , then equation ( 1 ) and the geometric parameters deduced in 4.2 give @xmath61 for the mass of p cygni s inner shell nebula . this is only an order - of magnitude estimate owing to the factor of @xmath12 uncertainties in the geometric filling factor and the electron density . is this mass estimate plausible ? we can perform a sanity check by guessing what we think the nebular mass _ should _ be based on the density contrast of the nebula and the duration of the outburst . for a constant spherical post - eruption wind with @xmath62=3@xmath1010@xmath63 m@xmath3 yr@xmath16 ( najarro et al.1997 ) , the particle density at the radius of the nebula is @xmath1100 @xmath43 . thus , the nebula has an overdensity factor of roughly 60 compared to the stellar wind at the same radius , presumably marking a huge increase in the mass - loss rate during the 1600 a.d . eruption . based on this factor of 60 overdensity of the nebula , we might expect the mass - loss rate during eruption to have been @xmath11.8@xmath1010@xmath64 m@xmath3 yr@xmath16 . this seems reasonable , since a mass - loss rate much greater than 10@xmath65 m@xmath3 yr@xmath16 would be required for a significant pseudo photosphere to form in the wind ( smith et al . 2004 ) , as expected during an lbv outburst ( e.g. , appenzeller 1986 ) . comparably large values of @xmath62 would also be required to form dust in the wind and obscure the star . so , if mass loss occurred at this rate during the 55 yr duration of p cygni s 1600 a.d . eruption ( humphreys et al . 1999 ) , then the nebular mass should be about 0.1 m@xmath3 . this is in fortuitously - good agreement with equation ( 3 ) . this conjecture depends on several assumptions and is not conclusive it merely shows that the mass we derived is sensible . in particular , this suggests that our assumed filling factor is not seriously wrong . given the high mass - loss rate during the 1600 outburst and the dip in the historical lightcurve after the eruption , it is reasonable to assume that significant quantities of dust formed during the event . dust residing at the radius of the [ fe ii ] shell observed in our spectra should have a temperature of 60 ( @xmath66)@xmath67 k , where @xmath66 depends on the grain size . for large grains like those seen in @xmath5 car and other lbvs ( smith et al . 2003 ) , we should therefore expect dust temperatures of @xmath6860 k in p cyg s nebula , emitting primarily at wavelengths longer than 30 @xmath0 . a total dust mass @xmath69 emits a flux of @xmath70 where @xmath71 is the grain opacity , @xmath72 is the planck function , and @xmath73=1.7 kpc . in the 60 @xmath0 _ iras _ band , for example , 60 k dust with @xmath7410@xmath64 m@xmath3 ( assuming a gas : dust mass ratio of 100 ) and @xmath71(60 @xmath0)=87 @xmath75 g@xmath16 ( draine 2003 ) would emit a flux of roughly 24 jy . waters & wesselius ( 1986 ) reported 60 @xmath0 excess emission ( flux in excess of the stellar wind emission , possibly due to cool dust ) of roughly 90% , corresponding to an excess of only @xmath11.2 jy . this would seem to indicate that either 1 ) in our analysis above we have overestimated the mass by more than an order of magnitude , or 2 ) that dust grains have been depleted in p cygni s nebula by more than an order of magnitude compared to the typical ism fraction . the extreme strength of infrared [ fe ii ] lines that we observe in the nebula suggests that grains are in fact depleted so that fe is released into the gas phase . we found a similar ionized gas to dust ratio ( about 1000 ) in the nebula of ry scuti ( smith et al . 2002 ; gehrz et al . 2001 ) . if the shell mass is 0.1 m@xmath3 and it is expanding at an average speed of 136 km s@xmath16 , then the kinetic energy of p cygni s shell is about 2@xmath1010@xmath76 ergs . this represents the minimum amount of mechanical energy imparted to the ejecta during the 1600 a.d . the luminous energy radiated during that eruption was roughly 2.5@xmath1010@xmath77 ergs ( humphreys et al . thus , a small fraction ( @xmath481% ) of the total energy was used to accelerate the ejecta during this event . of course , we must keep in mind that this fraction may be somewhat higher , since the nebula appears to have decelerated to 3/4 of its initial ejection speed . in any case , the ratio of kinetic to luminous energy associated with p cygni s eruption ( @xmath481% ) is still far less than the eruption of @xmath5 carinae in the 1840 s , when ke / lt@xmath681 ( smith et al . 2003 ) . while the total kinetic energy associated with p cygni s 17th century eruption is about 4000 times less than the 1840 s eruption of @xmath5 carinae , it comes much closer to the mechanical energy of @xmath5 car s second eruption in 1890 , which was about 8@xmath1010@xmath76 ergs ( smith 2005 ) . the similar amounts of kinetic energy in these two outbursts is underscored by the similar properties in the two nebulae . p cygni s shell and the little homunculus of @xmath5 car ( produced in that 1890 outburst ; ishibashi et al . 2003 ; smith 2005 ) both have extremely bright [ fe ii ] emission ( smith 2002a ; 2005 ) , both have little or no detectable thermal - ir emission from dust ( smith et al.2003 ; waters & wesselius 1986 ) , they have similar amounts of mass , and both have similar expansion velocities slower than the present - day stellar wind that is overtaking them . the mass and kinetic energy of p cygni s nebula are important for understanding the mechanism behind giant eruptions of lbvs and extragalactic @xmath5 car analogs , as well as their role in mass loss during the late evolution of the most massive stars . p cygni and @xmath5 car are the only two lbvs in our galaxy for which a giant outburst has been observed directly . compared to @xmath5 car ( smith et al . 2003 ) , the mass and kinetic energy of p cygni s nebula are much lower , and so these two objects define a range of likely values for the mass and mechanical energy of extragalactic @xmath5 car eruptions . the low mass and relatively low dust content of the nebula around p cygni are probably more representative of lower - luminosity events like sn1954j ( v12 in ngc 2403 ; tamann & sandage 1968 ; smith et al . 2001b ; van dyk et al . 2005 ) , while the massive and dusty homunculus around @xmath5 car itself is more akin to the most extreme events like sn1961v in ngc1058 ( van dyk et al . 2002 ; filippenko et al . 1995 ; goodrich et al . 1989 ) . this study raises an important question in the evolution of very massive stars : _ as massive stars shed their outer layers to become wr stars , is most of the mass lost in a steady stellar wind or in brief eruptions ? _ in the case of @xmath5 car , the huge mass of the homunculus ( @xmath6810 m@xmath3 ) indicates that most post - main - sequence mass loss happens during brief outbursts ( smith et al . is this also true for p cygni ? the total mass inside the nebular shell of p cygni , filled by the stellar wind since the 1600 eruption , is only 0.012 m@xmath3 ( assuming a steady mass loss rate of @xmath62=3@xmath1010@xmath63 m@xmath3 yr@xmath16 ; najarro et al.1997 ) . thus , p cygni has lost about 10 times more mass during a brief outburst than it has in the time since that eruption but obviously this depends on the typical time period between such eruptions , which could be longer than the 400 years since p cygni s last outburst . meaburn et al . ( 1996 ) have studied an outer shell with a dynamical age of roughly 2100 yr , presumably containing a similar amount of mass to that of the inner shell . the mass lost since that time by the quiescent stellar wind is only about 0.07 m@xmath3 , supporting the idea that most mass loss occurs during major outbursts . since giant outbursts like those of p cygni , @xmath5 car , and the extragalactic @xmath5 car analogs may dominate the total mass lost on the way toward the wr phase , it is more than a little embarrassing for the study of stellar evolution that we do not yet know how these eruptions are triggered or what supplies their energy . our study of the near - ir spectrum of p cygni s nebula has led to the main conclusions below : \1 . the near - ir spectrum of p cygni is dominated by collisionally - excited lines of [ fe ii ] , mainly from the three lowest terms . it closely resembles the near - ir spectrum of the homunculus nebula around @xmath5 car , except that p cygni s nebula lacks the molecular hydrogen seen there ( smith 2002b ) . \2 . using 17 near - ir lines of [ fe ii ] we have determined empirical values for the transition ratios and einstein a coefficients for lines between the lowest 3 terms of fe@xmath78 ( @xmath9 , @xmath8 , and @xmath7 ) . these empirical values are in better agreement with the calculations of ns88 than with those of q96 , but in some cases our values differ substantially from both . the ratio @xmath212567/@xmath216435 is of particular importance , as the two bright lines share a common upper level and are often used to deduce the reddening . we find an experimentally - determined value for this intrinsic line ratio of 1.49 ( adopting @xmath79=1.89 ; lamers et al . 1983 ) ; higher than previous values . \4 . from high - resolution spectra of [ fe ii ] , the geometric parameters of p cygni s inner shell are as follows : the shell radius is 10@xmath8010@xmath81 cm , the average expansion velocity is 136 km s@xmath16 , the dynamic age is about 530 yr , and the geometric filling factor for [ fe ii]-emitting filaments appears to be roughly 0.2 . \5 . based on the present expansion speed compared to the speed of the stellar wind , the dynamic age of the nebula , and the lack of any bright nebulosity inside the main shell , we conclude that the 10-radius [ fe ii]-emitting shell around p cygni is the major product of the 1600 a.d . outburst . \7 . based on models for the observed [ n ii ] @xmath26583/[n i ] @xmath25200 line ratio , we find a likely hydrogen ionization fraction of 0.80.95 for a wide range of possible excitation scenarios , while he is mostly neutral . the total mass of the nebula is about 0.1 @xmath82 , subject to errors in the filling factor and the electron density . this , combined with the weak thermal - ir emission from dust , suggests that the gas : dust mass ratio exceeds the canonical value of 100 by an order of magnitude . destruction of dust grains that released fe into the gas phase may help explain the very bright [ fe ii ] emission in p cygni s nebula . we are grateful to richard green for granting a half - night of director s discretionary time to obtain optical spectra of p cygni s nebula at kpno , and to robert d. gehrz for assistance during the irtf / spex observing run . we thank paco najarro and john hillier for sharing results prior to publication regarding their recent work on radiative transfer modeling of p cygni s atmosphere and wind . n.s.was supported by nasa through grant hf-01166.01a from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas5 - 26555 . 1.03199 & he i & 3.4@xmath1010@xmath83 & 1.0@xmath1010@xmath83 + 1.03931 & [ n i ] & 4.9@xmath1010@xmath83 & 1.5@xmath1010@xmath83 + 1.04612 & [ ni ii ] & 4.4@xmath1010@xmath83 & 2.0@xmath1010@xmath83 + 1.05428 & n i & 7.4@xmath1010@xmath83 & 1.9@xmath1010@xmath83 + 1.07112 & [ ni ii ] & 4.2@xmath1010@xmath83 & 1.4@xmath1010@xmath83 + 1.08277 & he i & 2.17@xmath1010@xmath84 & 1.3@xmath1010@xmath83 + 1.09123 & he i+[ni ii ] & 5.7@xmath1010@xmath83 & 1.3@xmath1010@xmath83 + 1.09392 & pa@xmath85 & 1.54@xmath1010@xmath84 & 2.0@xmath1010@xmath83 + 1.10603 & ? & 6.4@xmath1010@xmath83 & 1.7@xmath1010@xmath83 + 1.11644 & ? & 4.9@xmath1010@xmath83 & 1.5@xmath1010@xmath83 + 1.12262 & ? & 4.7@xmath1010@xmath83 & 1.5@xmath1010@xmath83 + 1.12890 & o i & 5.9@xmath1010@xmath83 & 1.7@xmath1010@xmath83 + 1.13477 & [ fe ii ] & 1.38@xmath1010@xmath84 & 2.0@xmath1010@xmath83 + 1.23249 & n i & 2.9@xmath1010@xmath83 & 0.5@xmath1010@xmath83 + 1.24819 & [ fe ii ] & 4.7@xmath1010@xmath83 & 0.8@xmath1010@xmath83 + 1.25642 & [ fe ii ] ( @xmath86 ) & 1.904@xmath1010@xmath87&0.8@xmath1010@xmath83 + 1.27005 & [ fe ii ] ( @xmath88 ) & 5.4@xmath1010@xmath83 & 0.5@xmath1010@xmath83 + 1.27855 & [ fe ii ] ( @xmath89 ) & 1.12@xmath1010@xmath84 & 0.9@xmath1010@xmath83 + 1.28177 & pa@xmath90 & 2.58@xmath1010@xmath84 & 1.0@xmath1010@xmath83 + 1.28521 & fe ii & 2.1@xmath1010@xmath83 & 0.4@xmath1010@xmath83 + 1.29396 & [ fe ii ] ( @xmath91 ) & 2.64@xmath1010@xmath84 & 0.6@xmath1010@xmath83 + 1.29749 & [ fe ii ] ( @xmath92 ) & 4.2@xmath1010@xmath83 & 0.7@xmath1010@xmath83 + 1.30014 & [ fe ii ] & 1.4@xmath1010@xmath83 & 0.4@xmath1010@xmath83 + 1.31615 & o i & 1.51@xmath1010@xmath84 & 1.0@xmath1010@xmath83 + 1.32028 & [ fe ii ] ( @xmath93 ) & 5.27@xmath1010@xmath84 & 1.4@xmath1010@xmath83 + 1.32746 & [ fe ii ] ( @xmath94 ) & 1.78@xmath1010@xmath84 & 1.0@xmath1010@xmath83 + 1.53309 & [ fe ii ] ( @xmath95 ) & 2.23@xmath1010@xmath84 & 0.2@xmath1010@xmath83 + 1.59905 & [ fe ii ] ( @xmath96 ) & 8.6@xmath1010@xmath83 & 0.3@xmath1010@xmath83 + 1.63735 & he i & 1.3@xmath1010@xmath83 & 0.3@xmath1010@xmath83 + 1.64319 & [ fe ii ] ( @xmath97 ) & 1.519@xmath1010@xmath87&0.9@xmath1010@xmath83 + 1.65759 & he i & 2.6@xmath1010@xmath83 & 0.5@xmath1010@xmath83 + 1.66369 & [ fe ii ] ( @xmath98 ) & 4.5@xmath1010@xmath83 & 0.5@xmath1010@xmath83 + 1.67664 & [ fe ii ] ( @xmath99 ) & 2.00@xmath1010@xmath84 & 0.6@xmath1010@xmath83 + 1.68121 & br11 & 2.7@xmath1010@xmath83 & 0.5@xmath1010@xmath83 + 1.68713 & fe ii & 5.9@xmath1010@xmath83 & 0.5@xmath1010@xmath83 + 1.69731 & he i & 2.2@xmath1010@xmath83 & 0.3@xmath1010@xmath83 + 1.70099 & he i & 4.0@xmath1010@xmath83 & 0.5@xmath1010@xmath83 + 1.71101 & [ fe ii ] ( @xmath100 ) & 2.7@xmath1010@xmath83 & 0.3@xmath1010@xmath83 + 1.73674 & br10 & 4.2@xmath1010@xmath83 & 0.5@xmath1010@xmath83 + 1.74497 & [ fe ii ] ( @xmath101 ) & 3.9@xmath1010@xmath83 & 0.6@xmath1010@xmath83 + 1.79736 & [ fe ii ] ( @xmath102 ) & 3.9@xmath1010@xmath83 & 0.9@xmath1010@xmath83 + 1.80033 & [ fe ii ] ( @xmath103 ) & 9.6@xmath1010@xmath83 & 0.8@xmath1010@xmath83 + 1.80963 & [ fe ii ] ( @xmath104 ) & 3.62@xmath1010@xmath84 & 1.2@xmath1010@xmath83 + 2.05835 & he i & 7.3@xmath1010@xmath83 & 1.0@xmath1010@xmath83 + 2.14273 & mg ii ( ? ) & 3.6@xmath1010@xmath83 & 0.7@xmath1010@xmath83 + 2.16547 & br@xmath85 & 8.2@xmath1010@xmath83 & 1.1@xmath1010@xmath83 + 2.22333 & [ fe ii ] & 1.8@xmath1010@xmath83 & 0.6@xmath1010@xmath83 + 2.30786 & [ ni ii ] ( @xmath20 ) & 2.7@xmath1010@xmath83 & 0.7@xmath1010@xmath83 + 2.36907 & [ ni ii ] ( @xmath20 ) & 2.3@xmath1010@xmath83 & 0.7@xmath1010@xmath83 + 2.40651 & ? & 4.1@xmath1010@xmath83 & 1.2@xmath1010@xmath83 + 10 - 1/10 - 2 1.25668/1.32055 & 3.80 & 3.82 & 3.66 & 3.68 & 3.62 & 3.61 & 0.10 + 10 - 1/10 - 6 1.25668/1.64355 & 1.04 & 1.36 & 0.87 & 1.15 & 1.25 & 1.25 & 0.01 + 10 - 1/10 - 7 1.25668/1.80939 & 5.17 & 6.75 & 4.17 & 5.44 & 5.24 & 5.25 & 0.18 + 10 - 2/10 - 6 1.32055/1.64355 & 0.27 & 0.36 & 0.24 & 0.31 & 0.35 & 0.35 & 0.01 + 10 - 2/10 - 7 1.32055/1.80939 & 1.36 & 1.77 & 1.14 & 1.48 & 1.45 & 1.46 & 0.06 + 10 - 6/10 - 7 1.64355/1.80939 & 4.99 & 4.97 & 4.77 & 4.75 & 4.19 & 4.20 & 0.14 + 11 - 2/11 - 3 1.24819/1.29427&&&&&0.35 & 0.35 & 0.06 + 11 - 3/11 - 4 1.29427/1.32778 & 1.74 & 1.64 & 1.70 & 1.61 & 1.48 & 1.48 & 0.09 + 11 - 3/11 - 6 1.29427/1.53347 & 0.75 & 0.94 & 0.67 & 0.84 & 1.18 & 1.18 & 0.03 + 11 - 3/11 - 7 1.29427/1.67688 & 1.03 & 1.30 & 0.88 & 1.11 & 1.32 & 1.32 & 0.05 + 11 - 3/11 - 8 1.29427/1.80002 & 1.51 & 1.89 & 1.25 & 1.56 & 2.75 & 2.75 & 0.24 + 11 - 4/11 - 6 1.32778/1.53347 & 0.43 & 0.57 & 0.39 & 0.52 & 0.80 & 0.80 & 0.05 + 11 - 4/11 - 7 1.32778/1.67688 & 0.59 & 0.79 & 0.52 & 0.69 & 0.89 & 0.89 & 0.06 + 11 - 4/11 - 8 1.32778/1.80002 & 0.87 & 1.15 & 0.73 & 0.97 & 1.86 & 1.85 & 0.19 + 11 - 6/11 - 7 1.53347/1.67688 & 1.37 & 1.38 & 1.31 & 1.31 & 1.11 & 1.12 & 0.03 + 11 - 6/11 - 8 1.53347/1.80002 & 2.01 & 2.00 & 1.86 & 1.85 & 2.33 & 2.32 & 0.19 + 11 - 7/11 - 8 1.67688/1.80002 & 1.47 & 1.46 & 1.42 & 1.41 & 2.09 & 2.08 & 0.18 + 12 - 4/12 - 5 1.27878/1.29777 & 2.30 & 2.28 & 2.28 & 2.26 & 2.67 & 2.67 & 0.49 + 12 - 4/12 - 7 1.27878/1.59947 & 0.73 & 0.86 & 0.64 & 0.74 & 1.31 & 1.30 & 0.11 + 12 - 4/12 - 8 1.27878/1.71113 & 2.78 & 3.27 & 2.33 & 2.74 & 4.14 & 4.15 & 0.57 + 12 - 4/12 - 9 1.27878/1.79710 & 1.62 & 1.89 & 1.33 & 1.55 & 2.89 & 2.87 & 0.70 + 12 - 5/12 - 7 1.29777/1.59947 & 0.32 & 0.38 & 0.28 & 0.33 & 0.49 & 0.49 & 0.08 + 12 - 5/12 - 8 1.29777/1.71113 & 1.21 & 1.43 & 1.02 & 1.21 & 1.55 & 1.56 & 0.31 + 12 - 5/12 - 9 1.29777/1.79710 & 0.71 & 0.83 & 0.58 & 0.69 & 1.08 & 1.08 & 0.31 + 12 - 7/12 - 8 1.59947/1.71113 & 3.79 & 3.81 & 3.66 & 3.69 & 3.16 & 3.19 & 0.37 + 12 - 7/12 - 9 1.59947/1.79710 & 2.22 & 2.21 & 2.09 & 2.09 & 2.20 & 2.21 & 0.51 + 12 - 8/12 - 9 1.71113/1.79710 & 0.58 & 0.58 & 0.57 & 0.57 & 0.70 & 0.69 & 0.18 + 13 - 5/13 - 8 1.27035/1.66377 & 0.92 & 1.02 & 0.77 & 0.86 & 1.20 & 1.20 & 0.17 + 13 - 5/13 - 9 1.27035/1.74493 & 1.85 & 2.05 & 1.52 & 1.69 & 1.38 & 1.38 & 0.25 + 13 - 8/13 - 9 1.66377/1.74493 & 2.02 & 2.01 & 1.97 & 1.96 & 1.15 & 1.15 & 0.22 + lccc 10 - 1 12566.80 & 4.74@xmath1010@xmath64 & 4.83@xmath1010@xmath64 & 4.83@xmath1010@xmath64 + 10 - 2 13205.54 & 1.31@xmath1010@xmath64 & 1.33@xmath1010@xmath64 & 1.35@xmath1010@xmath64 + 10 - 6 16435.50 & 5.98@xmath1010@xmath64 & 4.65@xmath1010@xmath64 & 4.26@xmath1010@xmath64 + 10 - 7 18093.95 & 1.32@xmath1010@xmath64 & 1.03@xmath1010@xmath64 & 1.07@xmath1010@xmath64 + 11 - 2 12481.9 & & & 0.35@xmath1010@xmath64 + 11 - 3 12942.68 & 1.98@xmath1010@xmath64 & 1.94@xmath1010@xmath64 & 1.94@xmath1010@xmath64 + 11 - 4 13277.76 & 1.17@xmath1010@xmath64 & 1.21@xmath1010@xmath64 & 1.32@xmath1010@xmath64 + 11 - 6 15334.71 & 3.12@xmath1010@xmath64 & 2.44@xmath1010@xmath64 & 1.74@xmath1010@xmath64 + 11 - 7 16768.76 & 2.49@xmath1010@xmath64 & 1.94@xmath1010@xmath64 & 1.63@xmath1010@xmath64 + 11 - 8 18000.16 & 1.82@xmath1010@xmath64 & 1.43@xmath1010@xmath64 & 0.81@xmath1010@xmath64 + 12 - 4 12787.76 & 2.45@xmath1010@xmath64 & 2.25@xmath1010@xmath64 & 2.66@xmath1010@xmath64 + 12 - 5 12977.73 & 1.08@xmath1010@xmath64 & 1.00@xmath1010@xmath64 & 1.00@xmath1010@xmath64 + 12 - 7 15994.73 & 4.18@xmath1010@xmath64 & 3.28@xmath1010@xmath64 & 2.20@xmath1010@xmath64 + 12 - 8 17111.29 & 1.18@xmath1010@xmath64 & 0.92@xmath1010@xmath64 & 0.72@xmath1010@xmath64 + 12 - 9 17971.04 & 2.12@xmath1010@xmath64 & 1.67@xmath1010@xmath64 & 1.06@xmath1010@xmath64 + 13 - 5 12703.46 & 3.32@xmath1010@xmath64 & 2.91@xmath1010@xmath64 & 2.91@xmath1010@xmath64 + 13 - 8 16637.66 & 4.75@xmath1010@xmath64 & 3.73@xmath1010@xmath64 & 2.68@xmath1010@xmath64 + 13 - 9 17449.34 & 2.47@xmath1010@xmath64 & 1.95@xmath1010@xmath64 & 2.39@xmath1010@xmath64 + llccc v@xmath105 ( heliocentric ) & km s@xmath16 & & & 38 + radial expansion velocity & km s@xmath16 & 120 & 152 & 136 + radius ( @xmath28 and @xmath29 ) & arcseconds & 7.8 & 9.7 & 8.8 + radius ( @xmath28 and @xmath29 ) & log@xmath106 cm & 17.3 & 17.4 & 17.35 + dynamic age ( r / v ) & years & 531 & 524 & 527 + filling factor & & & & 0.2 + avg . electron density & @xmath43 & & & 6000 + h ionization fraction & & & & 0.8 - 0.95 + mass & m@xmath3 & & & 0.1 + kinetic energy & log@xmath106 ergs&&&46.3 + ke / lt & & & & @xmath480.01 +

we combine new vla d array hi data of ngc 2613 with previous high resolution data to show new disk - halo features in this galaxy . the global hi distribution is modeled in detail using a technique which can disentangle the effects of inclination from scale height and can also solve for the average volume density distribution in and perpendicular to the disk . the model shows that the galaxy s inclination is on the low end of the range given by chaves & irwin ( 2001 ) and that the hi disk is thin ( = 188 pc ) , showing no evidence for halo . numerous discrete disk - halo features are observed , however , achieving heights up to 28 kpc from mid - plane . one prominent feature in particular , of mass , m and height , 22 kpc , is seen on the advancing side of the galaxy at a projected galactocentric radius of 15.5 kpc . if this feature achieves such high latitudes because of events in the disk alone , then input energies of order 10 ergs are required . we have instead investigated the feasibility of such a large feature being produced via buoyancy ( with drag ) within a hot , pre - existing x - ray corona . reasonable plume densities , temperatures , stall height ( 11 kpc ) , outflow velocities and ages can indeed be achieved in this way . the advantage of this scenario is that the input energy need only be sufficient to produce blow - out , a condition which requires a reduction of three orders of magnitude in energy . if this is correct , there should be an observable x - ray halo around ngc 2613 . # 1 # 1([eq:#1 ] )

astro-ph0212072

high latitude hi in disk galaxies is increasingly being recognized as an important tracer of disk activity and dynamics . the gas scale height is a measure of time integrated activity in the disk and discrete features trace the level of local activity . the disk - halo interface represents a severe transition between two very different environments : the high density , low velocity dispersion disk and the low density , high velocity dispersion halo . globally , it may provide information on the shape of the dark matter halo ( e.g. olling & merrifield 2000 ) . in a few recent cases , lagging hi halos have also been detected , such that the high latitude gas is rotating @xmath4 20 - 50 km s@xmath7 more slowly than the underlying disk ( schaap , sancisi , & swaters 2000 , fraternali et al . 2001 , lee et al . 2001 , rand 2000 and tllmann et al . explanations for such lags have focussed mainly on galactic fountains ( cf . bregman 1980 , swaters et al . 1997 , schaap et al . 2000 , collins et al . 2002 ) . edge - on galaxies are particularly useful in the study of the disk - halo region and , in this paper , we especially wish to investigate how discrete disk - halo features can be used as probes of the environment into which they are emerging . the edge - on galaxy , ngc2613 ( d = 25.9 mpc ) is a good candidate for such a study . in paper i ( chaves & irwin 2001 ) , we presented very large array ( vla ) c array observations of this galaxy . the resulting c array zeroth moment map over an optical image is shown in fig . [ carray ] , illustrating the presence of six symmetrically placed vertical extensions in this galaxy . in this paper , we present new vla d array hi data of ngc 2613 and examine them separately as well as in combination with the previously obtained c array data . the observations are presented in sect . [ observations ] , the results , including a description of extremely large high latitude hi features and a model of the global hi distribution are given in sect . [ results ] , in sect . [ discussion ] , we present the discussion , including a feasibility study of a buoyancy model , and a summary is given in sect . [ conclusions ] . details of the c array data acquisition and mapping have been given in paper i. the d array data were taken on sept . 15 , 2000 . the flux calibrator was 3c 48 and the phase calibrator was 0837 - 198 . the flux calibrator was also used as a bandpass calibrator and was observed once for 17 minutes at the beginning of the observing period . the phase calibrator was observed in 3 minute scans alternating with 25 minute scans on ngc 2613 . the total on - source observing time was 2.4 hours and on - line hanning velocity smoothing was applied . the data were edited , calibrated , and fourier transformed using natural weighting , cleaned , and continuum subtracted in the usual way using standard programs in the astronomical image processing system ( aips ) of the national radio astronomy observatory . the resulting ( ra , dec , velocity ) cube was then examined for broad scale structure . the calibrated d array uv data were then combined with the previously obtained c array uv data ( on - source observing time = 15.3 hours , see paper i ) to improve the signal to noise ( s / n ) ratio . since the high resolution c array data have already been presented , we here concentrate on the broader scale structure by re - mapping the c+d array data , using natural weighting , over a uv range corresponding to the d array data only , i.e. 5 k@xmath8 . that is , we show the lower resolution results ( at a typical d array resolution ) , but with the highest s / n possible . table 1 lists observing and map parameters for d array and for the combined c+d array data . note that , although we show images for the combined data set , we have inspected and analyzed all data sets both independently and in combination . note that a full description of the c array results and relevant images have been given in paper i and we have reproduced the zeroth moment c array map in fig . [ carray ] . the remaining the zeroth and first moment maps from the combined data set are shown in fig . [ moments ] , both rotated so that the x axis is parallel to the major axis of the galaxy . vertical slices have been taken perpendicular to the major axis at positions numbered 1 through 10 in fig . [ moments]a and these are shown as position - velocity ( pv ) plots in fig . [ pv ] . the pv slices in the last two panels of fig . [ pv ] show averages over the receding and advancing sides of the galaxy . for the first time , we see that there is a tail of emission on the south - eastern ( left , fig . [ moments]a ) side of the galaxy which resembles tidal tails seen in galaxy interaction simulations . this tail is likely produced via an interaction with the companion , eso 495-g017 , to the north - west , whose systemic velocity is separated from that of ngc 2613 by only 143 km s@xmath7 . the tail is at negative velocities with respect to the receding side of the galaxy from which it originates and shows a strong velocity gradient along its length , becoming negative with respect to the galaxy s systemic velocity as well , i.e. the velocities become forbidden . this can occur in the case of interactions but is difficult to explain any other way . given this velocity structure , the tail must be trailing from the eastern , receding side of ngc 2613 and is therefore in front of the galaxy . the anomalous emission seen around -200 km s@xmath7 in slice 2 , as well as the average over the receding side of the galaxy ( second last panel ) of fig . [ pv ] belongs to this tail as well . there is excess emission on the north - west ( right , fig . [ moments]a ) side of the galaxy in the vicinity of the companion , but we have found no systematic velocities associated with that emission or any other sign of a corresponding redshifted feature . in paper i , using the c array data alone , we identified and characterized the properties of 6 , symmetrically placed extensions labelled f1 to f6 in fig . [ carray ] . [ moments]a , which includes the lower d array resolution data , now also reveal these same features on broader scales , some of which are blended with other features . since a single cutoff level must be applied over the whole galaxy to create this image , this moment map is best used to see a global view of the disk - halo features and identify where new features might be present . the full _ z _ extents and , of course , the velocities of the individual features are best measured from pv slices perpendicular to the major axis . we have inspected many pv slices over the map and show selected slices ( labelled 1 through 10 in fig . [ moments]a ) in fig . [ pv ] . in the new data , the previously seen features f1 through f6 occur at the same velocities and the same positions as shown in fig . [ carray ] , after taking into account the different beam sizes and the addition of broader scale emission . please refer to both figs . [ moments]a and [ pv ] for a more thorough discussion of these features , presented below . f1 appears similar to what has previously been seen ( fig . [ carray ] ) , but at lower resolution . the pv plot for f1 can be seen in slice 2 which reveals it as a distinct feature at a velocity of 250 km s@xmath7 but it can also be seen in slice 1 as well , extending to a projected _ z _ height of at least 100@xmath9 ( 12.6 kpc ) `` above '' ( i.e. positive _ z _ ) mid - plane . a newly seen extension immediately to the west of f1 , sampled by slice 3 , may actually be associated with or part of f1 , since it is seen in pv space as a small arc at about the same velocity as f1 . f2 ( see slice 1 ) shows several velocity components and is more complex than previously seen . like f1 , f2 extends to at least 100@xmath9 at a velocity of 210 km s@xmath7 . it is not clear whether the other components may be blended with emission from the tidal tail . f3 reveals more structure than the extension in fig . [ carray ] and is sampled by slice 4 which reveals it as a 2-pronged arc above the disk centered at a velocity of 100 km s@xmath7 . such double peaks in velocity space are typical of such features and have been seen in other galaxies ( see , e.g. lee & irwin 1997 , lee et al . 2001 ) . the extension to the left of the f4 label in fig . [ moments]a ) extends to a height 225@xmath9 ( 28 kpc ) below ( i.e. negative _ z _ ) mid - plane in this map . as can be seen in slice 3 of the pv map , this feature occurs at a velocity of 50 km s@xmath7 and shows no velocity gradient with height . f4 reaches the highest _ z _ of all the extensions and , to our knowledge , reaches higher latitudes than any previously seen in an edge - on galaxy , exceeding even the starburst galaxy , ngc 5775 , whose known hi features extend to @xmath4 7 kpc from the plane ( lee et al . this is remarkable , given that ngc 2613 ( log l@xmath10 = 10.00 ) has a massive ( 0.1 @xmath11 100 m@xmath3 ) star formation rate that is a factor of 3 lower than that of ngc 5775 and a supernova input energy rate per unit area that is a factor of 7 less than ngc 5775 ( irwin et al . 1999 ) . feature f5 is sampled by slices 8 and 9 and can most clearly be seen in slice 9 as a vertical feature above the plane in pv space at a velocity of -300 km s@xmath7 . whereas it appeared as a single extension in fig . [ carray ] , it can now be seen to have a counterpart 55@xmath9 ( 7 kpc ) in projection farther out along the disk which is sampled by slice 10 . the counterpart clearly occurs at the same velocity and also shows no velocity gradient with height . these two features may form the base of a large loop or cylinder , similar to what has been seen in galaxies like ngc 5775 ( lee et al . 2001 ) . the most interesting and well - defined feature , however , is associated with f6 . for example , slices 6 , 7 , 8 and 9 pass through a feature below the major axis which extends 22 kpc ( 175@xmath9 ) in projection , from the plane and centers at -307 km individual channel maps of this feature are shown in fig . [ channels ] . we will refer to this as the -307 km s@xmath7 feature , rather than f6 since , as revealed by the new data , the feature is more extensive than what was originally seen in fig . [ carray ] . this feature will be discussed extensively in sect . [ 307feature ] . this disk - halo emission is faint ( 3 - 4 @xmath12 typically , note contour spacings in fig . [ pv ] ) but we consider the above - discussed features to be real because , _ 1 ) _ their spatial morphology is typical of the kinds of extensions and loops seen in other edge - on systems , _ 2 ) _ even though the feature might be faint far from the disk , there is evidence for disturbed contours ` beneath them ' even at highly significant contour levels ( see the 5th and 6th contours of fig . [ moments]a , for example ) , _ 3 ) _ at least the 6 major features , f1 to f6 can be seen in both independent data sets , _ 4 ) _ the features generally occur over several independent beams and/or velocity channels , and _ 5 ) _ in pv space , the features are distinct , yet smooth and connect continuously to the disk , unlike artifacts produced by a badly cleaned beam ( see , e.g. the arc above the disk seen in slice 3 of fig . [ pv ] at @xmath4 270 km s@xmath7 or the larger structures in slices 9 and 10 ) . in addition , _ 6 ) _ the receding and advancing panels of panels of fig . [ pv ] ) show _ averages _ over the entire advancing and receding sides of the disk , rather than just selected individual slices as shown elsewhere in the figure . in these panels , therefore , we have basically averaged together signal with noise and weak extensions would be diluted ( i.e. `` beam - averaged '' , where the beam is the size of half the galaxy ) . yet these panels still show the extensions quite clearly , especially on the advancing side where the disk - halo emission is dominated almost entirely by the -307 km s@xmath7 feature . this feature is seen up to the 4th contour which is a 6@xmath12 detection in the diluted averaged panel . finally , _ 7 ) _ for at least 4 of the 6 features , there is evidence from independent observations that extraplanar emission is present . for example , our radio continuum image ( see fig . 9 of paper i ) shows extended emission at the positions of f3 , f4 , f5 , and f6 . also , an early hi image by bottema ( 1989 ) at lower resolution shows extended hi emission at roughly the same locations . the zeroth moment map of fig . [ moments]a also shows a number of low intensity disconnected " features at high latitudes ( for example , the emission around -300@xmath9 , -200@xmath9 ) . we do _ not _ claim that all of these are real , but we have chosen to display them for the sake of comparison with future observations and because some of the emission shown in independent lower resolution images ( e.g. bottema 1989 , irwin et al . 1999 ) is so extended . an example would be the ` knot ' of emission between f3 and f5 which is sampled by slice 6 . for this feature , the velocities are discordent with respect to the underlying disk ( see fig . [ pv ] ) and therefore do not satisfy the criteria listed above . particularly interesting about these very high latitude features is that there is little or no evidence for any lag in velocity with _ z _ , even to very high distances from the place . the features seen below the plane in slices 3 , 8 , and 9 and features seen above the plane in slices 9 and 10 show no velocity gradient . the early data of bottema ( 1989 ) also showed that the high latitude hi was co - rotating . there is some evidence , in our data , for curvature in the smaller features ( cf . slices 3 & 4 above the plane ) , but one could not make any case for a lag . the only slice which shows some evidence for lagging gas is slice 1 , but this slice is also closest to the tidal tail , confusing the interpretation . the averages of the advancing and receding sides of the galaxy also do not support the presence of a lagging halo . here , we see an asymmetry in the sense of many disk - halo features on the receding side , but the advancing side is dominated by the single -307 km s@xmath7 feature . none of the features , however , show a decline with _ z_. we investigate this further in the next section . we have modeled the global hi density and velocity distributions for ngc 2613 following irwin & seaquist ( 1991 ) and irwin ( 1994 ) . this approach models all spectra in the hi cube by adopting a volume density distribution in and perpendicular to the plane , a rotation curve and velocity dispersion ( if required ) , and a position and orientation on the sky . given trial input parameters describing these distributions , the routine then varies the parameters , examining the residuals , until the best fit solution is found . various sections of the galaxy can be modeled , including galactocentric rings in the galaxy s plane . in the current version , we can also isolate slices parallel to the galaxy s plane . this permits modeling the halo , independent of the disk , which is important in this study in the event that a lagging hi halo might be present . while there is no evidence for lags in the discrete features shown in fig . [ pv ] , it is possible that the smooth , low intensity halo might be better fit with a lagging velocity distribution , similar to what has been found for ngc 2403 ( fraternali et al . the routine , called cubit , interfaces to classic aips ; potential users are requested to contact the first author for the code . for ngc 2613 , the adopted in - plane density distribution is a gaussian ring , described by : @xmath13 , where @xmath14 is the galactocentric radius , @xmath15 is the galactocentric distance of the center of the ring , @xmath16 is the in - plane density at the center of the ring , and @xmath17 is the scale length . @xmath17 can have different values for points @xmath18 ( @xmath19 ) and for points @xmath20 ( @xmath21 ) . the distribution perpendicular to the plane is given by the exponential , @xmath22 , where @xmath1 is the perpendicular distance from midplane and @xmath23 is the exponential scale height . the rotation curve is described by the brandt curve , @xmath24 , where @xmath25 is the galaxy s systemic velocity , @xmath26 is the peak of the rotation curve occurring at a galactocentric radius of @xmath27 and @xmath28 is the brandt index which is a shape parameter . a gaussian velocity dispersion , @xmath29 can also be applied , if needed . note that a velocity dispersion will result from all contributions to line widening along the entire line - of - sight direction , for example , any non - circular motions along the plane , streaming motions due to spiral arms and local turbulence . thus , @xmath29 may be larger than what might be expected from a local value . the above distributions result in 10 free parameters : @xmath16 , @xmath15 , @xmath19 , @xmath21 , @xmath23 , @xmath25 , @xmath26 , @xmath27 , @xmath28 , and @xmath29 . in addition , there are 4 more orientation parameters : the position of the galaxy center , ra(0 ) , dec(0 ) , the major axis position angle , pa , and the inclination , i. over the region of the cube occupied by ngc 2613 , there are 552 independent data points ( beam / velocity - channel resolution elements ) above a 2 @xmath12 intensity level in the combined c+d cube , so the 14 parameters are well constrained , over all . certain parameters may show a larger variation , however , depending on the nature of the distribution ; for example , @xmath27 tends to have a larger error since the point at which the rotation curve changes from rising to flat is not as well constrained by the data as some of the other parameters . table 2 lists the best fit modeled values for ngc 2613 for the combined c+d array data . we also modeled the c array data and d array data alone ; the error bars of table 2 reflect the variations in the parameters over the different arrays that result . given the range of resolution over these arrays , from 26.1@xmath9 @xmath30 20.2@xmath9 ( 3.3 @xmath30 2.5 kpc ) to 96.5@xmath9 @xmath30 42.9@xmath9 ( 12.1 @xmath30 5.4 kpc ) the error bars on the modeled parameters are very small . the best fit model is shown as dashed curves superimposed on the data in fig . [ pv ] . taking only residuals ( data cube - best fit model cube ) greater than 3@xmath12 , the root - mean - square value is 1.77 mjy beam@xmath7 and the average relative error is @xmath4 10% . a zeroth moment map made from the cube of the residuals of the best fit is shown in fig . [ residuals ] . to determine whether there are changes in i , pa , and @xmath23 with galactocentric radius , r , we also modeled the galaxy in concentric rings . for example , either a lower value of i with increasing r ( a warp ) or a higher value of @xmath23 with increasing r ( i.e. a flare ) could conspire to make the halo of a galaxy appear to lag . we find that the inclination does not decrease with radius ( there is a slight increase ) . there is indeed evidence for flaring such that , in each data set , @xmath23 systematically increases with r. the most significant change is measured for the d array data alone such that the mean scale height between a radius of 140@xmath9 and 300@xmath9 ( 17.6 to 38 kpc ) is 5.5@xmath9 ( 693 pc ) compared to a value of 1.7@xmath9 ( 214 pc ) over the whole galaxy . thus , the disk is quite thin with a modest flare at large galactocentric radii . in order to model the halo alone , a variety of lower and upper cutoffs to _ z _ were applied and these data were modeled in a similar fashion . ( note that _ z _ refers to the actual distance from the galaxy s midplane , not just the projected distance from the major axis . ) the number of independent data points is reduced typically by about a factor of 3 , depending on which _ z _ limits are chosen , but we have also held some of the parameters ( the central position , position angle and systemic velocity ) fixed . for the halo models , we find _ no _ evidence that @xmath26 could be lower at higher _ z_. the same conclusion is reached if the c array and d array data are modeled independently . the absence of a lagging hi halo in ngc 2613 will be discussed in sect . [ no - lag ] . a distinct feature occurring at -307 km s@xmath7 ( see the advancing panel of fig . [ pv ] and fig . [ channels ] ) has been mentioned in sect . [ disk - halo ] and extends to a ( negative ) _ z _ height of 22 kpc . ( this feature is also distinguishable on the major axis pv slice of paper i ) . the channel maps show a complex structure below the plane , although two spurs dominate . the feature is very well defined in pv space and contributions to the feature can be seen in slices 6 through 9 which covers a range of 13 kpc in projected distance along the disk of ngc 2613 ( fig . [ moments]a ) . there is also a distinct height ( _ z _ @xmath4 90@xmath9 = 11 kpc ) at which the feature begins to widen abruptly in velocity ( fig . [ pv ] ) and also ( approximately ) spatially ( fig . [ channels ] ) , a point we return to in the model of sect . [ buoyancy ] . further support for hi at high _ z _ at this position comes from early hi observations by bottema ( 1989 ) which show an hi blob " extending to @xmath4 25 kpc from the plane and sharing the rotation of the underlying disk . in addition , our low resolution 20 cm continuum map also shows a single large feature at this position extending 21 kpc from midplane ( irwin et al . thus , we will treat the -307 km s@xmath7 feature to be a single , coherent structure , as implied by its appearance in pv space , centered at a projected galactocentric distance of 15.5 kpc . this feature occurs over a restricted region of observed radial velocity ( @xmath31 = -307 @xmath32 30 km s@xmath7 ) . assuming that the feature originates in the disk ( see sect . [ very - high ] ) , this should allow us to place limits on the position along the line of sight from which it originates since the rotation curve of the galaxy is known . the galactocentric radius is given by @xmath33 ( ignoring sin @xmath34 = 0.98 ) where @xmath35 is the projected galactocentric radius and @xmath36 is the circular velocity at @xmath14 which is approximately constant in this part of the galaxy ( table 2 ) . allowing a @xmath3230 km s@xmath7 range on @xmath31 , the feature could therefore have originated from @xmath37 kpc or anywhere within @xmath32 7 kpc from this value along the line of sight , corresponding to a range of @xmath14 between 15.5 and 17 kpc . since the projected size of the feature is larger than this range ( fig . [ channels ] ) , it is likely that there has been some expansion with distance from the plane . we can also put a limit on the age of the feature . gas which leaves the disk at an earlier time should have a component of line of sight velocity corresponding to the motion of the underlying disk at that time . as the disk rotates , gas which leaves later may have a different line of sight velocity component . consequently , there may be a gradual change in observed velocity with height ( increasing or decreasing ) depending on geometry . since we see no such change , we can use the velocity half - width of the feature , 30 km s@xmath7 , to place a limit on the angle over which the underlying disk has rotated over the age of the feature . we find that the feature must have formed over a timescale which is less than the time required for the disk to rotate @xmath32 24@xmath38 at galactocentric radii between 15.5 and 17 kpc . from the circular velocity in this region , the corresponding timescale is 4.2 @xmath30 10@xmath39 yr . these arguments will depend on how the ejected gas may or may not be coupled to the disk and should be considered order of magnitude only . nevertheless , the result is similar to the kinematic ages found for expanding supershells in other galaxies ( cf . king & irwin 1997 , lee & irwin 1997 , lee et al . basically the `` straightness '' of the feature in pv space requires that the feature have an age @xmath40 the rotation period . the mass of this feature , as measured in pv space to the level at which the feature blends with the disk , is ( @xmath41 m@xmath3 . approximating the feature as a cylinder of height , 22 kpc , width and depth , 13 kpc , the mean density is @xmath42 @xmath43 . ( reducing the diameter of the cylinder to 7.3 kpc ( 150@xmath9 ) results in a mean density of @xmath44 @xmath43 . ) if the feature has reached a _ z _ height of 22 kpc in 4.2 @xmath30 10@xmath39 yr , the implied mean outflow velocity is 512 km s@xmath7 . the corresponding hi mass outflow rate would be 1.9 m@xmath3 yr@xmath7 . we can also compute the potential energy of this feature ( see lee et al . 2001 ) which requires a knowledge of the mid - plane stellar density and scale height . comparing the rotation curve of ngc 2613 ( chaves & irwin 2001 ) to that of the milky way ( sofue & rubin 2001 ) , ngc 2613 is a factor of 4 more massive , and factor of 2 larger in radius . ngc 2613 also has a thinner disk , if we take the thickness of the hi disk ( 188 pc , table 2 ) to be representative of the stellar thickness . scaling the mid - plane stellar density from the galactic value ( 0.185 m@xmath3 pc@xmath45 ) yields 0.74 m@xmath3 pc@xmath45 for ngc 2613 and adopting 100@xmath9 ( 12.6 kpc ) ( about half the full _ z _ extent ) to be representative of the height of the feature above the plane , we find a potential energy of 1.5 @xmath30 10@xmath5 ergs for the -307 km s@xmath7 feature . thus , energy of this order is required to transport the cool gas to the heights observed . although very large , this value is not unlike that determined for the largest supershells in other edge - on galaxies ( cf . an input energy of @xmath4 3 @xmath30 10@xmath5 ergs for ngc 3556 , king & irwin 1997 ) . we return to this issue in sect . [ buoyancy ] . there are now several galaxies for which there is evidence for lagging halos . in ngc 891 ( i @xmath46 88.6@xmath38 ) , the rotation curve of the hi halo between _ z _ = 1.4 to 2.8 kpc reveals a peak velocity @xmath4 25 km s@xmath7 lower than in the disk ( swaters et al . 1997 ) . in ngc 5775 ( i = 86@xmath38 ) , the ionized gas velocities decrease with _ z _ to heights of 5 - 6 kpc above which the velocity remains roughly constant ( rand 2000 ) . the hi in this galaxy shows more complex structure , though lagging gas tends to dominate at _ z _ heights up to 5 to 6 kpc . above this region , the velocities are more nearly constant or the features break up into clumps which are seen over a wide range in velocity ( lee et al . models of ngc 2403 ( inclination of 61@xmath38 ) suggest that a lagging halo extends to _ z _ @xmath4 3 kpc ( schaap et al . 2000 ) . thus , from the sparse data available , the region of the lag extends to typically 3 to 6 kpc above the plane above which ( if gas is detected at all ) there is nt strong evidence for a global lag . given the presence of lagging halos in these other galaxies and also some theoretical expectation of lags ( cf . bregman 1980 , benjamin 2002 , collins et al . 2002 ) , the absence of a lagging velocity gradient along the discrete features ( sect . [ disk - halo ] ) and the absence of a global lag ( sect . [ models ] ) in ngc 2613 require comment . firstly , given the inclination of this galaxy ( 79@xmath38 , table 2 ) and its hi radius of @xmath4 35 kpc , the discrete features must reach a height of _ z _ @xmath47 6.5 kpc before they will be seen beyond the projection of the disk . if it is true that lags only exist up to 3 to 6 kpc and above this there is no longer a lag ( see above ) , then lags ( if they exist ) along the discrete features would be projected against the background disk . this would make a velocity gradient ( if present ) impossible to detect along the discrete features . the ` straightness ' of the -307 km s@xmath7 feature in velocity ( fig . [ pv ] ) to such high latitudes beyond the projection of the disk is , however , quite remarkable . if this feature emerges into a hot x - ray corona , as suggested in sect . [ buoyancy ] , then either the velocity of the broad - scale x - ray corona is not appreciably different from the underlying disk ( although the velocity dispersion is likely much higher ) , or the timescale over which it could appreciably affect this gas column is greater than the age of the feature . as for a globally lagging halo , our kinematical model is indeed capable of detecting such a lag , even if it is projected against the disk ( see sect . [ models ] ) provided there is sufficient high latitude emission that halo gas can be detected at all . what we have found , however , is that ngc 2613 does not have an hi thick disk or halo . the vertical exponential scale height ( 188 pc , table 2 ) indicates that the disk is , in fact , thin . if we create a model galaxy using the parameters of table 2 , it is straightforward to show that emission from all high latitude ( _ z _ @xmath47 1 kpc ) hi in fact falls below the map noise . thus , ngc 2613 does not have a detectable lagging halo because it does not have a halo at all . as a further comparative test , we consider the hi halo of another galaxy at the same distance , ngc 5775 ( i = 86@xmath38 , d = 25 mpc ) , for which we detected an hi halo with exponential scale height of 9.14@xmath9 ( 1.1 kpc ) using the same technique ( irwin 1994 ) . if we create a model for the lower inclination ( i = 79@xmath38 ) ngc 2613 in which the vertical density distribution declines exponentially from its in - disk value with the same scale height as ngc 5775 , we find that it could easily have been detected , even at the lower inclination of ngc 2613 . this result illustrates the importance of using a model , rather than pv slices alone , in drawing conclusions about the possible existence of lagging halos . for example , the model can disentangle inclination effects from the effects of a thick disk . for ngc 2613 , since the modeled inclination is now known to be on the low end of the range quoted in chaves & irwin ( 2000 ) , our earlier conclusion of a lag with _ z _ can now be largely explained by projection against the background disk . this is succintly illustrated via the dashed curves in fig . [ pv ] which show that our thin disk model provides an almost perfect fit to the data . we have argued elsewhere ( paper i ) , largely on the basis of above / below plane symmetry ( fig . [ carray ] ) , that the observed features are internally generated and represent outflows . while these new data show that the galaxy is indeed interacting , it is unlikely that most of the observed features are produced via cloud impacts since the cloud would have to pass completely through the disk , forming similar structures on both sides , in contrast to what is expected theoretically ( see santilln et al . anomalous velocities are also more likely in the case of impacts , whereas the velocities observed in ngc 2613 are typical of the underlying disk . thus , the new data are consistent with the outflow interpretation . we can not , however , rule out the possibility of some impacting clouds in this system and it is also possible that the interaction may assist in the process of disk - halo dynamics , for example via stimulating a starburst or instabilities such as the parker instability . if the features are internally generated , their very large sizes ( _ z _ up to 28 kpc , sect . [ disk - halo ] imply exceptional energies ( @xmath4 10@xmath5 ergs ) which are difficult to conceive of , especially at large distances from the nuclear vicinity and in a galaxy which is not a starburst . this energy problem has been known since the first detections of heiles shells in the milky way ( heiles 1979 , 1984 ) and has generated suggestions as far ranging as multiple supernovae and stellar winds ( heckman 2001 ) , gamma ray bursters ( efremov et al . 1998 , loeb & perna 1998 ) , and jet bubbles ( gopal - krishna & irwin 2000 ) to help explain the high energies . in the following section , we suggest that the exceptional heights achieved by the features in ngc 2613 may be as much a result of the environment into which the features emerge , as to the input source itself . the -307 km s@xmath7 feature ( fig . [ pv ] , advancing ) is remarkable in its well defined structure , in the very high latitude ( 22 kpc ) that it achieves , and in its sudden and dramatic widening and break up in velocity at a height of @xmath4 11 kpc . in individual slices ( see slices 8 and 9 , in particular ) , this velocity widening gives the feature a mushroom - like appearance and is reminiscent of buoyant gas rising through a higher density medium . a precedent for this kind of behaviour is the 350 pc galactic mushroom discovered in canadian galactic plane survey data and which has been interpreted in terms of buoyant outflow ( english et al . 2000 ) . given this similarity , we here consider whether buoyant outflow could explain the observed high latitude discrete hi features in ngc 2613 . we present the following as a feasibility study only to see whether the results provide a reasonable match to the data using realistic parameters . we consider a match to the -307 km s@xmath7 feature only , at this time , since it is the clearest case amongst the disk halo features in ngc 2613 ; however , if the model is correct , it must clearly apply to the other features as well . a similar development has been presented by avillez & mac low ( 2001 ) for smaller features like the galactic mushroom , but we here consider much larger scales ( many kpc ) , and also include the effects of drag . the scenario envisioned is one in which a hot , x - ray emitting corona already exists around the galaxy , possibly set up via venting through previous fountain or chimney activity . some event or events occur within the disk which are sufficiently energetic to produce blow - out . the hi is already entrained or swept up in some way by the time the outflow emerges into the halo . an initial velocity ( i.e. the velocity at blow - out ) could be present but is not included here . thus , the velocities achieved in the plume are a result only of buoyant forces . the attraction of this model is that , rather than requiring that all of the energy in the hi plume be supplied by the instigating event in the disk ( @xmath48 ergs , sect . [ 307feature ] ) , we require only enough energy to produce the blow - out condition ( e.g. @xmath49 ergs , tomisaka 1998 ) . the remaining energy is extracted from the pressure gradient in the hot coronal gas . ultimately , the galaxy s potential itself is providing the energy source . although we do not investigate the details of the entrainment or sweeping up of hi , we assume that the hot gas inside the plume carries the cool hi with it and that the density of the hi declines with _ z _ in a fashion similar to the hot buoyant plume material . in this feasibility study , we do not consider structure in the corona or outflow plume , other than the cylindrical geometry chosen for the plume , and also neglect the effects of shocks . the integrity of the hi in the presence of hot gas will be considered in the next section . the stellar / mass density distribution of the thin disk is described by : @xmath50 where @xmath51 is the scale height of the stellar disk , and @xmath52 is the stellar density at mid - plane . integration of poisson s equation : @xmath53 over the regime , @xmath54 , together with the above density distribution , results in a gravitational acceleration which is constant with @xmath1 and has a magnitude : g= 8g_*(0)z _ * the hot coronal gas ( subscripted , @xmath55 ) is taken to be isothermal at temperature , @xmath56 , with an exponential fall - off in both pressure and density , respectively : p_c(z)=p_c(z_*)e^-z / h _ c(z)=_c(z_*)e^-z / h where @xmath57 is the scale height . this hot coronal distribution starts at the top of the ( hi + stellar ) thin disk , i.e. at @xmath51 where the outflow just blows out of the thin layer . the scale height is given by h = kt_c/(gm_p ) where @xmath58 is boltzmann s constant , @xmath59 is the mean molecular weight and @xmath60 is the mass of the proton . we consider a pure hydrogen gas ( @xmath59 = 1 ) for simplicity . inside the plume ( subscripted , @xmath34 ) , we consider the gas to be adiabatic . the temperature gradient is : = ( 1 - 1 ) t_idp_i for a plume in pressure equilibrium with its surroundings , p_i(z)=p_c(z_*)e^-z / h where , again , we take the plume base to occur at @xmath51 . substituting eqn . and its derivative into eqn . yields : t_i(z)=t_i(z_*)e^((1-)zh ) and from eqn . , eqn . and the perfect gas law : _ i(z)=_i(z_*)e^ - ( z ) the equation of motion ( force per unit volume ) of the plume material is : _ i(z)dv = _ c(z ) g- _ i(z)g - c_d_c(z)v^2z - 2v the first term on the right hand side denotes the buoyancy force , the second term gives the weight of the plume material , the third term specifies the drag force against the upper surface ( assuming a case in which the effects of this force can propagate through the column ) where @xmath61 is the dimensionless drag coefficient , and the fourth term represents the viscous drag on the cylinder sides , where r is the cylinder radius and @xmath62 is the coefficient of viscosity ( @xmath63 ) . both @xmath61 and @xmath62 depend on the reynolds number : @xmath64 assuming a pure ionized hydrogen gas , where l is a scale length above which motion is damped by viscous effects and @xmath65 where @xmath66 is the coronal electron density . taking @xmath67 km s@xmath7 ( sect . [ 307feature ] ) and @xmath68 k , we find for size scales , @xmath69 , of order several kpc , @xmath70 is in the range @xmath71 . in this range of @xmath70 , @xmath72 for an incompressible fluid with cylindrical geometry . the viscosity coefficient includes both turbulent and molecular terms , i.e. @xmath73 , where @xmath74 = @xmath75 and @xmath76 , where @xmath77 is the critical reynolds number which designates the value of @xmath70 at which the flow becomes turbulent . @xmath78 is not known and depends on the geometry and nature of the interface , but typically has values between 1 and 100 . we can consider whether the drag force on the cylinder sides ( 4th term of eqn . ) is appreciable in comparison to the drag on the top of the cylinder ( 3rd term of eqn . ) . taking @xmath79 , the ratio of the 3rd to 4th drag terms becomes [ @xmath80 . since @xmath81 is of order @xmath1 , then for @xmath82 , even @xmath78 up to 100 ensures that the drag at the top will dominate over that at the sides . therefore the 4th term in eqn . is small and will be neglected . substituting @xmath83 , @xmath84 , and @xmath85 into eqn . and rearranging yields : = ( z _ * ) e^ -0.4zh gv- gv- ( z _ * ) e^-0.4zh vz this equation was integrated numerically for the input parameters shown in table 3 , yielding the curves of @xmath86 and @xmath87 shown in fig . [ vz ] and fig . [ zt ] , respectively . the stellar scale height , @xmath51 is set to 188 pc ( table 2 ) and we consider a coronal temperature , @xmath88 k ( models 1 to 4 ) which is comparable to that found from x - ray observations of ngc 253 ( pietsch et al . 2000 ) as well as a temperature which is a factor of 2 higher ( model 5 ) . since the midplane stellar density of 0.74 m@xmath3 pc@xmath45 is a value which has been scaled from galactic values , given the size and mass of ngc 2613 ( see sect . [ 307feature ] ) , we also consider a slightly lower value ( models 3 and 4 ) . the behaviour of the curves depends only on the density ratio at the base of the corona , @xmath89 , rather than the individual densities , but we can additionally fix the density at the bottom of the corona , @xmath90 to be equal to the hi density at the top of the thin disk at the position of the plume ( see table 2 ) , providing constraints upon the density within and outside of the plume . the peaks of the velocity curves ( fig . [ vz ] ) indicate where the acceleration of the plume material goes to zero ; we define the @xmath1 height at which this occurs to be @xmath91 . this position should correspond to the point at which the plume widens in velocity space , observationally determined to be @xmath4 11 kpc . models 1 and 2 ( fig . [ vz ] ) show the effect of changing the initial density ratio , such that the lower ratio ( model 2 ) results in a lower maximum velocity and a lower @xmath91 ( 7.4 kpc as compared to 11.9 kpc ) . a comparison of models 1 and 3 or of models 2 and 4 show the effect of decreasing the midplane mass density . this lowers the gravitational acceleration ( eqn . ) which lowers both the buoyancy and the weight of the plume material . it also increases the scale height ( eqn . ) which increases both the buoyancy and the drag . the net effect is higher @xmath91 at lower @xmath92 for the range of parameters given here . a comparison of models 1 and 5 shows the effect of increasing the coronal gas temperature . the higher temperature increases the scale height alone ( eqn . ) which , again , increases both buoyancy and drag . since drag is velocity dependent , the net effect is that the peak velocities achieved are lower for higher coronal temperatures . clearly , models 1 , 4 , and 5 provide an adequate match to the observed @xmath91 . model 1 , however , results in a mean plume density ( between @xmath51 and @xmath91 ) of only @xmath93 @xmath43 whereas the hi density alone is @xmath4 @xmath94 ( sect . [ 307feature ] ) , and is therefore not realistic . models 4 and 5 both result in reasonable fits to the known observational parameters , with model 5 slightly preferred because of its higher internal plume density ( higher than that of the hi alone ) and lower internal temperature . note that the mean internal temperature derived here ( @xmath95 k ) is comparable to the hot ( t = 1.2 kev = @xmath96 k ) component of the outflow in ngc 253 ( pietsch et al . the peak velocities derived here ( @xmath4 250 to 300 km s@xmath7 ) are lower than the gross estimate of 500 km s@xmath7 computed in sect . [ 307feature ] since we only model the feature up to the stall height ( 11 kpc ) rather than over its total length . the timescales of @xmath4 5 to 6 @xmath30 10@xmath39 yr ( fig . [ zt ] ) are in good agreement with the estimate found from the kinematical structure of the plume ( 4@xmath9710@xmath39 yrs ) . the mass outflow rate to the stall point is @xmath4 1 m@xmath3 yr@xmath7 . mass flow continuity requires that the plume of model 5 should increase in radius by a factor of 1.8 between _ z _ heights of 4.7 and 11.1 kpc . it is unlikely that the hi will be uniformly distributed at its mean density of @xmath4 @xmath98 @xmath43 but may exist in denser clouds or clumps whose sizes and distribution are not described by this simple model . in general , however , the situation will be not unlike that of those high velocity clouds ( hvcs ) which are in the hot halo of the milky way and the same issues regarding whether or not the clouds can remain neutral must be considered . photoionization of the hi by starlight should be negligible , considering the high galactic latitudes achieved . as pointed out by murali ( 2000 ) , the relevant interactions are hi - proton and hi - electron interactions for which the interaction cross - sections are in the range @xmath99 to @xmath100 @xmath101 for a relative velocity of order 200 km s@xmath7 as is appropriate here . the mean free path into the hi cloud is then l = 1/(@xmath102 ) = 0.3 to 3 pc or smaller if the hi is clumped . this is considerably smaller than the size scale of the plume and therefore these ionizing interactions should be minor . the most important interaction will be heating due to thermal conductivity leading to the evaporation of clouds . the classical mass evaporation rate , applicable to the case in which the mean free path is small in comparison to cloud size , is given , for approximately spherical clouds , by @xmath103 = 2.75 @xmath30 10@xmath104 t@xmath105 r@xmath106 @xmath107 g s@xmath7 , where t is the temperature of the external medium , r@xmath106 is 1/2 of the largest dimension of the cloud in parsecs , and @xmath107 is a parameter which measures the inhibition of heat flux due to the magnetic field and cloud geometry ( cowie & mckee 1979 , cowie et al . 1981 ) . we will assume @xmath107 = 1 ( no inhibition ) which maximizes the evaporation rate . using t = 2.5 @xmath30 10@xmath108 k and r = 5 @xmath30 10@xmath109 pc we find @xmath103 = 0.021 m@xmath110 yr@xmath7 ( to within factors of a few , given the different geometry ) . the timescale for complete evaporation of the 8@xmath9710@xmath39 m@xmath110 hi plume at a constant rate is then m/ @xmath103 = 4 @xmath30 10@xmath111 yr . since this is two orders of magnitude larger than the age of the plume , we expect that the plume will not evaporate over its lifetime . a caveat , however , is that since m/ @xmath103 @xmath112 r@xmath113 , if hi is distributed in many smaller clouds , then the evaporation time could approach the age of the plume . if the hi is indeed reaching such high values of _ z _ because of buoyancy , then hi disk - halo features probe the parameters of the halo gas , as suggested in sect . [ introduction ] . we have so far considered only whether buoyancy is feasible . there may be other dynamics at work as well ( for example , an initial velocity at blow - out or magnetic fields ) and the proposed hot corona is also unlikely to have the smooth distributions postulated here . nevertheless , in the context of the model , it is interesting to predict the x - ray luminosity of the corona in the vicinity of the plume . the results of table 3 show that it is feasible to transfer large masses to high galactic latitudes via buoyancy in a postulated x - ray corona . this drastically reduces the computed input energy requirements since it is no longer necessary to eject large masses to high altitudes . it is only necessary to achieve blow - out through the thin hi disk of ngc 2613 . the conditions required for blow - out have been investigated by a variety of authors ( see tomisaka 1998 , for example ) but energy requirements are typically of order 10@xmath127 ergs , rather than the 10@xmath5 ergs that would normally be required for the -307 km s@xmath7 feature . while the details of the interaction between the hi and hot outflowing gas are beyond the scope of this paper , it is important to ask whether there would originally have been sufficient hi in the disk from which the 8 @xmath30 10@xmath39 m@xmath3 in the -307 km s@xmath7 feature could have been swept up . assuming cylinderical geometry , a disk region of radius , 6.5 kpc ( sect . [ 307feature ] ) and using the modeled density and thickness of the disk ( table 2 ) , the available hi mass is @xmath4 3 @xmath30 10@xmath128 m@xmath3 suggesting that @xmath4 26% of the disk mass is swept up . this fraction would be higher for cone - like outflow . we should also consider whether the mass outflow rate is consistent with that of a galactic fountain . the estimated hi mass outflow rate ( sect . [ 307feature ] , sect . [ buoyancy ] ) is @xmath4 1 to 2 m@xmath3 yr@xmath7 . thus , the combined hi + hot gas outflow rate will be of order several m@xmath3 yr@xmath7 . collins et al . ( 2002 ) have estimated the mass flow rate for the diffuse ionized gas ( dig ) component in a galactic fountain , assuming a ballistic model of gas clouds . they find global values of @xmath129 = 22 @xmath130 m@xmath3 yr@xmath7 for ngc 891 and @xmath129 = 13 @xmath130 m@xmath3 yr@xmath7 for ngc 5775 , where @xmath131 is the filling factor . if these global values apply to a 20 kpc radius disk , taking @xmath131 = 0.2 and scaling to the disk area of the -370 km s@xmath7 feature yields mass outflow rates of 1.4 and 2.3 m@xmath3 yr@xmath7 which are comparable to what we estimate , above . it is not yet clear , however , whether these galactic fountain values can be directly scaled to the relevant regions of ngc 2613 . it may be that some additional source of pressure is still required , for example , magnetic fields in the form of a parker instability . if such fields continue to rise into the corona , the inclusion of this magnetic pressure would relax ( i.e. lower ) the density or temperature requirements internal to the plume ( table 3 ) . new vla d array data of ngc 2613 have been combined with previous higher resolution observations ( chaves & irwin 2001 ) to show a more extensive hi distribution than previously observed . the galaxy is now seen to have a tidal tail on its eastern side due to an interaction with its companion , eso 495-g017 , to the north - west . the three - dimensional hi distribution in ngc 2613 has been modeled following irwin & seaquist ( 1991 ) and irwin ( 1994 ) , a method which allows the volume density distribution to be determined as well as the scale height and inclination to be disentangled . we find that the inclination of the galaxy ( 79@xmath38 ) is on the low end of the range given in chaves & irwin ( 2001 ) and the model now shows that there is no hi halo in ngc 2613 . rather , the global hi distribution is well fit by a thin disk of exponential scale height , @xmath132 = 188 pc . the use of such a model is very important in drawing conclusions about the presence or absence of a global halo in a galaxy of this inclination . previous reports of a lagging halo from pv slices alone can largely be attributed to projection against the background disk . while there is no significant global hi halo in ngc 2613 , there are more discrete disk - halo hi features than previously detected and these hi features achieve extremely high latitudes . even though a tidal interaction is occurring , we suggest that most of the discrete kpc - scale features have been produced internally rather than from impacting clouds , although we do not rule out the possiblity of the companion galaxy having some indirect effect ( e.g. triggering instabilities ) . the presence of many discrete features may be related to the fact that the global hi disk is thin , favouring blow - out . the observed _ z _ heights are quite remarkable ( e.g. up to 28 kpc ) . the -307 km s@xmath7 feature , in particular , below the plane on the advancing side , reaching 22 kpc in @xmath133 height and of total mass , ( 8 @xmath32 2 ) @xmath30 10@xmath39 m@xmath3 , is very obvious and well - defined in pv space . its center is likely close to its projected radius of 15.5 kpc and it extends over a large ( @xmath32 7 kpc radius ) projected galactocentric radius . if this feature has achieved its @xmath133 height as a result of internal processes , then extremely large energies are required , @xmath4 10@xmath5 ergs . given the very high input energies required for the -307 km s@xmath7 feature , its resemblance to smaller buoyant features ( cf . the galactic mushroom , english et al . 2000 ) , and the fact that x - ray halos are being found around an increasing number of star forming spiral galaxies , we have carried out a feasibility study as to whether this feature can be interpreted as an adiabatic buoyant plume . the observed hi would be carried out by a hot , low density outflowing gas and , after having achieved blowout , would rise through a hot pre - existing x - ray corona . a reasonable example ( model 5 ) , has a mean plume temperature and density of @xmath134 k and 5.5 @xmath30 10@xmath45 @xmath43 rising into a hot isothermal corona of temperature and mean density , @xmath135 k and 0.035 @xmath43 , respectively . these conditions produce a stall height of 11 kpc which is where the observed plume widens in velocity and position space . the coronal density at the stall height is 2.8 @xmath30 10@xmath45 @xmath43 . the maximum outflow velocity in this model is 290 km s@xmath7 and it reaches the stall height in 5.4@xmath9710@xmath39 yrs . this model shows that , even with buoyancy alone ( and there may be additional sources of pressure such as magnetic fields ) , hi can reach these extreme _ z _ heights . the advantage of such a model is that the energy requirements from the initial event are drastically reduced to being only what is required for blowout , a reduction of several orders of magnitude . the energy is largely extracted from the gravitational potential of the galaxy rather than the initial event within the disk . the behaviour of the plume should sample the parameters of the x - ray corona . the predicted x - ray luminosity suggests that the corona should be observable at heights below the stall height although we expect that the distribution of x - ray emission may not be as smooth as assumed by the model . ji wishes to thank the natural sciences and engineering research council of canada for a research grant . we are grateful to dr . r. n. henriksen for fruitful and envigorating discussions . thanks also to mustapha ishak for assistance with maple . benjamin , r. 2002 , in seeing through the dust " , asp conf series , ed . a. r. taylor , t. l. landecker , & t. willis bottema r. 1989 , , 225 , 358 bregman j. n. 1980 , , 236 , 577 chaves t. a. , & irwin j. a. 2001 , , 557 , 646 ( paper i ) collins j. a. , rand r. j. , duric n. , & walterbos r. a. m. 2000 , , 536 , 645 collins j. a. , benjamin r. a. , & rand r. j. 2002 , in press cowie l. l. , & mckee c. f. 1977 , , 211 , 135 cowie l. l. , mckee c. f. , & ostriker j. p. 1981 , , 247 , 908 de avillez m. a. , & berry d. l. 2001 , , 328 , 708 duric n. , irwin j. , & bloemen h. 1998 , , 331 , 428 efremov y. n. , elmegreen b. g. , & hodge p. w. 1998 , , 501 , l163 english j. , taylor a. r. , mashchenko s. y. , irwin j. a. , basu s. , & johnstone d. 2000 , , 533 , l25 fraternali f. , oosterloo t. , sancisi r. , & van moorsel g. 2001 , astro - ph 0110369 gopal - krishna , & irwin j. a. 2000 , , 361 , 888 heckman t. 2001 , in gas and galaxy evolution , asp conf . series , vol . 240 , ( san francisco : asp ) , ed . j. e. hibbard , m. p. rupen , & j. h. van gorkom , 345 heiles c. 1979 , , 229 , 533 heiles c. 1984 , , 55,585 irwin j. a. 1994 , , 429 , 618 irwin j. a. , english j. , & sorathia b. 1999 , , 117 , 2102 irwin j. a. & seaquist e. r. 1991 , , 371 , 111 , erratum , 415,415 king d. l. , & irwin j. a. 1997 , newa , 2 , 251 lang k. 1999 , astrophysical formulae ( springer : berlin ) lee s .- w , & irwin j. a. 1997 , , 490 , 247 lee s .- w . , irwin j. a. , dettmar r .- j . , cunningham c. t. , golla g. , & wang q. d. 2001 , , 377 , 759 loeb a. , & perna r. 1998 , , 503 , l135 lotz w. 1967 , , 14 , 207 murali c. 2000 , , 529 , l81 olling r. , & merrifield m. r. 2000 , , 311 , 361 pietsch w. , vogler a. , klein u. , & zinnecker h. 2000 , , 360 , 24 rand r. j. 2000 , , 494 , l45 santilln a. , franco j. , martos m. , & kim j. 1999 , , 515 , 657 schaap w. e. , sancisi r. , & swaters r. a. 2000 , , 356 , l49 sofue y. & rubin v. 2001 , , 39 , 137 swaters r. a. , sancisi r. , & van der hulst j. m. 1997 , , 491 , 140 tomisaka k. 1998 , , 298 , 797 tllmann r. , dettmar r .- j . , soida m. , urbanik m. , & rossa j. 2000 , , 364 , l36 lcc no . velocity channels & 63 & 63 + velocity resolution ( km s@xmath7 ) & 20.84 & 20.84 + total bandwidth ( mhz ) & 6.25 & 6.25 + synthesized beam & & + major @xmath30 minor axis ( @xmath9 @xmath30 @xmath9 ) @ pa ( @xmath38 ) & 96.5 @xmath30 42.9 @ -21.6 & 47.1 @xmath30 32.1 @ -8.2 + rms noise / channel ( mjy beam@xmath7 ) & 0.93 & 0.45 + rms noise / channel ( k ) & 0.14 & 0.18 + ra ( j2000 ) ( h m s ) & 08 33 22.8 @xmath32 0.3 + dec ( j2000 ) ( @xmath38 @xmath136 @xmath9 ) & -22 58 29 @xmath137 + pa ( @xmath38 ) & 114.2 @xmath138 + i ( @xmath38 ) & 79.2 @xmath139 + v@xmath140 ( km s@xmath7 ) & 1663 @xmath141 + v@xmath142 ( km s@xmath7 ) & 304 @xmath32 4 + r@xmath142 ( @xmath9 ) & 112 @xmath143 + m & 0.51 @xmath144 + @xmath29 ( km s@xmath7 ) & 17 @xmath32 5 + n@xmath145 ( @xmath43)@xmath146 & 0.43 @xmath147 + r@xmath145 ( @xmath9 ) & 140 @xmath32 5 + d@xmath148 ( @xmath9 ) & 95 @xmath149 + d@xmath150 ( @xmath9 ) & 81 @xmath151 + h@xmath152 ( @xmath9 ) & 1.5 @xmath153 + 5 + @xmath56 ( @xmath154 k ) & 2.5 & 2.5 & 2.5 & 2.5 & 5.0 + @xmath52 ( @xmath155 ) & 0.74 & 0.74 & 0.50 & 0.50 & 0.74 + h@xmath156 & 1.4 & 1.4 & 2.1 & 2.1 & 2.8 + @xmath89 & 100 & 20 & 100 & 20 & 10 + @xmath157 ( @xmath43 ) & 0.15 & 0.15 & 0.15 & 0.15 & 0.15 + @xmath158 ( @xmath43 ) & 0.0015 & 0.0075 & 0.0015 & 0.0075 & 0.015 + @xmath91 ( kpc ) & 11.9 & 7.4 & 17.5 & 11.0 & 11.1 + @xmath159 ( @xmath43 ) & 0.015 & 0.025 & 0.016 & 0.026 & 0.035 + @xmath160 ( @xmath98 @xmath43 ) & 0.031 & 0.76 & 0.036 & 0.80 & 2.8 + @xmath161 ( @xmath98 @xmath43 ) & 0.27 & 2.1 & 0.28 & 2.2 & 5.5 + @xmath162 ( @xmath163 k ) & 6.8 & 2.0 & 6.9 & 2.0 & 1.2 +

on 2008 may 2 , _ chandra _ observed the x - ray spectrum of boo ( g8 v+k4 v ) , resolving the binary for the first time in x - rays and allowing the coronae of the two stars to be studied separately . with the contributions of boo a and b to the system s total x - ray emission now observationally established ( 88.5% and 11.5% , respectively ) , consideration of mass loss measurements for gk dwarfs of various activity levels ( including one for boo ) leads to the surprising conclusion that boo b may dominate the wind from the binary , with boo a s wind being very weak despite its active corona . emission measure distributions and coronal abundances are computed for both stars and compared with _ _ measurements of other moderately active stars with g8-k5 spectral types , all of which exhibit a narrow peak in emission measure near , indicating that the coronal heating process in these stars has a strong preference for this temperature . as is the case for the sun and many other stars , our sample of stars shows coronal abundance anomalies dependent on the first ionization potential ( fip ) of the element . we see no dependence of the degree of `` fip effect '' on activity , but there is a dependence on spectral type , a correlation that becomes more convincing when moderately active main sequence stars with a broader range of spectral types are considered . this clear dependence of coronal abundances on spectral type weakens if the stellar sample is allowed to be contaminated by evolved stars , interacting binaries , or extremely active stars with , explaining why this correlation has not been recognized in the past .

1005.3281

at a distance of only 6.7 pc , the primary of the @xmath0 boo binary system , @xmath0 boo a ( g8 v ) , is one of the two brightest `` moderately active '' coronal x - ray sources in the sky . the only star of this coronal activity level with a comparable x - ray flux at earth is @xmath3 eri ( k2 v , @xmath4 pc ) . as such , these two stars are of crucial importance for establishing the coronal properties of stars that are significantly more active than the sun , but which are not in the extremely active regime that would be represented by the rs cvn binaries or very young stars with rotation periods of only a few days . @xcite find that the coronal emission measure ( em ) distributions of @xmath3 eri and @xmath0 boo are similar to those of bright active regions on the sun . they note that the x - ray emission from these stars can therefore be explained by stellar surfaces that are completely covered with solar - like active regions , suggesting that @xmath3 eri and @xmath0 boo a represent a maximum of solar - like activity for coronal stars , whereas stars with even higher x - ray luminosities and higher coronal temperatures represent a completely different regime of coronal activity . this interpretation places @xmath3 eri and @xmath0 boo a at an interesting coronal transition point . further evidence that @xmath3 eri and @xmath0 boo a lie near a transition in the nature of coronal activity comes from stellar wind measurements , which are based on analyses of h i ly@xmath5 absorption from wind - ism interaction regions @xcite . at low activity levels mass loss increases with activity , with @xmath3 eri having a mass loss rate 30 times that of the sun ( i.e. , @xmath6 , where @xmath7 m@xmath8 yr@xmath9 ) . however , the somewhat more active @xmath0 boo system has a surprisingly weak wind with only @xmath10 , leading to the suggestion that the observed increase of mass loss with activity ends abruptly at an x - ray surface flux of about @xmath11 ergs @xmath12 s@xmath9 , with @xmath3 eri and @xmath0 boo on opposite sides of this dividing line @xcite . the wind measurements suggest a slight modification of the proposal that @xmath3 eri and @xmath0 boo a represent an extreme of truly solar - like activity . they suggest that @xmath3 eri is the true extreme , while @xmath0 boo a actually lies beyond it at the low end of the non - solar - like activity regime . evidence that @xmath0 boo a s magnetic topology is fundamentally different from the sun comes from spectropolarimetric measurements of photospheric fields on @xmath0 boo a. @xcite find evidence for both a global dipole field component much stronger than that of the sun ( @xmath13 g ) , and a strong toroidal field component ( @xmath14 g ) that has no clear solar analog . we also note that the activity level of @xmath0 boo a is roughly where one starts to see polar starspots @xcite , which also have no solar analog . @xcite find evidence for high latitude spots on @xmath0 boo a , though not necessarily `` polar spots . '' perhaps the strong large - scale fields found by @xcite envelope the star and inhibit wind flow , explaining why @xmath0 boo s wind is surprisingly weak . consistent with this idea , @xcite present evidence that as stellar rotation and magnetic activity increase , stars tend to store a larger fraction of their magnetic energy in large scale fields , which could in principle inhibit mass loss . with this background in mind , we here present an analysis of the x - ray spectrum of @xmath0 boo a based on new observations from _ chandra _ , with the goal of seeing whether the apparent changes in coronal character noted above find clear expression in the coronal spectrum of @xmath0 boo a , and coronal properties inferred from it . a similar spectrum of @xmath3 eri has already been analyzed , so a central goal of this paper is to compare the @xmath0 boo a spectrum with that of @xmath3 eri and similar stars ( * ? ? ? * hereafter wl06 ) . other coronal spectra of @xmath0 boo have been obtained by previous x - ray and euv missions ( e.g. , * ? ? ? * ; * ? ? ? * ) , but the new _ chandra _ data are superior in many ways . one of them is that _ chandra_/letgs observations allow the x - ray spectrum to be observed at high spectral resolution over a very broad spectral range of @xmath15 . this allows the detection of many coronal emission lines that have never been previously detected before from @xmath0 boo a. another particulary important improvement is that _ chandra _ s superb spatial resolution allows us for the first time to truly address the issue of @xmath0 boo s binarity . for @xmath0 boo a has a companion , @xmath0 boo b ( k4 v ) , which is only 6@xmath16 away , and _ no x - ray or uv observation of @xmath0 boo has ever truly resolved the binary . _ table 1 lists various properties of the two @xmath0 boo stars . the faster rotation and larger size of the primary lead to the expectation that @xmath0 boo a should dominate the coronal emission from the system . detailed analyses of iue data and a rosat / hri image of @xmath0 boo do indeed suggest that @xmath0 boo a dominates the binary s uv and x - ray emission @xcite , so the usual assumption of simply assigning all flux to the primary should be a decent approximation . however , @xmath0 boo b s spectral type and rotation period are not that different from the x - ray luminous @xmath3 eri ( k2 v ; @xmath17 days ) . thus , @xmath0 boo b should be contributing a non - negligible flux to the system s total x - ray emission . the older observations simply do not have sufficient spatial resolution to quantify how much . in short , the new _ chandra _ data presented here allow the first x - ray detection of @xmath0 boo b , and they ensure that the brighter spectrum of @xmath0 boo a is uncontaminated by any emission from the companion . the @xmath0 boo binary was observed on 2007 may 2 for 97.2 ksec with _ chandra _ s letgs configuration , combining the low energy transmission grating ( letg ) and the component of the high resolution camera detector intended for use in spectroscopy ( hrc - s ) . the zeroth - order letgs image of @xmath0 boo is displayed in figure 1 , showing how cleanly the stars are separated in the _ chandra _ data , representing the first resolution of the binary in x - rays . we measure a position angle and stellar separation for the binary of @xmath18 and @xmath19 , respectively , in reasonable agreement with the @xmath20 and @xmath21 values expected from the system s orbital elements @xcite . the zeroth order image provides the simplest way to measure the contribution of each star to @xmath0 boo s total x - ray emission . the primary dominates the emission from the system , accounting for 88.5% of the total counts . we also use the zeroth order image to assess the variability of the two stars during the course of the daylong observation . figure 2 shows light curves for the two @xmath0 boo components , indicating significant variability for both stars . for @xmath0 boo a , there appears to be a long duration flare near the beginning of the observation , and a much shorter duration event later on . in extracting spectra from the raw _ chandra _ data , we initially tried to separate flaring and quiescent times . however , we were unable to discern any clear differences between the resulting noisy flare spectrum and that of quiescence . thus , in the final spectral extraction procedure described below , we simply extract a single spectrum for each star , including both quiescent and flaring times . the standard pipeline processing of letgs data does not properly separate the spectra of the two stars , so the spectral extraction must be performed in a more manual fashion . in a previous paper we have described in detail how we extracted separate letgs spectra of stars in two other binary systems , 36 oph and 70 oph ( wl06 ) . we follow a very similar procedure here , which is now described in abbreviated form . the data are processed using version 4.1.2 of the ciao software . the processing includes a removal of background counts using the standard `` light '' pulse height background filter available in ciao , but we also account for time - dependent gain corrections using software outside of ciao , corrections which would later be added to a subsequent official ciao release ( version 4.2 ) . by reducing background counts from our data , this filtering significantly improves the signal - to - noise of our final spectra . the processing initially yields a zeroth order image of the target at the aimpoint , which we have already discussed ( see fig . 1 ) , and two essentially identical spectra dispersed in opposite directions along the long axis of the hrc - s detector , a plus order and a minus order spectrum . the spatial resolution in the cross - dispersion direction worsens at long wavelengths where the spectrum is furthest from the focal point . we use a conservatively broad source extraction window with a width of 25 pixels centered on each stellar spectrum for @xmath22 . the window is broadened for @xmath23 . to avoid overlapping source windows , for each star we expand the window only in the direction away from its companion star by 30 pixels . at the longest wavelengths , we have to accept that there is some small degree of unresolvable blending . we use 90 pixel background windows on each side of the stellar spectra to determine a wavelength dependent background to subtract from the spectra of the two stars . finally , the plus and minus order spectra are coadded to yield the final product shown in figure 3 . the clear separation of the @xmath0 boo binary allows the first precise and unambiguous measurement of the x - ray luminosity of both stars . the system s x - ray luminosity reported in the rosat all - sky survey is @xmath24 @xcite . dividing this emission between @xmath0 boo s two stars according to the contribution ratio suggested by figure 1 yields the rosat x - ray luminosities reported in table 1 . this new knowledge of the coronal activity of both @xmath0 boo stars has interesting ramifications for the interpretation of @xmath0 boo s wind measurement . as mentioned in 1 , @xmath0 boo has a measured wind strength of @xmath10 , but the measurement is for the combined wind of the system , and there is no way to tell from the observations the contribution of each star to this total @xcite . figure 4 plots mass loss rate as a function of coronal x - ray emission for all main sequence stars with measured winds , analogous to a figure from @xcite . but in this figure we plot separate points for @xmath0 boo a and @xmath0 boo b , for three different assumptions about how the collective wind is divided between the two stars . we now know for a fact that @xmath0 boo b is very similar to @xmath3 eri , 36 oph , and 70 oph in terms of its coronal activity . but these stars all have rather high mass loss rates . the only way that @xmath0 boo b can be consistent with even the weakest of these winds , that of 36 oph , is if @xmath0 boo b accounts for practically all of @xmath0 boo s wind . this would leave the more active @xmath0 boo a with a wind that is possibly weaker than that of the sun , despite its active corona , strengthening the case for the existence of a wind dividing line near an x - ray surface flux of @xmath11 ergs @xmath12 s@xmath9 ( see fig . 4 ) , where stronger large scale magnetic fields begin to inhibit mass loss . even if @xmath0 boo a accounts for nearly all of the binary s wind , it would still be weak compared to that of @xmath3 eri , and then one would have to explain why @xmath0 boo b apparently has a wind over an order of magnitude weaker than that of many other stars of equal activity . we come to the surprising conclusion that the most likely interpretation of the available data is that @xmath0 boo b dominates the wind of the @xmath0 boo system . this conclusion is also consistent with the idea that a wind dividing line exists at an x - ray surface flux of about @xmath11 ergs @xmath12 s@xmath9 ( see 1 ) , separating less active stars with strong winds and more active stars with weaker winds @xcite . this activity dependent wind dividing line is not to be confused with the spectral type dependent wind - corona dividing line long known to exist for red giant stars @xcite . it must be noted that our conclusions rely on a very limited number of wind detections . further support for these interpretations should be sought from more wind measurements of main sequence stars , particularly for active stars . we use version 6.0 of the chianti atomic database to identify emission lines in our spectra @xcite . the identified lines are shown in figure 3 and listed in table 2 . table 2 lists counts for the detected lines , measured by direct integration from the spectra . line formation temperatures are quoted in the third column of the table , based on maxima of contribution functions computed using the ionization equilibrium computations of @xcite . uncertainties quoted in table 2 are 1@xmath25 . for the fainter @xmath0 boo b source , only the strongest lines are detected , but we quote 2@xmath25 upper limits for the nondetections , based on summing in quadrature the uncertainties in 5 bins surrounding the location of the line , and then multiplying this sum by two . measurements are made for all lines that appear visually to be at least marginally detected . however , cases in which the uncertainty implies a @xmath26 detection must be considered questionable . after the emission measure analysis described in 4.4 is completed , we confirm _ a posteriori _ that these questionable detections are at least plausible based on the derived emission measure distributions and abundances , which are constrained primarily by the stronger emission lines in the analysis . many of the emission lines are blends . we list in table 2 all lines that we believe contribute a significant amount of flux to the feature based on the line strengths in the chianti database . this determination is reassessed after em distributions are estimated , and model spectra can be computed from them and compared with the data . a few of the features identified in figure 3 are blends of lines of different species ( see table 2 ) . although we measure counts for these blends and list them in the table , these measurements are not used in the emission measure analysis . coronal electron densities can be estimated for the two @xmath0 boo stars directly from the o vii @xmath2721.8 line measurements in table 2 . there are in fact several he - like triplets in the observed spectral range that are useful for density measurements : si xiii @xmath276.7 , mg xi @xmath279.2 , ne ix @xmath2713.6 , and o vii @xmath2721.8 . however , insufficient spectral resolution , low signal - to - noise , and blending issues complicate any attempt to use the si xiii , mg xi , and ne ix lines for this purpose @xcite . for this reason , we focus only on the stronger , better separated o vii lines here . the specific line ratio of interest is that of the forbidden line to the intersystem line : @xmath28 . using the results of collisional equilibrium models from @xcite , the measured line ratios for @xmath0 boo a and @xmath0 boo b translate to upper limits of @xmath29 and @xmath30 , respectively , as reported in table 3 . the @xmath0 boo a limit is consistent with values that @xcite estimate from euve spectra . densities of about @xmath31 are consistent with spectra of @xmath3 eri , 36 oph ab , and 70 oph ab as well , which have activity levels similar to that of the @xmath0 boo stars ( wl06 ) . thus , there is no clear evidence for coronal density differences among moderately active stars observed by _ chandra_. among moderately active gk dwarfs with intrinsic x - ray luminosities of @xmath32 , the two easiest to observe from earth are @xmath0 boo a ( g8 v ) and @xmath3 eri ( k2 v ) , due to their close proximity . we can see how the coronal temperatures and abundances of these two key stars compare simply by plotting ratios of their coronal line luminosities as a function of line formation temperature , as is done in figure 5 . the @xmath0 boo a line luminosities in the figure originate from the line counts measured in table 2 . the @xmath3 eri luminosities are computed from line counts similarly listed in wl06 . figure 5 shows that nearly all the line ratios are above one at all temperatures , indicative of @xmath0 boo a being the more active star . the ratios are clearly temperature dependent , with the ratio rising from @xmath33 at @xmath34 to @xmath35 at @xmath36 . this implies a hotter corona for the more active @xmath0 boo a. this is consistent with previous observations demonstrating correlations between coronal activity and coronal temperature @xcite . it is intriguing how smoothly and linearly the line ratios rise with temperature . the lines in figure 5 are divided into those with low first ionization potential ( fip ) and those with high fip . in the solar corona and solar wind , elemental abundances are found to be dependent on fip . relative to the photosphere , elements with low fip ( fe , mg , si , etc . ) are generally found to have coronal abundances that are enhanced relative to elements with high fip ( c , n , o , ne , etc . ) evidence for this effect has been found for some stars of low to moderate activity @xcite . however , on very active stars the fip effect tends to be either absent , or sometimes an inverse fip effect is observed , where low - fip elements have coronal abundances that are _ lower _ than photospheric @xcite . thus , for main sequence stars the picture is of the fip effect generally decreasing as activity increases @xcite . but there are exceptions to this trend , and figure 5 is itself inconsistent with a tight correlation between activity and fip effect , as the lower flux ratios of the high - fip ions in figure 5 demonstrate that @xmath0 boo a s corona exhibits a stronger fip effect than the less active @xmath3 eri . we will return to this issue in 4.6 . in order to quantify the coronal abundances and temperature distributions , we perform an emission measure analysis for @xmath0 boo a and b based on our measurements of their coronal emission lines . the em analysis uses version 2.6 of the pintofale software developed by @xcite . this analysis follows closely that of wl06 in our study of letgs spectra from @xmath3 eri , 36 oph , and 70 oph , so we do not describe it again in great detail here . figure 6 shows em distributions ( in units of @xmath37 ) for the two @xmath0 boo stars , and compares them with distributions that we previously computed for five other moderately active g8-k5 dwarfs ( wl06 ) . these distributions are computed using chianti emissivities @xcite , and ionization balance calculations from @xcite . our line measurements are corrected for interstellar absorption , but the interstellar h column densities towards these stars , including @xmath0 boo ( see table 1 ) , are low enough that this is not an important correction . for elements with detected lines we can compute coronal abundances in the em analysis . however , line measurements alone only allow _ relative _ abundances to be computed , and only the shape of the em distribution can be inferred as opposed to its absolute value . initially , we assume a solar photospheric abundance for fe from @xcite , and all other abundances are computed relative to fe . the derived @xmath0 boo abundance ratios are listed in table 3 . in order to compute an _ absolute _ fe abundance , and to properly normalize the em distribution , the line - to - continuum ratio must be assessed . figure 7 illustrates this process for the case of @xmath0 boo a. the figure shows a synthetic spectrum generated from the emission measure distribution in figure 6 , assuming an absolute fe abundance of @xmath38}=0.7{\rm [ fe / h]}_{\odot}$ ] , which is judged to lead to the best fit . this is the same as the ratio found for @xmath3 eri ( wl06 ) . for @xmath0 boo b , we find a similar value of @xmath38}=0.8{\rm [ fe / h]}_{\odot}$ ] ( see table 3 ) . these results are used to properly normalize the em distributions in figure 6 . figure 3 displays synthetic spectra of @xmath0 boo a and b based on the em distributions . in addition to the dominant first - order spectrum , we also take into account orders 25 in generating synthetic spectra . both figures 3 and 7 explicitly show the contributions of these higher orders to the total spectrum . in logarithmic terms , our measurement for @xmath0 boo a implies @xmath39}=-0.15 $ ] , where @xmath40 is here the solar photospheric abundance of fe . this measurement can be compared with the @xmath39}=0.1^{+0.2}_{-0.15}$ ] measurement of @xcite based on euve spectra . at this point it is worth noting that the contributions of weak , unresolved , and unidentified lines to the observed `` continuum '' could easily lead to underestimates of @xmath38}$ ] , so our @xmath38}$ ] measurements could be conservatively regarded as lower limits . given that our measurement is lower than that of @xcite , perhaps there are more lines missing and unaccounted for within the letgs spectral region than in the euve spectral region analyzed by @xcite , presumably leading to an underestimate of @xmath38}$ ] on our part . nevertheless , both our measurement and that of @xcite show a higher fe abundance in the corona than in the photosphere , where @xmath41}=-0.26 $ ] ( see table 1 ) , consistent with a solar - like fip effect ( see 4.3 ) . in following sections that focus on comparing the coronal abundances of different stars , we avoid the systematic errors involved in computing absolute abundances by comparing only relative abundances , specifically abundances relative to fe . the seven em distributions in figure 6 are superficially similar in appearance , with a dramatic rise at about @xmath34 . but perhaps the most interesting feature is that all of the distributions show an em peak near @xmath1 , a peak that is particularly convincing for the two brightest stars , @xmath0 boo a and @xmath3 eri . the coronal heating on these moderately active stars seems to favor this particular temperature . the em distribution that @xcite derived for @xmath0 boo a from euve data agrees nicely with our _ chandra _ distribution . their peak at @xmath1 is not quite as narrow or prominent , but this may just be a result of the ion - by - ion method that @xcite use to quantify their emission measures . as there is some overlap in the euve and _ chandra_/letgs spectra in the @xmath42 range , there are some lines in common to these two analyses . the most precise way to compare the activity level of @xmath0 boo a during the euve and letgs observations is to directly compare the line fluxes rather than the derived em distributions . the letgs / euve line flux ratios have a weighted mean and standard deviation of @xmath43 . an assessment of the relative flux calibration of euve and letgs by @xcite using white dwarf spectra suggests calibration errors in the opposite direction , with letgs / euve ratios of @xmath44 . another systematic error that works in the opposite direction is the fact that , unlike letgs , the euve fluxes will be contaminated by emission from @xmath0 boo b. thus , the below unity value of the average letgs / euve line ratio for @xmath0 boo a is likely indicative of genuine stellar variability , with the star being somewhat fainter during the letgs observation . however , the very modest level of variability suggested by the letgs / euve comparison is hardly surprising , as the coronal emissions of cool main sequence stars are known to be quite variable on all kinds of timescales . finally , we can compute broadband x - ray luminosities directly from the letgs spectra . we first subtract the higher order contributions to the spectra estimated from the em analysis ( i.e. , the green line in figure 7 ) . we can then combine the corrected spectra with the known first order effective area curve and exposure time to convert counts in our spectra to fluxes . from the flux spectra we can compute x - ray luminosities for any wavelength region within the spectrum . in table 1 , we list luminosities computed in this way for the rosat - like @xmath45 kev ( @xmath46 ) bandpass . the total @xmath0 boo system luminosity of @xmath47 computed in this way agrees reasonably well with the rosat all - sky survey measurement of @xmath24 @xcite . the first order spectra suggest @xmath0 boo a accounts for about 84.6% of @xmath0 boo s total x - ray flux , compared to the 88.5% contribution to counts in the zeroth order image . the discrepancy is due in part to the difference between count measurements and flux , the latter of which requires consideration of the effective area of the instrument . the discrepancy is also due in part to differences in the effective areas for the zeroth order image and first order spectra . in 4.4 , we compared the em distributions of seven moderately active g8-k5 dwarfs observed with letgs . we also wish to compare the coronal abundances of these seven stars , in order to first determine if a fip effect exists for these stars , and if so to then compare its magnitude within this stellar sample . our calculation of abundances in the em analysis assumes that abundances are the same at all temperatures throughout the corona . figure 5 suggests that this is probably a good assumption , as it shows that the differential fip effect between @xmath0 boo a and @xmath3 eri is the same throughout the @xmath48 temperature range that dominates the coronal emission measure distributions of these stars . figure 8 compares the coronal abundances of our sample of stars , plotting the abundances relative to fe , as listed in table 3 . but the search for a fip effect relies on knowing not just the coronal abundance by itself , but the coronal abundance relative to the photospheric abundance . thus , in figure 8 we subtract photospheric abundances from the coronal ones . however , because photospheric abundances are imperfectly known , taking them into account involves numerous assumptions , which we now describe . in table 1 , photospheric abundances measured for @xmath0 boo a are listed for c , o , mg , si , and fe . these abundances are quoted relative to solar . the @xcite analysis from which these abundances are obtained involves a line - by - line comparison of solar and stellar photospheric absorption lines , so these stellar abundances are _ fundamentally _ relative values . for the purposes of figure 8 , we require absolute photospheric abundances relative to fe , log [ x / fe]@xmath49 , so we have to combine the abundances in table 1 with some assumed solar abundances . in figure 8a , we use solar abundances from @xcite . there are two elements for which no stellar photospheric abundances are available for @xmath0 boo a : n and ne . since these are high - fip elements , we assume that their abundances relative to the sun are the same as that of another high - fip element , o. in other words , we assume @xmath50 = \log [ ne / ne_{\odot } ] = \log [ o / o_{\odot}]=-0.09}$ ] ( see table 1 ) . for @xmath0 boo b there are no independent photospheric abundance measurements , so we simply have to assume @xmath0 boo b is identical to @xmath0 boo a in this respect . similar assumptions have been made for the photospheric abundances of @xmath3 eri , 70 oph ab , and 36 oph ab ( see wl06 ) . the low - fip and high - fip data points in figure 8a are connected with separate dotted lines . these dotted lines indicate the following curious abundance patterns for all the stars : 1 . mg abundances are always higher than those of si , 2 . ne abundances are always higher than the other high - fip abundances , and 3 . c and n abundances are intermediate between those of o and ne . are these abundance patterns telling us something about how coronal fractionation is operating in these moderately active stars ? or are these patterns just a product of assumptions made about the photospheric abundances ? at least for the high - fip elements , the latter of these two explanations seems most likely . in figure 8b , we explore this further by changing our assumed photospheric abundances . this figure is like figure 8a , but instead of @xcite solar abundances , we use those of @xcite , with the exception of ne . photospheric ne abundances provide unique problems , because there are no photospheric ne absorption lines to measure , which means that solar ne abundance measurements actually have to rely on emission lines from the corona , where elements are subject to fractionation processes like those that yield the fip effect . in figure 8b , we change the assumed photospheric abundance of ne drastically by assuming that the average coronal ne / o ratio of @xmath51 measured by @xcite _ for active stars _ actually represents the real _ photospheric _ abundance of _ all _ cool main sequence stars , as opposed to the much lower solar coronal ne / o ratio of @xmath52 @xcite . in other words , rather than assuming that ne is fractionated from o in active stars but not in the sun , in figure 8b we assume that ne is being fractionated in the atmospheres of inactive stars like the sun , but not in active stars . theoretical support for this interpretation is provided by @xcite , and a higher solar photospheric ne abundance could also help resolve discrepancies between solar interior models and helioseismological observations @xcite . with the revised assumptions regarding photospheric abundances discussed above , the high - fip abundance patterns seen in figure 8a are largely absent in figure 8b , and the high - fip dotted lines are closer to horizontal . this result could be used as evidence in favor of the photospheric abundance assumptions being made in figure 8b , as opposed to the @xcite solar abundances being assumed to make figure 8a . although changing the assumed solar abundances seems to be able to significantly reduce the discrepancies among the high - fip elements , such is not the case for the low - fip elements , where there is little difference in the discrepancy between mg and si in figures 8a and 8b . thus , presumably coronal fractionation processes really are enhancing stellar mg abundances relative to si in the coronae of these moderately active stars . @xcite propose that coronal fractionation originates from ponderomotive forces induced by the propagation of alfvn waves through the chromosphere . these models demonstrate the potential for fractionating mg relative to si in a manner consistent with the observed mg / si ratios . in their sample of early g dwarfs , @xcite also generally find high mg / si ratios . but in a sample of m dwarfs , all of which happen to show inverse fip effects , @xcite find _ low _ mg / si ratios , suggesting this behavior may be spectral type dependent . for the high - fip dotted lines in figure 8 , the lower the line , the higher the coronal fe abundance is relative to the high - fip elements , and therefore the stronger the inferred fip effect . significant fip effect is apparent for many of the stars , with 70 oph a s corona possessing the strongest fip bias . but there is a significant amount of scatter in the level of fip effect found for this sample of stars . figure 8 suggests the following sequence of decreasing fip bias : 70 oph a , @xmath0 boo a , 36 oph b , 36 oph a , @xmath0 boo b , @xmath3 eri , 70 oph b. the last of these stars , 70 oph b , appears to actually have a slight _ inverse _ fip effect . there is some spread of x - ray luminosity within our sample of stars , but there is no correlation at all between x - ray luminosity and fip effect in this sample . the most luminous star , @xmath0 boo a , shows a strong fip effect , but the second most luminous , @xmath3 eri , shows only a weak fip effect at most . the relatively strong fip effect of the most active star in our sample also runs contrary to the general anti - correlation between stellar activity and fip effect that has been reported in the past ( see 4.3 ) . since the fip effect on the sun is latitude dependent , with the effect being strong near the equator and minimal in the polar regions , one might imagine that perhaps stellar orientation could be a factor in the fip bias observed from a star , assuming coronal abundances on these more active stars also exhibit some sort of latitude dependence . however , since coronal emission is optically thin , at least half of a star s total emission should be visible from a star regardless of orientation , so it is hard to imagine orientation effects explaining the roughly factor - of-3 variation in fip bias seen in figure 8 . in our previous paper ( wl06 ) we have emphasized the remarkable difference between 70 oph a and 70 oph b , which represent the fip effect extremes in figure 8 despite being similar stars in the same binary system . one possible explanation that we proposed is that perhaps the fip effect is a time - dependent phenomenon . the solar example provides a precedent for this , with newly emerged active regions having close to photospheric abundances and only acquiring a fip bias on timescales of days @xcite . thus , for the stars in figure 8 with little or no apparent fip bias ( e.g. , 70 oph b , @xmath3 eri ) , perhaps the visible surface of the star at the time of observation was dominated by young active regions , whereas the stars with strong observed fip effects ( e.g. , 70 oph a , @xmath0 boo a ) were observed with older active regions . the time - dependent interpretation can only be tested by observing these stars again , preferably more than once . but the limited amount of evidence available so far does _ not _ support this interpretation . abundances computed from euve spectra taken at different times from our letgs data seem to be very consistent with our letgs results , showing a strong fip effect for @xmath0 boo a , and only a weak fip effect , if any , for @xmath3 eri @xcite . @xcite looked for abundance variations in multiple observations of the very active k dwarf ab dor , with the observations encompassing a range of activity levels , and found no such variability . there does not seem to be a strong case for activity dependence , time dependence , or orientation effects being responsible for the variation in fip bias seen in figure 8 . there does seem to be a case for there being a spectral type dependence . the two earliest type stars ( @xmath0 boo a and 70 oph a ) have the strongest solar - like fip effect , while the three latest type stars ( 70 oph b , @xmath0 boo b , and @xmath3 eri ) have the weakest . this is explored further in figure 9 , where we first compute the average of the high - fip dotted lines in figure 8b , a quantity we simply call the `` fip bias , '' and then plot this quantity versus spectral type . in order to broaden the spectral range under consideration , in figure 9 we supplement our measurements with those of early g stars from @xcite and m dwarfs from @xcite . in utilizing measurements from these other papers , we were careful to compute the `` fip bias '' in the same manner that we have for our sample of stars , with the same assumptions about solar photospheric abundances as in figure 8b . for the @xcite sample , stellar photospheric abundances are available from @xcite , which we take into account , but for the m dwarfs of @xcite there are no such measurements and we have no recourse but to assume that the photospheres of these stars possess solar photospheric abundances . finally , we have added a point for the sun , using abundances listed in @xcite . figure 9 shows a remarkably tight correlation of fip bias with spectral type , with the earlier type stars having solar - like fip effects and the later stars having inverse fip effects , though the lack of photospheric abundance measurements for the m dwarfs means that conclusions about them are more uncertain . the reversal point seems to happen near spectral type k5 v. to the best of our knowledge , this is the first time a strong spectral type dependence of fip bias has been claimed . why has this seemingly strong correlation not been recognized in the past ? answering this question requires clearly defining what the sample of stars in figure 9 represents . in short , these are inactive to moderately active main sequence stars with x - ray luminosities of @xmath53 . for solar - like main sequence stars such stars have rotational periods of @xmath54 days and ages of @xmath55 gyr @xcite most surveys of coronal x - ray emission are dominated by interacting binaries , evolved stars , and other targets that represent extremes of coronal activity , which are commonly observed by x - ray telescopes because they are so bright and easy to detect . but by avoiding such extremes and focusing solely on more normal stars , we find a very different empirical view of the fip effect . gone is the kind of activity dependence on which previous authors have focused ( see 4.3 ) , replaced instead with a strong spectral type dependence . in order to emphasize the lack of activity dependence in figure 9 , consider the two least active stars in the sample : the sun ( g2 v ) and proxima cen ( m5.5 ve ) . the two g1 stars right next to the sun in the figure have x - ray luminosities almost two orders of magnitude higher than the sun . but there is no difference in `` fip bias '' . likewise , three of the four stars right next to proxima cen in the figure have x - ray luminosities about two orders of magnitude higher than it . once again , no difference in fip bias . there are two stars in the @xcite sample that are not included in figure 9 , because they have @xmath2 and therefore violate our policy of avoiding extremes . these two stars are 47 cas b and ek dra , which both have rotation periods of under three days . this rapid rotation will probably exist for all main sequence stars with @xmath2 . if we plotted points for 47 cas b and ek dra in figure 9 , they would lie well above the relation defined by the other stars . rather than proposing that magnetic activity is starting to operate somewhat differently at the activity level of 47 cas b and ek dra , thereby yielding a different fip bias , we instead propose that the changes in fip effect are ultimately driven by changes to fundamental stellar properties induced by the rapid stellar rotation . rotation as rapid as that possessed by 47 cas b and ek dra will presumably affect the atmospheric and convection zone properties of the star to some extent ( e.g. , * ? ? ? interacting binaries and evolved stars will also have substantial differences between their atmospheric properties and those of main sequence stars with identical spectral types . figure 9 suggests to us that fip bias is tied directly to basic stellar properties of the star , and not any simple coronal property like total x - ray flux or magnetic field strength . within the theoretical paradigm of @xcite , in which alfvn wave propagation through the chromosphere drives coronal abundance anomalies , it is possible to imagine that changes in basic stellar properties such as radius , surface gravity , or convective motions could affect these waves and thereby affect the coronal fip bias . _ in short , we propose that the spectral type dependence of fip bias in figure 9 provides a more fundamental insight into the nature and cause of the fip effect than the activity trends that have been explored in the past . _ we have used _ chandra_/letgs observations to study the coronae of the @xmath0 boo binary , and we have compared our results with a previous study of five other moderately active gk dwarfs : @xmath3 eri , 70 oph ab , and 36 oph ab . our findings are summarized as follows : 1 . the two stellar components of @xmath0 boo are resolved for the first time in x - rays . we find that @xmath0 boo a accounts for 88.5% of the counts in the zeroth order image . for comparison , spectral analysis implies an 84.6% contribution to the system s x - ray flux in the rosat - like 0.1 - 2.4 kev bandpass . 2 . with @xmath0 boo b detected for the first time in x - rays , we now know it has an x - ray luminosity comparable to stars known to have strong stellar winds , leading to the conclusion that @xmath0 boo b may account for most of the modest wind observed from the system , with the more active @xmath0 boo a possessing a surprisingly weak wind despite its active corona . this strengthens the case for the presence of a `` wind dividing line '' separating moderately active stars with strong winds and very active stars with weak winds , possibly due to the increasing strength of large scale global magnetic fields that inhibit mass loss . 3 . our emission measure analysis of @xmath0 boo a and b finds evidence for a narrow peak in em at @xmath1 for both stars , a peak also seen for the other moderately active g8-k5 dwarfs in our sample , indicating that the coronal heating mechanism operating on these stars favors this temperature . the emission measure analysis provides us with coronal abundances relative to fe . looking for coronal abundance anomalies requires correcting for stellar photospheric abundances , which in turn involves assuming solar photospheric abundances , since the stellar abundances we use are measured relative to solar . when we simply assume solar abundances from @xcite , for our sample of stars we find that the high - fip element abundances are inconsistent with each other . this inconsistency is decreased significantly for all our stars if we instead assume a photospheric ne abundance consistent with stellar coronal measurements from @xcite , and solar abundances from @xcite for other elements . this is an argument in favor of these latter photospheric abundances , unless there are actual coronal fractionation processes taking place among the high - fip elements that we are erroneously removing . 5 . among low - fip ions , the coronal mg / si ratio is consistently higher than the photospheric ratio for all of our g8-k5 stars , presumably an effect of actual coronal fractionation . an enhancement of mg relative to si in the corona is predicted by many of the models of @xcite , which are designed to explain solar / stellar coronal abundance anomalies . however , although @xcite also finds high mg / si ratios for early g dwarfs , coronal mg / si ratios appear to be _ lower _ than photospheric for m dwarfs @xcite , suggesting a spectral type dependence for this coronal abundance ratio . 6 . with regards to the fip effect , our sample of stars exhibit a range of fip bias , with 70 oph a and @xmath0 boo a having the strongest solar - like fip effect , @xmath3 eri showing little or no fip effect , and 70 oph b possessing a slight inverse fip effect . expanding our sample of stars even further to include early g stars from @xcite and m dwarfs from @xcite , we find that the fip bias correlates beautifully with spectral type , but only if extremely active stars with @xmath2 are excluded . . the tight correlation of fip bias with spectral type suggests to us that the nature and causes of the fip effect can ultimately be traced directly to fundamental surface properties . among normal main sequence stars , these change in a very predictable manner with spectral type , explaining the tight correlation in figure 9 . no such correlation exists for extremely active stars , because many are evolved stars with disparate photospheric characteristics , and in other cases the stellar atmospheres will be affected by rapid rotation or the effects of close binarity . support for this work was provided by the national aeronautics and space administration through chandra award number go8 - 9003z issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of the national aeronautics and space administration under contract nas8 - 03060 . allende prieto , c. , barklem , p. s. , lambert , d. l. , & cunha , k. 2004 , a&a , 420 , 183 antia , h. m. , & basu , s. 2005 , apj , 620 , l129 arnaud , m. , & rothenflug , r. 1985 , a&as , 60 , 425 asplund , m. , grevesse , n. , sauval , a. j. , & scott , p. 2009 , ara&a , 47 , 481 . audard , m. , behar , e. , gdel , m. , raassen , a. j. j. , porquet , d. , mewe , r. , foley , c. r. , & bromage , g. e. 2001 , a&a , 365 , l329 audard , m. , gdel , m. , sres , a. , raassen , a. j. j. , & mewe , r. 2003 , a&a , 398 , 1137 bahcall , j. n. , basu , s. , & serenelli , a. m. 2005 , apj , 631 , 1281 ball , b. , drake , j. j. , lin , l. , kashyap , v. , laming , j. m. , & garca - alvarez , d. 2005 , apj , 634 , 1336 barnes , t. g. , evans , d. s. , & moffett , t. j. 1978 , mnras , 183 , 285 beuermann , k. , burwitz , v. , & rauch , t. 2006 , a&a , 458 , 541 brinkman , a. c. , et al . 2001 , a&a , 365 , l324 brown , b. p. , browning , m. k. , brun , a. s. , miesch , m. s. , & toomre , j. 2008 , apj , 689 , 1354 dere , k. p. , landi , e. , mason , h. e. , monsignori fossi , b. c. , & young , p. r. 1997 , a&as , 125 , 149 dere , k. p. , landi , e. , young , p. r. , del zanna , g. , landini , m. , & mason , h. e. 2009 , a&a , 498 , 915 dorren , j. d. , gdel , m. , & guinan , e. f. 1995 , apj , 448 , 431 drake , j. j. , & kashyap , v. 2001 , apj , 547 , 428 drake , j. j. , laming , j. m. , & widing , k. g. 1997 , apj , 478 , 403 drake , j. j. , peres , g. , orlando , s. , laming , j. m. , & maggio , a. 2000 , apj , 545 , 1074 drake , j. j. , & testa , p. 2005 , nature , 436 , 525 feldman , u. , & laming , j. m. 2000 , phys . , 61 , 222 grevesse , n. , & sauval , a. j. 1998 , space sci . rev . , 85 , 161 gdel , m. 2004 , astron . astrophys . , 12 , 71 gdel , m. , et al . 2001 , a&a , 365 , l336 gdel , m. , guinan , e. f. , & skinner , s. l. 1997 , apj , 483 , 947 hartmann , l. , schmidtke , p. c. , davis , r. , dupree , a. k. , raymond , j. , & wing , r. f. 1979 , apj , 233 , l69 huenemoerder , d. p. , canizares , c. r. , drake , j. j. , & sanz - forcada , j. 2003 , apj , 595 , 1131 huenemoerder , d. p. , canizares , c. r. , & schulz , n. s. 2001 , apj , 559 , 1135 kashyap , v. & drake , j. j. 1998 , apj , 503 , 450 kashyap , v. , & drake , j. j. 2000 , bull . india , 28 , 475 laming , j. m. 2004 , apj , 614 , 1063 laming , j. m. 2009 , apj , 695 , 954 laming , j. m. , & drake , j. j. 1999 , apj , 516 , 324 laming , j. m. , drake , j. j. , & widing , k. g. 1996 , apj , 462 , 948 liefke , c. , ness , j. -u . , schmitt , j. h. m. m. , & maggio , a. 2008 , a&a , 491 , 859 linsky , j. l. , & haisch , b. m. 1979 , apj , 229 , l27 maggio , a. , sciortino , s. , vaiana , g. s. , majer , p. , bookbinder , j. , golub , l. , harnden , f. r. , jr . , & rosner , r. 1987 , apj , 315 , 687 mazzotta , p. , mazzitelli , g. , colafrancesco , s. , & vittorio , n. 1998 , a&as , 133 , 403 mckenzie , d. l. 1987 , apj , 322 , 512 ness , j. -u . , schmitt , j. h. m. m. , burwitz , v. , mewe , r. , raassen , a. j. j. , van der meer , r. l. j. , predehl , p. , & brinkman , a. c. 2002 , a&a , 394 , 911 noyes , r. w. , hartmann , l. w. , baliunas , s. l. , duncan , d. k. , & vaughan , a. h. 1984 , apj , 279 , 763 pandey , j. c. , & srivastava , a. k. 2009 , apj , 697 , l153 perryman , m. a. c. , et al . 1997 , a&a , 323 , l49 petit , p. , et al . 2005 , mnras , 361 , 837 petit , p. , et al . 2008 , mnras , 388 , 80 porquet , d. , mewe , r. , dubau , j. , raassen , a. j. j. , & kaastra , j. s. 2001 , a&a , 376 , 1113 sanz - forcada , j. , affer , l. , & micela , g. 2009 , a&a , 505 , 299 sanz - forcada , j. , maggio , a. , & micela , g. 2003 , a&a , 408 , 1087 schmelz , j. t. , nasraoui , k. , roames , j. k. , lippner , l. a. , & garst , j. w. 2005 , apj , 634 , l197 schmitt , j. h. m. m. 1997 , a&a , 318 , 215 schmitt , j. h. m. m. , & liefke , c. 2004 , a&a , 417 , 651 sderhjelm , s. 1999 , a&a , 341 , 121 strassmeier , k. g. 2002 , astronomische nachrichten , 323 , 309 telleschi , a. , gdel , m. , briggs , k. , audard , m. , ness , j. -u . , & skinner , s. l. 2005 , apj , 622 , 653 toner , c. g. , & gray , d. f. 1988 , apj , 334 , 1008 widing , k. g. , & feldman , u. 2001 , apj , 555 , 426 wood , b. e. 2004 , living rev . solar phys . , 1 , 2 , url : http://www.livingreviews.org/lrsp-2004-2 wood , b. e. , & linsky , j. l. 2006 , apj , 643 , 444 ( wl06 ) wood , b. e. , mller , h. -r . , zank , g. p. , linsky , j. l. , & redfield , s. 2005a , apj , 628 , l143 wood , b. e. , redfield , s. , linsky , j. l. , mller , h. -r . , & zank , g. p. 2005b , apjs , 159 , 118 young , p. r. 2005 , a&a , 444 , l45 lccc other name & hd 131156 & hd 131156b & + spect . type & g8 v & k4 v & + dist . ( pc ) & 6.70 & 6.70 & 1 + radius ( r@xmath8)&0.83 & 0.61 & 2 + p@xmath56 ( days ) & 6.2 & 11.5 & 3 + @xmath57 & 28.86 & 27.97 & 4 + @xmath57 & 28.91 & 28.17 & + @xmath58 ( @xmath59 ) & 5 & 5 & 5 + @xmath60 & 17.92 & 17.92 & 6 + @xmath61}$ ] & @xmath62 & ... & 7 + @xmath63}$ ] & @xmath64 & ... & 7 + @xmath65}$ ] & @xmath66 & ... & 7 + @xmath67}$ ] & @xmath62 & ... & 7 + @xmath68}$ ] & @xmath66 & ... & 7 + lcccc si xiii & 6.648 & 6.99 & @xmath69 & @xmath70 + si xiii & 6.688 & 6.99 & & + si xiii & 6.740 & 6.99 & & + mg xii & 8.419 & 7.11 & @xmath71 & @xmath72 + mg xii & 8.425 & 7.11 & & + mg xi & 9.169 & 6.80 & @xmath73 & @xmath74 + mg xi & 9.231 & 6.80 & & + mg xi & 9.314 & 6.79 & & + ne x & 12.132 & 6.87 & @xmath75 & @xmath76 + ne x & 12.138 & 6.87 & & + fe xvii & 12.264 & 6.62 & @xmath77 & @xmath78 + fe xxi & 12.285 & 6.98 & & + ne ix & 13.447 & 6.58 & @xmath79 & @xmath80 + ne ix & 13.553 & 6.58 & & + ne ix & 13.699 & 6.58 & @xmath81 & @xmath82 + fe xviii & 14.203 & 6.74 & @xmath83 & @xmath84 + fe xviii & 14.208 & 6.74 & & + fe xvii & 15.015 & 6.59 & @xmath85 & @xmath86 + fe xvii & 15.262 & 6.59 & @xmath87 & @xmath88 + o viii & 15.176 & 6.65 & & + fe xix & 15.198 & 6.83 & & + o viii & 16.006 & 6.63 & @xmath89 & @xmath90 + fe xviii & 16.005 & 6.73 & & + fe xviii & 16.072 & 6.73 & & + fe xvii & 16.778 & 6.58 & @xmath91 & @xmath92 + fe xvii & 17.053 & 6.58 & @xmath93 & @xmath94 + fe xvii & 17.098 & 6.58 & & + o vii & 18.627 & 6.34 & @xmath95 & @xmath96 + o viii & 18.967 & 6.59 & @xmath97 & @xmath98 + o viii & 18.973 & 6.59 & & + o vii & 21.602 & 6.32 & @xmath99 & @xmath100 + o vii & 21.807 & 6.32 & @xmath101 & @xmath102 + o vii & 22.101 & 6.31 & @xmath103 & @xmath104 + n vii & 24.779 & 6.43 & @xmath105 & @xmath106 + n vii & 24.785 & 6.43 & & + c vi & 33.734 & 6.24 & @xmath107 & @xmath108 + c vi & 33.740 & 6.24 & & + s xiii & 35.667 & 6.43 & @xmath109 & @xmath110 + si xi & 43.763 & 6.25 & @xmath111 & @xmath112 + si xii & 44.019 & 6.44 & @xmath113 & @xmath114 + si xii & 44.165 & 6.44 & @xmath115 & @xmath116 + si xii & 45.521 & 6.44 & @xmath117 & @xmath118 + si xii & 45.691 & 6.44 & @xmath119 & @xmath120 + fe xvi & 46.661 & 6.43 & @xmath121 & @xmath122 + fe xvi & 46.718 & 6.43 & & + si xi & 49.222 & 6.24 & @xmath123 & @xmath124 + fe xvi & 50.361 & 6.43 & @xmath125 & @xmath126 + fe xvi & 50.565 & 6.43 & @xmath127 & @xmath128 + si x & 50.524 & 6.15 & & + si x & 50.691 & 6.15 & @xmath129 & @xmath130 + si xi & 52.298 & 6.24 & @xmath131 & @xmath132 + fe xv & 52.911 & 6.32 & @xmath133 & @xmath134 + fe xvi & 54.127 & 6.43 & @xmath135 & @xmath136 + fe xvi & 54.710 & 6.43 & @xmath137 & @xmath138 + mg x & 57.876 & 6.22 & @xmath139 & @xmath140 + mg x & 57.920 & 6.22 & & + fe xv & 59.405 & 6.32 & @xmath141 & @xmath142 + fe xvi & 62.872 & 6.42 & @xmath143 & @xmath144 + mg x & 63.295 & 6.21 & @xmath145 & @xmath146 + fe xvi & 63.711 & 6.42 & @xmath147 & @xmath148 + fe xvi & 66.249 & 6.42 & @xmath149 & @xmath128 + fe xvi & 66.357 & 6.42 & & + fe xv & 69.682 & 6.32 & @xmath150 & @xmath151 + fe xv & 69.941 & 6.32 & @xmath152 & @xmath140 + fe xv & 69.987 & 6.32 & & + fe xv & 70.054 & 6.32 & & + mg ix & 72.312 & 6.02 & @xmath153 & @xmath154 + fe xv & 73.472 & 6.32 & @xmath155 & @xmath156 + fe xvi & 76.497 & 6.42 & @xmath157 & @xmath158 + ne viii & 88.082 & 5.96 & @xmath159 & @xmath160 + ne viii & 88.120 & 5.96 & & + fe xviii & 93.923 & 6.68 & @xmath161 & @xmath162 + ne viii & 98.116 & 5.94 & @xmath163 & @xmath164 + ne viii & 98.260 & 5.94 & @xmath165 & @xmath166 + fe xix & 101.550 & 6.80 & @xmath167 & @xmath168 + fe xviii & 103.937 & 6.68 & @xmath169 & @xmath170 + fe xix & 108.355 & 6.79 & @xmath171 & @xmath164 + fe xxii & 117.180 & 7.03 & @xmath172 & @xmath173 + fe xx & 121.845 & 6.88 & @xmath174 & @xmath175 + fe xxi & 128.752 & 6.95 & @xmath176 & @xmath177 + fe xx & 132.840 & 6.88 & @xmath178 & @xmath179 + fe xxiii & 132.906 & 7.12 & & + fe ix & 171.073 & 5.95 & @xmath180 & @xmath181 + cccc @xmath182 & @xmath183 & @xmath184 & 9.4 + @xmath185 & 0.7 & 0.8 & 1 + @xmath186}$ ] & @xmath187 & @xmath188 & @xmath189 + @xmath190}$ ] & @xmath191 & @xmath192 & 1.02 + @xmath193}$ ] & @xmath194 & ... & 0.42 + @xmath195}$ ] & @xmath196 & @xmath197 & 1.33 + @xmath198}$ ] & @xmath199 & @xmath200 & 0.58 + @xmath201}$ ] & @xmath202 & @xmath203 & 0.08 + @xmath204}$ ] & @xmath205 & @xmath206 & 0.05 + @xmath207}$ ] & @xmath208 & ... & @xmath209 +

in this note we develop tools to study the cauchy problem for the system of thermo - elasticity in higher dimensions . the theory is developed for general homogeneous anisotropic media under non - degeneracy conditions . for degenerate cases a method of treatment is sketched and for the cases of cubic media and hexagonal media detailed studies are provided .

1109.3853

while isotropic thermo - elasticity is a well - known and well - established subject ( see , e.g. , the book of jiang song - racke @xcite and references therein ) only very few results are available for the case of anisotropic media . among them are the theses of borkenstein @xcite for cubic media and doll @xcite for the case of rhombic media together with the authors treatments @xcite , @xcite , all in two space dimensions . in this paper the system of anisotropic thermo - elasticity in three ( and more ) dimensions , i.e. , [ eq : cp ] @xmath0 for the elastic displacement @xmath1 and temperature difference @xmath2 to the equilibrium state , will be considered . the system couples the hyperbolic elasticity equation with the parabolic heat equation . the operator @xmath3 describes the elastic properties of the underlying medium , while @xmath4 denotes its thermal conductibility . the constant @xmath5 describes the thermo - elastic coupling . basic assumptions of our theory are @xmath6 , @xmath7 together with * @xmath8 , @xmath9 , is a 2-homogeneous matrix - valued symbol ; * @xmath10 is a real - analytic function of @xmath11 , @xmath12 ; * @xmath13 is self - adjoint and positive . in general we can _ not _ assume that @xmath14 is non - degenerate in the sense that @xmath15 for all @xmath11 ( as done for the two - dimensional case in @xcite ) . all basic examples show degeneracies in dimensions @xmath16 . _ isotropic media _ @xmath17 with lam constants @xmath18 and @xmath19 . the matrix @xmath14 is positive as long as @xmath20 and @xmath21 . the eigenvectors of @xmath14 are multiples of @xmath22 and @xmath23 and thus invariant under rotations of frequency space . [ expl2]_cubic media _ @xmath24 described by parameters @xmath18 , @xmath19 and @xmath25 . later we will describe the assumptions made on these parameters and the resulting spectral properties of the matrix function @xmath14 more precise . in the case of three space dimensions , the matrix @xmath14 is positive if and only if @xmath20 , @xmath26 together with @xmath27 . in three space dimensions this will be one of our main examples . we can replace the constant @xmath25 on the diagonal by @xmath28 in . this yields so - called _ rhombic media_. the behaviour of rhombic media is close to that of cubic media if the parameters are of similar size , in general there will appear exceptional situations . see , e.g. , @xcite or @xcite for a discussion of this effect in two space dimensions . _ hexagonal media _ are another particularly interesting case for three space dimensions . since we want to come back to them later on we introduce the corresponding operator . it is given by @xmath29 where @xmath30 contains the @xmath31 structure constants @xmath32 , @xmath33 , @xmath34 , @xmath35 and @xmath19 and @xmath36 is of a particular form , @xmath37 even the first ( non - trivial anisotropic ) example , the case of cubic media in three space dimensions , has degenerate directions in which @xmath14 has double eigenvalues . later on we will analyse this example in detail . we call a direction @xmath11 ( elastically ) _ non - degenerate _ if @xmath38 holds true for this direction @xmath22 . the set of non - degenerate directions is an open subset of @xmath39 . for non - degenerate directions the treatment of @xcite transfers almost immediately and gives a representation of solutions . we will sketch the results in section [ sec2 ] . in section [ sec3 ] we consider special degenerate directions and discuss the examples of cubic and hexagonal media . dispersive estimates for solutions are given in section [ sec4 ] . in the neighbourhood of degenerate directions they are essentially based on estimates developed by liess @xcite , @xcite for the treatment of anisotropic acoustic equations . for the following we consider a simply connected open subset @xmath40 of @xmath39 , where the symbol @xmath14 has @xmath41 distinct ( and real ) eigenvalues . we denote these eigenvalues in ascending order as @xmath42 by analytic perturbation theory , see @xcite , we know that these eigenvalues are real - analytic and that we find corresponding normalised eigenvectors @xmath43 depending analytically on @xmath44 . collecting them in the unitary matrix @xmath45 we can diagonalise the matrix @xmath14 @xmath46 in our treatment we will not make use of analyticity directly , instead our use of perturbation theory will be based on @xcite und @xcite and uses only smooth dependence . this will be of interest for generalisations later on . therefore , whenever we use analyticity , we will explicitly state that . we use @xmath47 to reduce the thermo - elastic system to a system of first order . for this we denote by @xmath48 and @xmath49 the partial fourier transforms of @xmath50 and @xmath51 with respect to the spatial variables and consider @xmath52 as usual @xmath53 and @xmath9 . then @xmath54 satisfies a first order system of ordinary differential equations , which has an apparantly simple structure . straightforward calculation shows that @xmath55 holds true with coefficient matrix @xmath56 where @xmath57 and @xmath58 following the conventions of @xcite we denote these functions @xmath59 as the _ coupling functions _ of the thermo - elastic system associated to the elastic operator @xmath3 . they play a prominent rle for the description of the time - asymptotic behaviour of solutions . this reflects the fact that they couple the homogeneous first order entries in @xmath60 with the second order lower right corner entry . note , that @xmath61 zeros of the coupling functions are of particular importance . following ( * ? ? ? 1 ) we define : a non - degenerate direction @xmath11 is called * _ hyperbolic _ if one of the coupling functions vanishes ; more precisely , it is called _ hyperbolic with respect to the eigenvalue @xmath62 _ if @xmath63 ; * _ parabolic _ if all coupling functions are non - zero . in the anisotropic case the set of hyperbolic directions is ( generically and the system is therefore decoupled . this case is reduced to the study of the lower dimensional blocks , one is a hyperbolic system the other one a thermo - elastic system of lower dimension . this is , e.g. , the case for hexagonal media , see section [ sec3.3 ] . ] ) a lower dimensional subset of @xmath39 . in order to decide whether a direction is hyperbolic or parabolic we can employ the following proposition . we denote for a matrix @xmath64 and a vector @xmath22 by @xmath65 the corresponding cyclic subspace , i.e. the span of the trajectory of @xmath22 under the action of the matrix @xmath64 . [ prop:2.1 ] the following statements are equivalent : 1 . the cyclic subspace of @xmath22 has dimension @xmath66 , i.e. , @xmath67 . exactly @xmath68 of the coupling functions vanish in @xmath22 . hence , a non - degenerate direction @xmath11 is parabolic if and only if @xmath69 and therefore @xmath70 if we represent @xmath22 in the eigenbasis of @xmath14 we obtain @xmath71 and therefore @xmath72 if @xmath68 of the coupling functions vanish , then @xmath73 must be in the span of the @xmath74 with @xmath75 and thus the cyclic subspace is at most of dimension @xmath66 . on the other hand , the first @xmath66 vectors in the trajectory are linearly independent since the corresponding matrix in the basis representation with respect to @xmath76 is just the van der monde matrix associated to the eigenvalues of @xmath14 for non - vanishing coupling functions and therefore regular . at first we collect some of the spectral properties of the matrix @xmath60 which are directly related to the characteristic polynomial of @xmath60 . [ prop22 ] the following identies hold true : @xmath77 furthermore , the matrix @xmath60 has a purely real eigenvalue for @xmath78 if and only if the direction @xmath9 is hyperbolic . if it is @xmath79-hyperbolic , then @xmath80 . the proof of the last fact is fairly straightforward and consists of separating real and imaginary parts of the characteristic polynomial . note that for all parabolic directions we can divide the characteristic polynomial by @xmath81 to obtain @xmath82 this formulation allows to consider the neighbourhoods of hyperbolic directions . assume for this that the set of hyperbolic directions with respect to @xmath62 @xmath83 is a regular submanifold of @xmath40 . if we consider the corresponding _ hyperbolic _ eigenvalues @xmath84 of @xmath60 in a neighbourhood of @xmath85 , i.e. the eigenvalues which satisfy @xmath86 for fixed @xmath87 , equation gives a precise description of the behaviour of the imaginary part of these eigenvalues . the proof is a straightforward generalisation from ( * ? ? ? [ prop:2.3 ] the non - tangential limit @xmath88 exists and is non - zero for all @xmath78 . furthermore , @xmath89 we decompose @xmath60 into homogeneous components @xmath91 of degree @xmath92 and @xmath93 , respectively . for sufficiently small @xmath87 we expect the eigenvalues of @xmath60 to be close to the eigenvalues of @xmath94 . for parabolic directions the ( non - zero ) eigenvalues of @xmath95 can be determined from the equation @xmath96 which follows directly from with @xmath97 . it can be solved ( e.g. graphically , see figure [ fig1 ] for @xmath98 ) to obtain the distinct eigenvalues @xmath99 , @xmath100 , , @xmath101 ordered as @xmath102 for parabolic directions.,width=302 ] for hyperbolic directions a similar result holds true . in the case of hyperbolic directions @xmath62 eigenvalues move to @xmath103 . according to the choice of the coupling constant @xmath5 different cases occur : 1 . if @xmath104 is large then @xmath105 , the other inequalities are unchanged ; 2 . if @xmath104 is small then @xmath106 and the other inequalities remain true . the critical threshold between these two cases is @xmath107 where @xmath95 has the _ double _ eigenvalue @xmath108 . following the conventions from @xcite we define : we denote a hyperbolic direction w.r.to @xmath62 as @xmath5-degenerate if holds true . for the following treatment we exclude @xmath5-degenerate hyperbolic directions and assume instead that for all hyperbolic directions in @xmath40 condition is not satisfied for the corresponding index @xmath79 . then the following statement is apparent . [ prop:2.4 ] let @xmath22 be not @xmath5-degenerate . then the matrix @xmath95 has @xmath109 distinct real eigenvalues @xmath110 for all @xmath44 . proposition [ prop:2.4 ] allows to apply the standard diagonalisation scheme ( see ( * ? ? ? 2.1 ) ) to @xmath91 as @xmath111 . hence , eigenvalues , eigenprojections and all their derivatives have full asymptotic expansions as @xmath111 . the proof is almost identical to that from ( * ? ? ? 2.5 ) and is omitted . [ prop:2.5 ] for all not @xmath5-degenerate directions @xmath112 the eigenvalues and eigenprojections of @xmath60 have full asymptotic expansions as @xmath111 . the main terms are given by @xmath113 with [ eq : b - def ] @xmath114 furthermore , @xmath115 if and only if @xmath22 is hyperbolic with respect to the eigenvalue @xmath62 . note , that @xmath116 implies @xmath117 recall that by proposition [ prop22 ] eigenvalues of @xmath60 can only be real along hyperbolic directions ( and then they are exactly the trivial real eigenvalues ) . in combination with the fact that eigenvalues of @xmath60 are continuous in @xmath118 we obtain : for all parabolic directions @xmath112 we have @xmath119 . the same is true as long as @xmath22 is _ not _ hyperbolic w.r.to @xmath62 . in this case the two - step procedure developed in ( * ? ? ? 2.2 ) , ( * ? ? 2.6 ) applies in analogy . essential assumption is the non - degeneracy of @xmath14 . we omit the proof and cite the corresponding result only . [ prop:2.6 ] for all non - degenerate directions the eigenvalues and eigenprojections of the matrix @xmath60 have full asymptotic expansions as @xmath120 . the first terms are given by @xmath121 despite the fact that we used the same notation for the eigenvalues as @xmath111 and @xmath120 , we do not claim that they are indeed the same functions of @xmath118 . this is only true for hyperbolic eigenvalues near hyperbolic directions , in general there might be multiplicities in between and there might be no consistent notation for these functions . [ cor:2.8 ] for all parabolic directions @xmath9 the eigenvalues of @xmath60 satisfy @xmath122 for @xmath123 . the same is true for parabolic eigenvalues in hyperbolic directions . in particular , we see by the asymptotic expansions that the eigenvalues of @xmath60 are simple for large and also for small values of @xmath87 . furthermore , we see that the hyperbolic eigenvalues are always separeted ( i.e. if multiplicities occur in hyperbolic directions , they involve only parabolic eigenvalues ) . the asymptotic expansions of propositions [ prop:2.5 ] and [ prop:2.6 ] allow to draw conclusions for the behaviour of the imaginary part . we collect them for later use . the first result is apparent . [ prop:2.9 ] on any compact set of parabolic directions we have the uniform estimates @xmath124 for all eigenvalues of @xmath60 and arbitrary @xmath125 . the next statement is concerned with a tubular neighbourhood of a compact subset of a regular submanifold @xmath85 of hyperbolic eigenvalues w.r.to @xmath62 . it is only of interest how the corresponding hyperbolic eigenvalues @xmath84 behave , the others still satisfy proposition [ prop:2.9 ] . uniformly on any tubular neighbourhood of a compact subset of @xmath85 of non-@xmath5-degenerate directions the corresponding hyperbolic eigenvalues @xmath84 satisfy the estimates @xmath126 by proposition [ prop:2.3 ] we know that @xmath127 for some function @xmath128 . our aim is to estimate @xmath128 . the left hand of this formula has a full asymptotic expansion as @xmath90 and @xmath120 . therefore , also the right hand side has one and it follows that @xmath129 thus , the desired estimate follows by a compactness argument as soon as we have a uniform lower / upper bound for @xmath130 . the representation of @xmath131 in proposition [ prop:2.5 ] in combination with implies @xmath132 which is clearly bounded and ( uniformly ) positive on any compact subset of @xmath85 ( where we have to use that @xmath133 is not @xmath5-degenerate ) . we will draw several conclusions what we have obtained so far and what we still have to consider in the remaining part of this treatise . if we consider the special case of cubic media in three space dimensions degenerate directions are given by @xmath134 with @xmath135 ( eight directions , corresponding to the corners of a cube ) or @xmath136 for some @xmath137 ( six directions , corresponding to its faces ) . this can be calculated directly , corresponding eigenspaces are @xmath138 and @xmath139 , or concluded by the cubic symmetry is invariant under the hexaeder group , i.e. the symmetry group of a cube . thus , eigenspaces must be transferred in an appropriate way , which implies that symmetries of order 3 or 4 can only be realised by higher dimensional eigenspaces . ] of @xmath140 in this particular case . see figure [ fig2 ] . to obtain the hyperbolic directions we apply proposition [ prop:2.1 ] and look for the action of @xmath22 under @xmath14 . we obtain that 1 . a direction @xmath22 is hyperbolic if and only if @xmath141 thus the set of hyperbolic directions is the union of nine great circles on @xmath142 ; 2 . @xmath143 for all 26 intersection points of these great circles , 14 of them are excluded as being degenerate . except for these 14 points on @xmath142 we obtained an almost complete description of the spectrum of @xmath60 . we know full asymptotic expansions of eigenvalues for small and large frequencies @xmath87 , estimates for the imaginary part of them and similar statements for eigenprojections . this information allows to draw conclusions on the large time behaviour of solutions , e.g. energy and dispersive estimates . this can be done similar to the treatment of @xcite , see section [ sec4 ] . the remaining degenerate directions appear in two types , which can be interchanged by the action of the symmetry group . the study of these degenerate directions is what is left open so far and will be the main point of section [ sec3 ] . . the midpoints of edges c are non - degenerate , but hyperbolic with respect to two different eigenvalues.,width=226 ] if we consider the special case of isotropic media , @xmath144 , we see that @xmath145 and corresponding eigenspaces are @xmath146 ( corresponding to @xmath147 ) and @xmath23 ( corresponding to @xmath19 ) . all directions are ( elastically ) degenerate . however , we still find locally smooth systems of eigenvectors . all directions are hyperbolic and the hyperbolic eigenvalue @xmath19 has multiplicity @xmath148 . therefore the system @xmath149 decouples into a diagonal part of size @xmath150 and a full @xmath151 block and is given after a rearrangement of the entries as @xmath152 this block structure corresponds to the helmholtz decomposition of vector fields applied to the elastic displacement . if @xmath153 the lower block cancels and we obtain wave equations with speed @xmath154 for the components of @xmath50 . otherwise , if we cancel the upper block we obtain the @xmath151 system corresponding to one - dimensional thermo - elasticity with its well - known properties . for completeness we mention some results on the one - dimensional system @xmath155 we assume @xmath156 . following our strategy we can rewrite this problem as first order system . the corresponding symbol @xmath60 is given by @xmath157 its eigenvalues satisfy asymptotic expansions for @xmath111 and @xmath158 . propositions [ prop:2.5 ] and [ prop:2.6 ] apply with @xmath159 and @xmath160 therefore , by proposition [ prop:2.5 ] [ eq:2.43 ] @xmath161 as @xmath111 and by proposition [ prop:2.6 ] @xmath162 as @xmath163 . the essential information for large time estimates is given by the behaviour of the imaginary part . it follows that @xmath164 for @xmath165 for certain constants and @xmath166 for hexagonal media in three space dimensions the situation is ( surprisingly ) simpler than for cubic media . the elastic operator defined by is invariant under rotation around the @xmath167-axis ( taking into account a corresponding rotation of the reference frame for vectors ) and therefore it suffices to understand its cross sections in the @xmath168@xmath169 plane . we will sketch some of the properties of the corresponding symbol @xmath14 . following proposition [ prop:2.1 ] we obtain 1 . that @xmath170 such that _ all _ directions @xmath171 are hyperbolic . the corresponding eigenspace is ( generically ) given by multiples of @xmath172 such that the hyperbolic eigenvalue is @xmath173 2 . it remains to look for directions with two hyperbolic eigenvalues . they satisfy @xmath174 . this is true , if @xmath175 or if @xmath176 or if @xmath177 provided the latter expression is non - negative . except in the limiting case @xmath178 , the coupling functions vanish to first order along the corresponding circle . if @xmath179 all directions are hyperbolic with two hyperbolic eigenvalues and if @xmath180 coupling functions vanish to third order . the matrices @xmath14 are invariant under rotation . introducing spherical coordinates on @xmath181 @xmath182 and using a corresponding ( moving ) basis for vectors given by @xmath183 ( sign chosen to make them smoothly dependent on @xmath184 ) decomposes @xmath14 into ( 1,2)-block - diagonal structure ( independent of the co - ordinate @xmath185 ) . the scalar block corresponds to the eigenvalue , while the @xmath186 block has trace @xmath187 and determinant @xmath188 . if @xmath189 , the @xmath186 block has distinct eigenvalues for all @xmath190 and therefore the only degenerate directions are directions where this block has as one of its eigenvalues . this happens if and only if the right hand side of is non - negative and on the circle defined by that equation . thus , the previously developed theory is applicable for all directions except the degenerate ones @xmath176 or . the always existent hyperbolic eigenvalue leads to a decoupling of the thermo - elastic system into two scalar blocks and a ( at least formally ) 2d thermo - elastic system . due to rotational invariance , it suffices to treat the cut @xmath191 for handling of degenerate directions . this will be sketched later . we want to study neighbourhoods of degenerate directions for some particular cases . to study degenerate directions in full generality is beyond the scope of this paper . we relate our approach to the type of singularity of the corresponding _ fresnel surface _ @xmath192 this surface is in general @xmath41-sheeted and for all non - degenerate directions these sheets are given by @xmath193 while in degenerate points the surface is self - intersecting . for the importance of these surfaces in elasticity theory and some interesting properties of them we refer to duff @xcite or the investigations from musgrave @xcite , @xcite and miller - musgrave @xcite . , @xmath194 and @xmath195 for the picture on the left ( cubic ) and @xmath196 , @xmath197 , @xmath198 , @xmath199 and @xmath200 for the picture on the right ( hexagonal).,title="fig:",width=264 ] , @xmath194 and @xmath195 for the picture on the left ( cubic ) and @xmath196 , @xmath197 , @xmath198 , @xmath199 and @xmath200 for the picture on the right ( hexagonal).,title="fig:",width=264 ] we remark only one of the general properties of @xmath201 here . if @xmath140 is polynomial in @xmath118 then the surface @xmath201 is algebraic of degree @xmath202 and therefore any straight line intersecting @xmath201 has at most @xmath202 intersection points with @xmath201 . in particular , if the inner sheet @xmath203 does not touch any of the the outer sheets , it has to be strictly convex . if we investigate isolated degenerate directions or regular manifolds of degenerate directions of codimension greater than one we are faced with two major obstacles . generically , eigenvectors of @xmath14 can not be chosen continuously in a neighbourhood of the degenerate direction and therefore a reformulation as system of first order as in is problematic . this problem is related to higher - dimensional perturbation theory of matrices . it is well - known that in the one - dimensional situation eigenspaces are continuous ( see , e.g. , the book of kato , @xcite ) and it can be resolved by introducing polar co - ordinates / normal co - ordinates around the degenerate directions and a system related to can be formulated on a corresponding blown - up space ( see , e.g. , below ) . second obstacle are the multiplicities itself . eigenvalues and eigenvectors of the constructed system of first order do not possess asymptotic expansions in powers of @xmath87 as @xmath87 tends to @xmath99 or @xmath204 . however , especially in the three - dimensional setting we can write full asymptotic expansions in the distance to the degeneracy uniform in all remaining co - ordinates . we will discuss the application of this general strategy in detail for conic singularities of the fresnel surface appearing for the case of cubic media and give the corresponding results for uniplanar singularities afterwards . finally we will consider hexagonal media and show that they are much simpler in their analytical structure . , on the right for @xmath205 . the parameters are chosen as @xmath206 , @xmath207 and @xmath208.,title="fig:",width=188 ] , on the right for @xmath205 . the parameters are chosen as @xmath206 , @xmath207 and @xmath208.,title="fig:",width=267 ] the fresnel surface for cubic media has eight conic singularities which are related by the symmetries of the underlying medium . it suffices to consider one of them and we choose @xmath209 . near this direction we introduce polar co - ordinates @xmath210 on the sphere @xmath142 by @xmath211 they allow to blow up the singularity by looking at @xmath212 instead of @xmath213 as local model of @xmath142 . in order to simplify notation , we apply a diagonaliser @xmath214 of @xmath215 to our matrices . for this we choose the unitary matrix @xmath216 ( corresponding to the vectors chosen already in ) . the matrix @xmath217 has a full asymptotic expansion as @xmath218 and can be written as @xmath219 with matrices @xmath220 now we can apply the block - diagonalisation procedure ( again following ( * ? ? ? * sec . 2.2 ) ) to obtain the behaviour of eigenvalues and eigenprojections of @xmath221 as @xmath218 for all @xmath185 . we restrict consideration to the case where @xmath222 , such that @xmath223 has two different eigenvalues . [ prop:3.1 ] the eigenvalues @xmath224 and the corresponding eigenprojections of @xmath225 have uniformly in @xmath185 full asymptotic expansions as @xmath218 . the main terms are given by @xmath226 the exceptional case when @xmath227 corresponds to isotropic media and is therefore excluded . in all other cases the two sheets @xmath228 and @xmath229 form a double cone on the fresnel surface @xmath230 . hence , the statement explains the notion of conical singularity . note , that the linear terms are independent of @xmath185 and therefore the cone is approximately a spherical cone near the conic point . we will only shortly review the main steps . first we ( 1,2)-block - diagonalise @xmath221 ( modulo @xmath231 for any @xmath232 we like ) . the diagonaliser we are going to construct has the form @xmath233 and as in ( * ? ? ? 2.2 ) its terms are given by recursion formulae . for @xmath234 we divide the off-(block-)diagonal terms of @xmath235 by the difference of the corresponding diagonal entries of @xmath223 . this gives as first term @xmath236 and allows to cancel the off-(block-)diagonal entries of @xmath235 . we skip the further construction and move to the next step . since the lower @xmath186 block of @xmath237 has distinct eigenvalues ( namely @xmath238 ) we can now perform a diagonalisation scheme in the subspace corresponding to this block ( modulo @xmath231 ) . again we restrict ourselves to the main terms . a unitary diagonaliser of the @xmath186-block can be chosen as the unitary matrix @xmath239 after transforming with that matrix we apply the recursive scheme to diagonalise further . note that after applying @xmath240 the matrix is diagonal modulo @xmath241 and therefore , @xmath242 determines the main terms of a diagonaliser of the matrix @xmath225 and we can deduce the statements about the eigenvalue asymptotics . let @xmath243 be the diagonaliser of @xmath225 constructed in proposition [ prop:3.1 ] . then we consider @xmath244 with @xmath245 and @xmath246 the diagonal matrix containing the square roots @xmath247 of the eigenvalues of @xmath225 . the vector @xmath54 satisfies the first order system @xmath248 with @xmath249 given by @xmath250 and @xmath251 . the coupling functions @xmath252 are the components of the vector @xmath253 . from proposition [ prop:3.1 ] we know that they have asymptotic expansions as @xmath218 . _ since @xmath254 by our construction it follows that @xmath255 we know that the coupling functions vanish along three great circles through @xmath256 . we see that the numbering of the eigenprojections is not consistent along the circles . the coupling functions @xmath257 and @xmath258 vanish both in the degenerate direction . _ since we do not assume that @xmath243 is unitary the relation @xmath259 does not hold for these coupling functions . however , @xmath260 is unitary and therefore @xmath261 as already observed . the coefficient matrix @xmath262 has the same structure as @xmath60 in section [ sec2 ] . therefore , we can conclude similar statements on eigenvalues and their behaviour by ( a ) investigating the characteristic polynomial and ( b ) block - diagonalising for small and large @xmath87 , respectively . 1 . @xmath263 and @xmath264 . 2 . @xmath262 has purely real eigenvalues for @xmath265 if and only if @xmath266 , i.e. , @xmath267 or @xmath268 . 3 . @xmath269 has the real eigenvalues @xmath270 and three eigenvalues satisfying @xmath271 if @xmath123 and @xmath272 if @xmath273 . 4 . the quotient @xmath274 involving the hyperbolic eigenvalues @xmath275 of @xmath262 is smooth and non - vanishing for fixed values of @xmath87 . we consider only part ( 2 ) to ( 4 ) . the characteristic polynomial of @xmath276 is given by an expression like . if we assume that eigenvalues are purely real we can split the expression into real and imaginary part . we consider the imaginary part first , which leads to @xmath277 therefore , real eigenvalues have to coincide with the square roots of eigenvalues of @xmath140 . if we assume @xmath278 is a root of the characteristic equation , we can divide by the corresponding factor and obtain if @xmath279 ( and therefore @xmath64 is non - degenerate ) @xmath280 if @xmath267 the characteristic polynomial factors as @xmath281 with @xmath282 and @xmath283 . the first factor resembles one - dimensional thermo - elasticity ( with @xmath284 ) and gives three roots with positive imaginary parts subject to and . finally ( 4 ) follows by collecting the two related terms in the characteristic equation of form . the imaginary part of the quotient is given by @xmath285 in hyperbolic / degenerate directions and therefore non - zero . the quotient may be used to determine asymptotic expansions of the hyperbolic eigenvalue and its imaginary part as @xmath218 for fixed @xmath87 and @xmath286 . we will follow a different strategy and diagonalise as @xmath218 uniform on bounded @xmath118 . note first , that @xmath288 is independent of @xmath185 and just the system of one - dimensional thermo - elasticity extended by four additional diagonal entries . since we need to understand this system first , we are going to recall some facts about the one - dimensional theory . as @xmath87 becomes small / large we already gave asymptotic expansions of eigenvalues in section [ ssec:1d ] . the bit of information which is still missing is contained in the following lemma . the coefficient matrix @xmath60 of the one - dimensional thermo - elastic system given in has for @xmath78 and under the natural assumptions @xmath156 only simple eigenvalues . note that the characteristic polynomial of this matrix @xmath60 is given by @xmath289 which is invariant under the transform @xmath290 and has alternating imaginary and real coefficients . from that we conclude that the only possible solutions are of the form @xmath291 , @xmath292 and @xmath293 for certain real @xmath294 . furthermore , from the general theory of section [ sec2 ] it is clear that @xmath295 . thus , the only possibility for multiplicities to occur is if @xmath296 . plugging in @xmath296 and multiplying the corresponding linear factors gives @xmath297 comparing coefficients with the above polynomial implies that @xmath298 , which contradicts to the positivity of all quantities involved . we write the coefficient matrix @xmath299 as sum of homogeneous components in @xmath300 @xmath301 and @xmath302 , @xmath303 , @xmath304 and @xmath305 . as a direct consequence of the previous lemma in combination with the asymptotic expansions of section [ ssec:1d ] we obtain [ prop:3.4 ] assume , that @xmath222 and @xmath306 . then the matrix @xmath307 has uniformly separated eigenvalues in @xmath308 , @xmath309 ( where @xmath310 are of constant multiplicity two ) . now we can apply several steps of diagonalisation based on the scheme of ( * ? ? ? first we apply the diagonaliser of the main part . this has only effects on the two entries related to @xmath311 and the last row / column and determines the eigenvalues @xmath312 and @xmath313 modulo @xmath314 . furthermore , proposition [ prop:3.4 ] allows to @xmath315-block - diagonalise modulo @xmath231 , @xmath232 arbitrary . finally we can investigate the remaining @xmath186-blocks and diagonalise again because the @xmath300-homogeneous entries @xmath316 are distinct ( trivially uniform in @xmath87 and @xmath185 ) . [ prop:3.5 ] assume , that @xmath222 and @xmath306 . the eigenvalues of @xmath299 have uniformly in @xmath87 and @xmath185 full asymptotic expansions as @xmath218 . the first main terms are given as @xmath317 where @xmath318 and @xmath319 are the eigenvalues of the one - dimensional thermo - elastic system with propagation speed @xmath311 and the signs @xmath320 and @xmath321 are independent of each other . the statement holds true uniform in @xmath87 . however , it is only of use as long as the error terms @xmath322 are smaller than the size of the eigenvalues . for @xmath90 the eigenvalues of the one - dimensional thermo - elastic system behave like @xmath323 and @xmath324 hence , the statement of is only of use if @xmath325 , i.e. if @xmath326 . for @xmath120 we know similarly @xmath327 and @xmath328 , which in turn implies that the expansion determines the behaviour of the eigenvalues . this restriction is by no means a severe one ; the expansion is only of interest for the degenerate eigenvalues @xmath329 ( for which no such restriction appears ) . to complete the picture we want to give some comments on expansions for small and large values of @xmath87 under the same assumptions as in proposition [ prop:3.5 ] . using the ideas from @xcite we can employ the ( block ) diagonalisation scheme to separate the three non - degenerate eigenvalues from the two degenerate ones asymptotically and give full asymptotic expansions for them as @xmath87 tends to zero or infinity . the obtained expressions coincide with the formulae from propositions [ prop:2.5 ] and [ prop:2.6 ] . it remains to understand the behaviour of the remaining @xmath330-blocks . this can be done directly by solving the characteristic polynomial as in ( * ? ? ? 2.7 ) or by a second diagonalisation scheme . we focus on the latter idea and consider the case of small @xmath87 first . the @xmath330-blocks have the form @xmath331 with a function @xmath332 . if we restrict the consideration to the zone @xmath333 the remainder can be written as @xmath334 and the standard diagonalisation scheme applied to the last two terms gives full asymptotic expansions in powers of @xmath335 as @xmath336 , @xmath337 a similar idea applies for large @xmath87 in the zone @xmath338 based on @xmath339 with a function @xmath340 it gives asymptotic expansions in powers of @xmath341 as @xmath342 . the fresnel surface for cubic media has six uniplanar singularities . again they are equivalent and it suffices to consider the neighbourhood of @xmath343 . we introduce polar co - ordinates near @xmath256 . let @xmath344 and @xmath345 . then we set @xmath346 and use an asymptotic expansion of @xmath14 as @xmath218 @xmath347 with coefficients @xmath348 to deduce properties of the eigenvalues and eigenprojections of @xmath14 near @xmath256 . we restrict considerations to the case when @xmath349 . then the following statement follows again by the two - step diagonalisation procedure ( like in the conical case and as developed in @xcite , @xcite ) . [ prop:3:6 ] assume @xmath222 , @xmath349 and @xmath350 . then the eigenvalues @xmath62 and the corresponding eigenprojections have uniformly in @xmath185 full asymptotic expansions as @xmath218 . the main terms are given by [ eq:3:29 ] @xmath351 where @xmath352 this statement reflects what we mean by an uniplanar singularity . two of the eigenvalues coincide up to second order . we follow the diagonalisation scheme . @xmath223 is already diagonal , @xmath235 does not contain ( 1,2)-block diagonal entries . to get expansions for the eigenvalues we have to apply two steps of block - diagonalisation . first we treat @xmath235 by the aid of @xmath353 such that @xmath354 block - diagonalises the matrix modulo @xmath314 . this preserves @xmath223 and @xmath355 and gives the new @xmath93-homogeneous component @xmath356 the starting terms of the expansion of the first eigenvalue can be read directly from these matrices . for the remaining two we have to diagonalise the lower @xmath330 block . this block has the form @xmath357 with @xmath358 , @xmath359 from . eigenvalues of this matrix are uniformly separated if the condition @xmath360 is satisfied . under this assumption the full diagonalisation scheme works through and the main terms can be calculated directly and give . for completeness we also give a unitary diagonaliser of the matrix , namely @xmath361 where @xmath362 , @xmath363 . expressions are extended by continuity . again we use the diagonaliser @xmath243 of @xmath225 constructed in proposition [ prop:3:6 ] to reformulate the thermo - elastic system as system of first order . formulae and give the corresponding representation . _ 1 . _ since @xmath364 in the notation of the proof of proposition [ prop:3:6 ] it follows that the coupling functions satisfy @xmath365 since @xmath366 the function @xmath367 vanishes only for @xmath368 , @xmath369 , while @xmath370 vanishes for @xmath371 , @xmath369 . _ note that in contrast to the conic situation the eigenvalues coincide to second order in the degenerate direction , while the coupling functions still vanish to first order ( if we approach the degeneracy from parabolic directions ) . we write the coefficient matrix @xmath299 as sum of homogeneous components in @xmath300 , cf . . this gives @xmath372 and @xmath373 has entries on the diagonal , in the last row and last column . in order to apply a block - diagonalisation as @xmath218 we assume that the matrix @xmath307 has as many distinct eigenvalues as possible . this is ensured if @xmath374 , @xmath375 and we can ( 1,2,1,2,1)-block - diagonalise this matrix family . note , that due to the structure of the last rows and columns , the system decouples modulo @xmath314 into a one - dimensional thermo - elastic system and one containing the elastic eigenvalues . the coupling comes only into play for the @xmath376 entries . [ prop:3.7 ] assume @xmath374 , @xmath375 . then the eigenvalues and eigenprojections of @xmath299 have full asymptotic expansions as @xmath218 . the main terms are given by @xmath377 where @xmath378 and @xmath379 are eigenvalues of the one - dimensional thermo - elastic system with parameter @xmath380 . the signs @xmath320 and @xmath321 are independent and the parameters @xmath358 and @xmath359 are given by . similar to proposition [ prop:3.5 ] this statement is uniform in @xmath87 . it will be of particular importance for us that the hyperbolic eigenvalues @xmath381 coincide up to second order in @xmath300 with the corresponding ( roots of ) eigenvalues of the elastic operator . this will be the key observation to transfer stationary phase estimates from elastic systems to the thermo - elastic one . finally we want to discuss the case of hexagonal media . the elastic operator defined by is invariant under rotations in @xmath167-direction . we will make use of this fact and reduce considerations to a two - dimensional situation corresponding to cross - sections of the fresnel surface . . the parameters are chosen as @xmath198 , @xmath382 , @xmath383 , @xmath384 , @xmath208 . the complete surface is obtained by rotation along the vertical axis.,width=188 ] as already pointed out in section [ sec:2:hex ] degenerate directions are @xmath385 , which are uniplanar . they could be handled similar to the cubic case , but rotational invariance makes estimates simpler . there are further circles of degenerate directions if @xmath386 we exploit rotational symmetry and consider the system only in the frequency hyperplane @xmath191 . then it is possible to express the eigenvectors @xmath387 corresponding to eigenvalues @xmath62 smoothly and the thermo - elastic system can be reformulated as system of first order in full analogy to the general treatment in section [ sec2 ] . the derived asymptotic expansions for eigenvalues and the description of their behaviour transfers away from the degeneracy and has to be equipped with an additional description near these degenerate directions . apart from the @xmath388-axis it is possible to find smooth families of eigenvectors @xmath387 of @xmath14 . this follows directly from rotational invariance combined with one - dimensional perturbation theory of matrices , @xcite . if we assume that the frequency support of initial data and therefore of the solution @xmath50 , @xmath51 is conically separated from the uniplanar directions we can follow section [ sec2 ] and rewrite as first order system in @xmath389 with coefficient matrix @xmath60 given by and of ( 1,1,5)-block structure . in what follows , we will ignore the scalar hyperbolic blocks and consider the remaining @xmath390 matrix . based on the discussions from section [ sec:2:hex ] we know that this @xmath391 block is non - degenerate in the sense that its @xmath92-homogeneous part has distinct eigenvalues if @xmath189 . we assume this in the sequel . but this means that the theory of section [ sec2 ] is applicable and gives a full discription of eigenvalues and eigenprojections and we are done . near the uniplanar directions , i.e. , on the @xmath388-axis , we follow the same approach as for cubic media . we introduce polar co - ordinates around this direction and construct expressions for the corresponding asymptotics . there is one major simplification , due to rotational invariance the construction is independent of the angular variable . we will show how to use the information obtained in sections [ sec2 ] and [ sec3 ] to derive @xmath392@xmath393 decay estimates for solutions to thermo - elastic systems . some of the ideas we present are general in the sense that they can be applied to arbitrary space dimensions , however , our moin focus will be the three - dimensional case and the examples considered in section [ sec3 ] . the estimates we have in mind are micro - localised to ( a ) non - degenerate parabolic , ( b ) non - degenerate hyperbolic or ( c ) degenerate directions . the first two situations generalise the consideration of @xcite , @xcite taking also into account the estimates due to sugimoto @xcite , @xcite , while the treatment of degenerate directions is inspired by the work of liess @xcite , @xcite , @xcite . let first @xmath394 be supported in @xmath40 with @xmath395 on @xmath396 and @xmath397 a cut - off function satisfying @xmath398 for @xmath399 and @xmath400 for @xmath401 . we extend both @xmath190 and @xmath402 as 0-homogeneous functions to @xmath403 . then we consider the solution to the first order system @xmath404 with data @xmath405 . note , that this is well - defined and @xmath60 needs only to be defined on @xmath406 . the proof of this estimate is straightforward from the two - dimensional situation considered in @xcite . for small frequencies we write the solution @xmath54 as sum @xmath410 @xmath411 the corresponding eigenprojections . we know that @xmath412 and @xmath413 by proposition [ prop:2.5 ] . now each of the appearing terms can be estimated using the @xmath392@xmath414 boundedness of the fourier transform ( for @xmath415 $ ] ) and hlder inequality . similarly , the representation in combination with the bound @xmath416 gives exponential decay of @xmath417 and @xmath418 norms and this combined with sobolev embedding yields the desired estimate . for intermediate frequencies we may have to deal with multiplicities and resulting singularities of the spectral projections . instead of we use a spectral calculus representation which implies @xmath419 based on the compactness of the relevant set of frequencies @xmath118 and the bound on the imaginary part due to corollary [ cor:2.8 ] / proposition [ prop:2.9 ] . here , @xmath420 is a smooth curve encircling the family of parabolic eigenvalues for the relevant @xmath118 . if we consider hyperbolic directions we know that the parabolic eigenvalues are separated from the hyperbolic ones and we can use the spectral projection associated to the group of parabolic eigenvalues to separate them from the hyperbolic one(s ) . in this case the estimate of the above theorem is valid for the corresponding ` parabolic modes ' of the solution . so we can restrict consideration to hyperbolic eigenvalues near hyperbolic directions . we consider _ only _ the for us interesting case when hyperbolic directions form part of a regular submanifold of @xmath39 and coupling functions vanish to first order , i.e. , we assume that the corresponding coupling function @xmath421 satisfies @xmath422 this implies that @xmath423 is regular of dimension @xmath424 , the normal derivative @xmath425 never vanishes and @xmath426 defines a tubular neighbourhood of @xmath85 with a natural parameterisation . the desired dispersive estimate is related to geometric properties of the section @xmath427 of the fresnel surface lying directly over @xmath85 , @xmath428 here @xmath429 denotes the cone over @xmath85 . for dimensions @xmath430 we have to distinguish between different cases , depending on whether the cross - section @xmath427 of the fresnel surface satisfies a convexity assumption or not . by the latter we mean that any intersection of @xmath431 with a hyperplane tangent to @xmath432 is convex in a neighbourhood of @xmath427 . if this convexity assumption is satisfied ( or if @xmath98 and therefore @xmath433 ) , we define the convex sugimoto index of @xmath427 as maximal order of contact of @xmath427 with hyperplanes normal to @xmath432 . [ thm:4.2 ] assume that @xmath190 is supported in a sufficiently small tubular neighbourhood of the regular hyperbolic submanifold @xmath85 and that @xmath427 satisfies the convexity assumption . let further @xmath434 be defined as above . first , we outline the strategy of the proof . we split variables in the tubular neighbourhood of the regular hyperbolic submanifold @xmath85 , one coordinate being the defining function @xmath438 and the other parameterising points on @xmath85 . we have to combine a ( simple ) parabolic type estimate in normal directions taking care of the imaginary part of the phase with stationary phase estimates for the integration along @xmath85 . the stationary phase estimate is done first and follows the lines of @xcite , @xcite along with ( * ? ? ? it is sufficient to show the estimate for @xmath439 . we follow the treatment of brenner @xcite and decompose the fourier integral representing the corresponding hyperbolic modes of the solution @xmath54 into dyadic pieces . for large and intermediate frequencies this amounts to estimate for all @xmath440 @xmath441 with the notation @xmath442 , @xmath443 , @xmath444 with @xmath445 and @xmath446 . the amplitude @xmath447 arises from the spectral projector @xmath448 and the phase @xmath449 is complex - valued with @xmath450 uniform in @xmath451 and @xmath440 . if @xmath452 , @xmath453 or if @xmath454 is not near a direction from @xmath85 , the principle of non - stationary phase implies and gives a rapid decay . it suffices to restrict to @xmath454 corresponding to stationary points . we use the method of stationary phase to estimate the integral over @xmath85 , this can be done uniformly over @xmath118 and @xmath455 , provided @xmath300 is chosen small enough and yields an estimate of the form @xmath456 uniform in @xmath68 and @xmath457 . in order to obtain this estimate we apply ruzhansky s multi - dimensional van der corput lemma , @xcite , based on the uniformity of the sugimoto index @xmath458 for small @xmath455 and the uniform bounds on the appearing amplitude . similar to @xcite the imaginary part of the phase can be incorporated in the estimate for the amplitude . integration over @xmath455 yields a further decay of @xmath459 , while integrating over @xmath118 and using @xmath460 yields @xmath461 hence , we need @xmath462 ( compared to the elasticity or wave equation with @xmath463 ) to apply brenner s argument and obtain the desired estimate for the high frequency part . the required regularity follows from using sobolev embedding for small @xmath464 . the treatment of small frequencies is somewhat simpler . we do not apply a dyadic decomposition , but still have to use a stationary phase argument along @xmath85 combined with the behaviour of the imaginary part of the phase away from it , @xmath465 without proof we comment on the non - convex situation . if the convexity assumption is violated we have to replace the convex sugimoto index by a corresponding non - convex one @xmath466 . this is defined as the maximum over the minimal contact orders of @xmath427 with hyperplanes normal to the cone @xmath432 , the maximum taken over all points of @xmath427 . the price we have to pay for non - convexity is a loss of decay . assume that @xmath190 is supported in a sufficiently small tubular neighbourhood of the regular hyperbolic submanifold @xmath85 and that @xmath427 does not satisfy the convexity assumption . let further @xmath467 be the non - convex sugimoto index . because of its importance later on we remark that in our applications to three - dimensional thermo - elasticity the manifolds @xmath85 are parts of circles on @xmath142 , i.e. can be seen as intersections of @xmath142 with a cone . so we have to look at the corresponding sections of the fresnel surface . in this case @xmath469 is just the maximal order of tangency between the curve @xmath427and its tangent lines . if the curvature of this curve is nowhere vanishing , then @xmath470 . furthermore , algebraicity of @xmath201 of order @xmath471 implies that the highest order of contact is @xmath471 and therefore @xmath472 is the admissible range of these indices . for cubic media there are two types of regular hyperbolic submanifolds . one is up to symmetry given by the circle @xmath175 on @xmath142 and the corresponding eigenvalue is equal to @xmath19 . thus the section of the fresnel surface is just a circle and therefore its curvature is nowhere vanishing . similarly , for intersections of the fresnel surface with the plane @xmath473 we obtain the hyperbolic eigenvalue @xmath474 . it is a simple calculation and it remains to check that @xmath475 , see @xcite for such a calculation . ] to show that the curvature of the corresponding section of the fresnel surface is nowhere vanishing as soon as @xmath476 and @xmath477 . hence , @xmath470 in both cases . the following statement resembles ( * ? ? ? * thm . 1.5 ) . in ( 3 ) a stronger decay rate is obtained for some conic degenerations , but they require a sufficiently bent cone and we can not guarantee that in our case . [ thm:4.3 ] assume @xmath478 , @xmath479 and @xmath480 are micro - locally supported in a sufficiently small conical neighbourhood of a conically degenerate point on @xmath481 . then the corresponding solution to the thermo - elastic system for cubic media satisfies the a - priori estimate @xmath482 for @xmath436 $ ] , @xmath437 and @xmath483 . the main idea is that the proof of @xcite uses polar co - ordinates around the singularities of the fresnel surface similar to our treatment in section [ sec3 ] . stationary phase arguments are applied in tangential direction and are uniform for small radii , while the final estimate follows after integration over the remaining variables . it suffices to prove the statement for @xmath484 , the small time estimate is a direct consequence of sobolev embedding theorem in combination with the obvious energy estimate . similar to the hyperbolic estimate discussed before , we apply a dyadic decomposition of frequency space ( localised to a small conic neighbourhood of the degenerate direction ) . the estimate for single dyadic components follows @xcite resp . 1.5 ) ; the only thing we have to check is that the necessary assumptions are satisfied uniform with respect to @xmath87 and @xmath485 . we consider @xmath486 where @xmath487 denotes the point with polar co - ordinates @xmath210 near the conic degenerate direction and @xmath443 . the amplitude @xmath488 arises from the spectral projector ( given in terms of the diagonaliser ) constructed in the blown - up polar co - ordinates and @xmath489 corresponds to the dyadic decomposition . the complex phase @xmath490 is described in proposition [ prop:3.5 ] . its imaginary part is non - negative and vanishes to second order in @xmath267 as well as for three hyperbolic manifolds emanating from the conic degenerate point . again we may treat this imaginary part as part of the amplitude and apply stationary phase estimates for the integral with respect to @xmath185 . as the approximation of the phase modulo @xmath241 is independent of @xmath185 and uniform in @xmath87 this yields @xmath491 uniform in @xmath87 , @xmath68 and @xmath492 . there is no further benefit from the imaginary part ( as there can not be a lower bound with respect to @xmath300 ) and integrating with respect to @xmath87 and @xmath300 concludes the estimate for @xmath493 . similarly , we estimate the small frequency part @xmath494 such that brenner s method again yields the desired decay estimate . the treatment of uniplanar degeneracies follows @xcite . we have to make one further additional assumption related to the shape of certain curves on the fresnel surface near the degenerate point . to be precise , we either require that @xmath495 for @xmath496 an open neighbourhood of the uniplanarly degenerate point and for any plane @xmath497 sufficiently close and parallel to the common tangent plane at the unode . this condition is equivalent to the technical assumption ( 1.12 ) made in @xcite if is violated , we need to consider sugimoto indices @xmath498 , i.e. , contact orders of these planar curves with its tangent planes combined with a uniformity assumption . under assumption the index is given by @xmath499 . for cubic media we have to use the statement of proposition [ prop:3:6 ] to determine the index @xmath500 . using the notation of , it suffices to calculate the indices of the indicator curves determined by @xmath501 . this yields @xmath502 in the nearly isotropic case we have @xmath499 , away from it @xmath503 . both are generic , while the borderline case with @xmath504 is not . the asymptotic construction of the eigenvalues and eigenprojections near the uniplanarly degenerate point of proposition [ prop:3.7 ] yields that the assumption is satisfied uniformly for the phase functions appearing in all dyadic components of the operator . [ thm:4.4 ] assume @xmath478 , @xmath479 and @xmath480 are micro - locally supported in a sufficiently small conical neighbourhood of a uniplanarly degenerate point on @xmath481 . let further @xmath500 be the index of the uniplanar point . then the corresponding solution to the thermo - elastic system for cubic media satisfies the a - priori estimate @xmath505 for @xmath436 $ ] , @xmath437 and @xmath483 . we will sketch the major differences to the treatment of conic degeneracies . we will again use polar co - ordinates and estimate corresponding dyadic components , where now @xmath490 is determined by proposition [ prop:3.7 ] . the imaginary part of @xmath490 vanishes to third order and is of no benefit , while the real part coincides to third order with the corresponding elastic eigenvalue . this allows to use estimates from @xcite and ( * ? ? ? * sect . 4 ) , the main difference to the previous situation is that we now use stationary phase estimates for both , the angular and the radial integral . the proof itself then coincides with the corresponding proof for cubic elasticity , cf . @xcite . using a change of variables the integral is written in the new variables @xmath506 ( i.e. , roughly @xmath507 ) and @xmath508 . in this form the phase splits and the crucial estimate is just a fourier transform of a density carried by the sheet of the fresnel surface ( with possible singularity in the unode ) . this is calculated by introducing _ distorted _ polar co - ordinates on the surface . as level sets we use cuts of the surface by planes parallel to the common tangent plane . then we will at first apply the method of stationary phase to the radial variable in these co - ordinates . these stationary points are non - degenerate and we use the obtained first terms in the asymptotics for a second stationary phase argument in the angular variables . the condition would imply again that stationary points are non - degenerate and we are done , while if is violated we use the lemma of van der corput instead to prove the estimate . it remains to collect all the estimates into a final statement for cubic media . parabolic directions are treated by lemma [ lem:4.1 ] ; hyperbolic manifolds away from degenerate points are covered by theorem [ thm:4.2 ] . the remaining 24 degenerate directions fall into either of the previously discussed categories and estimates follow from theorem [ thm:4.3 ] and [ thm:4.4 ] . the resulting estimates are collected in table [ tab1 ] . * acknowledgements . * the paper was inspired by many discussions with michael reissig and also ya - guang wang , who in particular raised the interest for dispersive decay rates for thermo - elastic systems and the applied decoupling techniques to deduce them . the author is also grateful to otto liess for pointing out some of his results on decay estimates for fourier transforms of measure carried by singular surfaces . printing of the 2nd ed . , grundlehren der mathematischen wissenschaften , 132 . springer verlag , berlin - heidelberg - new york , 1980 . o. liess . . , 6195 , 1991 . . , 329362 , 2004 . . in n. tose ( ed ) : recent trends in microlocal analysis . * 1412 * , 113 , 2005 . a. bannini and o. liess . . , 211232 , 2006 . o. liess . . , 127 , 2009 . m.j.p . musgrave . . , 339355 , 1954 . . , 356366 , 1954 . miller , m.j.p . musgrave . . , 352383 , 1957 . r. racke and song jiang . . chapman & hall / crc monographs and surveys in pure and applied mathematics . 112 . boca raton , fl : chapman & hall / crc . x , 308 p , 2000 . m. reissig and j. wirth . . , 127 , 2008 . m. ruzhansky . . , 75 77 , 2009 . m. ruzhansky and j. wirth . . , 941969 , 2011 . m. sugimoto . , 519531 , 1994 . . ( 4 ) , 521531 , 1996 . y .- , 683705 , 2003 . j. wirth . . , 2940 , 2008 . . , 895902 , 2009 . . p. 495504 . world - scientific 2009 .

bifurcations of solitary waves propagating along the interface between two ideal fluids are considered . the study is based on a hamiltonian approach . it concentrates on values of the density ratio close to a critical one , where the supercritical bifurcation changes to the subcritical one . as the solitary wave velocity approaches the minimum phase velocity of linear interfacial waves ( the bifurcation point ) , the solitary wave solutions transform into envelope solitons . in order to describe their behavior and bifurcations , a generalized nonlinear schrdinger equation describing the behavior of solitons and their bifurcations is derived . in comparison with the classical nls equation this equation takes into account three additional nonlinear terms : the so - called lifshitz term responsible for pulse steepening , a nonlocal term analogous to that first found by dysthe for gravity waves and the six - wave interaction term . we study both analytically and numerically two solitary wave families of this equation for values of the density ratio that are both above and below the critical density ratio . for , the soliton solution can be found explicitly at the bifurcation point . the maximum amplitude of such a soliton is proportional to , and at large distances the soliton amplitude decays algebraically . a stability analysis shows that solitons below the critical ratio are stable in the lyapunov sense in the wide range of soliton parameters . above the critical density ratio solitons are shown to be unstable with respect to finite perturbations . + pacs : 05.45.yv ; 47.55.-t ; 47.90.+a -30pt

physics0512006

the main goal of this paper is to study bifurcations for one - dimensional internal solitary waves propagating along the interface between two ideal fluids with different densities @xmath4 and @xmath5 . the lighter fluid with density @xmath5 lies above the heavier fluid with density @xmath4 : @xmath6 . these bifurcations occur if the solitary wave velocity @xmath7 coincides with the minimum phase velocity @xmath8 of linear internal waves . if the upper density @xmath5 is small enough compared to the lower density @xmath4 , then a bifurcation similar to that for pure gravity - capillary waves occurs @xcite . in this case solitary waves undergo a supercritical bifurcation at the critical velocity : their form approaches the form of the envelope solitons for the focusing one - dimensional nonlinear schrdinger equation ( 1d nlse ) @xcite . the soliton amplitude behaves universally near the critical velocity @xmath9 : it vanishes like @xmath10 . the width of the solitary wave increases proportionally to @xmath11 . as the density ratio @xmath12 increases , the character of the nonlinear interactions changes . the four - wave coupling coefficient decreases and vanishes at @xmath13 @xcite , that is when @xmath14 . such a value may not be obtained easily in a two - fluid configuration . however , it may be relevant in a three - layer ( or more ) configuration @xcite , or in nonlinear optics , where similar singularities can occur . above the critical ratio , solitary waves undergo a subcritical bifurcation : at the critical velocity , their amplitude jumps from zero for @xmath12 below @xmath15 , up to finite values when @xmath12 is above the critical density ratio . in order to describe such type of bifurcation , it is necessary to keep the next order terms beyond the classical 1d nlse . when the density ratio varies in a neighborhood of the critical density ratio , it is possible to use , as in the derivation of the classical nlse , perturbation theory assuming that the interacting wave amplitudes are small . at leading order , one needs to keep three kinds of terms . the first one , coming from the four - wave interaction , takes into account the so - called lifshitz invariant @xcite . this term is local relative to the amplitude of the soliton and its first spatial derivative and , as was shown in @xcite , appears from the expansion of four - wave interaction element under the assumption about its analytical dependence . however , for deep - water internal waves , as we show in this paper , there exists also a nonlocal term which has the same order of magnitude as the first one @xcite . its structure is similar to the dysthe term first found for water ( gravity ) waves @xcite ( see also @xcite ) . in the case of water waves , this term is responsible for the interaction of a narrow wave packet ( in @xmath16-space ) with mean flow , induced by the packet . the third term takes into account six - wave interactions . in order to find the six - wave coupling coefficient , one needs to calculate all possible renormalizations due to three- , four- and five - wave interactions and therefore this partial problem requires a lot of cumbersome calculations . for these calculations we use the hamiltonian formalism ( see the review @xcite , as well as the papers @xcite ) , which appears to be the most adequate method for this subject . in @xcite , a different method was used : the problem was reformulated as a spatial dynamical system and only the reversibility was exploited . it is necessary to underline also that the use of the hamiltonian approach to study solitary waves gives an appropriate framework for the temporal behavior of the dynamics of solitary waves , e.g. their stability . through this formalism , it is easy to perform different kinds of averaging and perturbations . second , it is crucial that by applying the hamiltonian technique the averaging equations of motion retain their original hamiltonian form . in particular , this is of great help for the investigation of soliton stability ( see , for instance , @xcite ) . consider the interface @xmath17 between two ideal incompressible fluids with respective densities @xmath18 and @xmath19 , in the presence of gravity ( with the acceleration @xmath20 acting down the vertical @xmath21axis ) and capillarity with interfacial tension @xmath22 . we shall assume that the lighter fluid with density @xmath19 occupies the region @xmath23 , and respectively the heavier fluid occupies the region @xmath24 . flows of both fluids are considered to be potential and two - dimensional . the fluid velocities are given by @xmath25 where the velocity potentials @xmath26 and @xmath27 satisfy laplace s equation @xmath28 these equations are subject to the following boundary conditions . far from the interface as @xmath29 @xmath30 on the interface @xmath17 the kinematic conditions hold : @xmath31 the dynamic condition reduces to the discontinuity of pressures across the interface due to capillarity : @xmath32 the use of bernoulli equations in each fluid allows to rewrite the latter equation in terms of potentials and their derivatives : @xmath33 the equations ( [ laplace])([bern ] ) conserve the total energy : @xmath34 where the kinetic energy is equal to @xmath35 and the potential energy is given by the expression @xmath36 as shown first in @xcite ( see also @xcite ) , the equations of motion ( [ kin ] ) and ( [ bern ] ) together with the laplace equations ( [ laplace ] ) represent a hamiltonian system . the hamiltonian coincides with the energy ( [ ham ] ) . the new variables @xmath37 and the interface shape @xmath38 are canonical conjugate variables : @xmath39 where @xmath40 . the given hamiltonian form generalizes zakharov s canonical form for free - surface hydrodynamics @xcite . a hamiltonian formulation of the problem of a free interface between two ideal fluids , under rigid lid boundary conditions for the upper fluid , was also given by benjamin & bridges @xcite . craig & groves @xcite give a similar expression , by using the dirichlet - neumann operators for both the upper and lower fluid domains ( see also @xcite ) . the hamiltonian can be expanded in series with respect to powers of the canonical variables . in this case the steepness of the interface plays the role of a small parameter of expansion . due to the conservation of the total mass for each fluid this expansion begins with the quadratic term . it is more convenient to work in fourier space . let us introduce the normal variables @xmath41 by means of the following formulas : @xmath42 where the density @xmath4 is set equal to unity and @xmath43 . in these formulas @xmath44\right)^{1/2}\ ] ] is the dispersion relation for linear internal waves and @xmath16 is the wave vector directed along the @xmath45axis ( 1d case ) . the transformation ( [ normal - var ] ) diagonalizes the quadratic part of the hamiltonian , @xmath46 as a result , the equations of motion in the new variables @xmath41 take the standard form @xcite : @xmath47 with the hamiltonian @xmath48 where @xmath49 is responsible for the nonlinear interactions between waves . in the given case of internal waves the expansion of @xmath49 in the wave amplitude will contain powers starting with the cubic terms . consider a solution of eq . ( [ k - rep ] ) in the form of a solitary wave propagating with constant velocity @xmath7 . then the potentials and the shape @xmath50 of the interface will depend on @xmath51 and @xmath52 through the combination @xmath53 . in particular , the inverse fourier transform of @xmath41 , @xmath54 will be a function of @xmath55 only . in fourier space such dependence implies an exponential dependence in time for the normal variables : @xmath56 where the time - independent amplitude @xmath57 is defined from the equation @xmath58 this equation can be casted into the following variational problem : @xmath59 where @xmath60 is the total momentum of the wave system . this means that a solution to this equation represents a stationary point of the hamiltonian @xmath61 for fixed momentum @xmath62 . a solution to eq . ( [ sol ] ) in the form of a solitary wave is possible if the difference @xmath63 is sign - definite . when the equation @xmath64 has real roots ( say @xmath65 ) , then , in accordance with @xmath66 , the solution of ( [ sol ] ) will be of the form : @xmath67 hence taking the first term as the zero approximation , after iteration the solution of this equation can be represented as an infinite series with respect to @xmath68 : @xmath69 in @xmath45space , this solution is equivalent to a periodic solution ( for details , see @xcite ) . physically , this criterion is very transparent . the equality ( [ cher ] ) is the resonance condition for cherenkov radiation of waves by an object moving with the velocity @xmath7 . due to such radiation a solitary wave will lose its energy and therefore can not be steady . for the internal wave dispersion ( [ dispersion ] ) the maximum solitary wave velocity @xmath7 coincides with the minimum phase velocity of linear waves : @xmath70 it occurs when @xmath71 ^{1/2}. \label{crit = k}\ ] ] at this point the values of the linear frequency and critical velocity are @xmath72 where @xmath73 is the atwood number . as the maximum solitary wave velocity is approached , the amplitude @xmath57 given by ( [ sol ] ) reaches a very sharp maximum at the point @xmath65 , where the straight line @xmath74 touches the dispersion curve @xmath75 : @xmath76 ^{-1}f_k . \label{res}\ ] ] here @xmath77 and @xmath78 is the positive - definite second derivative of @xmath79 taken at @xmath65 : @xmath80 . hence one can see that as @xmath81 the width of the distribution @xmath82 tends to zero , which corresponds to the peak at @xmath65 becoming narrower and narrower . due to the nonlinearities of the wave system this peak generates multiple harmonics near @xmath83 with integer @xmath84 . if the amplitude of this peak is small ( if , for instance , it vanishes smoothly while approaching the critical velocity ) then we can use the perturbation theory that consists in expanding @xmath85 through its harmonics : @xmath86 here the small parameter @xmath87 and the `` slow '' coordinate @xmath88 are introduced , so that @xmath89 is the amplitude of the envelope of _ n_-th harmonic . the assumption that the solitary wave amplitude vanishes continuously at @xmath90 means that the leading term of the series in eq . ( [ expand ] ) corresponds to the first harmonic , and all other harmonics are small with respect to the parameter @xmath91 . this is the condition under which the nonlinear schrdinger equation is derived ( see , for example , @xcite ) . in this case , at leading order in @xmath91 , we obtain the stationary nlse ( compare with @xcite ) @xmath92 where @xmath93 is related to the matrix element @xmath94 of four - wave interactions ( see below ) as @xmath95 > from now on , we drop the subscript @xmath96 for @xmath97 . in complete correspondence with ( [ var ] ) , the envelope equation ( [ snls ] ) can be recasted in the following variational problem : @xmath98 where @xmath99 is the number of waves or the wave action . the ( averaged ) hamiltonian @xmath100 is given by the expression @xmath101 in this approximation the leading term in the interaction hamiltonian @xmath49 has the form @xmath102 the tilde denotes renormalization of the vertex @xmath103 due to the interaction with the zeroth and second harmonics , corresponding to the cubic terms in the hamiltonian @xmath61 . thus , after averaging , the soliton solution is a stationary point of the ( averaged ) hamiltonian for fixed @xmath104 . > from eq . ( [ snls ] ) one can see that the localized solution is possible if the coupling coefficient @xmath93 is negative ( focusing nonlinearity ) . we recall that in our case @xmath105 . in this case ( [ snls ] ) can be rewritten in dimensionless variables as follows : @xmath106 its soliton solution @xmath107 is given by @xmath108 hence it follows that while approaching the critical velocity the soliton amplitude vanishes like @xmath109 and the soliton width grows as @xmath110 . the latter means that our approximation improves when approaching the critical velocity : the wave becomes more monochromatic and nonlinearity weaker . this approximation becomes exact at the critical velocity . as we show in the next section such a situation occurs for all interfacial solitary waves when the density ratio is less than the critical value @xmath111 ( see for example @xcite ) . while increasing @xmath0 the four - wave coupling coefficient @xmath112 remains negative up to the critical ratio , where it vanishes . for @xmath2 the coefficient @xmath112 becomes positive , so that the nonlinear interaction in ( [ snls ] ) changes its character , from focusing to defocusing . in this case , in order to have solitary wave solutions , one needs to keep next order terms beyond the classical nonlinear schrdinger equation ( [ snls ] ) , which should provide existence of localized solutions in the form of solitary waves . such solutions were computed numerically using the full water - wave equations by laget & dias @xcite . bridges et al . @xcite computed finite - amplitude travelling waves near the transition from focusing to defocusing . the simplest weakly nonlinear extension retains the terms due to six - wave interactions . such interactions should be of the focusing type in order to compensate for the defocusing four - wave interaction . from this consideration it becomes clear that the soliton amplitude undergoes a jump at the point @xmath90 . it is easy to estimate that such a jump will be proportional to @xmath113 . in order to obtain convergence of the hamiltonian series expansion , the jump must be small . in other words , such an expansion will be valid if the deviation of @xmath0 from its critical value @xmath1 is small enough . the appearance of the jump at the critical velocity means that the soliton undergoes subcritical bifurcation . such type of bifurcation is analogous to phase transition of first order . if the corresponding jump is small then we have the analogue of the first - order phase transition close to the second - order phase transition . for phase transitions such a situation occurs in a small neighborhood of the so - called tri - critical point . now we will give the general structure of the hamiltonian expansion corresponding to interfacial waves near the critical density ratio assuming the following two dimensionless parameters are small : @xmath114 as mentioned before , there are in this case three main contributions to nonlinear terms ( which , for instance , can balance dispersion , thus providing the existence of stationary solitary waves ) . two contributions come from the expansion of the four - wave interaction hamiltonian . because a stationary localized solution is assumed to be an envelope soliton , i.e. its spectrum remains narrow and concentrated near @xmath65 , we have to expand the four - wave matrix element @xmath115 , keeping the first - order terms that are linear in @xmath116 . as shown in the next section , this expansion contains local and nonlocal terms : @xmath117 the constants @xmath118 and @xmath119 have different parity relative to reflection @xmath120 . the coefficient @xmath121 changes its sign , but the coefficient @xmath122 retains its sign under this transform . the difference in parities between @xmath118 and @xmath119 gives different contributions to the averaged four - wave hamiltonian : @xmath123 \;dx . \label{wave4}\ ] ] here @xmath124 is the positive definite integral operator @xmath125 and @xmath126 is the hilbert transform : @xmath127 the fourier transform of the kernel of the operator @xmath124 is equal to @xmath128 . the third contribution , which is local in @xmath85 , corresponds to six - wave interactions : @xmath129 where @xmath130 is the corresponding coupling coefficient . the solitary wave shape in this case will be defined from the solution of the following variational problem : @xmath131 where the ( averaged ) hamiltonian is given by the expression @xmath132 the terms @xmath133 and @xmath134 are defined by eqs . ( [ wave4 ] ) and ( [ 6wave ] ) , respectively . the variational problem ( [ var2 ] ) can be considered as resulting from averaging the problem ( [ var ] ) over ` fast ' spatial oscillations . thus , in order to solve the variational problem ( [ var2 ] ) , we need to know four coefficients : @xmath135 and @xmath130 . one can easily see that the contributions from terms proportional to @xmath136 in @xmath133 and the six - wave hamiltonian can be determined independently , which makes calculations more simple . we begin our calculations with the four - wave matrix element @xmath137 in order to find @xmath138 and its `` derivatives '' @xmath121 and @xmath122 . the usual way to calculate matrix elements consists in expanding the hamiltonian in series with respect to powers of the canonical variables @xmath139 and @xmath38 . then one substitutes in each term @xmath140 of the hamiltonian the variables @xmath139 and @xmath38 expressed in terms of the normal amplitudes @xmath41 and @xmath141 with the help of the formulas ( [ normal - var ] ) , and finally one symmetrizes each term @xmath140 against all @xmath142 and @xmath143 . as a result one obtains @xmath144 the hamiltonian for interactions @xmath49 is represented as a sum of @xmath140 terms : @xmath145 and the matrix elements @xmath146 are symmetric with respect to all permutations inside of both groups of indices @xmath147 and @xmath148 . moreover , @xmath149 . to find the needed matrix elements , it is convenient to represent first the hamiltonian ( [ ham ] ) in the following form by integration by parts : @xmath150 \,dx , \label{e : form24}\ ] ] where @xmath151 up to the multiplier @xmath152 , @xmath153 coincides with the normal component of the velocity @xmath154 on the interface @xmath17 . the vector @xmath155 is the unit normal to the interface . thus , only @xmath153 needs to be expressed in terms of @xmath139 and @xmath156 . to find this dependence we first solve the laplace equations ( [ laplace ] ) for @xmath26 and @xmath27 , @xmath157 where @xmath158 are functions determined from the interface boundary conditions and @xmath159 is the integral operator defined in the previous section . the operator @xmath160 is defined through the infinite series : @xmath161 the boundary values @xmath162 of @xmath163 on the interface are expressed by means of the operators @xmath164 @xmath165a_{1,2}(x).\ ] ] hence by calculating the derivatives of the potentials @xmath166 at the interface @xmath17 we have the following expressions for @xmath167 : @xmath168 writing down the equality between @xmath169 and @xmath170 yields @xmath171 if in addition one uses the definition of @xmath139 , @xmath172 one has two relations to determine @xmath153 . let us introduce the operator @xmath173 this operator represents the green operator for one fluid ( the lower fluid ) that establishes the relation between @xmath169 and @xmath174 on the surface @xmath175 : @xmath176 with the help of @xmath177 , the kinetic energy of the lower fluid @xmath178 is expressed as follows : @xmath179 this relation can be taken as the definition of the green operator @xmath177 . according to this definition this operator is self - adjoint as it should be . of course , this fact can be verified also by direct calculations , for instance , by expanding @xmath177 with respect to powers of @xmath38 . in this case one needs first to expand the operator @xmath180 @xmath181 then the green operator @xmath177 is written as a series in powers of @xmath38 : @xmath182 where @xmath183 @xmath184 @xmath185 in the case of interfacial waves the green operator @xmath186 , defined by the relation @xmath187 , is constructed by solving the linear system ( [ v1=v2],[psi - int ] ) by means of the operator @xmath177 ( [ surf - green ] ) : @xmath188 here the green operator for the upper fluid @xmath189 is equal to @xmath190 . the latter formula means that @xmath191 hence one can see that the green operator @xmath186 is self - adjoint , as it should be . the total kinetic energy of two fluids is defined by the operator @xmath192@xmath193 the expansion of the green operator @xmath186 in powers of @xmath38 is expressed through the expansion of @xmath194 : @xmath195 where @xmath196 @xmath197 @xmath198 the inverse green operator @xmath199 is the combination @xmath200 hence the green operator @xmath186 is given through the following expansion : @xmath201 where @xmath202 @xmath203 , \label{g3}\ ] ] @xmath204 . \nonumber\end{aligned}\ ] ] this expansion of @xmath186 allows one to write down the expansion for the hamiltonian : @xmath205 where @xmath206 dx , \label{h-0}\ ] ] @xmath207 dx , \label{h-3}\ ] ] @xmath208\psi dx-\int \frac{\sigma \eta _ x^4}8dx . \nonumber\end{aligned}\ ] ] the hamiltonians @xmath209 and @xmath210 are expressed respectively through @xmath211 and @xmath212 ( [ g3],[g4 ] ) : @xmath213 @xmath214 note that the hamiltonian @xmath215 for the interfacial case was first calculated in the paper @xcite . for surface waves ( @xmath216 or @xmath217 ) the expression for @xmath218 in the form ( [ h-4 ] ) was presented in the papers @xcite , @xcite . after these calculations we can find the needed matrix elements . first , let us consider the hamiltonian expansion in fourier space ( [ h-0][h-6 ] ) . for @xmath219 and @xmath215 it gives @xmath220 \mid \eta _ k\mid ^2\right\ } dk , \label{hf-0}\ ] ] latexmath:[\[h^{(3)}=-\int \frac a{2\left ( \rho + 1\right ) } \left [ here the subscript @xmath96 in @xmath222 means @xmath223 and so on , while @xmath224 . with this notation the four - wave hamiltonian takes the form @xmath225 where @xmath226 hence it can be checked easily that the transition to the normal variables by means of ( [ normal ] ) diagonalizes the quadratic hamiltonian : @xmath227 so that the equations of motion are written in the standard form ( [ k - rep ] ) . substituting now the transformation ( [ normal ] ) into ( [ hf-3 ] ) gives the following expression for @xmath215 : @xmath228 here the matrix elements @xmath229 and @xmath230 are : @xmath231 + \left ( \frac{k_2\omega _ 3\omega _ 1}{2k_3k_1\omega _ 2}\right ) ^{1/2}\left [ |k_3||k_1|+k_3k_1\right ] \right\ } , \nonumber\end{aligned}\ ] ] @xmath232 \right . \label{v } \\ & & \ \left . + \left ( \frac{k_1\omega _ 2\omega _ 3}{2k_2k_3\omega _ 1}\right ) ^{1/2}\left [ |k_2||k_3|+k_2k_3\right ] + \left ( \frac{k_2\omega _ 3\omega _ 1}{2k_3k_1\omega _ 2}\right ) ^{1/2}\left [ -|k_3||k_1|+k_3k_1\right ] \right\}. \nonumber\end{aligned}\ ] ] the hamiltonian describing @xmath233 interacting waves has the form @xmath234 here the primitive ( non - renormalized ) 4-wave matrix element @xmath235 is given by the expression @xmath236 where @xmath237 in this section we presented only three matrix elements : @xmath229 , @xmath238 and @xmath239 . these matrix elements can be used not only for one - dimensional surface waves , but also after obvious generalizations to the general ( 2d ) case also . all the other matrix elements can be found by the same procedure as the one used to find @xmath229 , @xmath230 and @xmath239 . however , for solitons near the bifurcation velocity @xmath90 when the density ratio is close to the critical one , the complete knowledge of @xmath229 , @xmath230 and @xmath239 is sufficient . indeed we only need to know the coupling coefficients @xmath93 , @xmath240 @xmath122 and @xmath130 in ( [ wave4 ] ) and ( [ 6wave ] ) . a straightforward and classical way to find them is to use the diagram technique @xcite based on the renormalization of matrix elements . we will use this approach partially , only to calculate the constant @xmath112 . the three other coefficients ( @xmath121 , @xmath122 and @xmath130 ) can be found by using the procedure of averaging with respect to high frequencies . in the present case , it is the carrying frequency @xmath241 of the main harmonic . the amplitudes of all the other harmonics are assumed to be small in order to apply the perturbation technique based on the hamiltonian expansion . first we compute the coefficient @xmath93 . as is well known ( see @xcite for example ) , if three - wave interactions are not resonant , they can be excluded by canonical transformations that result in the renormalization of the high - order matrix elements . in particular , for @xmath233 interacting waves , such a renormalization yields @xmath242 thus , in the four - wave interaction vertex , there are three contributions : the first one comes from @xmath243 ( @xmath244 ) , the second contribution is connected with capillarity ( @xmath245 ) , and the last contribution stems from the renormalization ( [ matrix9 ] ) due to three - wave interactions . the latter written for @xmath246 ( @xmath247 ) represents the interactions with the zeroth and second harmonics . however the interaction with the zeroth harmonic vanishes because the three - wave matrix element @xmath230 tends to zero sufficiently rapidly if one of the wavenumbers @xmath248 tends to zero . therefore @xmath249 from ( [ u ] ) , ( [ v ] ) and ( [ t-4 ] ) one obtains @xmath250 @xmath251 @xmath252 substituting these expressions into ( [ crit - ratio ] ) gives @xmath253 where the square of the critical atwood number @xmath254 is equal to @xmath255 . this gives for the critical value of @xmath12 @xmath256 in agreement with the paper @xcite . for @xmath257 , the four - wave coupling coefficient is negative , and the corresponding nonlinearity is of the focusing type . in this case , solitary waves near the critical velocity @xmath8 are described by the stationary nlse ( [ snls ] ) and undergo a supercritical bifurcation at @xmath90 @xcite . for @xmath258 the coupling coefficient changes sign and the bifurcation becomes subcritical . to find the soliton shape in this case we need to calculate three more coefficients : @xmath240 @xmath122 and @xmath130 . the first two coefficients are defined from the expansion of @xmath259 near @xmath260 . to find the coefficient @xmath122 we shall use the averaging procedure . there are two contributions to @xmath261 the first one comes from the hamiltonian ( [ h-4 ] ) . hence one can see that nonlocality arises from two terms in @xmath133 : @xmath262 here the brackets @xmath263 denote average with respect to short - wave oscillations so that @xmath264 but @xmath265 . the latter follows after substituting @xmath38 and @xmath139 expressed in terms of the envelope amplitude @xmath85 : @xmath266 ^{1/2}\left ( \psi e^{-i\omega _ 0t+ik_0x}+\psi ^{*}e^{i\omega _ 0t - ik_0x}\right ) , \label{normal1 } \\ \psi & = & -i\left [ \frac{(1+\rho ) \omega _ 0}{2k_0}\right ] ^{1/2}\left ( \psi e^{-i\omega _ 0t+ik_0x}-\psi ^{*}e^{i\omega _ 0t - ik_0x}\right ) . \nonumber\end{aligned}\ ] ] it is interesting to note that these approximations of the exact formulas ( [ normal ] ) have an accuracy of @xmath267 where @xmath82 is the width of the main peak , since at the critical wavenumber @xmath65 @xmath268 in particular , computing the averages @xmath269 and @xmath270 from ( [ normal1 ] ) gives zero for the first mean and @xmath271 as a result , the nonlocal hamiltonian ( [ h - nl ] ) has the form @xmath272 it is important that @xmath273 is negative definite . this means that this term provides focusing . another contribution to @xmath122 comes from the interaction with the zeroth harmonic , which for gravity waves is responsible for the mean flow induced by the wave packet . the same statement is also valid for interfacial waves . consider the three - wave hamiltonian ( [ h-3 ] ) and average with respect to fast oscillations with frequency @xmath241 , under the assumption that @xmath38 and @xmath139 contain two parts : @xmath274 ^{1/2}\left ( \psi e^{-i\omega _ 0t+ik_0x}+\psi ^{*}e^{i\omega _ 0t - ik_0x}\right ) + \widetilde{\eta } , \label{separation } \\ \psi & = & -i\left [ \frac{(1+\rho ) \omega _ 0}{2k_0}\right ] ^{1/2}\left ( \psi e^{-i\omega _ 0t+ik_0x}-\psi ^{*}e^{i\omega _ 0t - ik_0x}\right ) + \widetilde{\psi } . \nonumber\end{aligned}\ ] ] in ( [ separation ] ) , @xmath275 and @xmath276 are low - frequency quantities responsible for the mean flow induced by the high - frequency wave packet , concentrated at @xmath65 . substituting ( [ separation ] ) into the three - wave hamiltonian ( [ h-3 ] ) and then averaging yields @xmath277 this hamiltonian describes the interaction of the wave packet with the mean flow @xmath278 . to complete the system one needs to add the quadratic hamiltonian , @xmath279 dx,\ ] ] where the first part relates to the dispersion of the wave packet and the second one describes long linear gravity waves . in this case , the total hamiltonian is the sum of these two terms : @xmath280 note that @xmath281 ( subscript @xmath282 means low frequency ) is only one part of the full hamiltonian . it is used to find the coefficient @xmath122 , which can be found independently . the hamiltonian equations of motion for @xmath283 are written similarly to ( [ canonical ] ) and ( [ k - rep ] ) : @xmath284 @xmath285 combining the last two equations gives @xmath286 this leads to the following equation for @xmath85 : @xmath287 in the context of gravity waves the nonlinear term in this equation was first found by dysthe ( @xmath217 ) @xcite . it is of focusing type . this corresponds to the second nonlocal contribution to the hamiltonian : @xmath288 combined with ( [ nl-4 ] ) it gives the final form for the nonlocal interaction hamiltonian : @xmath289 hence the coefficient @xmath122 is given by @xmath290 this coefficient being multiplied by @xmath291 , it corresponds to an attraction between waves ( focusing nonlinearity ) . let us now consider the coefficient @xmath121 , which is responsible for the local four - wave interactions ( [ wave4 ] ) . it is possible to develop the same scheme as the one we used in calculating the coefficient @xmath122 . another way is to calculate the first derivative of the matrix element @xmath259 with respect to its arguments at the wavenumber @xmath260 ( of course , excluding the nonlocal interaction which has been defined already ) . we skip all these calculations and present directly the final answer : @xmath292 so that @xmath293 the six - wave coupling coefficient @xmath130 is more complicated to compute . the procedure requires applying the diagram technique ( which , in fact , is a usual perturbation theory based on the use of multi - scale expansion methods ) . the corresponding calculations are presented in appendix a. they give @xmath294 where @xmath295 . thus , the six - wave interaction term also leads to a focusing nonlinearity . in the previous section , we computed the coefficients that are needed to analyze solitary wave solutions and their bifurcations near the critical density ratio . as pointed out in section ii , solitary wave solutions represent stationary points of the hamiltonian for fixed momentum @xmath62 . when the critical velocity is approached , the solitons are transformed into envelope solitons . the envelope solitons , like the original ones , also represent stationary points but of the mean hamiltonian , averaged with respect to fast oscillations , latexmath:[\[\overline{h}=\int \left [ \frac{\omega _ 0}{4k_0 ^ 2}|\psi _ x|^2+\frac{\mu}{2 } for fixed momentum . after averaging the momentum becomes the number of waves @xmath104 @xmath297 multiplied by @xmath298 . the variational problem ( [ var2 ] ) for the envelope solitons is then written as follows : @xmath299 the soliton shape in this case is governed by the corresponding lagrange - euler equation : @xmath300 ) are small compared with @xmath301 . however , in the soliton solution both dispersion and nonlinearity occur at a level comparable with @xmath302 , where @xmath303 . it follows , for instance , from the variational problem ( [ var3 ] ) or from the equation ( [ soliton1 ] ) for the soliton shape . equation ( [ soliton1 ] ) is a pseudo - differential equation . besides differentiation it contains the integral operator @xmath124 that introduces some complexity in its analysis . the equation ( [ soliton1 ] ) can be simplified by introducing the amplitude @xmath304 and the phase @xmath305 of @xmath306 . substituting @xmath307 into eq . ( [ soliton1 ] ) and separating real and imaginary parts give the following equation for the phase : @xmath308 incidentally , the phase equation for the classical nlse ( [ dnls ] ) is @xmath309 since @xmath310 as @xmath311 , the constant is equal to zero and there is no dependence on @xmath51 of the phase as opposed to the present case . in nonlinear optics , the dependence on @xmath51 of the phase @xmath305 is called pulse chirp . by means of the relation ( [ phase - eq ] ) , the phase is excluded from the equation for the amplitude @xmath304 : @xmath312 where the six - wave coupling coefficient @xmath130 is renormalized as @xmath313 substituting the rescaling @xmath314 ) yields @xmath315 where all primes have been omitted . the value of @xmath112 is @xmath316 for @xmath317 and @xmath318 for @xmath319 . from the scaling ( [ scaling ] ) it is seen that the new @xmath320 takes small values when the old @xmath91 goes to @xmath321 , and , respectively , large values when the old @xmath322 approaches @xmath321 . below the critical density ratio ( @xmath316 ) all nonlinear terms are focusing and , thus , in this case soliton solutions exist in the whole range of ( new ) @xmath320 . for small @xmath320 , the last two terms in eq . ( [ norm - r ] ) can be neglected and solitons are those of the classical nlse : @xmath323 thus , at the critical velocity @xmath8 , solitons with @xmath324 undergo supercritical bifurcation : their amplitude vanishes like @xmath325 . when @xmath91 increases , the soliton width decreases while the soliton amplitude grows . this behavior is confirmed with the numerical computation of solutions to this equation ( see below ) . this property can also be observed with @xmath326 ( i.e. in the absence of nonlocal nonlinearity ) . then this equation admits the following analytical solution expressed in terms of elementary functions @xcite,@xcite : @xmath327 .\end{aligned}\ ] ] this solution is valid for both signs of @xmath93 . it is seen that this solution approaches the nls soliton ( [ nls - soliton ] ) when @xmath91 goes to zero , if @xmath316 ( focusing nonlinearity ) . for @xmath318 the solution ( [ r - soliton ] ) at @xmath328 transforms into a soliton with algebraic decay at infinity : @xmath329 such solitons have been studied in @xcite . if @xmath330 , the soliton solution decays exponentially as @xmath331 for both signs of @xmath112 ( latexmath:[$\psi \propto e^{-\lambda @xmath318 ( defocusing nonlinearity ) , an asymptotic analysis of eq . ( [ norm - r ] ) shows that the soliton solution will decay algebraically , similarly to ( [ power ] ) : @xmath333 where @xmath334 is the number of waves on this soliton solution . this asymptotic behavior suggests to seek for solutions in a form similar to ( [ power ] ) : @xmath335 where @xmath336 and @xmath337 are unknown constants . substituting this ansatz into eq . ( [ norm - r ] ) and assuming @xmath328 yields @xmath338 setting equal terms that are proportional to @xmath339 and @xmath340 , one obtains two equations for determining @xmath336 and @xmath337 : @xmath341 the solution of this system with @xmath342 is @xmath343 for this solution the number of waves is given by the expression @xmath344 this is the critical soliton solution with finite amplitude at @xmath328 . in other words , solitons above the critical density ratio ( @xmath345 ) undergo a subcritical bifurcation . the numerical value of @xmath334 is @xmath346 , which is good agreement with ( [ crit - n ] ) . when the density ratio is equal to the critical one ( @xmath347 ) , equation ( [ eq = r ] ) admits another rescaling , different from ( [ scaling ] ) : @xmath348 under this new scaling equation ( [ eq = r ] ) becomes @xmath349 where tildes have been omitted and @xmath119 is given by the last relation in ( [ scaling ] ) . ( [ critnls ] ) can be related to the critical stationary nls equation . when @xmath326 this equation is nothing else than the quintic nls equation which , as is well - known ( see , for instance , @xcite or the review @xcite ) , belongs to the class of critical nls equations . however , this property persists in the presence of nonlocal nonlinearity . it follows , for instance , if one considers the behavior of the number of waves @xmath104 on the parameter @xmath320 for the soliton solution . indeed , the change @xmath350 reduces eq . ( [ critnls ] ) to a form not containing @xmath351 : @xmath352 consequently , the number of waves for soliton solutions is independent on @xmath91 : @xmath353 as @xmath354 , we also come up asymptotically with the same equation ( [ critnls ] ) . in this limit , and independently on @xmath355 the main balance of the term @xmath356 in equation ( [ norm - r ] ) comes from the last two last terms in its left - hand side . the latter means that we have convergence of two soliton branches corresponding to different signs of @xmath93 as @xmath354 . it is easy to show also that the hamiltonian evaluated on the soliton solution of ( [ critnls ] ) is equal to zero . however , for solitons with @xmath345 , the hamiltonian takes positive values , while for @xmath324 ( focusing case ) the hamiltonian of solitons happens to be negative . this is a simple consequence of the variational problem ( [ var3 ] ) . indeed , consider the scaling transformation that keeps the number of waves @xmath357 unchanged : @xmath358 where @xmath304 is a soliton solution of equation ( [ norm - r ] ) . under this transform the hamiltonian @xmath100 ( [ averh ] ) becomes a function of the scaling parameter @xmath336 : @xmath359 because the soliton solution is a stationary point of @xmath100 ( [ var3 ] ) , @xmath360 and therefore @xmath361 hence we have @xmath362 thus , at @xmath347 , @xmath363 . if @xmath324 , @xmath364 and @xmath365 if @xmath345 . solitons of both equations ( [ norm - r ] ) and ( [ critnls ] ) with different values of @xmath91 were found numerically for both signs of @xmath93 as well as for @xmath347 . in order to find them , we used the method suggested by v.i . petviashvili @xcite . the idea of the petviashvili method is based on the solution of diffusive type equations with the introduction of a new time @xmath366 . stationary solutions of this extended equation coincide with the soliton solution of the original equation that we are looking for . in the present case we developed the following algorithm . the iteration scheme to find solutions connects values @xmath367 at the @xmath368th time step with @xmath369 by the relation : @xmath370 \right\ } ^{(n)}m\left\ { r^{(n)}\right\ } . \label{iteration}\ ] ] here @xmath371 is the time step and @xmath372 is the functional of @xmath373 of the form @xmath374 ^{1/8},\ ] ] where @xmath375 is the @xmath376-norm . it is seen from this equation that its stationary solution satisfies the snls equation ( [ norm - r ] ) . the presence of the norm @xmath377 in the denominator of @xmath378 provides attraction to the nontrivial solution of ( [ norm - r ] ) , with @xmath379 the efficiency of this scheme has been demonstrated and it provides very fast convergence . the iteration scheme ( [ iteration ] ) works very well for @xmath345 and @xmath347 . for @xmath380 however , it was convenient to use in ( [ iteration ] ) the factor @xmath381 in the form : @xmath382 ^{1/4}.\ ] ] figures 1 and 2 show typical soliton shapes for @xmath383 when @xmath384 . numerically we checked that solitons decay at infinity like @xmath385 . with parameters @xmath316 , @xmath386.,width=453,height=302 ] with parameters @xmath318 , @xmath386.,width=453,height=302 ] for @xmath328 and @xmath318 the iteration scheme gave the soliton dependence ( fig . 3 ) , which is in agreement with the analytical result ( [ r - power],[power - par ] ) at very high accuracy ( for example , the factor @xmath387 at the stationary solution , presented in the paper , was equal to 1 within an error less than @xmath388 ) . the time step @xmath389 was equal to @xmath390 . all derivatives in the equations , as well as the action of the integral operator @xmath159 , were calculated by means of the standard fft program . the initial conditions for the iteration procedure ( [ iteration ] ) were taken in the form of solitons ( [ r - soliton ] ) with @xmath326 , which provide exponential decay while approaching the ends of the numerical interval on @xmath51 . we chose a symmetric numerical domain @xmath391 $ ] , with @xmath392 for all runs except the case @xmath328 where we took @xmath393 the soliton shape in this particular case is presented in fig . 3 . with parameters @xmath318 , @xmath328.,width=453,height=302 ] fig . 4 shows the dependence of @xmath104 on the soliton parameter @xmath91 . it is seen that the lower soliton branch and the upper soliton branch converge at large @xmath91 . between the upper and the lower curves we have a straight line corresponding to the critical solitons with @xmath394 . on @xmath91 for @xmath395,width=453,height=302 ] the same separation takes place for soliton amplitudes ( fig . 5 ) . at the point @xmath328 solitons undergo bifurcations . , width=453,height=302 ] for negative @xmath93 we have a supercritical soliton bifurcation , while a subcritical occurs when @xmath345 , with the soliton amplitude jump given by ( [ power - par ] ) . in this section we study the stability of the solitons that we analyzed in the previous section . in order to do that , we use the method based on lyapunov s theorem . applying this theorem to the present hamiltonian system means the following : a soliton considered as a stationary point of the hamiltonian @xmath100 for a fixed number of waves @xmath104 is stable if it realizes a minimum ( or maximum ) of the hamiltonian . if a stationary point represents a saddle point then the corresponding soliton is expected to be unstable . the latter is only an indication that solitons are unstable . for instance , the classical counterexample is the hamiltonian @xmath396 for a system with two degrees of freedom where the saddle point @xmath397 is stable . in fact this type of indication is already available if one looks at the scaling ( [ scaling - r],[scalingh ] ) for which @xmath398 where @xmath399 is the solitary wave solution . hence , it is seen that for negative @xmath93 the function @xmath400 is bounded from below and its minimum , @xmath401 corresponds to the soliton . on the contrary , for @xmath345 , this function has a maximum equal to @xmath365 , which is unbounded from below as @xmath402 . moreover , in the latter case , it is possible to see that the stationary point corresponding to the soliton solution is a saddle point . it follows immediately if , in addition to the scaling transformation , one considers the gauge transformation @xmath403 under which @xmath404 for @xmath324 , however , this transformation shows that the soliton solution remains the minimum point . we reach the same conclusion if we apply the vakhitov - kolokolov criterion @xcite to the soliton solutions . the criterion states that , if @xmath405 then solitons are unstable and they are stable in the opposite case . this criterion has a simple physical interpretation . the quantity @xmath406 is related to the energy of the soliton as a bound state . if by adding one particle ( i.e. increasing @xmath104 ) this level shifts towards the continuous spectrum , then obviously such bound state will be unstable . as we saw in the previous section the derivative @xmath407 is negative for @xmath345 and becomes positive when @xmath324 . we would like to emphasize that the vakhitov - kolokolov criterion was derived for the classical nls equation and , strictly speaking , can not be applied to our system . thus , we have again indications that support stability of the lower soliton branch and , respectively , instability for the upper branch . note that this stability indication is consistent with that for the cubic nls solitons ( [ nls - soliton ] ) which , as is well - known , are stable ( see , e.g. @xcite ) . now we show that the lower soliton branch ( @xmath408 ) is stable : solitons from this branch indeed realize a minimum of the hamiltonian @xmath100 for a fixed number of waves @xmath104 . the dimensionless hamiltonian @xmath100 written in terms of amplitude and phase reads @xmath409 dx . \label{ham1}\ ] ] the last term here is positive definite and vanishes exactly on the stationary solitons . thus , we now should prove that the minimum of @xmath100 is reached on the soliton obtained as solution of eq . ( [ norm - r ] ) . consider the integral @xmath410 dx\ ] ] and get its estimate through the integrals @xmath411 and @xmath104 . as was shown in @xcite , we have the following estimate in the absence of nonlocal terms : @xmath412 where @xmath413 and @xmath414 is the soliton solution for the quintic ( critical ) nls equation , @xmath415 this solution can be easily found and gives @xmath416 . for the integral @xmath417 , we can write the following set of inequalities : @xmath418 this inequality can be made sharper . to do that , consider the functional @xmath419 its minimum value will give the best constant @xmath420 . to find it one needs to determine a minimizer among all stationary points of the functional @xmath421 . the stationary points of @xmath421 are defined from the solutions of the equation @xmath422 the minimizer for @xmath421 is given by the ground soliton solution of this equation . it is a symmetric function without nodes . hence the best constant @xmath423 thus , @xmath424 this inequality allows to find the criterion for the hamiltonian to be bounded from below . substituting ( [ ineq ] ) into ( [ ham1 ] ) for @xmath324 yields @xmath425 \int r_x{}^2dx-\frac{1}{2}\int r^4\,dx . \label{ineq1}\ ] ] hence it is seen that the hamiltonian is bounded from below if @xmath426 or @xmath427 the final step in proving the hamiltonian boundedness is based on the estimate for the integral @xmath428 . according to @xcite @xmath429 substitution of this inequality into the estimate ( [ ineq1 ] ) results in the desired bound for @xmath100 : @xmath430 .\ ] ] it turns out that the numerical value of @xmath431 is almost the same as the critical number @xmath432 , defined by solitons with @xmath347 . thus , solitons from the lower branch which satisfy the criterion ( [ ineq2 ] ) are stable not only with respect to small perturbations but also against finite ones . concerning the solitons from the upper branch , they are all unstable with respect to finite disturbances . in summary , we would like to emphasize once more the difference between @xmath433 and @xmath434 although the derivative @xmath407 is positive for the whole lower soliton branch and according to the vakhitov - kolokolov criterion this branch is expected to be stable . up to now it is an open question but we think that it is so and @xmath433 and @xmath434 should coincide . thus , we have analyzed the behavior of solitons near the critical density ratio @xmath435 . above @xmath15 solitons undergo a subcritical bifurcation with the amplitude jump proportional to @xmath436 . therefore our theory based on the hamiltonian expansion works when this jump is small , i.e. in a small vicinity of @xmath435 . in order to describe solitons a generalized nls equation is derived based on the hamiltonian average over fast oscillations with the carrying frequency corresponding to the critical soliton velocity @xmath437 the generalized nls equation contains three kinds of nonlinear terms . the first one is nothing else than the nonlinear frequency shift taking into account six - wave nonlinear interactions . this is the local term . another nonlinearity is responsible for steepening the envelope solitons ( this is the so - called lifshitz term @xcite ) . and finally we have found the nonlocal contribution familiar to that calculated by dysthe for gravity waves . this dysthe term is focusing as well as the six - wave nonlinear interaction term . within the generalized nls equation we analyzed the stability of solitons . in particular , we have found a region of wave intensity , @xmath438 , where solitons are stable . their stability is based on the boundedness of the hamiltonian from below . for solitons above @xmath15 we have shown their possible instability , at least , instability with respect to finite perturbations . an interesting problem is the nonlinear stage of this instability , whether it can lead to the collapse of solitons . the latter question is important not only from the point of view of interfacial waves but also because there exists some similarity between the generalized nls equation derived in this paper and those describing the behavior of short optical pulses in fibers from the femtosecond range of pulse durations . the authors thank a.i . dyachenko for valuable discussions concerning the numerical aspects of this paper . two authors ( da and ek ) thank the centre de mathmatiques et de leurs applications ( cmla ) of ecole normale suprieure of cachan , where this paper was initiated . this paper was performed in the framework of the nato linkage grant est.clg.978941 . the work of da and ek was also supported by the russian foundation of basic research and by the program of russian academy of sciences `` mathematical methods in nonlinear dynamics '' . 99 m. s. longuet - higgins , _ capillary - gravity waves of solitary type on deep water _ , j. fluid mech . * 200 * , 451 - 470 ( 1989 ) . g. iooss and k. kirchgssner , _ bifurcation dondes solitaires en prsence dune faible tension superficielle _ , c. r. acad . paris , srie i * 311 * , 265 - 268 ( 1990 ) . vanden - broeck and f. dias , _ gravity - capillary solitary waves in water of infinite depth and related free - surface flows _ , j. fluid mech . * 240 * , 549 - 557 ( 1992 ) . f. dias and g. iooss , _ capillary - gravity solitary waves with damped oscillations _ , physica d * 65 * , 399 - 423 ( 1993 ) . m. s. longuet - higgins , _ capillary - gravity waves of solitary type and envelope solitons on deep water _ , j. fluid mech . * 252 * , 703 - 711 ( 1993 ) . t. r. akylas , _ envelope solitons with stationary crests _ , phys . fluids * 5 * , 789 - 791 ( 1993 ) . f. dias and g. iooss , _ capillary - gravity interfacial waves in deep water _ , eur . j. mech . b / fluids * 15 * , 367 - 390 ( 1996 ) . rusas and j. grue , _ solitary waves and conjugate flows in a three - layer fluid _ , eur . j. mech . b / fluids * 21 * , 185 - 206 ( 2002 ) . l. d. landau and e. m. lifshitz , _ statistical physics , part 1 _ ( pergamon press , new york ) [ russian original : nauka , moscow , 1995 , p. 521 ] . e. a. kuznetsov , _ hard soliton excitation : stability investigation _ , zhetf * 116 * , 299 - 317 ( 1999 ) [ jetp * 89 * , 163 - 172 ( 1999 ) ] . s. m. sun , _ some analytical properties of capillary - gravity waves in two - fluid flows of infinite depth _ , proc . a * 453 * , 1153 - 1175 ( 1997 ) . k. b. dysthe , _ note on a modification to the nonlinear schrdinger equation for application to deep water waves _ , proc . a * 369 * , 105 - 114 ( 1979 ) . m. j. ablowitz , j. hammack , d. henderson and c. m. schober , _ modulated periodic stokes waves in deep water _ , lett . * 84 * , 887 - 890 ( 2000 ) ; _ long - time dynamics of the modulational instability of deep water waves _ , physica * d 152 - 153 * , 416 - 433 ( 2001 ) . v. e. zakharov and e. a. kuznetsov , _ hamiltonian formalism for nonlinear waves _ , usp . fiz . nauk * 167 * , 1137 - 1168 ( 1997 ) [ physics - uspekhi * 40 * , 1087 - 1116 ( 1997 ) ] . v. e. zakharov and e. a. kuznetsov , _ optical solitons and quasisolitons _ , zh . ksp . . fiz . * 113 * , 1892 ( 1998 ) [ jetp * 86 * , 1035 - 1046 ( 1998 ) ] . a. sinitsyn and v. m. kontorovich , in : studies of turbulent structure of ocean , ed . by ozmidov , mgi , sevastopol ( 1975 ) ( first symposium `` issledovanie melkomaschtabnoi okeanicheskoi turbulentnosti '' , kaliningrad , 1974 ) ; v. m. kontorovich , _ a possible role of internal waves in occurence of small - scale turbulence in stratified ocean . vuzov , radiofizika * 19 * , 872 - 879 ( 1976 ) . e. a. kuznetsov and m. d. spector , _ existence of hexagon relief on the surface of the liquid dielectrics in an external electric field _ , zh . ksp . * 71 * , 262 ( 1976 ) [ sov . jetp * 44 * , 136 ( 1976 ) ] . e. a. kuznetsov , m. d. spector and v. e. zakharov , _ surface singularities of ideal fluid _ , phys.lett . * 182a * , 387 - 393 , ( 1993 ) . f. dias and t. bridges , _ geometric aspects of spatially periodic interfacial waves _ , stud . appl . math . * 93 * , 93 - 132 ( 1994 ) . v. e. zakharov , _ stability of periodic waves of finite amplitude on the surface of a deep fluid _ , zh . prikl . * 9 * , 86 - 94 ( 1968 ) [ j. appl . * 9 * , 190 - 194 ( 1968 ) ] ; v. e. zakharov and n. n. filonenko , _ the energy spectrum for stochastic oscillations of a fluid surface _ , doclady akad . nauk sssr * 170 * , 1292 - 1295 ( 1966 ) [ sov . * 11 * , 881 - 884 ( 1967 ) ] . t. b. benjamin and t. j. bridges , _ reappraisal of the kelvin - helmholtz problem . i. hamiltonian structure _ , j. fluid mech . * 333 * , 301 - 325 ( 1997 ) . w. craig and m. d. groves , _ normal forms for wave motion in fluid interfaces _ , wave motion * 31 * , 21 - 41 ( 2000 ) . w. craig , p. guyenne , and h. kalisch , _ hamiltonian long wave expansions for free surfaces and interfaces _ , comm . pure appl . math . * 58 * , 1587 - 1641 ( 2005 ) . zakharov , _ kolmogorov spectra in weak turbulence problems _ , in : handbook of plasma physics , vol . 2 , basic plasma physics , eds . galeev , r.n . sudan ( elsevier , north - holland , 1984 ) , pp . 3 - 36 . a. c. newell , _ solitons in mathematics and physics _ ( siam , philadelphia , 1985 ) . f. dias and e. a. kuznetsov , _ on the nonlinear stability of solitary waves for the fifth - order korteweg - de vries equation _ , phys . letters a * 263 * , 98 - 104 ( 1999 ) . o. laget and f. dias , _ numerical computation of capillary - gravity interfacial solitary waves _ , j. fluid mech . * 349 * , 221 - 251 ( 1997 ) . t. j. bridges , p. christodoulides and f. dias , _ spatial bifurcations of interfacial waves when the phase and group velocities are nearly equal _ , j. fluid mech . * 285 * , 121 - 158 ( 1995 ) . a. i. dyachenko , a. o. korotkevich , and v. e. zakharov , _ weak turbulent kolmogorov spectrum for surface gravity waves _ , phys . lett . * 92 * , 134501 ( 2004 ) . v. e. zakharov , v. s. lvov , _ the statistical description of the nonlinear wave fields _ , izv . vuzov , radiofizika * 18 * , 1470 - 1487 ( 1975 ) . g. iooss , _ existence dorbites homoclines un quilibre elliptique , pour un systme rversible _ , c. r. acad . paris * 324 * , 993 - 997 ( 1997 ) . e. a. kuznetsov and s. k. turitsyn , _ talanov transformations for selffocusing problems and instability of waveguides_. phys.lett . * 112a * , 273 ( 1985 ) . e. a. kuznetsov , a. m. rubenchik , and v. e. zakharov , _ soliton stability in plasmas and fluids _ , phys . rep . * 142 * , 103 ( 1986 ) . v. i. petviashvili , _ on the equation for unusual soliton , _ fizika plasmy , * 2 * , 469 - 472 ( 1976 ) [ sov . j. plasma phys . , * 2 * , 247 - 250 ( 1976 ) ] ; _ non - one - dimensional solitons _ , in : nonlinear waves gaponov - grekhov , moscow , nauka , 1979 , pp . 5 - 21 . n. g. vakhitov , a. a. kolokolov , _ stationary solutions of the wave equation in media with saturated nonlinearity _ , vuz radiofizika * 16 * , 1020 - 1028 ( 1973 ) [ radiophys . * 16 * , 783 ( 1973 ) ] . as pointed out in section iii , the calculation of the nonlinear coupling coefficients @xmath121 , @xmath122 and of the six - wave coefficient @xmath130 can be performed independently for each coefficient . therefore for the coefficient @xmath130 this problem reduces to the calculation of the nonlinear frequency shift for the main harmonic with @xmath65 and @xmath439 when the first contribution to the nonlinear frequency shift proportional to @xmath440 is equal to zero . second , in such a case , it is enough to consider its limit of a monochromatic wave , instead of a quasi - monochromatic wave . this means that one needs to develop the perturbation theory assuming that @xmath441 where the leading order corresponds to the main harmonic @xmath442 and amplitudes of all other ( combined ) harmonics are supposed to be small in comparison with the first harmonic amplitude . after this remark we introduce the small parameter @xmath443 so that @xmath444 where @xmath445 is a slow time and @xmath446 is the envelope of the main harmonic . hence it is easy to see that all needed combined harmonics @xmath447 become functions of the slow time @xmath103 and take in their expansion powers of the parameter @xmath443 : @xmath448 note that in this case nonlinear interactions with the zeroth harmonic do not give any contribution because the corresponding matrix elements vanish by the same law as for three - wave interaction . the equations of motion for amplitudes @xmath447 follow from the general equation ( [ k - rep ] ) for normal amplitude @xmath41 : @xmath449 c_n =- i\frac{\partial \overline{h}_{\mathrm{int}}}{\partial c_n^{*}}. \label{app-4}\ ] ] here @xmath450 is mean value of the interaction hamiltonian after substitution expressions ( [ app-1 ] ) into ( [ h - int ] ) and average . @xmath450 contains 19 matrix coefficients needed to determine the constant @xmath130 : @xmath451 where @xmath452 , @xmath453 , and so on . these 19 coefficients are calculated in accordance with the rules explained in the third section . as is seen from the equation of motion ( [ app-4 ] ) the time derivative of @xmath447 is small compared with the second term @xmath454 c_n$ ] , except the main harmonic where this term vanishes . in its turn , the equation for @xmath455 contains terms proportional to @xmath456 and @xmath457 . the terms of the third order ( @xmath458 @xmath456 ) are nothing more than the relation ( [ crit - ratio ] ) serving for definition of the critical atwood number : @xmath459 the equation of motion for @xmath85 appears in the fifth order of @xmath443 : @xmath460 where all matrix elements are defined without factors of @xmath461 as they had according to the definition of section iii ( compare with ( [ u ] ) , ( [ v ] ) , [ t-4 ] ) ) . here the amplitudes @xmath462 are found with the help of equations ( [ app-4 ] ) . this is a pure algebraic procedure which we performed with the help of computer . as the result of substitution of @xmath463 into the above equation we arrive at the equation : @xmath464 the corresponding nonlinear term is focusing .

we extend the concept of the functional renormalization for quantum many - body problems to nonequilibrium situations . using a suitable generating functional based on the keldysh approach , we derive a system of coupled differential equations for the-particle vertex functions . the approach is completely general and allows calculations for both stationary and time - dependent situations . as a specific example we study the stationary state transport through a quantum dot with local coulomb correlations at finite bias voltage employing two different truncation schemes for the infinite hierarchy of equations arising in the functional renormalization group scheme .

cond-mat0609457

the reliable calculation of physical properties of interacting quantum mechanical systems presents a formidable task . typically , one has to cope with the interplay of different energy - scales possibly covering several orders of magnitude even for simple situations . approximate tools like perturbation theory , but even numerically exact techniques can usually handle only a restricted window of energy scales and are furthermore limited in their applicability by the approximations involved or the computational resources available . in addition , due to the divergence of certain classes of feynman diagrams , some of the interesting many - particle problems can not be tackled by straight forward perturbation theory . the situation becomes even more involved if one is interested in properties off equilibrium , in particular time - dependent situations . a standard approach for such cases is based on the keldysh formalism@xcite for the time evolution of green functions , resulting in a matrix structure of propagators and self - energies . this structure is a direct consequence of the fact that in nonequilibrium we have to calculate averages of operators taken not with respect to the ground state but with respect to an arbitrary state . therefore , the gell - mann and low theorem@xcite is not valid any more . other approaches attempt to treat the time evolution of the nonequilibrium system numerically , for example using the density matrix renormalization group@xcite , the numerical renormalization group ( nrg)@xcite or the hamiltonian based flow - equation method.@xcite the keldysh technique shows a big flexibility and one therefore can find a wide range of applications such as transport through atomic , molecular and nano devices under various conditions,@xcite systems of atoms interacting with a radiation field in contact with a bath,@xcite or electron - electron interaction in a weakly ionized plasma.@xcite one powerful concept to study interacting many - particle systems is the rather general idea of the renormalization group@xcite ( rg ) , which has also been applied to time - dependent and stationary nonequilibrium situations recently.@xcite in the rg approach one usually starts from high energy scales , leaving out possible infrared divergences and works ones way down to the desired low - energy region in a systematic way . however , the precise definition of `` systematic way '' does in general depend on the problem studied . in order to resolve this ambiguity for interacting quantum mechanical many - particle systems in equilibrium , two different schemes attempting a unique , problem independent prescription have emerged during the past decade . one is wegner s hamiltonian based flow - equation technique,@xcite the second a field theoretical approach , which we want to focus on in the following . this approach is based on a functional representation of the partition function of the system and has become known as functional renormalization group ( frg).@xcite . a detailed description of the various possible implementations of the frg and its previous applications in equilibrium can be found e.g. in refs . and . in the present work we extend the frg to nonequilibrium by formulating the problem on the real instead of the imaginary time axis using the functional integral representation of the action on the keldysh contour . within a diagrammatic approach a similar set of frg flow equations has already been derived by jakobs and schoeller and applied to study nonlinear transport through one - dimensional correlated electron systems.@xcite we believe that this method will enable us to treat a variety of nonequilibrium problems within a scheme which is well established in equilibrium and in contrast to other approaches is comparatively modest with respect to the computer resources required . our framework for nonequilibrium will turn out to be sufficiently general to allow for a treatment of systems disturbed by arbitrary external fields ( bias voltage , laser field , etc . ) , which can be constant or time - dependent . for classical many - body problems the frg ( also called nonperturbative renormalization group ) was used to study nonequilibrium transitions between stationary states.@xcite as a simple but nontrivial application to test the potential and weakness of our implementation of a nonequilibrium frg we choose the single impurity anderson model ( siam).@xcite this model represents the paradigm for correlation effects in condensed matter physics and is at the heart of a large range of experimental@xcite and theoretical investigations.@xcite it furthermore is the standard model for the description of the transport properties of interacting single - level quantum dots . the paper is organized as follows . in sec . [ formalism ] and the appendices we extend the equilibrium frg to treat general nonequilibrium quantum many - body problems . as an example we study the finite bias , stationary transport through an anderson impurity in sec . [ appli ] . within the nonequilibrium extension we obtain a system of coupled tensor equations , which represent the flow of the different components ( in the keldysh space ) of the self - energy and the vertex function . we discuss two approximations derived from the frg scheme that have been successfully applied to the equilibrium situation.@xcite a summary and outlook in sec . [ summary ] concludes the paper . we start from the standard definition of the two - time green function @xmath1 in the interaction picture , where @xmath2 comprises a set of single - particle quantum numbers and time @xmath3 . the time evolution operators are given as @xmath4 \label{eq : anti_time_evolve } s^{-1 } & = & \tilde{\rm t } \exp\left\{i\int\limits^{\infty}_{-\infty}v_i(t)dt\right\}\;\;,\end{aligned}\ ] ] with @xmath5 the usual time ordering and @xmath6 the anti time - ordering operator . the interaction term @xmath7 is arbitrary , including possible explicit time dependence . [ cb][cb]@xmath8 [ ct][ct]@xmath9 [ ct][ct]@xmath3 [ rc][rc]@xmath10 [ lc][lc]@xmath11 in a nonequilibrium situation , the propagation of the system from @xmath12 is not any more equivalent to the propagation from @xmath13 , i.e. one has to distinguish whether the time arguments in eq . ( [ eq : standard_gf ] ) belong to the former or latter.@xcite this scheme is usually depicted by the keldysh double - time contour shown in fig . [ fig : keldysh_contour ] , where @xmath8 represents the propagation @xmath14 ( upper branch of the keldysh contour ) and @xmath9 the propagation @xmath13 ( lower branch of the keldysh contour ) . consequently , one has to introduce four distinct propagators , namely the time - ordered green function @xmath15 where @xmath16 , the anti time - ordered green function @xmath17 with @xmath18 and @xmath19 where @xmath20 for bosons and @xmath21 for fermions . concerning the keldysh indices we here follow the notation of ref . @xmath22 and @xmath23 take into account the excitation spectrum while @xmath24 and @xmath25 describe the thermodynamic state of the system . only three of the green functions are independent and one commonly introduces the linear combinations @xmath26\\ \label{eq : gadv } g^{\rm a}(\xi',\xi ) & = & \theta(t - t')\left[g^{-+}(\xi',\xi)+\zeta g^{+-}(\xi',\xi)\right]\\ \label{eq : gkeldysh } g^{\rm k}(\xi',\xi ) & = & g^{-+}(\xi',\xi)+g^{+-}(\xi',\xi)\;\;,\end{aligned}\ ] ] named retarded , advanced and keldysh component , respectively . to derive a nonequilibrium frg scheme for an interacting many - body problem along the lines of refs . and one needs a formulation that allows to express the @xmath0-particle vertices as functional derivatives of a generating functional . here we use the approach by kameneev.@xcite to set up a functional integral representation of the generating functional we define the matrix @xmath27 and the short hand notation @xmath28 where @xmath29 is a spinor of fields ( grassmann for fermions or complex for bosons ) with @xmath30 having a time from the upper branch of the keldysh contour and @xmath31 a time from the lower . later it will also prove useful to fourier transform from time @xmath3 to frequency @xmath32 . one then has to replace @xmath3 in @xmath2 by @xmath32 . the integrals over @xmath2 and @xmath33 stand for summations over the quantum numbers and integrations over time or frequency . the following steps can be performed with @xmath2 either containing time or frequency . the generalization of the functional integral representation of the partition function to nonequilibrium is@xcite @xmath34^{-1}\psi \right ) - is_{\rm int}\left(\ { \bar \psi \ } , \{\psi\}\right ) \right\ } } \ ; , \;\ ; \xi_0= \int{\mathcal d } \bar \psi \psi \exp{\left\{\left(\bar\psi,\left[\hat{g}_0\right]^{-1}\psi\right)\right\}}\;\;.\end{aligned}\ ] ] the matrix @xmath35 denotes the propagator of the noninteracting part of the many - body problem and @xmath36 is the ( arbitrary ) interaction term . to construct a generating functional for @xmath0-particle green functions , one introduces external source fields @xmath37 and @xmath38 @xmath39^{-1 } \psi\right ) - is_{\rm int}(\{\bar \psi \ } , \ { \psi \ } ) - \left ( \bar \psi , \eta\right ) - \left(\bar \eta , \psi \right ) \bigg\ } \;.\ ] ] the ( connected ) @xmath0-particle green function @xmath40 can then be obtained by taking functional derivatives @xmath41 with @xmath42 } \ ; .\end{aligned}\ ] ] the derivatives in eq . ( [ eq : gmdef ] ) are taken with respect to the spinors @xmath37 and @xmath43 , i.e. the resulting green function @xmath40 is a tensor of rank @xmath44 in the keldysh indices . in the following we adopt the notation that quantities with explicit index @xmath0 are tensors of rank @xmath44 ; without index @xmath0 we denote the propagator and self - energy , both carrying a hat to point out their matrix structure . introducing the fields @xmath45 we can perform a legendre transformation @xmath46^{-1}\phi\right ) \ ; , \end{aligned}\ ] ] to the generating functional of the one - particle irreducible vertex functions @xmath47 ( tensors of rank @xmath44 in the keldysh indices ) @xmath48 in contrast to the usual definition@xcite of @xmath49 for convenience ( see appendix a ) we added a term @xmath50^{-1 } \phi\right)$ ] in eq . ( [ eq : gammadef ] ) . the relations between the @xmath40 and @xmath47 in imaginary time can be found in text books@xcite and can straightforwardly be extended to real times on the keldysh contour . for example , for the one - particle green function one obtains @xmath51^{-1 } \right]^{-1}_{\xi',\xi } \end{aligned}\ ] ] where @xmath52^{-1 } - \hat{\sigma}\right)^{-1}_{\xi',\xi } , \end{aligned}\ ] ] with the proper one - particle self - energy @xmath53 . this implies the relation @xmath54 . we have a matrix structure not only with respect to @xmath2 and @xmath33 ( as in equilibrium ) , but also with respect to the keldysh indices . how does this additional structure manifests itself in the green functions ( [ eq : gmdef ] ) ? for @xmath55 we have two derivatives with respect to spinor fields , i.e. the structure of a tensor product , which can be made explicit by using a tensor product notation @xmath56 leading to the matrix @xmath57 \frac{\delta^2 { \mathcal w}^c } { \delta \bar \eta_{+}(\xi')\delta \eta_{-}(\xi ) } & \frac{\delta^2 { \mathcal w}^c } { \delta \bar \eta_{+}(\xi')\delta \eta_{+}(\xi ) } \end{array}\right)\nonumber\\[2 mm ] & = & \left(\begin{array}{cc } g^{--}(\xi',\xi ) & g^{-+}(\xi',\xi ) \\[2 mm ] g^{+-}(\xi',\xi ) & g^{++}(\xi',\xi ) \end{array}\right)\;\;.\end{aligned}\ ] ] up to now we extended standard text book manipulations@xcite from imaginary time to the real - time keldysh contour . we emphasize that no translational invariance in time is assumed . to derive the frg flow equations in nonequilibrium we can follow the steps of ref . . in eqs . ( [ eq : partitionfunction ] ) and ( [ eq : genfunct ] ) we replace the noninteracting propagator by a propagator @xmath58 depending on a parameter @xmath59 $ ] and require @xmath60 i.e. at the starting point @xmath61 no degrees of freedom are `` turned on '' while at @xmath62 the @xmath63 free problem is recovered . in models with infrared divergences @xmath63 can be used to regularize the problem . in equilibrium this is often be achieved by implementing @xmath63 as an infrared cutoff in momentum or energy . one of the advantages of the frg approach over other rg schemes is that one is not restricted to these choices and other ways of introducing the parameter @xmath63 have turned out to be useful for equilibrium problems.@xcite all that is required to derive the fundamental flow equations ( [ eq : gamma1f2 ] ) and ( [ eqq1 ] ) ( see below ) are the conditions eq.([eq : demands ] ) . in our application of the nonequilibrium frg to the steady state transport through an interacting quantum dot it is natural to implement @xmath63 as an energy cutoff . however , such a choice must not be the natural one in cases where one is interested in studying transient behavior . in this situation the propagator and the vertex functions in general depend on several spatial and time variables and there is no obvious momentum or energy cutoff scheme . within the frg several ways of introducing @xmath63 can be worked out , compared and the one best suited for the problem under investigation can be identified . through @xmath58 the quantities defined in eqs . ( [ eq : partitionfunction ] ) to ( [ eq : vertexfundef ] ) acquire a @xmath63-dependence . taking the derivative with respect to @xmath63 one now derives a functional differential equation for @xmath64 . from this , by expanding in powers of the external sources , an infinite hierarchy of coupled differential equations for the @xmath65 is obtained . although the steps in the derivation are formally equivalent to ref . , because of the real - time formulation additional factors @xmath66 and signs appear in several places . we thus believe that it is helpful to present the details of the derivation for the present approach , which is done in appendix a. [ br][br]@xmath2 [ br][br]@xmath33 diagrammatic form of the flow equation for @xmath67 . the slashed line stands for the single scale propagator @xmath68.,title="fig:",scaledwidth=20.0% ] in particular , for the flow of the self - energy one finds the expression @xmath69 \ ; , \end{aligned}\ ] ] which can be visualized by the diagram in fig . [ fig : gamma1 ] . the trace in eq . ( [ eq : gamma1fl ] ) is meant to run over all quantum numbers and time respectively frequency . in eq . ( [ eq : gamma1fl ] ) appears the so - called single scale propagator ( the slashed line in fig . [ fig : gamma1 ] ) @xmath70^{-1 } \ ; .\end{aligned}\ ] ] and the quantity @xmath71 denotes the matrix obtained by keeping the indices @xmath2 and @xmath33 fixed . [ br][br]@xmath2 [ br][br]@xmath33 diagrammatic form of the flow equation for @xmath72 . the slashed line stands for the single scale propagator @xmath68 , the unslashed line for @xmath73.,title="fig:",scaledwidth=45.0% ] we thus arrive at an expression that is formally identical to eq . ( 19 ) in ref . . the difference appears in the matrix structure , which now also contains the index components for the branches of the keldysh contour . to make this explicit , we write out eq . ( [ eq : gamma1fl ] ) with respect to the keldysh structure @xmath74 where @xmath75 and @xmath76 denote keldysh indices and take the values @xmath77 . apparently , the derivative of @xmath78 is determined by @xmath79 and the two - particle vertex @xmath80 . thus an equation for @xmath81 is required . the structure of this equation turns out to be similar to eq . ( 21 ) of ref . . here we only show the diagrams representing it in fig . [ fig : gamma2 ] , the full expression is given in eq . ( [ eqq1 ] ) . the differential equation for @xmath82 does not only contain @xmath67 implicitly via the propagators and @xmath72 , but also the three - particle vertex @xmath83 . the flow of the three - particle vertex depends on the four - particle vertex etc . it is generically impossible to solve the full set of infinitely many coupled differential equations . in applications one has to truncate it , and this is usually done at order @xmath84 , i.e. one replaces all vertices with @xmath85 by their initial values , which for problems with a two - particle interaction means @xmath86 for @xmath85 . even within this truncated system the remaining set of differential equations must typically further be approximated to allow for an analytical or a numerical solution.@xcite at the end of the frg flow , that is for @xmath62 , the @xmath87 present approximations to the many - body vertex functions from which observables can be computed ( for examples see sec . [ appli ] ) . we again note that a similar set of flow equations was derived by jakobs and schoeller using a diagrammatic approach.@xcite the derivation of the flow equations in the previous section and the appendices was completely general and they can be used to study time dependent or stationary properties of systems at finite temperature @xmath88 or @xmath89 . a simple but nonetheless nontrivial application is nonequilibrium , stationary transport through an interacting single - level quantum dot at @xmath89 . for a stationary situation , time translational invariance holds and one can fourier transform to frequency space . moreover , because the quantum dot is a zero - dimensional structure , no additional degrees of freedom except spin have to be taken into account , thus considerably reducing the complexity of the problem . nonequilibrium theory of single - level quantum dots described by the siam ( see below ) has been a major subject of research over the past years . several techniques have been developed respectively applied , for example statistical@xcite , perturbative,@xcite rg based@xcite and numerical procedures.@xcite in particular the rg based calculations were up to now restricted to situations away from the kondo regime , either by looking at features in the mixed - valence state @xcite or in strong magnetic fields.@xcite despite this tremendous ongoing effort , relatively little is known about nonequilibrium properties , even in the stationary state , of the model . on the other hand , the very detailed knowledge of the _ equilibrium _ properties@xcite makes it possible to interpret certain features or even dismiss approximations on the basis of this fundamental understanding . the standard model to describe transport through interacting quantum dots is the siam.@xcite experimentally,@xcite the dot region is attached to two external leads , and the current is driven by an applied bias voltage @xmath90 . furthermore , the filling of the dot can be controlled by a gate voltage @xmath91 . to keep the notation simple we use units with @xmath92 . the hamiltonian reads @xmath93 \end{aligned}\ ] ] in standard second quantized notation ( @xmath94 ) . as usual , @xmath95 is the wave vector of the band states in the leads and @xmath96 the spin . in addition , the index @xmath97 distinguishes the left and right reservoirs which can have different chemical potentials @xmath98 through an applied bias voltage @xmath99 . since we do not include a magnetic field lifting the spin - degeneracy of the spin up and down level , the one - particle green function and the self - energy are spin independent and we suppress the spin index . we assume the dispersions @xmath100 and hybridizations @xmath101 between dot and the left and right leads to be identical , and @xmath102 to be @xmath95-independent . equation ( [ eq : siam ] ) is written such that for @xmath103 we have particle - hole symmetry . the interaction in the hamiltonian is reduced to the dot site only . we can thus reduce the problem to a zero - dimensional one by `` integrating out '' the leads . for the flow equations we need the propagator @xmath104 at @xmath105 . to this end we use the dyson equation to obtain@xcite @xmath106^{-1}\;\;,\ ] ] where @xmath107 g^{+-}_{d,0 } & g^{++}_{d,0 } \end{array}\right)\;,\nonumber\\ \label{eq : hybridization } \hat{\sigma}_{\rm lead } & = & \frac{v^2}{2n}\sum_{\vec{k } l } \left(\begin{array}{cc } g^{--}_{\vec{k } l } & g^{-+}_{\vec{k } l } \\[2 mm ] g^{+-}_{\vec{k } l } & g^{++}_{\vec{k } l } \end{array}\right)\;\;.\end{aligned}\ ] ] to further evaluate eq . ( [ eq : hybridization ] ) , we have to insert the green functions of the free electron gas , given by @xmath108{l}\displaystyle \frac{1}{\omega-\varepsilon_{\vec{k}}+\mu_l+i\delta } \\[5 mm ] \displaystyle + 2i\pi f(\varepsilon_{\vec{k } } ) \delta(\omega-\varepsilon_{\vec{k}}+\mu_l)\;,\end{array}\\ g^{++}_{\vec{k } l}(\omega ) & = & -\left[g_{\vec{k } l}^{--}(\omega)\right]^{*}\;,\\ g_{\vec{k } l}^{-+}(\omega ) & = & 2i\pi f(\varepsilon_{\vec{k}})\delta(\omega-\varepsilon_{\vec{k}}+\mu_l)\;,\\ g_{\vec{k } l}^{+-}(\omega ) & = & -2i\pi f(-\varepsilon_{\vec{k } } ) \delta(\omega-\varepsilon_{\vec{k}}+\mu_l)\ ; , \end{aligned}\ ] ] into eq . ( [ eq : kop1 ] ) . this leads to @xmath109 } { ( \omega - v_g)^2+\gamma^2}\ ; , \\ \label{eq : kulp2}g_{d,0}^{++}(\omega)&=&-[g_{d,0}^{--}(\omega)]^*\;,\\ \label{eq : pulp1}g_{d,0}^{-+}(\omega)&=&i\frac{\gamma\left [ f_l(\omega)+f_r(\omega)\right]}{(\omega - v_g)^2+\gamma^2}\;,\\ \label{eq : tulp1}g_{d,0}^{+-}(\omega)&=&-i\frac{\gamma\left [ f_l(-\omega)+f_r(-\omega)\right]}{(\omega - v_g)^2+\gamma^2}\;,\end{aligned}\ ] ] where @xmath110 are the fermi functions of the leads and @xmath111 , with @xmath112 the density of states at the fermi level of the semi - infinite leads , represents the tunnel barrier between the leads and the impurity . finally , the quantities we want to calculate are the current @xmath113 through the dot and the differential conductance @xmath114 . for the model ( [ eq : siam ] ) the current is given by@xcite @xmath115 \left[g^{+-}_d(\epsilon)-g^{-+}_d(\epsilon)\right]\;,\end{aligned}\ ] ] where we already used that the left and right couplings are identical , i.e. @xmath116 , and summed over both spin directions . the interacting one - particle green function of the dot is denoted by @xmath117 and @xmath118 are the currents across the left and right dot - lead contacts respectively . equation ( [ eq : current ] ) is written in a somewhat unusual form , not employing the relation @xmath119 , where @xmath120 denotes the dot s one - particle spectral function . the reason for this will be explained below . another quantity of interest is @xmath121.@xcite obviously , since no charge is produced on the quantum dot , @xmath122 in the exact solution . using again the results of refs . and , the expression for @xmath123 becomes @xmath124 - 2\im m\sigma^{-+}(\omega)}{\tilde\delta(\omega)}\\[2 mm ] \tilde\delta(\omega ) & = & \left|\omega - v_g+i\gamma\left[1-f(\omega)\right]-\sigma^{--}(\omega)\right|^2+\left[\gamma f(-\omega)+\im m\sigma^{+-}(\omega)\right]\left[\gamma f(\omega)-\im m\sigma^{-+}(\omega)\right]\nonumber\\[2 mm ] \label{eq : fofw } f(\omega ) & = & f_l(\omega)+f_r(\omega)\end{aligned}\ ] ] where we used that @xmath125 and @xmath126 are purely imaginary . depending on the type of approximation used @xmath127 might either hold for all parameters@xcite or not.@xcite we note that fulfilling @xmath122 is , however , not sufficient for an approximation to provide reliable results . e.g. the self - consistent hartree - fock approximation fulfills @xmath122 , but nonetheless does not capture the correct physics even in equilibrium.@xcite before studying the coupled system for @xmath128 and @xmath82 we begin with the simpler case where we replace @xmath82 on the right hand side of eq . ( [ eq : gamma1f2 ] ) by the antisymmetrized bare interaction and consider only the flow of @xmath129 . the only nonzero components of the bare two - particle vertex are those with all four keldysh indices being the same ( @xmath130)@xcite @xmath131 with this replacement eq . ( [ eq : gamma1f2 ] ) reduces to @xmath132 within this approximation the self - energy is always time or frequency independent , and no terms off - diagonal in the keldysh contour indices are generated . it leads to at least qualitatively good results in equilibrium,@xcite and has the additional advantage that the flow equations can be solved analytically . as the last step we specify how the parameter @xmath63 is introduced . since we are interested in a stationary situation , i.e.the propagators only depend on the time difference @xmath133 , all equations can be transformed into frequency space and one natural choice is a frequency cutoff of the form @xmath134 with @xmath135.@xcite evaluating @xmath136 by means of the morris lemma@xcite results in @xmath137^{-1}-\hat\sigma^{\lambda}}\;\;.\ ] ] a straightforward calculation permits us thus to rewrite eq . ( [ eq : pot ] ) as @xmath138 where @xmath139 finally , the initial condition for the self - energy is @xmath140.@xcite we now focus on @xmath89 . in a first step we discuss the equilibrium situation , that is @xmath141 . then we obtain the decoupled system @xmath142},\ ] ] which can be solved analytically . we first note that with @xmath143^*$ ] both equations are equivalent . for @xmath144 we obtain with the definition @xmath145 the solution @xmath146 where @xmath147 and @xmath148 are the bessel functions of first and second kind . the desired solution of the cutoff free problem is obtained by setting @xmath62 , i.e. @xmath149 which is precisely the result eq . ( 4 ) obtained by @xcite it is , however , important to note that in this work , based on the imaginary - time formulation of the frg , the differential equation has a different structure . it is real and has a positive definite denominator . thus , while the solutions at @xmath62 are identical for the imaginary - time and real - time formulations , the flow towards @xmath62 will show differences . as we will see next , the complex nature of the differential equation ( [ eq : ap1 ] ) can lead to problems connected to its analytical structure when attempting a numerical solution . for small @xmath150 no particular problems arise . as an example the result for the flow of @xmath144 as function of @xmath63 for @xmath151 and @xmath152 obtained with a standard runge - kutta solver is shown in fig . [ fig : feq_ok ] . consistent with the analytical solution eq . ( [ eq : analytic0 ] ) , the imaginary part ( dashed line ) goes to zero as @xmath153 , while the real part ( solid line ) rapidly approaches the value given by formula ( [ eq : analytic0 ] ) . with @xmath154 for @xmath151 , @xmath152 , and @xmath155 . the full curve shows the real part , the dashed the imaginary part of @xmath144.,scaledwidth=40.0% ] with @xmath154 for @xmath156 , @xmath157 , and @xmath155 . the full and dashed curves show real and imaginary part obtained from the integration @xmath158 , the dashed - dotted and dotted curves real and imaginary part obtained from an integration @xmath159 , using the solution from ( [ eq : analytic0 ] ) as initial value for @xmath144 . the crosses denote the analytical solution ( [ eq : analytical_solution]).,scaledwidth=40.0% ] however , for larger values of @xmath150 the numerical solution becomes unstable in a certain regime of @xmath91 . a typical result in such a situation is shown in fig . [ fig : feq_not_ok ] . the different curves were obtained as follows : the full and dashed ones from the numerical solution starting with @xmath160 at @xmath161 , the dash - dotted and dotted by integrating the differential equation ( [ eq : ap1 ] ) backwards from @xmath62 with the correct solution for @xmath62 as given by formula ( [ eq : analytic0 ] ) as initial value . the crosses finally are the results from the analytical solution eq . ( [ eq : analytical_solution ] ) . evidently , there exists a crossing of different branches of solutions to the differential equation for @xmath162 and the numerical solution with starting point @xmath163 picks the wrong one as @xmath153 . the reason for this behavior is that for large @xmath164 there exists a certain @xmath165 such that @xmath166 with real @xmath167 , resulting in a pole in the differential equation ( [ eq : ap1 ] ) . for @xmath168 this pole does not appear for real @xmath167 , but as shown in fig . [ fig : solution ] @xmath169 changes sign at @xmath165 , which in turn induces a sign change on the right hand side of the differential equation , leading to the behavior observed in fig . [ fig : feq_not_ok ] . there also exists a second critical value @xmath170 such that for @xmath171 we find @xmath172 and the instability has vanished again . determined from ( [ eq : analytical_solution ] ) and @xmath173 for different values of @xmath164 and @xmath155 as function of @xmath91 . for @xmath156 and @xmath174 there exist an interval @xmath175 $ ] where @xmath176 , while for small @xmath164 or @xmath177 $ ] we always have @xmath172.,scaledwidth=40.0% ] obviously , this instability limits the applicability of the present approximation to sufficiently small values of @xmath164 . this is different from the imaginary - time approach by andergassen _ et al._,@xcite where this simple approximation leads to qualitative correct results even for values of @xmath164 significantly larger than @xmath49 . we now turn to the case of finite bias voltage @xmath90 . as a typical example , the flow of @xmath144 for @xmath151 ( full and dashed curves ) and @xmath178 ( dashed - dotted and dotted curves ) for @xmath152 at @xmath179 ( equilibrium ) and @xmath180 is shown in fig . [ fig : noneq_simple ] . since the results for @xmath181 are related to those for @xmath144 by @xmath143^*$ ] we do not show them here . the @xmath90 dependence of the curves for @xmath182 $ ] ( thick lines ) looks sensible . for @xmath183 an imaginary part of order @xmath184 is generated in the flow which does not vanish for @xmath185 ( see the thin dotted line ) . causality requires that the relation @xmath186\ ] ] must hold for the exact solution . because of @xmath187 , the finite imaginary part of @xmath188 leads to a breaking of the condition ( [ eq : causality ] ) to order @xmath184 at the end of the frg flow . this is consistent with the fact that by neglecting the flow of the vertex terms of order @xmath184 are only partially kept in the present frg truncation scheme . to avoid any confusion we emphasize that our rg method is different from any low - order perturbation theory . the weak breaking of causality can also be understood as a consequence of our approximation leading to a _ complex _ , energy - independent self - energy : the off - diagonal components , being related to the distribution functions for electrons and holes , respectively , in general have different support on the energy axis . the energy independence makes it impossible to respect this structure here . with @xmath63 for @xmath151 and @xmath178 for @xmath152 at @xmath179 and @xmath189 ( thick curves : real part ; thin curves : imaginary part ) . the curve for @xmath190 at @xmath151 and @xmath180 ( thin dashed line ) lies on top of the corresponding zero bias curve and is thus not visible.,scaledwidth=40.0% ] for our further discussion the order @xmath184 violation of eq . ( [ eq : causality ] ) means that we may not rely on relations like eqs . ( [ eq : gret])-([eq : gkeldysh ] ) but have to work with @xmath191 , thus the somewhat unusual formula ( [ eq : current ] ) . a naive application of @xmath192 and use of @xmath193 would have led to unphysical results . that working with @xmath194 is still sensible can be seen from a straightforward evaluation leading to @xmath195-\sigma^{--}\right|^2+\gamma^2f(\omega)f(-\omega)}\nonumber \\[5 mm ] & & f(\omega)= f_l(\omega)+f_r(\omega)\nonumber\end{aligned}\ ] ] which is purely imaginary with a definite sign . inserting the expression ( [ eq : gpmmgmp_simple ] ) into the formula ( [ eq : current ] ) , one can calculate the current and thus the conductance . since we are at @xmath89 , an explicit expression for the current of the cutoff free problem ( at @xmath62 ) can be obtained by noting that with @xmath196 , @xmath197 one has @xmath198 and @xmath199 for @xmath200 $ ] , which leads to @xmath201 with the abbreviations @xmath202 \label{eq : abbriv_alpha } \gamma^\ast & = & \sqrt{\gamma^2+(\im m\sigma^{--})^2}\;.\end{aligned}\ ] ] equation ( [ eq : current_simple ] ) for the current is equivalent to the noninteracting expression but with renormalized parameters @xmath203 and @xmath204 , which depend on the interaction as well as the bias and gate voltage . as function of @xmath91 for @xmath205 and several values of the bias voltage @xmath90.,scaledwidth=40.0% ] an example for the differential conductance as function of @xmath91 obtained from eq . ( [ eq : current_simple ] ) for @xmath205 and several values of @xmath90 is shown in fig . [ fig : conductance_simple ] , where @xmath206 ( after reintroducing @xmath207 and @xmath208 ) . increasing @xmath90 leads , as expected , first to a decrease of the conductance close to @xmath103 and later to a splitting of order @xmath90 . since we will discuss a more refined scheme including parts of the flow of the two - particle vertex next , we do not intend to dwell too much on the results of this simplest approximation . we note in passing that for @xmath103 due to particle - hole symmetry we obtain from the differential equation ( [ ap ] ) that @xmath209 independent of @xmath164 . consequently the current @xmath113 calculated via eq . ( [ eq : current_simple ] ) and the conductance are independent of @xmath164 , too , and given by the corresponding expressions for the noninteracting system . as we will see in the next section , this deficiency will be cured by the approximate inclusion of the vertex flow . in the present approximation the current conservation @xmath122 holds for all parameters as @xmath210 [ cf . ( [ eq : deltaj2 ] ) ] . a more refined approximation is obtained when we insert the flowing two - particle vertex @xmath82 as given by expression ( [ eqq2 ] ) in the calculation of the self - energy eq . ( [ eq : gamma1f2 ] ) . by this we introduce an energy - dependence of the self - energy.@xcite however , because the size of the resulting system of differential equations becomes extremely large if the full frequency dependence is kept ( for a discussion on this in equilibrium see ref . ) , we only keep the flow of the frequency independent part of the vertex , an additional approximation which has successfully been used in equilibrium.@xcite as a consequence we again end up with a frequency independent @xmath128 . the resulting expression for the self - energy ( see appendix b ) is @xmath211 where @xmath212 is the flowing interaction given by the expression ( [ eqq3 ] ) . as has been observed by karrasch _ et al._,@xcite this approximation leads to a surprisingly accurate description of the transport properties in equilibrium . in particular it is superior to the lowest order approximation including only the bare vertex . with @xmath63 for @xmath151 ( left panel ) and @xmath156 ( right panel ) at @xmath213 and @xmath155 . the full curves show the real part , the dashed the imaginary part of @xmath144 . the stars at the vertical axis denote the values as obtained from the imaginary - time frg.@xcite , scaledwidth=40.0% ] we again begin with the discussion of the solution to eq . ( [ eq : sigme_full ] ) in equilibrium . results for the flow of @xmath144 are presented in fig . [ fig : full_eq ] for @xmath151 ( left panel ) and @xmath156 ( right panel ) for @xmath213 . since @xmath143^*$ ] only one component is shown . the stars in fig . [ fig : full_eq ] denote the solutions of the imaginary - time equations taken from ref . . note that for @xmath156 and @xmath214 the simple approximation eq . ( [ ap ] ) showed an instability , while with the flowing vertex the system is stable even for these large values of @xmath164 and reproduces the correct equilibrium solutions for @xmath153.@xcite the reason for this is that the flow of the vertex reduces the resulting effective interaction below the critical value in the instability region.@xcite and @xmath215 with @xmath63 for @xmath151 ( left panel ) and @xmath156 ( right panel ) , @xmath216 and @xmath180 . the full curves show the real part , the dashed the imaginary part of @xmath144 , the dot - dashed the imaginary part of @xmath217 . the real part for the latter is zero.,scaledwidth=40.0% ] for the same parameters as in fig . [ fig : full_eq ] we present the resulting flow with finite bias @xmath180 in fig . [ fig : full_noneq ] . in addition to the curves for real ( solid lines ) and imaginary part ( dashed lines ) of @xmath218 a third curve is displayed , the imaginary part of @xmath219 ( dashed - dotted lines ) , which now is generated during the flow . note that @xmath220 and @xmath221^*$ ] . furthermore , we always find @xmath222 . for @xmath156 ( right panel in fig . [ fig : full_noneq ] ) we have rescaled @xmath223 by a factor @xmath224 to make it visible on the scale of @xmath144 . since @xmath225 is a complex energy independent quantity eq . ( [ eq : causality ] ) is again not fulfilled . we note that the error is still of order @xmath184 , but for fixed @xmath91 and @xmath90 it is significantly smaller than in the simplest truncation scheme discussed above . the energy - independence of the self - energy allows to derive an analytical expression for the current at @xmath89 similar to eq . ( [ eq : current_simple ] ) , which due to the appearance of @xmath219 now becomes @xmath226 with @xmath227 as in eq . ( [ eq : abbriv_vgstar ] ) and @xmath228 \gamma^\ast & = & \sqrt{\tilde\gamma^2+(\im m\sigma^{--})^2}\;;\end{aligned}\ ] ] where @xmath229 is taken at @xmath62 . thus , the only change to the expression ( [ eq : current_simple ] ) is a formal replacement @xmath230 in @xmath231 . equation ( [ eq : current_full ] ) is of the same structure as for the noninteracting case with @xmath91 and @xmath49 replaced by renormalized parameters . however , the two self - energy contributions @xmath218 and @xmath219 enter distinctively different in the expression for the current . while @xmath232 solely plays the role of an additional life - time broadening , @xmath233 directly modifies the tunneling rate both in the prefactor of @xmath113 _ and _ in the expression for the life - time broadening . a problem occurs when using the results of the present approximation in eq . ( [ eq : deltaj2 ] ) , leading to @xmath234 } { \left|\omega - v_g+i\gamma\left[1-f(\omega)\right]-\sigma^{--}\right|^2+\left[\gamma f(-\omega)+\im m\sigma^{+-}\right]\left[\gamma f(\omega)-\im m\sigma^{-+}\right]}\;\;.\ ] ] the requirement @xmath122 is only fulfilled for @xmath103 , because then @xmath209 and the integrand is asymmetric with respect to @xmath32 . thus our approximation of an energy - independent flowing vertex violates current conservation for @xmath235 in nonequilibrium . we verified that @xmath236 which is consistent with the fact that not all terms of order @xmath184 are kept in our truncated frg procedure . how does @xmath123 behave in the limit @xmath237 ? to see this we note that , because @xmath238 does not depend on the sign of @xmath90 and furthermore goes to zero as @xmath237 , @xmath239 . consequently , @xmath240 and hence the violation of current conservation vanishes in the linear response regime @xmath237 . ( after reintroducing @xmath207 and @xmath208 ) as function of @xmath90 for @xmath151 , @xmath241 and @xmath242 and @xmath103 . for @xmath156 we find a region of negative differential conductance in the region @xmath243 [ fig : g_vs_mu]).,scaledwidth=40.0% ] in fig . [ fig : current_full ] we show the current at @xmath103 as function of @xmath90 for @xmath151 , @xmath241 and @xmath242 . with increasing @xmath164 the current for intermediate @xmath90 is strongly suppressed . in addition there occurs a structure at low @xmath90 , which turns into a region of negative differential conductance with increasing @xmath164 . the appearance of such a shoulder in the current was observed in other calculations as well.@xcite however , whether the negative differential conductance we find for still larger values of @xmath164 ( c.f . [ fig : g_vs_mu ] ) is a true feature of the model or rather an artifact of the approximations used is presently not clear and should be clarified in further investigations . however , negative differential conductance has also been observed in a slave - boson treatment of the model.@xcite as function of @xmath90 for @xmath103 and various values of @xmath164 . for @xmath244 a distinct minimum around @xmath245 appears.,scaledwidth=40.0% ] keeping @xmath103 fixed , we can calculate the conductance @xmath114 as function of @xmath90 for different values of @xmath164 . the results are collected in fig . [ fig : g_vs_mu ] . in contrast to the simple approximation without flow of the vertex , the conductance is now strongly dependent on @xmath164 , except for @xmath155 , where due to the unitary limit at @xmath89 we always find @xmath246 . as already anticipated from the current in fig . [ fig : current_full ] , a minimum in @xmath247 starts to form around @xmath245 for @xmath244 , which is accompanied by a peak at @xmath248 . a similar behavior in the conductance was observed in a perturbative treatment,@xcite which in contrast to our current approximation involves the full energy - dependence in the self - energy . this at least qualitative agreement we of course can not resolve structures like the hubbard bands with an energy independent self - energy again supports our claim that despite the violation of the relation ( [ eq : causality ] ) we can obtain reasonable results from @xmath191 . as function of @xmath91 for different values of @xmath90 and @xmath151 ( upper panel ) and @xmath156 ( lower panel ) . note the extended plateau at @xmath155 for @xmath156 , which is a manifestation of the pinning of spectral weight at the fermi level.,scaledwidth=40.0% ] we finally discuss the variation of the conductance with @xmath91 for fixed @xmath164 and @xmath90 . we again emphasize , that for @xmath235 , @xmath122 only holds to leading order in @xmath164 . in fig . [ fig : g_vs_vg ] we present the curves for two different values of @xmath164 , namely @xmath151 ( upper panel in fig . [ fig : g_vs_vg ] ) and @xmath156 ( lower panel in fig . [ fig : g_vs_vg ] ) . in the former case , the variation of @xmath247 with @xmath90 is rather smooth , as is to be expected from the current in fig . [ fig : current_full ] . for large @xmath164 , we observe an extended plateau at zero bias , which is a manifestation of the fact that in the strong coupling regime a pinning of spectral weight at the fermi energy occurs . this feature is also observed in the imaginary - time frg as well as in nrg calculations.@xcite increasing @xmath90 quickly leads to a similarly extended region of negative differential conductance , which , assuming that this result is a true feature of the model , therefore seems to be linked to the `` kondo '' pinning . we note that it is unlikely that the appearance of the negative differential conductance is related to the breaking of current conservation at order @xmath184 as it also appears for @xmath103 where @xmath122 . for large @xmath90 multiple structures appear in @xmath247 , which are related to the energy scales @xmath90 and @xmath164 . starting from a generating functional proposed by kameneev@xcite we have derived an infinite hierarchy of differential equations for the vertex functions of an interacting quantum mechanical many - body system in nonequilibrium ( see also refs . and ) . the major difference to the imaginary - time formulation comes from the use of keldysh green functions . the indices @xmath249 and @xmath250 referring to the upper and lower branch of the ( real-)time contour lead to an additional matrix structure to the problem . our formulation is sufficiently general that it allows to treat bosonic and fermionic models with or without explicit time dependence and at @xmath251 . since the frg leads to an infinite hierarchy of coupled differential equations , one has to introduce approximations , at least a truncation at a certain level . typically one neglects the flow of the three - particle vertex . as has been demonstrated in ref . for the imaginary - time frg one can solve the remaining system of flow equations for simple models like the siam numerically , thus keeping the full energy - dependence . due to the fact that the vertex function carries three _ continuous _ frequency arguments in addition to the discrete quantum numbers of the system , such a calculation can become computationally quite expensive . to reduce the numerical effort , further additional approximations can be introduced . a particularly important and successful one is obtained by neglecting the energy dependence of the vertex functions,@xcite which already leads to a surprisingly accurate description of local and transport properties of interacting quantum dots in the linear response regime . we applied the nonequilibrium frg to the siam with finite bias voltage in the stationary state . it turned out , that for the simplest approximation where only the flow of the self - energy is kept , the analytic structure of the differential equation leads to problems in the numerical solution . in addition , this approximation leads to a violation of the causality relation ( [ eq : causality ] ) to order @xmath184 . the first problem was resolved by including the two - particle vertex in the flow at least up to the largest interaction considered here ( @xmath252 ) . at the present stage this was for computational reasons done by assuming it to be energy - independent , yielding again an energy - independent self - energy . although this approximation also violates eq . ( [ eq : causality ] ) to order @xmath184 for a fixed @xmath91 and @xmath90 the error is significantly smaller compared to the simplest scheme . we were able to obtain reasonable expressions and numerical results for the current and the conductance using the functions @xmath253 instead of @xmath254 in the current formula . we reproduced nonequilibrium features of the current and differential conductance known from the application of other approximate methods to the siam . in the more advanced truncation scheme and for @xmath255 , the current conservation @xmath256 only holds to leading order in @xmath164 . this defect can be traced back to the energy independence of the two - particle vertex , leading to finite , but energy - independent @xmath219 and @xmath257 . by a close inspection of formula ( [ eq : deltaj2 ] ) , however , one can see that this deficiency can for example be cured by assuming a coarse - grained energy dependence of the form @xmath258 -\frac{v_b}{2}<\omega<\frac{v_b}{2 } : & f(\omega)=1\;\rightarrow & \im m\sigma^{+-,\lambda}=-\im m\sigma^{-+,\lambda}=\mbox{const.}\\[3 mm ] \frac{v_b}{2}<\omega : & f(\omega)=0\;\rightarrow & \im m\sigma^{-+,\lambda}=0\;\mbox { , $ \im m\sigma^{+-,\lambda}$=const.}\end{aligned}\ ] ] and corresponding energy dependencies for @xmath218 and @xmath259 . here the const . might depend on @xmath63 but not on @xmath32 . such an approximation will quite likely also reestablish causality eq . ( [ eq : causality ] ) . work along this line is in progress . including finite temperature , magnetic field etc . in the calculations of transport is easily possible , as well as the extension to more complicated `` impurity '' structures ( see e.g.ref . ) . the latter aspect is typically a rather cumbersome step for other methods like perturbation theories or in particular numerical techniques . our present stage of work should mainly be seen as a `` proof of principle '' . already for the siam there remain a variety of fundamental things to do ; like the implementation of the full energy - dependence in the calculations , which then should cure the violation of charge conservation and causality ; or implementing a reasonable `` cutoff scheme '' for time - dependent problems off equilibrium , e.g. to calculate current transients . despite its current limitations we expect the real - time nonequilibrium formulation of the frg to be a useful tool to understand nonequilibrium features of mesoscopic systems , just like the imaginary - time version in the calculation of equilibrium properties . we acknowledge useful conversations with h. schoeller , s. jakobs , s. kehrein , j. kroha , h. monien , a. schiller , a. dirks , j. freericks , k. ueda , t. fujii , k. thygesen , and f.b . we thank h. schmidt for pointing out typos in an earlier version of our paper . this work was supported by the dfg through the collaborative research center sfb 602 . computer support was provided through the gesellschaft fr wissenschaftliche datenverarbeitung in gttingen and the norddeutsche verbund fr hoch- und hchstleistungsrechnen . although the following derivation is mainly equivalent to the one presented in ref . , the appearance of the factor @xmath66 in the real - time formulation of the generating functional leads to some changes in signs and prefactors @xmath66 in the equations . it thus appears to be helpful for the reader to repeat the derivation . as a first step we differentiate @xmath260 with respect to @xmath63 , which after straightforward algebra leads to @xmath261 considering @xmath262 and @xmath263 as the fundamental variables we obtain from eq . ( [ eq : gammadef ] ) @xmath264 applying the chain rule and using eq . ( [ flusswc ] ) this leads to @xmath265 where the last term in eq . ( [ eq : gammadef ] ) cancels a corresponding contribution arising in ( [ flusswc ] ) thus a posterior justifying the inclusion of this term . extending the well known relation@xcite between the second functional derivatives of @xmath49 and @xmath266 to nonequilibrium we obtain the functional differential equation @xmath267 where @xmath268 stands for the upper left block of the matrix @xmath269^{-1 } & i \frac{\delta^2 \gamma^{\lambda}}{\delta \bar \phi \delta \bar \phi } \\ i \zeta\frac{\delta^2 \gamma^{\lambda}}{\delta \phi \delta \phi } & -\left\ { i\frac{\delta^2 \gamma^{\lambda}}{\delta \phi \delta \bar \phi } + \left ( \left [ \hat g^{0,\lambda } \right]^{-1}\right)^t\right\ } \end{array } \right)^{-1 } \end{aligned}\ ] ] and the upper index @xmath3 denotes the transposed matrix . to obtain differential equations for the @xmath270 which include self - energy corrections we express @xmath271 in terms of @xmath272 instead of @xmath273 this is achieved by defining @xmath274 and using @xmath275^{-1 } - \zeta \gamma_1^{\lambda } \right]^{-1 } \;,\end{aligned}\ ] ] which leads to @xmath276 \ ; , \end{aligned}\ ] ] with @xmath277^t \end{array } \right ) \left ( \begin{array}{cc } { \mathcal u}_{\bar \phi , \phi } & i\frac{\delta^2 \gamma^{\lambda}}{\delta \bar \phi \delta \bar \phi } \\ ( i\zeta)\frac{\delta^2 \gamma^{\lambda}}{\delta \phi \delta \phi } & \zeta { \mathcal u}^{t}_{\bar \phi , \phi } \end{array } \right ) \right]^{-1 } \ ; .\end{aligned}\ ] ] it is important to note that @xmath278 as well as @xmath279 and @xmath280 are at least quadratic in the external sources . the initial condition for the exact functional differential equation ( [ gammaflussalt ] ) can either be obtained by lengthy but straightforward algebra , which we are not going to present here or by the following simple argument : at @xmath281 , @xmath282 ( no degrees of freedom are `` turned on '' ) and in a perturbative expansion of the @xmath283 the only term which does not vanish is the bare two - particle vertex . we thus find @xmath284 expanding @xmath285 eq . ( [ vtildedef ] ) in a geometric series @xmath286^t \frac{\delta^2\gamma}{\delta\phi\delta\phi}+ \ldots\end{aligned}\ ] ] and @xmath64 in a taylor series with respect to the external sources @xmath287 one can derive an exact infinite hierarchy of flow equations for the @xmath270 . for example , the flow equation for the single - particle vertex @xmath288 ( the self - energy ) can be obtained by taking the expansion ( [ eq : vtildeexp ] ) up to first order in @xmath289 , inserting it into ( [ gammaflussalt ] ) , replacing @xmath290 on both sides by the expansion ( [ gumma ] ) and comparing the terms quadratic in the external fields . this procedure leads to the expression ( [ eq : gamma1fl ] ) . to obtain the flow equation for @xmath128 we have expanded eq . ( [ vtildedef ] ) in a geometric series up to first order and @xmath64 ( [ gumma ] ) in the external sources up to @xmath84 . to find the flow equation for the vertex function , @xmath82 , we have to proceed one step further , that is expand ( [ vtildedef ] ) up to the second order and @xmath64 up to @xmath291 . after comparison of the terms with the same power in the fields we obtain an expression similar to the equilibrium one ( see eq . ( 15 ) of ref . ) @xmath292\nonumber\\ & - & g^{\eta \rho,\lambda}_{\xi_3,\xi_3 ' } s^{\nu \mu,\lambda}_{\xi_4,\xi_4'}\big [ \gamma^{\alpha \mu \gamma \eta,\lambda}(\xi_1',\xi_4';\xi_1,\xi_3 ) \gamma^{\rho \beta \nu \delta,\lambda}(\xi_3',\xi_2';\xi_4,\xi_2 ) \nonumber\\ & + & \gamma^{\alpha\rho\gamma\nu,\lambda}(\xi_1',\xi_3';\xi_1,\xi_4)\gamma^{\mu\beta\eta\delta,\lambda}(\xi_4',\xi_2';\xi_3,\xi_2)\nonumber\\ & - & \gamma^{\beta \mu \gamma \eta,\lambda}(\xi_2',\xi_4';\xi_1,\xi_3)\gamma^{\rho \alpha \nu \delta,\lambda}(\xi_3',\xi_1';\xi_4,\xi_2 ) \nonumber\\ \label{eqq1 } & - & \gamma^{\beta\rho\gamma\nu,\lambda}(\xi_2',\xi_3';\xi_1,\xi_4)\gamma^{\mu\alpha\eta\delta,\lambda}(\xi_4',\xi_1';\xi_3,\xi_2 ) \big ] \bigg)\;,\end{aligned}\ ] ] where the contribution involving @xmath293 was replaced by its initial value , i.e. @xmath294 . the main difference to equilibrium is the presence of mixed contractions for the keldysh indices of the matrices @xmath295 and @xmath296 and the four - index tensors @xmath82 . up to this point the derivation of the flow equation for the two - particle vertex was as general as the discussion in sec . [ formalism ] . we now consider the application to the steady state current through a quantum dot described by the siam presented in sec . [ appli ] . we thus go over to frequency space and introduce the cutoff @xmath63 as in eq . ( [ eq : matsubaracutoff ] ) . the single - particle quantum number in @xmath2 is given by the spin index . we can take advantage of spin and energy conservation which implies that the lower indices of @xmath297 and @xmath298 are equal , that is @xmath299 and @xmath300 . this does _ not _ hold for the keldysh indices . the integrals over @xmath301 and @xmath302 as well as the sums over @xmath303 and @xmath304 can then be performed trivially . using the morris lemma,@xcite we can rewrite the matrix product of @xmath297 and @xmath298 as @xmath305 with @xmath306^{-1 } - \hat \sigma^{\lambda}(\omega ) } \ ; .\end{aligned}\ ] ] the @xmath307-function can be used to perform another of the frequency integrals , and the remaining one can be evaluated because of energy conservation of the two - particle vertex . using the spin independence of the green function for zero magnetic field , this leads to comparing ( [ eqq2 ] ) with eq . ( 20 ) in ref . , were similar approximations were made in equilibrium , we see that we have two more terms because of the keldysh indices . in the first term appears @xmath309 , while its transpose @xmath310 enters everywhere else . using the antisymmetry of @xmath82 in the spin indices and spin conservation we can further simplify eq . ( [ eqq2 ] ) by introducing the flowing interaction @xmath311 defined as @xmath312 this leads to the flow equation @xmath313\nonumber \\ & \!\!\!-g_d^{\eta \rho,\lambda}(\omega ) g_d^{\nu \mu,\lambda}(\omega)&\!\!\!\big [ 2u^{\alpha \mu \gamma \eta,\lambda } u^{\rho \beta \nu \delta,\lambda}- u^{\alpha \mu \gamma \eta,\lambda } u^{\beta \rho \nu \delta,\lambda}- u^{\mu \alpha \gamma \eta,\lambda } u^{\rho \beta \nu \delta,\lambda}\nonumber\\ & \!\!\!&+2u^{\alpha \rho \gamma \nu,\lambda } u^{\mu \beta \eta \delta,\lambda}- u^{\alpha \rho \gamma \nu,\lambda } u^{\beta \mu \eta \delta,\lambda}- u^{\rho \alpha \gamma \nu,\lambda } u^{\mu \beta \eta \delta,\lambda}\nonumber\\ \label{eqq3 } & \!\!\!&- u^{\mu \beta \gamma \eta,\lambda}u^{\alpha \rho \nu \delta,\lambda } -u^{\rho \beta \gamma \nu,\lambda}u^{\alpha \mu \eta \delta,\lambda } \big ] \bigg)\;\;.\end{aligned}\ ] ]

we apply the wiener hermite ( wh ) expansion to the non - linear evolution of the large - scale structure , and obtain an approximate expression for the matter power spectrum in the full order of the expansion . this method allows us to expand any random function in terms of an orthonormal basis in the space of random functions in such a way that the first order of the expansion expresses gaussian distribution , and others are the deviation from the gaussianity . it is proved that the wh expansion is mathematically equivalent to the-expansion approach in the renormalized perturbation theory ( rpt ) . while exponential behavior in the high- limit has been proved for the mass density and velocity fluctuations of dark matter in the rpt , we prove the behavior again in the context of the wh expansion using the result of the standard perturbation theory ( spt ) . we propose a new approximate expression for the matter power spectrum which interpolates the low- expression corresponding to the 1-loop level in spt and the high- expression obtained by taking a high- limit of the wh expansion . the validity of our prescription is specifically verified by comparing with the 2-loop solutions of spt . the proposed power spectrum agrees with the result of-body simulation with accuracy better than or in a range of the baryon acoustic oscillation scales , where the wave number is about = 0.20.4 at . this accuracy is comparable to or slightly less than the ones in the closure theory , the fractional difference of which from the-body result is within . one merit of our method is that the computational time is very short because only single and double integrals are involved in our solution .

1210.1663

precise measurements of matter power spectrum in the large - scale structure are a powerful tool not only to investigate the details of the structure formation , but also to estimate the cosmological parameters . for example , the precise measurements of the baryon acoustic oscillation ( bao ) in the matter power spectrum observed by the sloan digital sky survey has emerged as a powerful tool to estimate cosmological parameters @xcite . also the observation of cosmic shear in the near future is expected to give a useful constraint on the nature of dark energy . obviously , proper understanding of the observed power spectrum becomes possible only if an accurate theoretical prediction is available which requires a good understanding of the non - linear evolution of dark matter perturbation , the relation between dark matter and baryonic matter ( bias effect ) and the redshift distortion effect . there have been various studies and much progress on the theoretical calculations of the power spectrum , but it is still useful and required to have more accurate theoretical treatment . in this paper , we give a new approach to describe the non - linear evolution of dark matter . it is called the `` wiener hermite ( wh ) expansion method '' , where the stochastic nature of the cosmological density perturbation is manifestly used and the stochastic variables are expanded in terms of an orthonormal basis in the space of stochastic functions . the method was developed in the 1970s for application to turbulent theory in fluid dynamics and applied to cosmological turbulent theory by one of the authors of this paper . the method gives us a coupled equation at each perturbative order , even at the first order in such a way that the lower order quantities are modified by higher order quantities . this is totally different from the usual perturbation theory where lower order quantities are never influenced by higher order quantities . thus , it gives us a prescription for the renormalization of higher order effects , and the precise meaning is described below . each expansion coefficient has a clear statistical meaning ; namely , the coefficients of the first , second and third terms in the expansion express the amplitude of gaussianity , the skewness and kurtosis , respectively . thus each term corresponds directly to an appropriate @xmath8-point correlation function . we mention here some details on the previous approaches relation to ours . it has been known for some time that the standard perturbation theory ( spt ) of cosmological perturbation can be analytically solved in the einstein - de sitter universe in integral forms @xcite . when it is considered up to the third order in spt ( 1-loop level ) , the analytical predictions describe the nonlinearity well at sufficiently high redshifts @xcite . however , the predictions are still insufficient at the observable low redshifts ( @xmath9 0 3 ) , and we need to consider further non - linear effects . furthermore , it is computationally expensive to deal with the higher order corrections in spt . therefore , various modification of spt have been proposed in the past . one of the main approach is the `` renormalized perturbation theory '' ( rpt ; @xcite ) , where the basic equations for fluid describing matter perturbation are rewritten in a convenient compact form in order to use a diagrammatic technique developed in quantum field theory @xcite . further modification have been considered , such as , e.g. , the `` closure theory '' @xcite , the `` time renormalization group '' approach @xcite , and the `` @xmath0-expansion approach '' using multi - point propagators @xcite . many other new methods have also been studied @xcite . on the other hand , there is also an approach to the large - scale structure in the framework of the lagrangian picture , called `` lagrangian resummation theory '' ( lrt ; @xcite ) . it will be shown that our approach is mathematically equivalent to the @xmath0-expansion approach , but it still has the features described above and gives us a very convenient expression for the matter power spectrum described below . in almost all modified perturbation theories , the resummation of nonlinear effects , which means the partial summation of the infinite order in spt , is considered . this implies that any modified perturbative expansion methods should be described in the context of spt . in this paper , we use only spt , and prove again various properties of cosmological perturbations , e.g. , their behavior in the small - scale limit ( high-@xmath1 limit ) proved in the context of rpt @xcite . since low-@xmath1 solutions can be safely computed using spt , the derivation of more precise solutions of cosmological perturbations by interpolating between the 1-loop results and the high-@xmath1 behavior @xcite has been attempted . however , some arbitrariness have remained for this prescription . to resolve this problem , we propose a unique interpolation between the low-@xmath1 solutions and the high-@xmath1 ones by assuming that the higher order solutions in perturbation theory are well approximated by the ones in the high-@xmath1 limit . then , we precisely compute only up to 1-loop level corrections in spt and replace the higher order corrections with the ones calculated in the high-@xmath1 limit . in this way we obtain an approximate full power spectrum , and the power spectrum shows a very good agreement with @xmath2-body results up to rather high-@xmath1 ( about @xmath10 0.20.4 @xmath11mpc@xmath12 ) within 1 % or 2 % accuracy . this paper is organized as follows . in section 2 , we first explain the stochastic properties which should be satisfied by the density and velocity perturbations of dark matter . in section 3 , we briefly review spt , which will be used later . then the wh expansion technique is explained in our context in section 4 . the relationship between spt and the wh expansion method is established there , and we also show the mathematical equivalence between the wh expansion and the @xmath0-expansion . in section 5 , we prove again the high-@xmath1 limit behavior of the cosmological perturbations in the context of spt and propose an approximate full power spectrum , where the lower order corrections are calculated only up to 1-loop levels in spt and the higher order corrections are replaced with the high-@xmath1 solutions . in section 6 , we compare our result with some other analytic predictions and @xmath2-body simulations . we compute the two - point correlation function in section 7 . we summarize our work and discuss future works in section 8 . after decoupling , baryon and dark matter fluctuations are tightly coupled by the gravitational force , and the evolution can then be described by pressureless fluid equations ( continuity equation and euler equation ) with the poisson equation for the newton gravity . thus our basic equations are as follows @xcite : @xmath13 = 0 , \notag \\ & \frac{\partial { \mathbf{v}}}{\partial t } + { \mathbf{v}}\cdot \nabla { \mathbf{v}}= -\nabla \phi , \notag \\ & \nabla^2 \phi + \lambda c^2 = 4 \pi g \rho , \label{ori}\end{aligned}\ ] ] where @xmath14 , @xmath15 , and @xmath16 denote the mass density , velocity and gravitational potential , respectively , and @xmath17 is the cosmological constant . when we transform the spatial coordinates as @xmath18 and redefine the velocity as @xmath19 , where @xmath20 is the scale factor and @xmath21 is the peculiar velocity , we can express eq . ( [ ori ] ) as @xmath22 = 0 , \label{ori1}\ ] ] @xmath23 { \mathbf{u}}(\tau,{\mathbf{x } } ) = - \nabla \phi(\tau,{\mathbf{x } } ) , \label{ori2}\ ] ] @xmath24 where the conformal time @xmath25 is defined as @xmath26 , and the conformal hubble parameter @xmath27 is defined as @xmath28 , where @xmath29 is the hubble parameter . we further defined the cosmological gravitational potential as @xmath30 . in the standard cosmological perturbation theory , physical quantities are decomposed into the background part and the perturbative part . the background part of the mass density @xmath31 is defined as @xmath32 where @xmath33 denotes the ensemble average and we used the translation symmetry of the ensemble average . on the other hand , the peculiar velocity @xmath21 has no background part because of rotation symmetry in the average sense . therefore , the perturbative part of the mass density and the peculiar velocity has the property that their ensemble average are zero by definition : @xmath34 averaging the above set of equations , we obtain the following background equations . @xmath35 @xmath36 integrating eq . ( [ h ] ) , we find the usual friedman equation , @xmath37 where the integral constant @xmath38 is interpreted as the spatial curvature . by subtracting the background equations from eq . ( [ ori1 ] ) ( [ ori2 ] ) , we find our basic equations in fourier space as follows ; @xmath39 @xmath40 where @xmath41 and @xmath42 denotes the divergence of velocity , and @xmath43 denotes the three - dimensional dirac delta distribution . we neglected the vorticity @xmath44 because the vorticity is zero if its initial value is zero , and even if its initial value is non - zero , it decays due to the expansion of the universe . the functions @xmath45 @xmath46 encode the nonlinearity of the evolution and satisfy the conditions @xmath47 note that this decomposition between background and perturbation is exact in newtonian gravity , namely there are no backreaction terms generated from the ensemble average in newtonian gravity , and the perturbative parts @xmath48 and @xmath49 obeying eqs . ( [ eq1 ] ) and ( [ eq2 ] ) naturally satisfy the stochastic condition in eq . ( [ stochastic ] ) : @xmath50 . more specifically , when the solutions for nonlinear equations are expanded perturbatively , the property that their ensemble averages are zero is not guaranteed in general , and thus a redefinition of the perturbation variables such as @xmath51 is necessary . in the case of cosmological spt , the average , such as @xmath52 being proportional to @xmath53 , is interpreted as the vacuum bubble diagram in the diagrammatical picture and contributes only in infinitely large scales . this means we really need to redefine of the cosmological background . however , we do not need to consider this prescription for perturbative variables in eq . ( [ eq1 ] ) and ( [ eq2 ] ) . this is a special feature of newtonian gravity . it will be interesting to see how the backreaction look like in the case of general relativistic gravity in our approach . we here explain spt very briefly which will be used later . the solutions in the case of an einstein de sitter universe , where @xmath54 and @xmath55 , can be described by analytically integral forms in spt . more explicitly , the solution may be written in the following perturbative form , @xmath56 where the scale factor @xmath20 is a growing mode solution in the linearized theory . when the scale factor @xmath20 is small , the series are dominated by their first term , that is , by linearized theory . the relation between the time - independent coefficients @xmath57 and @xmath58 is shown from the continuity equation ( [ eq1 ] ) as @xmath59 , and the time - independent linear power spectrum for @xmath60 is defined as @xmath61 where the amplitude of the wave vector is expressed as @xmath62 . then , the coefficients @xmath63 and @xmath64 are described as follows , @xmath65 @xmath66 where @xmath67 and @xmath68 and @xmath69 are completely symmetrized functions for the wave vectors @xmath70 . the functions @xmath68 and @xmath69 are constructed according to the following recursion relations ( @xmath71 ) @xcite : @xmath72 , \label{chikujif}\end{aligned}\ ] ] @xmath73 , \label{chikujig}\end{aligned}\ ] ] where @xmath74 , @xmath75 , and @xmath76 . for @xmath77 , we have @xmath78 @xmath79 the stochastic property @xmath50 is specifically shown from these solutions . when we consider the average of eq . ( [ solution - del ] ) , we only have to consider both coefficients @xmath80 and @xmath81 due to the linearity of the ensemble average . the ensemble average for @xmath80 is @xmath82 where @xmath83 is defined as @xmath84 when the function @xmath68 in eq . ( [ chikujif ] ) substitute into eq . ( [ shiki ] ) , @xmath85 is satisfied due to the dirac delta function in eq . ( [ shiki ] ) , and the functions @xmath86 and @xmath87 become zero from eq . ( [ albe ] ) . then , it is shown that the function @xmath68 in eq . ( [ shiki ] ) becomes zero , and @xmath88 . a similar analysis can be applied to @xmath49 . note that this stochastic property is independent of the initial conditions of @xmath89 , that is , the initial condition can have primordial non - gaussianity . now , we explain our expansion method for @xmath48 and @xmath49 . our expansion scheme should satisfy the following two properties . first , it is known observationally and theoretically that the cosmological perturbations in the universe have a nearly gaussian distribution . thus the first order in our expansion should express the gaussian distribution . second , the expansion scheme should respect the stochastic condition of the cosmological perturbations , @xmath50 . based on these two conditions , we adopt the wh expansion as our expansion method . in the wh expansion , the perturbation variables @xmath48 and @xmath49 are expanded as follows , @xmath90 @xmath91 where the functions @xmath92 @xmath93 are the stochastic variables . the first - order @xmath94 is a white noise function which satisfies the gaussian distribution , @xmath95 and we further define higher order bases @xmath92 @xmath96 in the expansion as follows : @xmath97 thus they become an orthonormal basis in the space of stochastic functions , where the ensemble average of @xmath92 is clearly zero . @xmath98 @xmath99 therefore , the stochastic property in eq . ( [ stochastic ] ) is satisfied by the definition of the wh expansion . the coefficients of the wh expansion are derived by averaging @xmath48 and @xmath49 after multiplying the stochastic variable @xmath92 : @xmath100 the power spectrum is described in the wh expansion by @xmath101 ^ 2 \notag \\ & = \sum_{r=0}^{\infty } p_{\rm wh}^{(r+1)}(z , k ) , \label{full_power}\end{aligned}\ ] ] where the contribution of the @xmath102th order in the wh expansion to the power spectrum @xmath103 is defined as @xmath104 ^ 2 . \label{hpower}\ ] ] since the solutions for any order of spt are given in an einstein de sitter universe analytically , any new expansion for @xmath48 and @xmath49 must be described in the context of spt . for the linear order , we assume @xmath105 ^ 2 , \label{}\ ] ] where the superscript and script indices for @xmath106 denote the order of the wh expansion and spt , respectively . it is straightforward to include the intrinsic non - gaussianities as higher order contributions in the wh expansion . from now on , we express @xmath107 . substituting the solutions in the spt eqs . ( [ spt ] ) , ( [ solution - del ] ) , and ( [ solution - the ] ) into eq . ( [ coef ] ) , we find the general relation of the solutions between spt and the wh expansion as follows : @xmath108 where @xmath109 this expression means that the density fluctuation with order @xmath102 in the wh expansion is the sum of all the density fluctuations with order @xmath110 in spt . we can understand the relation between spt and the wh expansion through a diagrammatical representation , where there is no non - dimensional coupling constant and the order of the loop is determined by the order of the linear power spectrum @xmath111 , that is , the @xmath8-loop contributions contains the terms proportional to @xmath112 . each order of the wh expansion includes all vertex loop contributions which come from the @xmath113 coefficients themselves . the order of the vertex loop is expressed by @xmath8 ( @xmath114 ) . on the other hand , the loop contributions from irreducible diagrams , where the loop order is expressed by @xmath115 ( @xmath116 ) , arise only after calculating the power spectrum in eqs . ( [ full_power ] ) and ( [ hpower ] ) ( see @xcite for details ) . the relation given in eq . ( [ coef ] ) corresponds to eq . ( 17 ) in @xcite . that is , the wh expansion method coincides with the @xmath0-expansion approach : @xmath117 for @xmath118 in eqs . ( [ wh - spt-1 ] ) ( [ wh - spt ] ) and ( [ gamma - wh ] ) , we have @xmath119 thus , the first order of the wh expansion , @xmath120 , corresponds to the propagator in rpt . furthermore , @xmath121 denotes the irreducible diagrams , and this is expressed in the @xmath0-expansion as @xmath122 note that since we focus only on spt , we consider only growing solutions , unlike rpt . although the wh expansion is defined and the interpretation is physically and mathematically useful for understanding the nonlinear evolution of dark matter , we need to truncate the expansion at some order , most probably at a lower order such as @xmath123 or @xmath124 in order to perform the actual calculation . however , the validity of the truncation is not guaranteed immediately . furthermore , the computational difficulties for the power spectrum increase very rapidly when we increase the order of truncation . in order to resolve these difficulties , we propose in this section an approximate semi - analytic expression for the full power spectrum including all order in the wh expansion by adopting the following assumption : the high - order solutions in spt become dominant in the high-@xmath1 limit . therefore , they are approximated well enough by the ones in the high-@xmath1 limit . here , we show the exponential behavior of the solutions using spt , which have been proved in rpt @xcite . we prove that the functions @xmath125 and @xmath126 take the following expression in the high-@xmath1 limit : @xmath127 with @xmath128 here , we define the high-@xmath1 limit as @xmath129 from now on , we shall simplify the notations : @xmath130 and @xmath131 . we prove this by induction in @xmath8 as follows . for @xmath132 , eq . ( [ theorem ] ) is clearly satisfied . for some @xmath8 , we assume @xmath133 then we show that the @xmath134 order satisfies the same limit . the functions @xmath135 and @xmath136 are given by eqs . ( [ chikujif ] ) and ( [ chikujig ] ) . then , let us examine which terms become dominant in the high-@xmath1 limit in these recursion relations . from eq . ( [ high - k ] ) , we keep only terms with scale dependence as @xmath137 . then , we have the terms proportional to @xmath138 , @xmath139 for @xmath140 , and @xmath141 and @xmath142 in the high-@xmath1 limit . this means that the recursion relation for @xmath135 in the high-@xmath1 limit becomes : @xmath143 \notag \\ & \hspace{1.5 cm } + 4\frac{c(r , m)}{c(n+r+1,m ) } \bigg[\sum_{m=1}^{r } g_{m}({\mathbf{k}}_m)\beta({\mathbf{k}}_{1m},{\mathbf{k}}+{\mathbf{k}}_{(m+1)r}+{\mathbf{p}}_{1n } ) g_{r+n+1-m}({\mathbf{k}},{\mathbf{k}}_{m+1},\dots,{\mathbf{k}}_{r},{\mathbf{p}}_n)\bigg ] \notag \\ & \hspace{1.5 cm } + ( 2(r+n)+3 ) \left ( \frac{n}{r+n+1 } \right ) \alpha({\mathbf{p}}_n,{\mathbf{k}}+{\mathbf{k}}_{1r}+{\mathbf{p}}_{1(n-1 ) } ) f_{r+n}({\mathbf{k}},{\mathbf{k}}_r,{\mathbf{p}}_{n-1 } ) \notag \\ & \hspace{1.5 cm } + 4\left ( \frac{n}{r+n+1 } \right ) \beta({\mathbf{p}}_n,{\mathbf{k}}+{\mathbf{k}}_{1r}+{\mathbf{p}}_{1(n-1 ) } ) g_{r+n}({\mathbf{k}},{\mathbf{k}}_r,{\mathbf{p}}_{n-1 } ) \bigg\ } , \label{recursion1}\end{aligned}\ ] ] where we denote @xmath144 and @xmath145 , and define as @xmath146 furthermore , the behavior of @xmath147 and @xmath148 in the high-@xmath1 limit is @xmath149 and eq ( [ recursion1 ] ) becomes @xmath150 \notag \\ & \hspace{2.5 cm } + 2 \bigg[\sum_{m=1}^{r } g_m({\mathbf{q}}_m)\beta(\tilde{{\mathbf{k}}}_{1},\tilde{{\mathbf{k}}}_2 ) g_{r+1-m}({\mathbf{q}}_{m+1},\dots,{\mathbf{q}}_{r+1 } ) \bigg ] \notag \\ & \hspace{2.5 cm } + ( 2(r+n)+3 ) n f_{r+1}({\mathbf{k}},{\mathbf{k}}_r ) + 2n g_{r+1}({\mathbf{k}},{\mathbf{k}}_r ) \bigg\ } , \label{recursion2}\end{aligned}\ ] ] where @xmath151 and @xmath152 , @xmath153 . from eq . ( [ chikujif ] ) and eq . ( [ chikujig ] ) , we can show the following relations . @xmath154 = ( r+1)f_{r+1}({\mathbf{k}},{\mathbf{k}}_r ) - g_{r+1}({\mathbf{k}},{\mathbf{k}}_r ) \label{ff}\end{aligned}\ ] ] @xmath155 = -\frac{1}{2 } \left ( 3f_{r+1}({\mathbf{k}},{\mathbf{k}}_r ) - ( 2r+3 ) g_{r+1}({\mathbf{k}},{\mathbf{k}}_r ) \right ) \label{gg}\ ] ] substituting eq . ( [ ff ] ) and eq . ( [ gg ] ) into eq . ( [ recursion2 ] ) , we can finally derive the following relation in the high-@xmath1 limit , @xmath156 similarly , for @xmath157 we can show the following relation @xmath158 this ends the proof . we calculate the coefficients of the wh expansion in the high-@xmath1 limit , @xmath159^n \notag \\ & = \exp\left ( -\frac{k^2 \sigma_v^2}{2 } \right ) \delta_l(z,|{\mathbf{k}}-{\mathbf{k}}_{1r}|)\delta_l(z , k_1 ) \dots \delta_l(z , k_r ) f_{r+1}({\mathbf{k}}-{\mathbf{k}}_{1r},{\mathbf{k}}_r ) \notag \\ & = \exp\left ( -\frac{k^2 \sigma_v^2}{2 } \right ) \delta^{(r+1)}_{r+1}(z,{\mathbf{k}}-{\mathbf{k}}_{1r},{\mathbf{k}}_1,\dots,{\mathbf{k}}_r ) , \label{exp}\end{aligned}\ ] ] where we have used eqs . ( [ wh - spt ] ) , ( [ theorem ] ) , and @xmath160 . we define @xmath161 as @xmath162 note that we define the @xmath163-dependent quantities such as @xmath164 and @xmath165 by multiplying the scale factor @xmath20 , but we assume that the scale factor can be replaced by the growth factor @xmath166 in the general cosmological models , for which @xmath167 : @xmath168 and @xmath169 . this relation in eq . ( [ exp ] ) is equivalent to eq . ( 42 ) in @xcite from the relation between @xmath170 and @xmath171 in eq . ( [ delta_gamma ] ) . then , we describe the full power spectrum in the high-@xmath1 limit as @xmath172 where @xmath173 ^ 2 \notag \\ & = ( r+1 ) ! \int \frac{d^3k_1}{(2\pi)^3 } \dots \int \frac{d^3k_{r}}{(2\pi)^3 } \left [ f_{r+1}({\mathbf{k}}-{\mathbf{k}}_{1r},{\mathbf{k}}_r ) \right]^2 p_l(z,|{\mathbf{k}}-{\mathbf{k}}_{1r}| ) p_l(z , k_1 ) \dots p_l(z , k_r ) . \label{tree}\end{aligned}\ ] ] @xmath174 includes the contributions from all the irreducible diagrams . here , we take further high-@xmath1 limits in eq . ( [ tree ] ) . @xmath175 using eq . ( [ theorem ] ) , we show @xmath176 ^ 2 \notag \\ & \to \frac{(k^2 \sigma_v^2)^r}{r ! } p_l(z , k ) . \label{further limit}\end{aligned}\ ] ] note that the limit of @xmath177 @xmath178 is equal to the approximation that the most effective region of @xmath179 to each integral is around @xmath180 . however , since @xmath179 have integral range of @xmath181 , there necessarily exist the case of @xmath182 in the integral . for @xmath183 @xmath184 , we have @xmath185 ^ 2 \notag \\ & \to ( r+1 ) ! \int \frac{d^3q}{(2\pi)^3}\frac{d^3k_2}{(2\pi)^3 } \dots \int \frac{d^3k_{r}}{(2\pi)^3 } [ \delta^{(r+1)}(z,{\mathbf{q } } , { \mathbf{k}},{\mathbf{k}}_2 , \dots,{\mathbf{k}}_{r})]^2 , \label{}\end{aligned}\ ] ] where we define @xmath186 as @xmath187 . this is equal to the case of the high-@xmath1 limit . the same analysis is applied to arbitrary @xmath179 . therefore , the factor @xmath102 is multiplied in eq . ( [ further limit ] ) . from eq . ( [ further limit ] ) and eq . ( [ high - k power ] ) , we finally derive @xmath188 surprisingly , the solutions in the high-@xmath1 limit cancel each other , and the full power spectrum reduces to the linear power spectrum . this fact is well known in the 1-loop level of spt @xcite , but it is interesting that this cancellation also applies for the dominant terms in the high-@xmath1 limit in the full power spectrum . of course , it is really not that the full power spectrum becomes the linear power spectrum in the high-@xmath1 limit , because we have chosen only dominant terms in the high-@xmath1 limit in the proof of eq . ( [ theorem ] ) and eq . ( [ further limit ] ) , and the subleading terms can also affect the full power spectrum even in the high-@xmath1 limit . this result implies that the nonlinear corrections for the power spectrum generally tend to cancel each other , and result in small corrections as specifically known for 1 and 2-loop cases of spt . we now propose an appropriate interpolation between the low-@xmath1 and high-@xmath1 solutions . the low-@xmath1-solutions are the 1-loop solutions in spt , while the high-@xmath1 solutions are given by eq . ( [ high - k power ] ) derived in the previous subsection . in order to have an expression applicable for the case of @xmath118 , we use eq . ( [ theorem ] ) in the following @xmath189 then , for @xmath71 , we show @xmath190 \left [ \int \frac{d^3p}{(2\pi)^3 } \gamma({\mathbf{p } } ) \gamma(-{\mathbf{p } } ) p_l(p)\right]^{n-1 } \notag \\ & = \delta^{(1)}_3(k)\frac{2}{2^nn!}\left[- \frac{k^2}{6\pi^2 } \int dp p_l(p ) \right]^{n-1 } , \label{}\end{aligned}\ ] ] where we have denoted the 1-loop correction term in spt as @xmath191 then , we derive the approximate solution of @xmath120 as @xmath192 . \label{h1}\end{aligned}\ ] ] where we have used the general growth factor @xmath193 instead of the scale factor @xmath20 . the contribution to the power spectrum @xmath194 in eq ( [ hpower ] ) is @xmath195 ^ 2 . \label{h1}\ ] ] for low-@xmath1 , we can expand the exponential term as @xmath196 , and this lead the 1-loop correction , @xmath197 while , for the high-@xmath1 limit , @xmath120 becomes coincident with @xmath198 due to the good convergence of @xmath199 . next , for @xmath200 , we use the approximation of eq . ( [ high - k power ] ) . here , we further approximate @xmath201 because it is expensive to compute the terms in the case of @xmath202 due to their large multiple integrals . using the following approximation from eq . ( [ theorem ] ) , @xmath203 we derive the approximate solution of @xmath103 as , @xmath204 ^ 2 \notag \\ & \to \exp\left ( -k^2 \sigma_v^2 \right ) ( r+1 ) ! \frac{2!}{(r+1 ) ! } \frac{2!}{(r+1 ) ! } \left ( \frac{r+1}{2 } \right ) \int \frac{d^3k_1}{(2\pi)^3 } \left [ \delta_{2}^{(2)}(z,{\mathbf{k}}-{\mathbf{k}}_1,{\mathbf{k}}_1 ) \right]^2 ( k^2 \sigma_v^2)^{r-1}\notag \\ & \to \exp\left ( -k^2 \sigma_v^2 \right ) \frac{p_{22}(z , k)}{k^2 \sigma_v^2 } \frac{1}{r!}(k^2 \sigma_v^2)^{r } , \label{wh r>1}\end{aligned}\ ] ] where we multiply in the factor @xmath205 for the same reason as in eq . ( [ further limit ] ) . we have denoted another 1-loop correction term in the spt as , @xmath206 ^ 2 p_l(z,|{\mathbf{k}}-{\mathbf{p}}| ) p_l(z , p ) . \label{}\ ] ] indeed , if for @xmath207 the exponential factor is expanded as @xmath208 , the 1-loop correction , @xmath209 , is reproduced in eq . ( [ wh r>1 ] ) . finally , we achieve the approximate full power spectrum , @xmath210 ^ 2 + \exp\left ( -k^2 \sigma_v^2 \right ) \frac{p_{22}(z , k)}{k^2 \sigma_v^2 } \sum_{r=1}^{\infty}\frac{1}{r!}(k^2 \sigma_v^2)^{r } , \notag \\ p_{\rm af}(z , k ) & \equiv & \left [ \delta_l(z , k ) - \frac{2\delta_3^{(1)}(z , k)}{k^2 \sigma_v^2 } \left ( \exp\left(-\frac{k^2 \sigma_v^2}{2}\right ) -1 \right ) \right]^2 + \frac{p_{22}(z , k)}{k^2 \sigma_v^2}\left [ 1 - \exp\left ( -k^2 \sigma_v^2 \right ) \right ] . \label{result1}\end{aligned}\ ] ] this is the main result of this paper . this gives an appropriate interpolation between the low-@xmath1 solutions and the high-@xmath1 limit ones . we can derive the approximate solutions of each order in the wh expansion and spt using approximation such as eq . ( [ wh1ap ] ) and eq . ( [ whrap ] ) , and therefore we call this method as the `` approximate full wiener hermite ( afwh ) '' expansion method or `` approximate full perturbation theory '' . we can easily compute the solution of eq . ( [ result1 ] ) numerically , because the solution has only single or double integrals . we compare the approximate full power spectrum , @xmath211 , in eq . ( [ result1 ] ) with some other analytic predictions and @xmath2-body simulations . we mainly use @xmath2-body simulations presented in @xcite , but in sec . [ closure ] we also use other @xmath2-body results with higher resolutions in @xcite . it is plotted for the cosmological models with the _ wilkinson microwave anisotropy probe _ ( @xmath212 ) five year @xcite . cosmological parameters : @xmath213 , @xmath214 , @xmath215 , @xmath216 , @xmath217 , @xmath218 . the @xmath2-body simulation data with low - resolutions and high - resolutions in @xcite and @xcite were created by a public @xmath2-body code _ their initial conditions were generated by the _ 2lpt _ code @xcite at @xmath219 and @xmath220 , respectively . while the @xmath2-body simulations with low - resolutions were computed with a cubic box of size @xmath221 containing @xmath222 particles , the @xmath2-body results with high resolutions , called l11-n11 and l12-n11 , contain @xmath223 particles and were computed by combining the results with different box sizes , 2,048 @xmath224 and 4096 @xmath224 . although the realization of the simulations with high resolutions is only 1 , the simulations with low resolutions have the output data of 30 independent realizations , and consider the correction of the finite - mode sampling by @xcite . therefore , the size of each error bar for @xmath2-body results with low - resolutions becomes hard to see visually . for details of the @xmath2-body simulation used in this paper , see taruya et al . @xcite and okamura et al . @xcite . @xcite proposed a simple scheme to interpolate between the low-@xmath1 and high-@xmath1 solutions , based on the @xmath0-expansion method . in the scheme , the solutions are regularized so that the low-@xmath1 solutions become the ones in spt and the high-@xmath1 solutions become eq . ( [ high - k power ] ) . we call this scheme `` regularized @xmath0-expansion '' . from eqs . ( 20 ) ( 51 ) in @xcite , its 1-loop solution for the power spectrum is given by @xmath225 . \label{reg}\ ] ] since the wh expansion and the @xmath0-expansion are completely equivalent to each other , we can understand this solution as the truncation up to the second order in the wh expansion and write @xmath194 using the regularized @xmath0-expansion as @xmath226_{\rm reg } = \exp\left ( -k^2 \sigma_v^2 \right ) \left [ \delta_l + \delta_3 + \frac{k^2 \sigma_v^2}{2 } \delta_l \right]^2 . \label{hreg1}\ ] ] the difference between eq . ( [ h1 ] ) and eq . ( [ hreg1 ] ) is manner of interpolating between the solutions . since the regularized @xmath0-expansion method is a heuristic scheme , we have used the approximation of eq . ( [ wh1ap ] ) . when we ignore the contributions of @xmath227 , we derive the solution of the approximate wh expansion corresponding to the regularized @xmath0-expansion in eq . ( [ reg ] ) as @xmath228 ^ 2 + \exp(-k^2 \sigma_v^2)p_{22}(z , k ) . \label{wh12}\ ] ] we plot these two solutions , @xmath229 and @xmath230_{\rm reg}/p_l$ ] , and their fractional difference , @xmath231_{\rm reg } ) /p_{l}$ ] , at @xmath232 in fig . [ fig : propagator ] . on bao scales ( @xmath233 ) , the fractional difference is within 1 % . at high @xmath1 , our solution becomes slightly larger than the regularized @xmath0-expansion . however , there is no means of investigating which results are more accurate in detail , because on such scales the amplitude of @xmath194 decays enough due to the exponential factor to not largely contribute to the full power spectrum . in fig . [ fig:1loop ] , we plot the various analytic solutions with the 1-loop level corrections and @xmath2-body simulation result ( blue dashed : spt , green dashed : regularized @xmath0-expansion , black solid : lrt , orange solid : @xmath234 in eq . ( [ wh12 ] ) ; red solid : afwh in eq . ( [ result1 ] ) ; and black symbols : @xmath2-body result ) at @xmath232 . ] @xmath235 where we denote as @xmath236 . although lrt is very similar to the regularized @xmath0-expansion and our result , the complete correspondence ( e.g. , the origin of the exponential factor ) is not trivial . ] to easily see the bao , we plot the ratio of power spectrum to a smooth reference spectrum , @xmath237 , where the function @xmath238 is the linear power spectrum calculated from the smoothed transfer function neglecting the bao feature in @xcite . to investigate the agreement with @xmath2-body results in more quantitative ways , we also plot the fractional differences between @xmath2-body simulations and the predicted power spectrum @xmath239 , i.e. , @xmath240/p(k)$ ] ( blue : @xmath2-body results versus 1-loop spt ; green : regularized @xmath0-expansion ; black : lrt ; orange : @xmath234 ; red : afwh in eq . ( [ result1 ] ) ) . the regularized @xmath0-expansion , lrt , and @xmath234 are very similar that we can hardly see any difference . their solutions improve the overestimation of spt , but decay at low @xmath1 soon because of their exponential factor . on the other hand , we can find that the main difference of our result from the previous works with 1-loop level is the higher order of the wh expansion , @xmath241 . because of these terms , the afwh in eq . ( [ result1 ] ) ( red solid line ) does not decay and keeps the values well around those from @xmath2-body simulations on bao scales . one merit of our interpolation is that we can directly compare our approximate solutions with ones of each order in the perturbation theory . in previous works , the validity of the predicted power spectra has been verified only by comparing with the @xmath2-body results . however , we can verify the validity of our approximations , such as eq . ( [ h1 ] ) and eq . ( [ wh r>1 ] ) by comparing with the 2-loop solutions in spt . the 2-loop corrections are given by @xmath242 ^ 2 , \label{}\ ] ] where each term is calculated , respectively , as @xmath243 ^ 2 p_l(z,|{\mathbf{k}}-{\mathbf{p}}_1-{\mathbf{p}}_2| ) p_l(z , p_1 ) p_l(z , p_2 ) . \label{2-loop}\end{aligned}\ ] ] on the other hand , we show the corresponding approximate solutions using eq . ( [ wh1ap ] ) and eq . ( [ whrap ] ) as follows , @xmath244_{\rm ap } = \frac{1}{2}\left ( - \frac{k^2 \sigma_v^2}{2 } \right ) p_{13 } , \\ \nonumber p_{24 } & \to & \left [ p_{24 } \right]_{\rm ap } = -(k^2 \sigma_v^2)p_{22 } , \\ p_{33 } & \to & \left [ p_{33 } \right]_{\rm ap } = \left ( \frac{k^2 \sigma_v^2}{2 } \right ) p_{22},\\ p_{\rm 2loop } & \to&\left [ p_{\rm 2loop } \right]_{\rm ap } = - \frac{k^2 \sigma_v^2}{4}p_{13 } - \frac{k^2 \sigma_v^2}{2}p_{22 } + \left [ \delta_3^{(1 ) } \right]^2 . \label{2-loop apro}\end{aligned}\ ] ] here , we have not considered the approximation of the term @xmath245 ^ 2 $ ] , because it is the square of the 1loop term and we can easily compute it . cc + in fig . [ fig:2loop ] , we plot the correct 2-loop solutions , their approximate solutions , and their fractional difference , @xmath246_{\rm ap } - p]/p_{\rm nw}$ ] , ( @xmath247 , @xmath248 , @xmath249 , and @xmath250 ) , in each panel . we plot the solutions up to @xmath251 at @xmath232 , because the 2-loop corrections give a good result up to about these scales ( see @xcite ) . for @xmath252 and @xmath248 in the top panels , the approximate solutions respectively , are overestimated and underestimated by about 5% at @xmath253 . on the other hand , for @xmath249 in the bottom left panel , the approximate solution coincides very well with the precise solution within 1% . one may think that since there is a large difference between @xmath252 and @xmath248 , our approximation is not valid . however , remember that each correction term in the perturbation theory tends to cancel out , resulting in small corrections . therefore , if the approximate solution for @xmath252 is overestimated , it would be natural that there is an underestimation in other solutions such as @xmath248 to cancel out the overestimated solutions . as a result , for the full 2-loop corrections in the bottom right panel , the fractional difference becomes within 1% up to @xmath254 . finally , we compare with the closure theory ( second born ) in @xcite , which is one of the best predictions at the moment . in addition , we plot high - resolution @xmath2-body simulations presented by @xcite . in fig . [ fig : closure ] , we plot the power spectra from closure theory ( purple dashed ) , afwh in eq . ( [ result1 ] ) ( red solid ) , and @xmath2-body results with error bars ( black symbols : low resolution ; green symbols : high resolution ) at some redshifts ( @xmath255 , @xmath256 , @xmath257 , @xmath258 ) . the range of plotted scales is @xmath259 . we also plot the fractional differences ( purple : @xmath2-body result with low resolution versus closure ; red : afwh ; green : @xmath2-body result with high - resolution versus afwh ) overall , the predictions of afwh tend to overestimate the @xmath2-body simulations at low-@xmath1 ( bao scales ) slightly , and then begin to underestimate at high @xmath1 . the overestimation is due to the fact that we computed only the 1-loop level in spt precisely . in fact , our approximate solutions are slightly larger than the 2-loop spt solutions as shown fig . [ fig:2loop ] . therefore , to derive more precise prediction on bao scales , we need to calculate up to the 2-loop level corrections . the reason for the underestimation is that either the expression for the high @xmath1 limit would not apply perfectly to the range of calculation or the subleading contributions on small scales would become effective . we would need to compute the higher order of the wh expansion without the approximation to derive the precise nonlinearity in the high-@xmath1 range . although our results certainly give slightly less accuracy than that of closure theory , the difference is controlled within 1% on bao scales . we compute the two - point correlation function calculated from the power spectrum in eq . ( [ result1 ] ) , which is given by @xmath260 usually , we are not able to compute the correlation function in spt , because the integrand function diverge . for example , in the 1-loop level of spt , the predicted solution has the scale dependence of @xmath261 at high @xmath1 , because the approximate solutions of @xmath262 and @xmath209 are proportional to @xmath261 . therefore , the integrand in eq . ( [ correlation ] ) has the scale - dependence of @xmath263 : @xmath264 where we have used the behavior of the linear power spectrum at high-@xmath1 , @xmath265 . this solution diverge at high @xmath1 , and we are not able to evaluate the integration in the range of @xmath266 . on the other hand , we are able to compute the correlation function for afwh because the solution has the scale dependence like the linear power spectrum at high @xmath1 : @xmath267 ^ 2 + \frac{p_{22}(z , k)}{k^2 \sigma_v^2 } , \notag \\ & \propto p_l(z , k ) . \label{}\end{aligned}\ ] ] in the first line , we dropped the terms including exponential factor . the scale - dependence of @xmath268 and @xmath209 at high-@xmath1 are proportional to @xmath269 and @xmath261 . as a result , @xmath270 ( red line in the right panel of fig . [ fig : afpt ] ) is proportional to @xmath271 ( black dashed in fig . [ fig : afpt ] ) , and the integrand of eq . ( [ correlation ] ) converge like the linear power spectrum . furthermore , as long as we focus on the bao scales in real space ( @xmath272 $ ] ) , the behavior of the power spectrum on small scales in fourier space ( @xmath273 $ ] ) contributes very little to the result of the correlation function . therefore , we may truncate the wh expansion up to the appropriate order so that the integrand function converge to zero due to the exponential factor and we can easily compute the correlation function : @xmath274 ^ 2 \notag \\ & + \exp(-k^2 \sigma_v^2 ) p_{22}\left ( 1 + \frac{1}{2}(k^2 \sigma_v^2 ) + \frac{1}{3 ! } ( k^2 \sigma_v^2)^2 + \frac{1}{4 ! } ( k^2 \sigma_v^2)^3 + \frac{1}{5 ! } ( k^2 \sigma_v^2)^4\right ) , \label{wh6th}\end{aligned}\ ] ] where we have truncated the wh expansion up to the sixth order . we plot the solution of eq . ( [ wh6th ] ) in fig . [ fig : afpt ] at @xmath232 , where the difference between the afwh in eq . ( [ result1 ] ) ( red line ) and the solution of eq . ( [ wh6th ] ) ( red dashed ) up to @xmath275 $ ] is not visible , and the solution behaves like the ones of closure theory at high-@xmath1 . in the right panel of fig . [ fig : afpt ] , we further plot the solutions of the fifth ( orange line ) and seventh ( orange dashed ) order in the wh expansion . here , we adopt the sixth order solution of the wh expansion to compute the correlation function in eq . ( [ correlation ] ) . in fig . [ fig : correlation ] , we plot the analytic predictions of the correlation function , @xmath276 ( red line : afwh ; black dashed : linear theory ; black symbols : @xmath2-body results ) , and the fractional difference between the predicted correlation functions from afwh and the @xmath2-body simulation results , @xmath277/\xi(r)$ ] . our predictions explain the displacement of the location of the bao peaks and the smoothing of their amplitudes due to the non - linear effects . as a result , the fractional difference against the @xmath2-body results is within 13% . almost the same results are derived even for the second order of the wh expansion in eq . ( [ wh12 ] ) , because on small scales there is very little contribution to the correlation function around the bao peak . this fact is also well known also in other modified perturbation theories ( e.g. , @xcite ) . we have applied the wh expansion to the evolution equation of dark matter in newtonian gravity . it diagrammatically corresponds to the classification of the power spectrum in which each order includes all of the vertex loop contributions . it is proved that the wh expansion is mathematically equivalent to the @xmath0-expansion approach in the multi - point propagators method . even if wh expansion method is physically and mathematically useful for understanding the non - linearity of the evolution of dark matter , the validity of the finite truncation of the expansion is not clear and the difficulty in the calculation will remain . to resolve these difficulties , we proposed a way to include the effect of all orders by assuming that the high non - linear solutions are well approximated by the ones in the high-@xmath1 limit . namely we calculate only low order terms precisely and replace the high order solutions with the ones in the high-@xmath1 limit . it has been known in rpt that the matter density and velocity fluctuations of dark matter are exponential in the high-@xmath1 limit . we proved again this behavior in the context of spt using the wh expansion by proving that the kernel functions @xmath278 and @xmath279 take the form of eq . ( [ theorem ] ) in the high-@xmath1 limit . using the approximate kernel functions @xmath278 and @xmath279 , we proposed an appropriate interpolation between high-@xmath1 and low-@xmath1 solutions , and the approximate full power spectrum in eq . ( [ result1 ] ) , which approximately include the full order of spt . we compared our results with some other analytic predictions ( e.g. , regularized @xmath0-expansion , lrt , spt , and closure theory ) and @xmath2-body simulation results . since the wh expansion is equivalent to the @xmath0-expansion and the regularized @xmath0-expansion bases on the @xmath0-expansion , we can describe the first order of the wh expansion , @xmath194 , using the regularized @xmath0-expansion in eq . ( [ hreg1 ] ) . one of the difference between our result and the regularized @xmath0-expansion is the manner of interpolating between the high-@xmath1 and low-@xmath1 solutions , but this difference slightly affects the predicted power spectrum . another difference is that we consider the higher order of the wh expansion approximately . as a result , even for the 1-loop level , the predicted power spectrum in eq . ( [ result1 ] ) does not decay due to the exponential factor as shown in fig . [ fig:1loop ] , and results in good agreement with the @xmath2-body simulation on bao scales . the validity of the various modified perturbation theory ( e.g. , lrt , rpt , closure theory , ) predictions is usually verified only by comparing with the @xmath2-body simulations . however , we can also verify our approximation by comparing with the solutions with the spt 2loop level . in fig.[fig:2loop ] , we showed that the fractional difference between the approximate solutions and the precise solutions with the spt 2-loop level is within 1 % on bao scales ( @xmath280 ) for the @xmath212 five year cosmological parameters at @xmath232 . we also compared with the closure theory which is one of the best prediction at a moment , and the accuracy of afwh in eq . ( [ result1 ] ) is comparable to or slightly less than the ones in the closure theory , with the fractional difference within 1% on bao scales . finally , we computed the two - point correlation function for afwh . we can compute the correlation functions because the predicted power spectrum in afwh converges like the one from linear theory . since the contributions on small scales do not affect the values of the correlation function , one may use eq . ( [ wh6th ] ) to compute the correlation function . this solution has the same behavior as eq . ( [ result1 ] ) on bao scales and decay on small scales due to the exponential factor in fig . [ fig : afpt ] . the predicted correlation functions from the afwh agree very well with the @xmath2-body simulation results , and the fractional difference is within @xmath281% . we could use and apply our results to various studies of the nonlinear evolution of dark matter ( e.g. , redshift distortion effect , bias effect , and bispectrum , etc . ) , because our prescription is easy and gives good results that are comparable to closure theory , and furthermore the computational time is very rapid . we would like to thank t. nishimichi and a. taruya for providing us with the numerical simulation results and useful comments and y. itoh , and t. okamura for useful discussion . this work is supported in part by the gcoe program `` weaving science web beyond particle - matter hierarchy '' at tohoku university and by a grant - in - aid for scientific research from jsps ( no . 24 - 3849 for nss and nos . 18072001 , 20540245 for tf ) as well as by core - to - core program `` international research network for dark energy . '' tf thanks luc branchet and the institute of astronomical observatory , paris for warm hospitality during his stay in the last stage of the present work . d. j. eisenstein , w. hu , and m. tegmark , _ cosmic complementarity : h(0 ) and omega(m ) from combining cmb experiments and redshift surveys _ , _ astrophys . j. _ * 504 * ( 1998 ) l57l61 , [ http://xxx.lanl.gov/abs/astro-ph/9805239[astro-ph/9805239 ] ] . collaboration , d. j. eisenstein _ et . _ , _ detection of the baryon acoustic peak in the large - scale correlation function of sdss luminous red galaxies _ , _ astrophys . j. _ * 633 * ( 2005 ) 560574 , [ http://xxx.lanl.gov/abs/astro-ph/0501171[astro-ph/0501171 ] ] . seo and d. j. eisenstein , _ probing dark energy with baryonic acoustic oscillations from future large galaxy redshift surveys _ , _ astrophys.j . _ * 598 * ( 2003 ) 720740 , [ http://xxx.lanl.gov/abs/astro-ph/0307460[astro-ph/0307460 ] ] . c. blake and k. glazebrook , _ probing dark energy using baryonic oscillations in the galaxy power spectrum as a cosmological ruler _ , _ astrophys.j . _ * 594 * ( 2003 ) 665673 , [ http://xxx.lanl.gov/abs/astro-ph/0301632[astro-ph/0301632 ] ] . k. glazebrook and c. blake , _ measuring the cosmic evolution of dark energy with baryonic oscillations in the galaxy power spectrum _ , _ astrophys.j . _ * 631 * ( 2005 ) 120 , [ http://xxx.lanl.gov/abs/astro-ph/0505608[astro-ph/0505608 ] ] . m. shoji , d. jeong , and e. komatsu , _ extracting angular diameter distance and expansion rate of the universe from two - dimensional galaxy power spectrum at high redshifts : baryon acoustic oscillation fitting versus full modeling _ , _ astrophys . j. _ * 693 * ( 2009 ) 14041416 , [ http://xxx.lanl.gov/abs/0805.4238[arxiv:0805.4238 ] ] . n. makino , m. sasaki , and y. suto , _ analytic approach to the perturbative expansion of nonlinear gravitational fluctuations in cosmological density and velocity fields _ , _ phys . rev . _ * d46 * ( 1992 ) 585602 . r. scoccimarro and j. frieman , _ loop corrections in non - linear cosmological perturbation theory ii . two - point statistics and self - similarity _ , _ astrophys . j. _ * 473 * ( 1996 ) 620 , [ http://xxx.lanl.gov/abs/astro-ph/9602070[astro-ph/9602070 ] ] . f. bernardeau , s. colombi , e. gaztanaga , and r. scoccimarro , _ large - scale structure of the universe and cosmological perturbation theory _ , _ phys . * 367 * ( 2002 ) 1248 , [ http://xxx.lanl.gov/abs/astro-ph/0112551[astro-ph/0112551 ] ] . d. jeong and e. komatsu , _ perturbation theory reloaded : analytical calculation of non - linearity in baryonic oscillations in the real space matter power spectrum _ , _ astrophys . j. _ * 651 * ( 2006 ) 619626 , [ http://xxx.lanl.gov/abs/astro-ph/0604075[astro-ph/0604075 ] ] . d. jeong and e. komatsu , _ perturbation theory reloaded ii : non - linear bias , baryon acoustic oscillations and millennium simulation in real space _ , _ astrophys . j. _ * 691 * ( 2009 ) 569595 , [ http://xxx.lanl.gov/abs/0805.2632[arxiv:0805.2632 ] ] . a. taruya , t. nishimichi , s. saito , and t. hiramatsu , _ non - linear evolution of baryon acoustic oscillations from improved perturbation theory in real and redshift spaces _ , _ phys . * d80 * ( 2009 ) 123503 , [ http://xxx.lanl.gov/abs/0906.0507[arxiv:0906.0507 ] ] . t. hiramatsu and a. taruya , _ chasing the non - linear evolution of matter power spectrum with numerical resummation method : solution of closure equations _ , _ phys . * d79 * ( 2009 ) 103526 , [ http://xxx.lanl.gov/abs/0902.3772[arxiv:0902.3772 ] ] . f. bernardeau , m. crocce , and r. scoccimarro , _ multi - point propagators in cosmological gravitational instability _ , _ phys . _ * d78 * ( 2008 ) 103521 , [ http://xxx.lanl.gov/abs/0806.2334[arxiv:0806.2334 ] ] . p. valageas , _ a new approach to gravitational clustering : a path- integral formalism and large - n expansions _ , _ astron . _ * 421 * ( 2004 ) 2340 , [ http://xxx.lanl.gov/abs/astro-ph/0307008 [ astro - ph/0307008 ] ] . t. matsubara , _ resumming cosmological perturbations via the lagrangian picture : one - loop results in real space and in redshift space _ , _ phys . _ * d77 * ( 2008 ) 063530 , [ http://xxx.lanl.gov/abs/0711.2521 [ arxiv:0711.2521 ] ] . t. matsubara , _ nonlinear perturbation theory integrated with nonlocal bias , redshift - space distortions , and primordial non- gaussianity _ , _ phys . * d83 * ( 2011 ) 083518 , [ http://xxx.lanl.gov/abs/1102.4619[arxiv:1102.4619 ] ] . t. matsubara , _ nonlinear perturbation theory with halo bias and redshift- space distortions via the lagrangian picture _ * d78 * ( 2008 ) 083519 , [ http://xxx.lanl.gov/abs/0807.1733 [ arxiv:0807.1733 ] ] . t. okamura , a. taruya , and t. matsubara , _ next - to - leading resummation of cosmological perturbations via the lagrangian picture : 2-loop correction in real and redshift spaces _ , _ jcap _ * 1108 * ( 2011 ) 012 , [ http://xxx.lanl.gov/abs/1105.1491[arxiv:1105.1491 ] ] . collaboration , e. komatsu _ et . _ , _ five - year wilkinson microwave anisotropy probe ( wmap1 ) observations : cosmological interpretation _ , _ astrophys . j. suppl . _ * 180 * ( 2009 ) 330376 , [ http://xxx.lanl.gov/abs/0803.0547[arxiv:0803.0547 ] ] . m. crocce , s. pueblas , and r. scoccimarro , _ transients from initial conditions in cosmological simulations _ , _ mon . not . * 373 * ( 2006 ) 369381 , [ http://xxx.lanl.gov/abs/astro-ph/0606505[astro-ph/0606505 ] ] . t. nishimichi _ et . _ , _ modeling nonlinear evolution of baryon acoustic oscillations : convergence regime of n - body simulations and analytic models _ , http://xxx.lanl.gov/abs/0810.0813[arxiv:0810.0813 ] .

we perform a detailed comparison of the phase - space density traced by the particle distribution in gadget simulations to the result obtained with a spherical vlasov solver using the splitting algorithm . the systems considered are apodized hnon spheres with two values of the virial ratio , and . after checking that spherical symmetry is well preserved by the-body simulations , visual and quantitative comparisons are performed . in particular we introduce new statistics , correlators and entropic estimators , based on the likelihood of whether-body simulations actually trace randomly the vlasov phase - space density . when taking into account the limits of both the-body and the vlasov codes , namely collective effects due to the particle shot noise in the first case and diffusion and possible nonlinear instabilities due to finite resolution of the phase - space grid in the second case , we find a spectacular agreement between both methods , even in regions of phase - space where nontrivial physical instabilities develop . however , in the colder case , , it was not possible to prove actual numerical convergence of the-body results after a number of dynamical times , even with particles . -1 cm [ firstpage ] gravitation methods : numerical galaxies : kinematics and dynamics dark matter

1504.07337

stars in galaxies and dark matter in the universe can be modeled in phase - space as self - gravitating collisionless fluids obeying the vlasov - poisson equations : @xmath5 where @xmath6 represents the phase - space density at position @xmath7 and velocity @xmath8 , @xmath9 is the gravitational potential , and @xmath10 is the gravitational constant . in general , these equations do not have simple analytical solutions . they are therefore often solved numerically . the most widely used numerical scheme is the @xmath2-body approach and there exist many different implementations , which mainly differ from each other in the way poisson equation is solved ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * for reviews on the subject ) . the @xmath2-body method attempts to sample the phase - space density by an ensemble of dirac functions that represent particles interacting with each other through gravitational force . in order to avoid numerical artefacts due to the @xmath11 divergence of the force at small distances , the gravitational potential is usually replaced by an effective one so that the force is smoothed at scales smaller than a softening parameter @xmath12 . this procedure corresponds to assuming that the particles are clouds of size @xmath13 interacting with each other . approximating the phase - space density with macro - particles , however , has its own limitation . in particular , the close @xmath2-body encounter is one of the most notable sources of numerical artefacts , in addition to more subtle collective effects induced by the discrete nature of the distribution of the particles ( see , e.g. * ? ? ? * ; * ? ? ? ? * ; * ? ? ? * ; * ? ? ? of course , the time integration scheme and the way to solve the poisson equation numerically are well - known sources of errors , even though not particular to the @xmath2-body method . there are several previous studies that discussed the limitations of the @xmath2-body results , including underestimating strong numerical artefacts , particularly in the cold case where the initial velocity dispersion is null ( see , e.g. , * ? ? ? * ; * ? ? ? * ) , and long - term nonlinear resonant modes induced by the discrete nature of the particles ( see , e.g. , * ? ? ? * ; * ? ? ? we also note that it is not yet obvious that the fine inner structure of dark matter halos is completely understood from physical and even numerical points of view , despite numerous intensive convergences studies of the @xmath2-body approach ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? it is therefore highly desired to develop alternative numerical methods to the traditional @xmath2-body approach so that one can understand better its validity and fundamental limitations . in the cold case , relevant to the current paradigm of cold dark matter scenario , the phase - space distribution function is supported by a three - dimensional sheet evolving in six - dimensional phase - space , which can be partitioned in a continuous way with an ensemble of tetrahedra as proposed in recent works ( see , e.g. , * ? ? ? * ; * ? ? ? unfortunately , the increasing complexity of the structure of the system during evolution requires more and more sampling elements , and the computational cost becomes prohibitive after several dynamical time - scales . in this article , we consider the warm case , in which the system presents a non - negligible initial local velocity dispersion component relative to gravitational potential energy . in this case , the phase - space distribution function has to be sampled on a 6-dimensional mesh , which makes again the computational cost very high . therefore , we shall restrict to spherical systems , hence reducing the actual number of dimensions of the dynamical setup to three . there exist many methods to solve the vlasov - poisson equations in the warm case , mainly developed in plasma physics . one of the most famous solvers is the splitting algorithm of @xcite and its numerous extensions ( see , e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * but this list is far from complete ) . this algorithm , that we shall adopt below , exploits directly the liouville theorem : the phase - space density @xmath14 is conserved along motion . then the equations of the dynamics during each time step are divided into `` drift '' and `` kick '' parts according to hamiltonian dynamics and are solved backwards : @xmath15 where @xmath16 is computed from @xmath17 . in practice the phase - space distribution function is sampled on a mesh , and each step is performed by using tracer particles located at mesh sites and following the equations of motion split as above . resampling of @xmath17 , @xmath18 and finally the phase - space distribution function at the next time step is performed by using an interpolation , e.g. based on the spline method . the splitting scheme was applied for the first time in astronomy in early 1980 s , to one dimensional systems @xcite , galactic disks @xcite and spherical systems @xcite . nevertheless , it has been almost forgotten since then except for a few contributions ( e.g. , * ? ? ? * ; * ? ? ? * ) that include a recent preliminary investigation of the algorithm in full 6-dimensional phase - space @xcite . as mentioned above , however , solving fully six - dimensional phase - space problems with sufficient accuracy is still very unrealistic now . in this article , therefore , we focus on spherical systems , where phase - space is only three dimensional : the three coordinates of interest are the radial position @xmath19 , the radial velocity @xmath20 and the angular momentum @xmath21 . following earlier works performed in the framework of one dimensional gravity ( see , e.g. , * * ) , we carry out a detailed comparison between an @xmath2-body code , gadget @xcite , and an improved version of the splitting algorithm implementation by @xcite , vlasolve . our goal is to check how well the particle distribution in gadget traces the phase - space density obtained from vlasolve , and to see how the results depend on various parameters of the simulations , in particular the number of particles in the @xmath2-body simulations and the spatial resolution in the vlasov code . we would however like to emphasize here that the purpose of this article is not to compare the performance of the two codes from the view - point of computational cost . while a fairly good physical insight is obtained through visual inspection of the resulting phase - space density plots , we also present a more quantitative comparison . to do so , we introduce correlators and entropic estimators based on a likelihood approach , ans ask whether the @xmath2-body simulations can be considered as local poisson realizations of the vlasov code phase - space density . because of our restrictive choice of the geometry of the system , it is important to simulate spherical configurations that are known to be stable against small anisotropic perturbations induced by the shot noise of the particles . indeed , we shall use the public treecode gadget without any specific modification to enforce spherical dynamics . although an alternative approach consisting in enforcing pure radial dynamics in gadget ( see , e.g. , * ? ? ? * ) may facilitate comparisons with the vlasov code , we do not adopt this approach in order to avoid any possible subtle biases in the analyses . in this respect , the hnon sphere @xcite is particularly suited for our purpose since it is known to preserve well its spherical nature during the course of dynamics even when being simulated with a @xmath2-body technique and , in particular , it is not prone to radial orbit instability ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in this configuration , the initial phase - space distribution function is isotropic and gaussian distributed in velocity space and given by @xmath22 with @xmath23 , the total mass of the system . in the simulations discussed in this article , we work in units where @xmath24 , and the initial radius of the hnon sphere and its total mass are chosen to be @xmath25 which fixes @xmath26 in equation ( [ eq : henons ] ) once the virial ratio is given . we shall consider `` warm '' and `` cold '' settings , which correspond to the initial virial ratio @xmath27 of @xmath28 and @xmath29 , respectively , where @xmath30 and @xmath31 are the total kinetic and potential energy of the system . the two classes of initial conditions exhibit distinct features , in particular concerning the metastable state to which the system relaxes through phase mixing . the warm system builds a core - halo structure , with the halo displaying a power - law profile @xmath32 ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? in contrast , the cold system develops a more concentrated smaller core ( see , e.g. , * ? ? ? * ; * ? ? ? * ) , but never reaches a strictly stationary regime because a significant fraction of the mass acquires positive energy and escapes from the system ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? this article is organized as follows . in [ sec : thesimus ] we describe our vlasov solver , vlasolve . section [ sec : gadpres ] provides information about the @xmath2-body runs and the parameters used in gadget . in [ sec : sphericity ] , we check that the @xmath2-body simulations stay indeed spherical during evolution . section [ sec : visu ] presents a visual inspection of the phase - space density , which is followed by a quantitative statistical analysis in [ sec : statana ] . finally , [ sec : discussion ] summarizes and discusses our present results . under spherical symmetry , the vlasov equation reads @xmath33 where @xmath20 is the radial component of the velocity , @xmath21 is the angular momentum , @xmath34 is the mass inside a sphere of radius @xmath19 . our code vlasolve solves equation ( [ eq : vl ] ) numerically with the splitting algorithm , following closely @xcite . phase space is discretized into a rectangular mesh of size @xmath35 for @xmath36 , @xmath37 , and @xmath38 . more specifically , we use a logarithmically equal interval for @xmath19 , a linearly equal interval for @xmath20 . the @xmath39-bin of the angular momentum slice corresponds to the interval @xmath40 $ ] and is represented by @xmath41 . we modify the splitting algorithm using the fact that the angular momentum is an invariant of the hamiltonian system . hence , one may treat each slice with a different value of @xmath21 in phase - space independently , except for gravitational coupling via the poisson equation . we include the inertial component of the force , @xmath42 , in the `` drift '' step ( equations [ eq : ck1 ] and [ eq : ck3 ] ) , while the `` kick '' step ( equation [ eq : ck2 ] ) corresponds solely to gravitational force : @xmath43 , \\ f^{**}(r , v , j)&=&f^*(r , v+ g m_r / r^2 \delta t , j ) , \\ f(r , v , j , t+\delta t)&=&f^{**}[r^*(-\delta t/2),v^*(-\delta t/2),j],\end{aligned}\ ] ] where @xmath44 and @xmath45 solve analytically the motion in absence of gravity starting from coordinates @xmath46 in phase - space ( see , e.g. , * ? ? ? * ) : @xmath47 ^ 2 + j^2 } { 2 h_{\rm k}}},\\ { v}^*(h ) & = & { \rm sgn}(v ) \sqrt{2 h_{\rm k}- \frac{j^2}{{r^*(h)}^2}},\end{aligned}\ ] ] with @xmath48 ( when @xmath49 , these equations are valid until @xmath50 ) . because a non - zero angular momentum bends the trajectories in @xmath51 space , the drift step requires a two - dimensional interpolation of the phase - space distribution function in @xmath51 space , while the kick step , which only modifies the velocities , can be completed with a one - dimensional interpolation . we follow @xcite , and carry out the interpolations using third - order splines . in this interpolation scheme , however , the positivity of the phase - space distribution function is not warranted , and numerical aliasing and diffusion effects are expected when the phase - space distribution function varies over scales of the order of , or smaller than , the mesh element size . in order to reduce such numerical artefacts , we modify equation ( [ eq : henons ] ) as follows : @xmath52 , \quad r \leq r_{\rm h } , \label{eq : apodization}\end{aligned}\ ] ] with @xmath53 . then we recompute @xmath54 in equation ( [ eq : henons ] ) so that the total mass remains unity . this apodization slightly changes the actual values of the virial ratio to @xmath55 and @xmath56 , although we shall still denote them by @xmath1 and @xmath57 just for simplicity . it may also modify the long - term dynamical properties of the original hnon sphere relative to what is expected . this is why we check again the extent to which the spherical nature of the system is retained in the @xmath2-body simulations ( [ sec : sphericity ] ) . adopting a logarithmic binning for @xmath19 is well suited for tracing small - scale features around the center of the system . this implies , however , that radii smaller than a finite minimum value @xmath58 are missing from the computing domain . a conventional trick to overcome the problem is to assume a reflecting boundary at @xmath59 ( see , e.g .. * ? ? ? * ; * ? ? ? * ) . usually , a systematic time - lag between orbits in this method is neglected : particles reaching the reflective kernel boundary instantly travel the @xmath60 distance through the central region , while they should actually take a finite time depending on their radial velocity and angular momentum . in vlasolve , we improve the reflecting sphere method by taking into account the actual time spent by particles travelling inside the region @xmath61 , which is made easily possible by neglecting the gravitational force . technical details about the implementation are provided in appendix [ sec : monr0 ] . to complete algorithmic details , appendix [ sec : para ] discusses the hybrid parallelization of vlasolve with openmp and mpi libraries . in this paper , we perform 4 simulation runs with different resolutions , each for @xmath3 and @xmath1 ( table [ tab : resovla ] ) . to cover the dynamical range of interest , the computing mesh uses @xmath62 , @xmath63 and @xmath64 . the maximum amplitude of the velocity is @xmath65 and 4 for @xmath66 and @xmath57 , respectively . with this choice of the parameters , the computational domains are sufficiently large to contain all the system up to the end of the simulations , which corresponds to @xmath67 for @xmath66 and @xmath68 for @xmath3 . as will be illustrated later in phase - space density plots , these final epochs are sufficient for the system to have relaxed at the coarse level to a meta - stable state through mixing . strictly speaking , this is not the case in the @xmath3 case because a fraction of the mass escapes from the system ( see , e.g. , * ? ? ? * ; * ? ? ? * ) , as already mentioned in the introduction . we adopt a constant time step @xmath69 throughout each simulation . just to stay on the conservative side , we choose a resolutely small value of @xmath69 , despite the increased computational cost . note however that excessively small time step might artificially increase diffusion effects related to successive interpolations of the phase - space distribution function @xcite . in appendix [ sec : resoeffects ] , a comparison among all the simulations is performed for @xmath3 . it indicates that diffusion and aliasing effects discussed earlier are indeed significant , despite the apodization of initial conditions , but do not seem to affect the dynamical properties of the system . note that is tempting to undersample angular momentum space since @xmath21 is an invariant of the dynamics . however , we show in this appendix that it is not wise to do so , because it can provoke nonlinear instabilities after a few dynamical times . we perform the @xmath2-body simulations using the latest version of the gadget-2 code @xcite . only the treecode part of this `` treepm '' algorithm is employed . the particle number is varied from @xmath70 to @xmath71 for @xmath3 and @xmath1 . we also run an additional simulation with @xmath4 for @xmath3 . we choose the parameters for gadget runs as follows : * the softening length of the gravitational force is set as @xmath72 , that is about @xmath73 of the initial mean interparticle distance @xmath74 ( this estimate neglects the effects of the apodization [ eq : apodization ] ) . * in gadget , each particle has its individual time step bounded by @xmath75 $ ] , where @xmath76 is the acceleration of the particle and @xmath77 is a control parameter . we choose @xmath78 and @xmath79 . * the tolerance parameter controlling the accuracy of the relative cell - opening criterion ( parameter designed by errtolforceacc in the documentation of gadget , see equation 18 of * ? ? ? * ) is set as @xmath80 . appendix [ app : instab ] presents the effects of changing these parameters on the phase - space distribution function for simulations with @xmath81 particles and a virial ratio of @xmath3 . these analyses , performed at @xmath82 , confirm that the parameters used for the simulations of this paper are reasonable . interestingly , changing the softening length by large factors does not influence much the results , as already noticed previously in the literature ( see , e.g. * ? ? ? * ) , as long as it is kept small enough . before presenting comparisons between gadget and vlasolve , it is necessary to make sure that the sphericity of the system is preserved in the gadget simulations because our vlasov runs are performed assuming exact spherical symmetry . figure [ fig : sphe ] shows , for different values of the number of particles @xmath2 , the evolution with time of the ratios @xmath83 and @xmath84 , where @xmath85 are the eigenvalues of the inertia tensor of the particle distribution . the dashed regions correspond to the one sigma zone obtained from an ensemble of 100 local poisson realizations of the spherical density @xmath86 , which is estimated from interpolation over spherical shells from the gadget particles . from the measurements in fig . [ fig : sphe ] , deviations from spherical symmetry due to the particle shot noise can be roughly scaled to @xmath87 where @xmath88 and @xmath89 are the variances of @xmath83 and @xmath84 obtained from the dispersion over the 100 realizations . note that equation ( [ eq : spherr ] ) is not intended to be accurate . the asphericity due to discreteness should depend on details of the density profile , as shown in fig . [ fig : sphe ] . while it would be possible to compute in a perturbative way the quantities in equation ( [ eq : spherr ] ) from statistical analysis of the inertia tensor assuming @xmath90 and using error propagation formulae , this is a cumbersome exercise far beyond the scope of this paper . we also note that another possible source of errors comes from the position of the center of the system . indeed , an inaccurate determination of the center obviously worsens the apparent agreement with spherical symmetry . in the measurements presented in fig . [ fig : sphe ] , the inertia matrix is not computed with respect to the center of gravity of the particle distribution , which can be affected by the fact that some particles can get far away from the system through @xmath2-body relaxation . instead , we determine the center of the system using an iterative procedure trying to optimize the match of the phase - space distribution function with that of the vlasov code , as detailed in [ sec : cordef ] . this procedure is not free from errors either , and may contribute to the fluctuations observed in the curves of fig . [ fig : sphe ] . inspection of fig . [ fig : sphe ] shows that the measured ratios @xmath83 and @xmath84 behave differently in the @xmath66 and @xmath3 simulations . in the @xmath66 case , the agreement of the measurements with the poisson prediction is in general good , with a slight trend to ellipticity , except for the top red curve and the bottom green curve where the deviation from spherical symmetry is larger than the poisson expectation . still , in the case of @xmath66 , the system remains to a very good approximation spherical for all values of @xmath2 , given the expected deviations due to pure statistical noise . the curves representing the eigenvalue ratios are more steady for @xmath3 than for @xmath66 , which might be slightly puzzling at first sight . however , a very plausible explanation of this difference is that the initial velocity dispersion is larger for @xmath66 than for @xmath3 , hence adding a more prominent random component to the time behavior of the deviation from sphericity . regarding @xmath3 , deviations from spherical symmetry are clearly more significant compared to local poisson expectations after @xmath91 , roughly the collapse time of the sphere . while the @xmath70 run exhibits a deviation larger than 10 percent , spherical symmetry is confirmed to be a good approximation for @xmath92 . finally , we also check deviations from spherical symmetry for subsets of particles in excursions corresponding to @xmath93 , where @xmath94 is the phase - space distribution function measured in the @xmath95 vlasolve runs . for each value of the virial ratio , two thresholds @xmath96 are chosen such that the excursions contained initially about 90 and 60 percent of the total mass ( see bottom panels of fig . [ fig : entro ] below ) . given the uncertainties in the measurements , the above conclusions still hold : the properties of the deviations from spherical symmetry , that we do not show here for simplicity , do not indeed depend significantly on radius . we only notice a slight improvement in the @xmath66 case when considering particles in the excursions . now we are ready to perform direct comparisons between the vlasov and @xmath2-body simulation results . for this purpose , we consider the phase - space density at different epochs ( figs . [ fig:0v5_12 ] to [ fig:0v1_all ] below ) . to be more specific , we plot the constant angular momentum slice of @xmath97 at @xmath98 , and its integral over the angular momentum : @xmath99 figures [ fig:0v5_12 ] and [ fig:0v5_allj ] plot @xmath100 and @xmath101 , respectively , for the vlasolve and gadget simulations of the warm case , @xmath66 . in both figures , snapshots at @xmath102 , 50 , 80 and 100 are plotted from left to right . the panels correspond to the vlasolve runs with @xmath103 and @xmath104 , the gadget runs with @xmath105 , @xmath106 and @xmath107 , from top to bottom . the overall conclusion of the visual inspection of figs . [ fig:0v5_12 ] and [ fig:0v5_allj ] is that the vlasov solver and the @xmath2-body code exhibit very good agreement with each other , probably even much more than expected . in particular , both results present a remarkably similar instability in the region @xmath108 , even in details , showing a surprising reliability of the conventional @xmath2-body approach for these particular initial conditions . however , before reaching this conclusion , one has to take into account several limiting factors . in particular , we should bear in mind the fact that the vlasolve simulations are subject to significant diffusion , which smears out fine details of the phase - space distribution function . this diffusion effect is clearly visible at @xmath109 , when comparing the outer filamentary structures observed in the vlasov simulations to the @xmath2-body result . putting aside this coarse - graining effect , the structures are exactly similar in both the @xmath2-body and vlasov simulations at @xmath110 , even including small gaps in the phase - space distribution function related to nonlinear instabilities that start building up . these instabilities grow further at later epochs . they are considerably smeared out in the @xmath104 vlasolve simulation but unquestionably present . adding resolution in @xmath51 space ( at the cost of resolution in @xmath21 ) improves the agreement with gadget , which confirms that the instabilities observed in the gadget simulations are physical and not of numerical nature . figure [ fig:0v5_12 ] indicates that lowering the number of particles in the @xmath2-body simulations may be interpreted as a coarse - graining : it makes finer details more fuzzy but still keeps global features of the phase - space density correctly . we also note that using a small number of slices in @xmath21 in the vlasov solver does not seem to alter the dynamical properties of the system despite the considerable level of aliasing it introduces . the situation is more complicated for the cold case , @xmath3 ( figs . [ fig:0v1_12 ] and [ fig:0v1_all ] ) . up to @xmath111 , the above conclusions for @xmath66 are still valid . however , some instabilities emerge at @xmath111 in the gadget simulations with @xmath112 particles as well as the @xmath113 vlasov run . until this epoch , the @xmath114 and the @xmath104 simulations agree perfectly with each other ( modulo the smearing effects already discussed above ) and present a smooth phase - space density without any sign of instability . on the other hand , the other simulations exhibit slightly irregular phase - space density . such a trend is easily seen in fig . [ fig:0v1_12 ] , even though not so obvious in the @xmath113 vlasov simulation . these irregularities appear as well in the @xmath114 simulations but at later epochs , and then develop in a dramatic way . a careful inspection of successive snapshots of the simulations indeed suggests that the onset of these irregular patterns comes later with increasing @xmath2 . as discussed in appendices [ app : instab ] and [ app : resvlaeff ] , these instabilities result from the discrete nature of the system in the @xmath2-body case , and from the aliasing effect due to sparse - sampling of the angular momentum space in the vlasov code . since the pattern of the instabilities changes significantly from one simulation to another unlike the @xmath66 case , they should be due to numerical , not physical , origin . as shown in appendix [ app : instab ] , their presence is very insensitive to the choice of softening , time step or parameters controlling force accuracy in gadget . they can therefore be reduced only by increasing the number of particles and the resolution in the gadget and vlasolve simulations , respectively . it is important to notice that even the @xmath4 result might be insufficient to describe properly the system at late epochs . in the @xmath104 vlasov simulation , the phase - space distribution function seems to be rather smooth at all times and the system is free of instability , contrarily to the @xmath66 case . however , it is difficult at this point to know if actual physical instabilities build up at late times in the @xmath3 case , because diffusion in the vlasov simulation might prevent the appearance of some unstable modes . while the irregular patterns observed in figs . [ fig:0v1_12 ] and [ fig:0v1_all ] are definitely of numerical nature , the fact that they develop so easily may indicate that the system is prone to react nonlinearly to small perturbations . uneven gaps between the filaments of the phase - space density can be observed at @xmath82 ( third column of fig . [ fig:0v1_12 ] ) , even in the @xmath104 vlasov simulation , and one might expect that they correspond to seeds of actual physical instabilities . in this respect , the system might actually develop , at some point , physical unstable modes . these results are quite suggestive of what was obtained previously with a spherical shell code for cold and self - similar systems @xcite . even with our @xmath4 particle simulation , it is not clear whether these unstable modes dominate over collective effects due to discreteness . a better understanding of the phenomenon would require a convergence study using even higher - resolution simulations . to perform a more accurate analysis , one can try to quantify to which extent the particle distribution in the @xmath2-body simulations can be considered as a local poisson process of the phase - space density calculated in the semi - lagrangian code . to do so , we use , in addition to entropic measurements described further , the following correlators , @xmath115 with @xmath116^k , \label{eq : cor2 } \\ \kappa_k & = & \int [ f(\omega)]^{k+1 } { \rm d}\omega . \label{eq : cor3}\end{aligned}\ ] ] in these equations , @xmath117 is a positive integer , @xmath94 the vlasolve phase - space density , @xmath118 the total mass , @xmath119 and @xmath120 , where @xmath121 , @xmath122 and @xmath123 are respectively the radial position , radial velocity and angular momentum of each particle of the gadget simulation . for a point set randomly sampling a smooth density distribution @xmath124 , the probability density @xmath125 of having a given particle at phase - space position @xmath126 is independent from the rest of the particle distribution and is simply proportional to @xmath127 : @xmath128 the density probability of having @xmath2 particles at respective positions @xmath129 , @xmath130 , @xmath131 , @xmath132 is given by @xmath133 ensemble averaging of @xmath134 under the law @xmath124 then reads @xmath135^k\ { \cal p}(\omega_1,\cdots,\omega_n)\ { \rm d}\omega_1 \cdots { \rm d}\omega_n , \nonumber \\ & & \\ & = & \frac{m}{n } \sum_{i=1}^n \int_{\omega } \frac{1}{m } [ f(\omega_i)]^{k } g(\omega_i ) { \rm d}\omega_i , \nonumber \\ & & \quad \quad \quad \quad \times \prod_{j \neq i } \int_{\omega_j } \frac{g(\omega_j)}{m } { \rm d}\omega_j \\ & = & \int [ f(\omega)]^{k } g(\omega ) { \rm d}\omega,\end{aligned}\ ] ] and @xmath136^{k } g(\omega ) { \rm d}\omega}{\int [ f(\omega)]^{k+1 } { \rm d}\omega}. \label{eq : ensavck}\ ] ] hence , if the distributions @xmath124 and @xmath94 coincide , i.e. , in our case , if gadget actually poisson samples the vlasolve phase - space density , one obtains @xmath137 after ensemble averaging . when increasing @xmath117 , more weight is given to regions in phase - space corresponding to larger values of @xmath94 . for a point process totally anticorrelated with @xmath94 , @xmath138 cancels , while its largest possible value is given by @xmath139 , when all the particles stay in the region where @xmath94 is maximal . an important issue is to compute properly the center of the system position in the gadget simulations . in order to do this , we find the coordinate origin _ maximizing _ @xmath140 , even though the result of such a procedure can potentially lead to @xmath141 , to optimize the match between concentrations of particles and local extrema of @xmath94 . the variance of @xmath138 can also be calculated in an analogous way to @xmath142 : @xmath143,\end{aligned}\ ] ] which reduces to @xmath144 when @xmath94 and @xmath124 coincide . in practice , we shall use the following estimator for this statistical error : @xmath145 , \label{eq : varck}\ ] ] where @xmath146 and @xmath134 are directly estimated from the @xmath2-body simulation . the log - likelihood that the gadget particle distribution locally poisson samples the vlasolve phase - space density @xmath94 can be written , following the reasoning that leads to equation ( [ eq : lll ] ) , @xmath147 . \label{eq : simplogl}\ ] ] however , the region @xmath148 where @xmath149 being of finite extent , one expects @xmath150 as soon as a particle escapes @xmath148 , which is very likely , due for instance to @xmath2-body relaxation . furthermore , the vlasov solver does not guaranty the positivity of @xmath94 . to take into account in a fair way both the defects of the @xmath2-body and the vlasov simulations , it is better to restrict to a region @xmath151 where @xmath94 is strictly positive : @xmath152 the log - likelihood of having @xmath153 particles in the region @xmath151 and the rest outside it ( leaving the freedom of the remaining particles to span all the space outside @xmath151 ) is given by a binomial law : @xmath154,\ ] ] where @xmath155 is the fractional mass inside @xmath151 in the vlasolve simulation . hence , equation ( [ eq : simplogl ] ) simply becomes @xmath156 + \ln { \cal l}_{\rm b}(q_{\rm th},\nu ) , \label{eq : loglcor}\ ] ] where @xmath157 is the number of particles of the @xmath158 simulation inside @xmath151 and @xmath159 . note that the distribution of particles which maximizes the first term in equation ( [ eq : loglcor ] ) corresponds again to the case where all the particles of @xmath151 stay in the region where @xmath94 is maximal , similarly to the case when the correlator @xmath138 is equal to its maximum possible value . clearly , this situation is not typical , but it is in fact the most likely to consider when it can take place : this is why we maximize @xmath140 to estimate the center of the @xmath2-body system , even though it might turn to be larger than unity . the expectation value of @xmath160 under the law @xmath94 can be obtained by ensemble averaging : @xmath161 { \rm d}\omega , \label{eq : likf}\\ s_{\rm b}(f_{\rm th } ) & \equiv & -\frac{1}{\nu n } \sum_{q=0}^{n } { \cal l}_{\rm b}(q,\nu ) \ln { \cal l}_{\rm b}(q,\nu).\end{aligned}\ ] ] in the limit @xmath162 , the quantity @xmath163 reduces to the gibbs entropy of the system , which explains the choice of notations . moreover , if @xmath90 and if the fractional mass @xmath155 inside the domain of interest @xmath164 is of order of unity , which is the case for our analyses , the term @xmath165 is in practice negligible compared to @xmath163 , so @xmath166 depends only weakly on the total number of particles , as expected . the variance of @xmath160 can be calculated likewise @xmath167 \\ & \simeq & \frac{1}{\nu n } \left\ { \int_{d_{\rm th}}\frac{f(\omega)}{m_{\rm th } } \ln^2\left [ \frac{f(\omega)}{m_{\rm th } } \right ] { \rm d}\omega - \nu [ s(f_{\rm th})]^2 \right\},\nonumber \\ & & \label{eq : errstat}\end{aligned}\ ] ] where we have neglected , following the arguments developed earlier , the contributions of @xmath168 to the error . to understand better the interest of using the statistics given by equation ( [ eq : loglcor ] ) , one can introduce the difference between the measured value of the log - likelihood and its expectation under the law @xmath94 : @xmath169,\ ] ] where @xmath170 is given by expression ( [ eq : loglcor ] ) calculated for @xmath171 extracted from a gadget simulation . the quantity @xmath172 estimates the magnitude of the difference between the underlying smooth phase - space density @xmath124 sampled by gadget and the vlasolve phase space density , @xmath94 : its ensemble average other many gadget realizations indeed reads , when neglecting the binomial term in equation ( [ eq : loglcor ] ) , @xmath173 \ln \left [ \frac{f(\omega)}{m_{\rm th } } \right ] { \rm d } \omega.\ ] ] under the assumption that the @xmath2-body simulation poisson samples the distribution @xmath94 , the magnitude of @xmath172 should be of the same order of @xmath174 . top panels of fig . [ fig : entro ] show the quantity @xmath163 as a function of time for the various vlasolve simulations we performed and two values of @xmath96 chosen such that approximately 90 percent and 60 percent of the total mass is initially inside the excursion @xmath164 , respectively . the quantity @xmath163 is a casimir invariant that is an integral over a function of @xmath94 and should thus be conserved during runtime if the code was perfect . this not the case because of diffusion and aliasing effects in @xmath51 space : deviation from conservation of @xmath175 happens shortly after collapse time . then there is a strong mixing phase during which @xmath175 increases , then possibly decreases , according to the value of @xmath96 , and finally reaches an approximate plateau . deviation from conservation of @xmath175 naturally happens sooner when resolution in @xmath51 space is smaller . resolution in @xmath21 space does not have much influence on @xmath176 because angular momentum is an invariant of the dynamics . however , as clearly shown in [ sec : visu ] and in appendix [ app : resvlaeff ] for @xmath3 , we already know that sparse sampling in @xmath21 space is not recommended since it can introduce some instabilities in the dynamics , even though this effect does not affect much our likelihood measurements . middle panels of fig . [ fig : entro ] show the quantity @xmath177 measured in gadget from the particles belonging to the excursion @xmath164 as a function of time , where @xmath160 is given by equation ( [ eq : loglcor ] ) . for a given value of the threshold @xmath96 , if the gadget simulations would actually behave like poisson realizations of the vlasolve ones , all the colored curves should be close to the solid line , which corresponds to @xmath175 . this is clearly not the case for small @xmath96 ( upper group of curves ) , except a early times . increasing the number of particles in the @xmath2-body simulation improves the agreement with the vlasov code for @xmath66 but does not seem to have a convincing impact in the @xmath3 case : for @xmath178 , all the @xmath2-body simulations converge to the same plateau somewhat below the vlasov code result . on the contrary , for @xmath179 and @xmath3 , the agreement between gadget and vlasolve is striking at all times , except may be for the @xmath70 simulation during the strong mixing phase . note also , that at late times , all the @xmath2-body simulations converge which each other , independently of @xmath96 and @xmath180 , except again for the @xmath70 simulation with @xmath3 , but we know that this latter presents significant deviations from spherical symmetry and should be probably discarded for the analyses performed here . to complete the analyses and understand better the results obtained for the log - likelihood , the fractional mass inside the excursions @xmath93 is shown in bottom panels of fig . [ fig : entro ] . again , this quantity is a casimir , so it should not change with time in the idealistic case . in practice , while it is difficult to predict the effects of aliasing on the @xmath181 mass inside @xmath164 , diffusion effects are more likely to decrease it , especially by dilution of filamentary structures that build up during the course of dynamics . in the @xmath66 case , most of the disagreement between gadget likelihood and its expectation given by vlasolve can be understood in terms of fractional mass : effects related to the discrete nature of the @xmath2-body simulations seem to spread particles away from @xmath164 . however this process is subtle and seems to remain local as suggested by visual inspection of figs . [ fig:0v5_12 ] and [ fig:0v5_allj ] . we also checked that it does not affect dramatically the projected density , @xmath86 . in the @xmath3 case , the interpretation of the results is slightly more complicated . for @xmath179 , the gadget fractional mass inside the excursion @xmath164 behaves similarly as in the @xmath66 case as a function of particle number . on the other hand , when examining the quantity @xmath182 , the @xmath2-body measurements converge with each other and with vlasolve much better , especially after relaxation . this means that particles left in @xmath164 are redistributed in a non trivial way , such that the effects of the excursion mass loss are compensated . for @xmath183 , even the @xmath4 gadget sample disagrees with the vlasolve simulation . clearly , the vlasov simulation becomes quickly defective in regions where @xmath94 is small . on the other hand , convergence of the gadget simulations at late times might be misleading . indeed , we noticed from visual inspection of figs . [ fig:0v1_12 ] and [ fig:0v1_all ] that some instabilities appeared in all of them as soon as @xmath184 , although later when @xmath2 is larger . interestingly , the measurements in the @xmath105 and @xmath4 simulations are nearly indistinguishable from each other , which is a sign that we are nevertheless close to numerical convergence . entropic measurements of fig . [ fig : entro ] are confirmed , at least partly , by fig . [ fig : correlators ] . in particular , a depression of which the depth depends on the number of particles in the @xmath2-body simulation appears on all the curves . when increasing @xmath2 , the amplitude of the depression decreases and the occurrence of its maximum amplitude is delayed , independently of the actual dynamical state of the system . again , it can certainly be attributed to collective effects due to poisson noise . overall agreement between @xmath2-body and vlasov codes improves when increasing the number of particles in the @xmath2-body simulation . for @xmath66 , this is rather independent of @xmath117 in equation ( [ eq : cor1 ] ) , i.e. of the fact of putting more or less weight to overdense regions in phase - space . in the @xmath3 case , putting aside the depression of which the depth depends on the number of particles , the correlator @xmath140 starts to decrease with time at @xmath185 . this can be mainly attributed to defects in the vlasov simulation in underdense regions as discussed earlier . for @xmath186 , which gives more weight to higher values of the phase - space density , the correlator indeed stays steady as a function of time ( again putting aside the @xmath2-dependent depression ) . however , one notices for @xmath187 a net increase with time of the correlator for the simulation with @xmath70 particles , but let us remind that this simulation presents significant deviations from spherical symmetry . in this paper we have compared the phase - space distribution function traced by the particle distribution in gadget simulations to the results obtained with our new vlasov code vlasolve for spherical systems , an improved version of the splitting algorithm of @xcite . for the specific comparison , we have chosen ( apodized ) hnon spheres , which are known to be insensitive to radial orbit instability and in particular to preserve the spherical nature of the system . the latter property is confirmed from simulations run with three - dimensional @xmath2-body codes . we considered two values of the initial virial ratio of the spheres , @xmath66 and @xmath3 , corresponding to `` warm '' and `` cold '' configurations , respectively . we have plotted detailed structures of the phase - space distribution functions varying the spatial / mass resolution of the numerical code in a systematic fashion . we have conducted further a quantitative analysis by introducing two new statistical tools . the first one is of entropic nature and corresponds to the log - likelihood quantifying to which extent the @xmath2-body results represent a local poisson sampling of the vlasov phase - space density . the second tool is a correlator of order @xmath117 , proportional to the integral over phase - space of the product between the vlasov phase - space density raised to the power @xmath117 and the particle distribution function . the overall conclusion is that both the vlasov and @xmath2-body methods agree remarkably well with each other , both from the visual and statistical points of view , if sufficient resolution is employed . given the completely different numerical approaches to collisionless dynamics , this is not trivial at all , and the degree of agreement that we have shown for the first time is perhaps even better than what had been expected before . this is reassuring for numerous previous results that have been almost exclusively obtained from the @xmath2-body method . nevertheless there are still unsolved subtle issues in details : * when performing a visual inspection of the phase - space distribution function in the cold case , @xmath3 , although still good at the coarse level , we find that the level of agreement between the @xmath2-body and the vlasov codes worsens at small scales after a few dynamical times . this is mainly due to collective effects induced by the shot noise of the particles in the @xmath2-body simulations ( and not to close particle encounters ) . even with @xmath4 particles , we are not able to prove numerical convergence of the @xmath2-body results . the comparison at this level , however , is made difficult by the fact that the vlasov code is significantly diffusive , which might prevent the development of a variety of physical unstable modes . * while the statistical tools do not provide as rich and intuitive information as visual inspection , they identify some subtle effects . in particular , when taking into account general trends due to diffusion in the vlasov code , significant for @xmath3 , we notice that the match between gadget and vlasolve worsens with time , then improves . the degree of the mismatch increases , and it shows up earlier , when reducing the number of particles in the @xmath2-body simulation . again , this may be ascribed to collective effects due to the shot noise of the particles . nevertheless , the very good match between the gadget simulations with @xmath105 and @xmath4 particles may suggest that convergence is nearly reached in terms of number of particles and information theory , even if it is not fully proved . it is worth mentioning again that the collective effect mentioned above is not related to @xmath2-body relaxation , but rather results from random poisson fluctuations . this can be formulated as follows ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * for similar arguments ) : a given particle at some distance @xmath19 from the center of the system feels a force proportional to the number @xmath188 of particles inside the sphere of radius @xmath19 . poisson fluctuations imply thus that there is a relative error of order of @xmath189 on this force . importantly , the inner number of particles @xmath188 changes with time with random fluctuations around the mean behavior : these fluctuations can be considered as a correlated random walk . indeed , because of the finite velocity dispersion , particles cross both inwards and outwards the frontier of the sphere of radius @xmath19 . a larger velocity dispersion weakens the amount of correlation , thus makes the errors on the force more random , which should have a fuzzy effect on the phase - space density , similarly as collisional relaxation : this is what we can expect for @xmath66 and as observed on fig . [ fig:0v5_allj ] . on the contrary , a smaller velocity dispersion makes the error on the force more _ systematic _ which should induce coherent distortions of the phase - space density : this is what we can expect for @xmath3 and confirmed by visual inspection of fig . [ fig:0v1_all ] . this effect has non - trivial consequences on the energy spectrum of the particles , particularly in cold configurations @xcite . it certainly explains as well the deviations between vlasolve and gadget observed when measuring the statistical estimators defined in this paper . according to @xcite , this _ collective _ effect is dominant over @xmath2-body relaxation , and , as confirmed by our detailed numerical tests in appendix [ app : instab ] , is not significantly influenced by softening . note as well that shot noise creates anisotropies in the system , i.e. deviations from spherical symmetry that may be eventually amplified . @xcite argue that this effect is subdominant compared to the radial component of the noise - induced perturbation when considering the collapse of an homogeneous sphere . although their calculation is performed only prior to collapse and in the cold case , we believe that the conclusion still remains valid for the kind of initial conditions studied in this paper , as suggested by our numerical experiments that seem to preserve well spherical symmetry . clearly , the collective effect due to particle shot noise is a real problem for simulations of close to cold spherical systems when it comes to examine fine structures of the phase - space density . we were not able to prove convergence of the phase - space density in the @xmath3 case even for an @xmath4 particle simulation . notably , this may have non - trivial consequences on the fine structure of simulated dark matter halos , where numerical convergence in terms of number of particles might not have been reached yet despite the numerous intensive studies . indeed , convergence toward the continuous limit might be much slower than expected , hence giving the false impression that it is achieved . we thank christophe alard , analle hall , jrme perez and simon prunet for useful discussions . this work has been funded in part by anr grant anr-13-monu-0003 and was granted access to the hpc resources of the institute for scientific computing and simulation financed by region ile de france and the project equip@meso ( anr-10-eqpx-29 - 01 ) overseen by the french national research agency ( anr ) as part of the `` investissements davenir '' program . y.s . gratefully acknowledges the support from grant - in aid for scientific research by jsps ( japan society for promotion of science ) no . 24340035 . aarseth s. j. , lin d. n. c. , papaloizou j. c. b. , 1988 , apj , 324 , 288 alard c. , colombi s. , 2005 , mnras , 359 , 123 barnes e. i. , lanzel p. a. , williams l. l. r. , 2009 , apj , 704 , 372 bertschinger e. , 1998 , ara&a , 36 , 599 besse n. , sonnendrcker e. , 2003 , jcoph , 191 , 341 besse n. , latu g. , ghizzo a. , sonnendrcker e. , bertrand p. , 2008 , jcoph , 227 , 7889 binney j. , 2004 , mnras , 350 , 939 boily c. m. , athanassoula e. , kroupa p. , 2002 , mnras , 332 , 971 campos pinto m. , 2011 , arxiv , arxiv:1112.1859 cheng c. z. , knorr g. , 1976 , jcoph , 22 , 330 colombi s. , 2001 , newar , 45 , 373 colombi s. , touma j. , 2008 , cnsns , 13 , 46 colombi s. , touma j. , 2014 , mnras , 441 , 2414 crouseilles n. , latu g. , sonnendrcker , e. , 2009 , jcoph , 228 , 1429 crouseilles n. , mehrenberger m. , sonnendrcker e. , 2010 , jcoph , 229 , 1927 dehnen w. , read j. i. , 2011 , epjp , 126 , 55 dolag k. , borgani s. , schindler s. , diaferio a. , bykov a. m. , 2008 , ssrv , 134 , 229 filbet f. , sonnendrcker e. , bertrand p. , 2001 , jcoph , 172 , 166 fujiwara t. , 1981 , pasj , 33 , 531 fujiwara t. , 1983 , pasj , 35 , 547 gott , j.r . iii , 1973 , apj , 186 , 481 gl y. , christlieb a. j. , hitchon w. n. g. , 2014 , jcoph , 270 , 711 hahn o. , abel t. , kaehler r. , 2013 , mnras , 434 , 1171 hall a. , 2015 , private communication hnon m. , 1964 , anap , 27 , 83 henriksen r. n. , widrow l. m. , 1997 , phrvl , 78 , 3426 hozumi s. , burkert a. , fujiwara t. , 2000 , mnras , 311 , 377 hozumi s. , fujiwara t. , kan - ya y. , 1996 , pasj , 48 , 503 huss a. , jain b. , steinmetz m. , 1999 , apj , 517 , 64 jing y.p . , suto , y. , 2000 , apjl , 529 , l69 jing y.p . , suto , y. , 2002 , apj , 574 , 538 joyce m. , marcos b. , sylos labini f. , 2009 , mnras , 397 , 775 melott a. l. , 2007 , arxiv , arxiv:0709.0745 melott a. l. , shandarin s. f. , splinter r. j. , suto y. , 1997 , apj , 479 , l79 mineau p. , feix m. r. , rouet j. l. , 1990 , a&a , 228 , 344 moore b. , governato f. , quinn t. , stadel j. , lake g. , 1998 , apj , 499 , l5 nishida m. t. , yoshizawa m. , watanabe y. , inagaki s. , kato s. , 1981 , pasj , 33 , 567 power c. , navarro j. f. , jenkins a. , frenk c. s. , white s. d. m. , springel v. , stadel j. , quinn t. , 2003 , mnras , 338 , 14 rossmanith j. a. , seal d. c. , 2011 , jcoph , 230 , 6203 roy f. , perez j. , 2004 , mnras , 348 , 62 shandarin s. , habib s. , heitmann k. , 2012 , phrvd , 85 , 083005 shoucri m. m. , gagne r. r. j. , 1978 , jcoph , 27 , 315 sonnendrcker e. , roche j. , bertrand p. , ghizzo a. , 1999 , jcoph , 149 , 201 splinter r. j. , melott a. l. , shandarin s. f. , suto y. , 1998 , apj , 497 , 38 springel v. , 2005 , mnras , 364 , 1105 springel v. , et al . , 2008 , mnras , 391 , 1685 springel v. , yoshida n. , white s. d. m. , 2001 , newa , 6 , 79 stadel j. , potter d. , moore b. , diemand j. , madau p. , zemp m. , kuhlen m. , quilis v. , 2009 , mnras , 398 , l21 sylos labini f. , 2012 , mnras , 423 , 1610 umeda t. , 2008 , ep&s , 60 , 773 van albada t. s. , 1982 , mnras , 201 , 939 watanabe y. , inagaki s. , nishida m. t. , tanaka y. d. , kato s. , 1981 , pasj , 33 , 541 yoshikawa k. , yoshida n. , umemura m. , 2013 , apj , 762 , 116 in this appendix , we explain how reflecting boundaries conditions with time delay are implemented in vlasolve . if the mass inside the sphere of radius @xmath58 is neglected , the trajectories followed by each test particle associated to a grid site that penetrates the sphere are fixed and do not depend on time . this property , combined with the fact that we use a constant time step , allows us to pre - compute these trajectories once and for all . the delayed central sphere method is then implemented by associating a linked list to each grid site whose associate test particle radial position @xmath19 half a time step backward in time is such that @xmath61 . each linked list contains as many elements as the number of time steps needed for the particle to travel a distance of @xmath60 and the @xmath190 element in the list stores the coordinates of the test particle @xmath191 time steps backward in time . before starting the simulation , we initialize each element coordinate and the corresponding value of the initial distribution function . for each time step , the value of each element is then simply updated by assigning to it the value of its successor while the last element value , whose coordinates fall inside the computing domain , @xmath192 , is interpolated . a comparison of the results obtained with the reflective central sphere to our improved delayed central sphere is shown on figure [ fig : delayed ] . the improvements are unquestionable . we implemented a hybrid shared and distributed memory version of vlasolve via the openmp and mpi libraries , respectively . shared memory parallelism is relatively straightforward to achieve in the spherically symmetric case , by taking advantage of the fact that the angular momentum @xmath21 is a conserved quantity . spline interpolations , which represent the most expensive part of the code , can thus be computed independently for each slice of constant @xmath21 . we therefore easily reach an almost perfect parallelization up to a number of tasks equal to the grid resolution @xmath193 of angular momentum space , which is typically larger than the number of available cores on a shared memory system . distributed memory parallelization via mpi is not as simple . indeed , spline interpolations are intrinsically non - local , which makes the parallelization along dimensions other than @xmath21 non trivial . sticking with the trivial parallelization described above unfortunately limits the maximum total number of processes running in parallel to @xmath193 , which is suboptimal . we overcome this limitation by performing mpi domain decomposition in @xmath51 space , following the approach of @xcite , who propose to localize the cubic spline interpolation to each domain by using hermite boundary conditions between the domains with an ad hoc reconstruction of the derivatives . figures [ fig : testvla ] and [ fig : testvla2 ] show , respectively for @xmath98 and integrated over angular momentum , the phase - space distribution function measured in vlasolve simulations with different resolutions . these simulations have been performed for a hnon sphere with initial virial ratio @xmath3 . beside the very good global agreement between the various runs , these figures bring out three effects , which increase when the resolution of the phase - space grid is reduced : * diffusion smearing out fine details that build up in phase - space during the course of dynamics , for instance clearly visible when one compares top to bottom middle panels of fig . [ fig : testvla ] . one concern with diffusion is that it might prevent the appearance of unstable modes . however , we did not perform any simulation in this work that would prove this . * aliasing due to artificial oscillations in the spline interpolation : for the problem studied here , aliasing becomes particularly visible after relaxation in the region above the large @xmath19 tail , but this does not have significant impact on the dynamics . * aliasing due to undersampling angular momentum space : it is visible at all times when one examines the phase - space distribution function integrated over angular momentum ( top panels of fig . [ fig : testvla2 ] ) and can have dramatic consequences on the dynamics . the two top lines of panels of fig . [ fig : testvla ] and [ fig : testvla2 ] , corresponding to a sparse sampling in @xmath21 space with only 32 slices , indeed show the appearance of an instability , which presents , on the third column of these figures , the same pattern whether @xmath194 or @xmath195 . this instability is not present in the simulations with higher resolution in @xmath21 , as shown by the two bottom lines of panels . note that the presence of this instability depends on initial conditions : for @xmath66 , we did not notice it for the time coverage considered , @xmath196 ( upper line of panels of figs . [ fig:0v5_12 ] and [ fig:0v5_allj ] ) . [ sec : resoeffects ] in [ sec : visu ] we noticed the presence of an instability in the @xmath3 @xmath2-body simulations . one aim of this appendix is to confirm that this instability is related to the number of particles used in the simulations and not to any other control parameter of the gadget code . in the same time , it is also an opportunity to check that our fiducial choice of the gadget control parameters , given in [ sec : gadpres ] , is correct . figure [ fig : testgad ] illustrates the main results of the tests we performed for simulations with @xmath106 particles . these tests consisted in changing the softening length of the force , the maximum time step value and the tolerance parameter @xmath197 controlling the errors on the force . improving the accuracy of the force calculation or dividing the maximum time step @xmath198 by a factor 50 , which corresponds to imposing @xmath199 , does not change the results . this is confirmed as well by the measurements of the correlators @xmath138 introduced in [ sec : statana ] , that we do not show here for simplicity . only the value of the softening parameter of the force @xmath12 has an impact on the dynamics for the tests we did . reducing @xmath12 by a factor 5 seems to slightly blur the phase - space density , although this effect is difficult to decipher , while increasing @xmath12 by a factor 10 sharpens the fine structures of the phase - space density . since @xmath12 controls the intensity of close encounters between particles , this is not surprising . note that increasing @xmath12 by a factor 10 is probably an exaggeration , because it _ worsens dramatically _ the match during the mixing phase between the @xmath2-body simulation and the vlasov code when examining the correlators @xmath138 , a sign that @xmath12 is probably getting too close to a physical characteristic scale of the system . by a factor ten gives @xmath200 , to be compared for example to the size of the core of the system after relaxation , @xmath201 . ] we indeed noticed that increasing @xmath12 only by a factor 5 does not have much impact , on the other hand , on @xmath138 . however , all these effects do not affect the amplitude of the large scale irregularities on the pattern of @xmath97 , which are present whatever value of @xmath12 . this is also a strong indication that close particle encounters are not at the origin of these irregularities . we can therefore only conclude that these irregularities and the associated nonlinear instability are the result of non trivial collective effects related to particle shot noise . this argument is also supported by the fact that in addition , the moment of their appearance is particle number dependent , as discussed in [ sec : visu ] . [ lastpage ]

during the past two years , the galactic black hole microquasar grs 1915 + 105 has exhibited a bewildering diversity of large amplitude , chaotic variability in x - rays . although it is generally accepted that the variability in this source results from an accretion disk instability , the exact nature of the instability remains unknown . here we investigate different accretion disk models and viscosity prescriptions in order to provide a basic explanation for the exotic temporal behavior in grs 1915 + 105 . we discuss a range of possible accretion flow geometries . based on the fact that the overall cycle times are very much longer than the rise / fall time scales in grs 1915 , we rule out the geometry of advection dominated accretion flow ( adaf ) or a hot quasi - spherical region plus a cold outer disk for this source . a cold disk extending down to the last inner stable orbit plus a hot corona above it , on the other hand , is allowed . we thus concentrate on geometrically thin ( though not necessarily standard ) shakura - sunyaev type disks ( shakura & sunyaev 1973 ; hereafter ss73 ) . we argue that x - ray observations clearly require a quasi - stable accretion disk solution at high accretion rates where radiation pressure begins to dominate , which excludes the standard-viscosity prescription . to remedy this deficiency , we have therefore devised a modified viscosity law that has a quasi - stable upper branch , and we have developed a code to solve the time - dependent equations to study such an accretion disk . via numerical simulations , we show that the model does account for several gross observational features of grs 1915 + 105 , including its overall cyclic behavior on time scales of 100 - 1000 s. on the other hand , the rise / fall time scales are not short enough , no rapid oscillations on time scales 10 s emerge naturally from the model , and the computed cycle - time dependence on the average luminosity is stronger than is found in grs 1915 + 105 . we then consider , and numerically test , several effects as a possible explanation for the residual disagreement between the model and the observations . a hot corona with the energy input rate being a function of the local cold disk state and a radius - dependent-parameter do _ not _ appear to be promising in this regard . however , a more elaborate model that includes the cold disk , a corona , and plasma ejections from the inner disk region allows us to reproduce several additional observed features of grs 1915 + 105 . we conclude that the most likely structure of the accretion flow in this source is that of a cold disk with a modified viscosity prescription , plus a corona that accounts for much of the x - ray emission , and unsteady plasma ejections that occur when the luminosity of the source is high . the disk is geometrically thin due to the fact that most of the accretion power is drained by the corona and the jet .

astro-ph9905371

compact x - ray sources exhibit a wide range of temporal variabilities ( from milliseconds to years ) . perhaps none of these is as exotic and diverse as the x - ray temporal variability observed from the black hole microquasar grs 1915 + 105 ( castro - tirado , brandt , & lund 1992 ; greiner , morgan , & remillard 1996 ; morgan , remillard & greiner 1997 ; muno , morgan & remillard 1999 ) . this object is one of two known galactic x - ray sources that exhibit superluminal radio jets ( mirabel & rodrigues 1994 ) . the combination of relativistic constraints and radio measurements at hi indicate that the source lies behind the sagittarius arm at a distance of 12.5 @xmath3 kpc ( mirabel & rodrigues 1994 ) . interstellar extinction limits optical / ir studies to weak detections at wavelengths less than 1 micron ( mirabel et al . the source is suspected to be a black hole binary because of its spectral and temporal similarities with the other galactic x - ray source with superluminal radio jets , gro j1655 - 40 ( zhang et al 1994 ) , which has a binary mass function indicative of a black hole system ( bailyn et al . 1995 ) . estimates for the mass of the compact object in grs 1915 + 105 range from 7 to 33 @xmath4 . even with the uncertainty in distance , its peak x - ray luminosity is unusually high , i.e. , @xmath5 ergs / sec , which is around the eddington luminosity for a @xmath6 @xmath4 object . in spite of several attempts it has proven especially illusive to interpret the x - ray light curves of grs1915 . it is not yet clear that even the basic time scales exhibited by the variability have been successfully explained . belloni et al . ( 1997a , b ) accounted for the observations with an empirical model in which the inner disk region `` disappears '' in the low count rate state , and is then replenished on a viscous time scale . the parameters of their model are : the inner disk radius , @xmath7 ; the corresponding effective temperature of the disk @xmath8 , and an ad - hoc non - thermal power law ( which is possibly produced in the disk corona ) . although no detailed physical model for the instability was given , very interesting patterns of behavior for @xmath7 and @xmath8 , as well as several other observables , were deduced from the data , and the shakura - sunyaev viscosity parameter was found to be unexpectedly low ( which may mean that the standard viscosity prescription is invalid for this source ) . the rather small values of @xmath7 found by these authors can be used to discriminate between different models of the accretion flow in grs 1915 ( see appendix and [ sect : previous ] ) . abramowicz , chen & taam ( 1995 ) suggested a model for the low frequency quasi - periodic oscillations ( qpo ) observed in selected x - ray binaries , in which a corona above the _ standard accretion _ disk leads to a mild oscillatory behavior . with some modifications , this model could reasonably be expected to account for at least some of the temporal variability in grs 1915 + 105 as well ( taam , chen & swank 1997 ) . however , it appears to us that the analysis of abramowicz et al . ( 1995 ) and taam et al . ( 1997 ) contains an error in the heating / cooling equation for the disk which , when corrected , constrains their model to have the same stability characteristics as a standard shakura - sunyaev disk , and is therefore unlikely to explain the grs 1915 + 105 observations ( see appendix a ) . we show more generally in appendix [ sect : geometry ] that neither a hot central region , nor an advection - dominated flow , nor a `` slim '' accretion disk are compatible with the observations of grs 1915 + 105 . ( `` slim '' accretion disk theory was developed in the most detail by abramowicz et al . 1988 ; it is similar to a standard thin shakura - sunyaev disk , except for the energy equation , which incorporates the radial advection of energy into the black hole . ) in this paper , we attempt to undertake a more systematic study of the variability patterns in grs 1915 + 105 within the context of the `` cold disk@xmath9hot corona '' picture . in [ sect : framework ] we present a general discussion that will guide us in our selection of a novel ( though somewhat ad - hoc ) prescription for the viscosity in cases where the radiation pressure is substantial ( [ sect : prescription ] ) . the details of our numerical algorithm to solve the time - dependent disk equations with the use of this new viscosity prescription are given in [ sect : code ] . in [ sect : results ] , we present the results of our time - dependent disk calculations . the calculated light curves are found to agree qualitatively with many observational features of grs 1915 + 105 . in particular , the characteristic cycle times and duty cycles are in reasonable agreement with the observations . moreover , the trend in the cycle time with the average accretion rate , @xmath10 has the correct sense . however , there are important disagreements as well . we therefore introduce a more elaborate model in [ sect : jet ] , where , in accordance with the observations ( e.g. , mirabel & rodrigues 1994 ) , the inner disk is allowed to expel some of its energy in the form of a non - steady jet . we assume that the ejected energy is not observed in x - rays , but rather that it ultimately produces radio emission . we show that this more elaborate model agrees with the grs 1915 + 105 observations much better , perhaps indicating that we are finally developing a zeroth order understanding of the geometry and the most important processes in this enigmatic source . in [ sect : previous ] we discuss our results in the light of the earlier work on grs 1915 , and in [ sect : discussion ] we summarize our conclusions . figure 1 shows four examples of typical x - ray light curves from grs 1915 obtained with the rxte satellite ( see , e.g. , morgan , remillard , & greiner 1997 ) . it appears that the source undergoes a limit - cycle type of instability ; the cycle times in panels ( a ) ( d ) are @xmath1 2400 , 60 , 1200 , and 800 sec , respectively . within the cycles shown in panels ( c ) and ( d ) there appear yet other quasi - regular oscillations of still shorter time scale . the shortest time scale over which a substantial change in the x - ray flux occurs is @xmath1 5 sec . we will associate the longer cycles with the viscous time scales of the inner accretion disk , whereas the more rapid behavior will be related to the thermal time scale . the x - ray spectrum of grs 1915 + 105 varies systematically with x - ray intensity ; usually the spectral hardness is strongly correlated ( or anti - correlated ) with intensity ( e.g. , belloni et al . 1997a , b ; taam et al . 1997 ; muno et al . more specifically , the spectrum is typically found to be composed of a multi - temperature disk component , hereafter the thermal component " , and a power - law component with a varying spectral index . finally , the source exhibits a wide range of qpos with central frequencies in the range of 0.01 - 10 hz . the amplitude and frequency of the qpos appear to be strongly correlated with the spectral state of grs 1915 + 105 ( chen et al . 1997 ; muno et al . 1999 ) . the idea of using a modified viscosity law to test time - dependent phenomena in accretion disks around black holes is not new ( e.g. , taam & lin 1984 ; chen & taam 1994 ; and further references below in this section ) . but let us examine in global terms how the various disk configurations relate to the behavior of grs 1915 + 105 in particular . in geometrically thin disks the advection of energy is not important . the viscous time scale is much longer than both the thermal and hydrostatic time scales ( see frank et al . 1992 ) , unless @xmath11 which is unlikely for this source ( belloni et al . 1997b ) . in this context , then , the equations for vertical hydrostatic equilibrium and energy balance between viscous heating and radiative losses should be valid . it is customary and instructive to plot the disk effective temperature @xmath12 versus the column density @xmath13 to investigate the stability properties of disks . in the one zone limit , the equation for hydrostatic equilibrium in the vertical direction is @xmath14 where @xmath15 is the mass of the black hole , @xmath16 is the scale height and @xmath17 is the radial distance from the central object . the gas pressure is given by @xmath18 ( for hydrogen rich material ) , where @xmath19 is the mid - plane temperature , @xmath20 is the proton mass , and the radiation pressure is @xmath21 . the energy balance equation is given by @xmath22 where @xmath23 is the vertical radiation flux , @xmath24 is the optical depth of the disk , @xmath25 , and @xmath26 is the radiative opacity ( assumed to be dominated by electron scattering opacity ) , @xmath27 is the stress tensor ( see , e.g. , shapiro & teukolsky , 14.5 ) , and @xmath28 is the keplerian angular frequency . as we said earlier , in the standard shakura - sunyaev formulation , the stress tensor is proportional to the total pressure , @xmath29 we can now combine equations ( [ eq1 ] ) ( [ ss ] ) to eliminate the variable h and thereby derive an expression to generate the @xmath12@xmath13 curve for any given viscosity . in figure ( [ fig : scurve ] ) we show several illustrative @xmath12@xmath13 curves to help motivate our selection of @xmath0 . the dashed curve represents the track that the standard shakura - sunyaev theory with @xmath30 const produces . such a model is thermally and viscously unstable when the slope of the @xmath12@xmath13 curve becomes negative ( e.g. , frank et al . 1992 ) , which might be a desirable result in the view of the instabilities observed in grs 1915 + 105 . however , the observed light curves require a quasi - stable accretion mode for the high - luminosity state as well , which is _ not predicted _ by the standard theory ( see also the end of app . b ) . this suggestion ( already discussed by belloni et al . 1997a , b and taam et al . 1997 ) is motivated by the fact that in some cases the upper luminosity state of the source lasts even longer than the low state , which is presumably the stable gas - pressure - dominated branch . a @xmath12-@xmath13 relation that has a stable high - luminosity state is shown in figure ( [ fig : scurve ] , solid curve ) . it has the characteristic `` s - shape '' , familiar from the studies of the thermal ionization instability in accretion disks ( e.g. , meyer & meyer - hofmeister 1981 , bath & pringle 1982 , cannizzo , chen & livio 1995 , and frank et al . 1992 , 5.8 ) or from studies of radiation - pressure driven instabilities by , e.g. , taam & lin ( 1984 ) and chen & taam ( 1994 ) . it is well known that parts of an s - curve that have a negative slope are unstable . in short , when the @xmath12-@xmath13 relation produces an s - curve , the unstable region of the disk oscillates between the two stable branches of the solution , which gives rise to the high and low luminosity states . so far , we have not considered flows where advection is important . as adafs ( advection dominated accretion flows ) will be discussed in [ sect : adaf ] , here we concentrate on the slim accretion disks ( abramowicz et al . these flows have a stable high @xmath10 solution even with the standard viscosity prescription due to the additional cooling ( energy transport ) via advection . however , the s - curves produced by these models have the upper stable branch for @xmath31 only ( see , e.g. , see fig . 2 in chen & taam 1993 ) . in other words , if one were to position the upper stable branch of the slim accretion disks on our figure ( [ fig : scurve ] ) , then it would look similar to the corresponding stable branch of the solid curve in the figure , except that it would be positioned higher by a factor of about @xmath32 ( at a given @xmath13 ) . thus , if one assumes that an unstable part of the disk makes a transition from the lower bend in the s - curve upwards to the stable branch , and that @xmath13 remains approximately constant during this process , then the accretion rate in the inner disk must increase by as much a factor of a few hundred to a thousand to reach the stable advective regime . while this may not be a problem in terms of the luminosity output , since most of the energy is advected inwards and the luminosity of the disk saturates at @xmath33 ( see fig . 1 in szuszkiewicz , malkan & abramowicz 1996 ) , the light curves produced will always be very spiky and the duty cycle of the high state is always much smaller than the values of @xmath34 that are often seen in grs 1915 + 105 . the physical cause of the spiky behavior is the vast difference between the accretion rates in the high and low states . as we will discuss later in [ sect : results ] , an outburst starts because `` too much '' mass ( column density @xmath13 ) has been accumulated above the maximum possible stable value of @xmath35 ( i.e. , @xmath13 at the lower bend of the s - curve in figure [ fig : scurve ] ) . if the high stable state accretion rate is very much larger than the low state accretion rate , it then takes only a very short time in the high state to remove this excess mass from the inner disk . in other words , it would not be possible to maintain mass conservation if the accretion rate persisted at the high state for very long . this is indeed seen from the temperature curves computed for slim disks by szuszkiewicz & miller ( 1998 , see their figures 7 - 9 ) . therefore , we believe that slim disks are not likely to adequately account for the behavior of grs 1915 + 105 , at least not if one seeks a theory capable of explaining all of the variability in this source . the geometry in our unstable disk model is that of a standard shakura - sunyaev configuration ( except for the viscosity law ) overlayed with a hot corona . here we describe the set of equations that we use to follow the temporal evolution of this disk ; these equations are applicable to any local viscosity law . a discussion of the viscosity itself is deferred until [ sect : prescription ] . the standard euler equations for conservation of mass and angular momentum in the disk ( see , e.g. , frank et al . 1992 ) can be combined in the usual way to yield the equation describing the evolution of the surface density : @xmath36 \right\}\ , , \label{eq4}\ ] ] where all of the variables , except for @xmath37 , were defined in the previous section . here @xmath37 is the viscosity that is related to @xmath0 by @xmath38 = @xmath39 . the time - dependent energy equation includes the heating and cooling terms of equation ( [ eq2 ] ) , but must also be able to take into account large radial gradients of temperature , and therefore includes a number of additional terms . the form of the energy equation that we use follows the formalism of abramowicz et al . ( 1995 ) with some modifications , and is given by @xmath40 \nonumber \\ = f^{+ } - f^{- } - { 2\over r}\ , { \partial(r f_r h)\over \partial r } + j , \label{eq5}\end{aligned}\ ] ] where @xmath41 , @xmath42 is the ratio of the radiation to the gas pressure , @xmath43 is the ratio of specific heats ( @xmath44 ) and @xmath45 is given in abramowicz et al . ( the radial velocity @xmath46 is given by the standard expression ( eq . 5.7 of frank et al . 1992 ) : @xmath47 . \label{vr}\ ] ] the terms on the left hand side of equation ( [ eq5 ] ) represent the full time derivative ( e.g. , @xmath48 ) of the gas entropy , while the terms on the right are the viscous heating , the energy flux in the vertical direction , the diffusion of radiation in the radial direction , and the viscous diffusion of thermal energy . following cannizzo ( 1993 ; and references cited therein ) , we take @xmath49 $ ] to be the radial energy flux carried by viscous thermal diffusion , where @xmath50 is the specific heat at constant pressure . @xmath51 is the accretion disk heating rate per unit area , and is given by @xmath52 the radiation flux in the radial direction is @xmath53 the overall cooling rate ( larger than simply that from radiative diffusion ) in the vertical direction is given by @xmath54 where @xmath55 is the fraction of the power that is transferred to the disk surface by mechanisms other than the usual radiation diffusion ( cf . abramowicz et al . 1995 , svensson & zdziarski 1994 ) . equations ( [ heatr ] ) and ( [ fvert ] ) differ from those used by abramowicz et al . ( 1995 ) for reasons that are detailed in appendix b. we note that in the most general sense , the quantity @xmath56 should be thought of as a factor correcting the vertical energy transport in the standard accretion disk theory . in our formulation of the problem , it is computationally irrelevant whether this fraction of energy is deposited in the corona and radiated as a non - thermal power - law component , or it contributes to the blackbody disk flux , since we are concerned only with the bolometric luminosity of the disk for now . the latter case could be realized if this additional energy transport were to deposit its energy just below the photosphere of the disk . of course , if our model produces light curves that are in reasonable agreement with those of grs 1915 + 105 , then a future study , one that would carefully delineate the different spectral components , will be warranted . equations ( [ eq4 ] ) and ( [ eq5 ] ) , together with the equation of hydrostatic equilibrium ( [ eq1 ] ) , yield a closed set of coupled , time - dependent equations for evolving the variables t and @xmath13 . for solving these equations we employ a simple explicit scheme with a variable time step . the time step is chosen to be always a fraction of the smallest thermal time scale in the disk ( i.e. , in its inner part ) . we also use a fixed grid in @xmath17 . the spacing is uniform in @xmath57 , and we take 100 to 300 radial bins . we will often refer to the radial coordinate in a dimensionless form , i.e. , @xmath58 , where @xmath59 is the schwarzschild radius . the outer boundary of the disk is chosen at a large radius , @xmath60 few hundred to a thousand , such that the disk at @xmath61 is always gas - pressure dominated . thus , the outer boundary conditions are given by @xmath62 and @xmath63 for @xmath64 , where @xmath65 and @xmath66 are the temperature and the column density in the shakura - sunyaev formulation at that radius for the given @xmath67 , @xmath10 and @xmath68 , where @xmath69 is the viscosity parameter for a gas - dominated disk ( see below ) . the inner boundary of the disk is fixed at @xmath70 , and the boundary conditions there are @xmath71 and @xmath72 . in this section we present our modified viscosity law that we anticipate will account for the unstable behavior of grs 1915 + 105 . the shakura - sunyaev viscosity prescription postulates that the viscous stress tensor @xmath27 is given by equation ( [ ss ] ) , where the parameter @xmath0 is defined as @xmath73 and where @xmath74 and @xmath75 are the turbulent eddy velocity and scale - length , respectively , @xmath76 is the sound speed , and @xmath77 is the volume averaged magnetic field energy density . obviously , this approach is useful only when @xmath0 remains approximately constant , i.e. , independent of the local thermodynamic variables in the disk , which is probably the case in a number of situations . whenever the physical state of the accreting gas changes considerably , it is hard to see why @xmath0 would not change . one confirmation of this is the well - known thermal ( ionization ) instability of accretion disks in dwarf nova systems ( e.g. , meyer & meyer - hofmeister 1981 ; bath & pringle 1982 ; cannizzo , chen & livio 1995 ; and frank et al . 1992 ; 5.8 ) . it has been shown in this case that the @xmath0-parameter needs to be larger on the hot stable branch than it is on the low branch of the solution . some more recent work suggests that , in addition , @xmath0 should have a power - law dependence on @xmath78 , e.g. , @xmath79 , in order to explain the observed exponential luminosity decline of the dwarf nova outbursts ( cannizzo et al . 1995 ; vishniac & wheeler 1996 ) . further , based on theoretical considerations , different authors have adopted several variants on the prescription for @xmath27 when @xmath80 . for example , lightman & eardley ( 1974 ; hereafter le74 ) , and stella & rosner ( 1984 ; hereafter sr84 ) suggested that @xmath81 where @xmath69 is a constant . their physical reasoning was based on the expectation that chaotic magnetic fields should be limited by the gas pressure only . in addition , sakimoto & coronitti ( 1989 ; hereafter sc89 ) showed that even if one assumes that magnetic fields are effectively generated by radiation - dominated disks , magnetic buoyancy quickly expels these fields from the disks . since the disk viscosity may be produced to a large extent by magnetic fields , this result implies that the @xmath0-parameter must decrease when the radiation pressure becomes dominant . an attractive feature of this viscosity prescription ( eq . [ le ] ) is the fact that it is stable even when @xmath80 , as indicated by the positive slope generated by this model for the track in @xmath12@xmath13 parameter space ( see the dotted curve in fig . [ fig : scurve ] ) . at the same time , however , this viscosity prescription ( to which we will refer as the lightman - eardley prescription ) has no unstable region at all , and thus fails to reproduce the unstable behavior observed in grs 1915 + 105 . of course , the conclusions of le74 , sr84 and sc89 should be regarded as qualitative , at best , since it is not yet feasible to describe the turbulence and magnetic fields immersed in a radiation - dominated fluid in a quantitative and model - independent way . in this paper , we adopt the view that it may still be possible to couple the radiation to the particles through collisions and thereby allow the radiation pressure to contribute to the viscosity when @xmath82 is not too great . for example , sr84 argued that the energy density in the chaotic magnetic fields can not exceed the particle energy density , since it is the latter that generates the magnetic fields in the first place . however , as long as @xmath83 , the magnetic field pressure does not exceed the gas pressure ( because then @xmath84 ) , so that the sr84 arguments are superfluous . to cast our suggestion into a mathematical form , we have adopted a viscosity prescription intermediate between those of shakura - sunyaev and lightman - eardley . our viscosity law is proportional to the total pressure for small to moderate values of @xmath42 , while for large values of @xmath85 , it becomes proportional to the gas pressure . we have devised the following simple formula for the viscosity prescription that smoothly joins these two limits , such that @xmath0 is given by @xmath86 where @xmath87 is an adjustable parameter . the s - curve plotted in figure ( [ fig : scurve ] ) was generated using this viscosity prescription with @xmath88 . there are essentially 5 free parameters of the model that need to be specified and/or systematically varied to ascertain the sensitivity of the results to their value : ( _ i _ ) the black hole mass , @xmath15 , ( _ ii _ ) the average accretion rate , @xmath10 , ( _ iii _ ) the viscosity parameter , @xmath69 , for the gas - dominated region , ( _ iv _ ) @xmath87 , the critical ratio @xmath82 , and ( _ v _ ) @xmath67 , the fraction of the luminosity transported vertically by mechanisms other than radiative diffusion . we adopt a mass of @xmath89 , which is typical of the measured masses of galactic black hole candidates in binary systems ( see , e.g. , barret et al . 1996 ) . for purposes of illustration , we chose a value of @xmath69 equal to @xmath90 , since it leads to viscous time scales comparable to those observed . similarly , we selected a value of @xmath91 , in part because it produces a reasonable s - curve . however , we checked a range of other values as well ( from @xmath92 to @xmath93 ) , and found that @xmath94 reproduces the grs 1915 + 105 observations most closely . we present a computed light curve and variations of the disk parameters through one complete cycle in figures ( [ fig : feqo ] ) and ( [ fig : temp_tau_distr ] ) , respectively , for one set of illustrative parameters : @xmath95 and @xmath96 . in figure ( [ fig : temp_tau_distr ] ) , panels ( a ) , ( b ) , ( c ) and ( d ) show the evolution of the disk effective temperature @xmath97 , the ratio of radiation to gas pressure ( @xmath85 ) , the disk thomson optical depth @xmath24 and the radial flow velocity @xmath98 . note that since we are currently concerned only with the luminosity integrated over all photon energies , we define @xmath97 to include the emissivity from _ both _ the corona and the disk , i.e. , @xmath99 , where @xmath100 is given by equation ( [ fvert ] ) . in each panel , the series of curves are for temporal increments of @xmath101 sec . the curve with the smallest luminosity @xmath102 ( where @xmath103 ) in the upper left panel of figure ( [ fig : temp_tau_distr ] ) corresponds to the state of the disk right after the end of an outburst . an outburst starts in the innermost region of the disk , when a few zones at the smallest radii quickly make a transition to the upper stable branch of the solution ( high state ) . although this is not obvious from fig . ( [ fig : temp_tau_distr ] ) , the viscosity of the high state is larger than it is in the low state , because @xmath104 , and so even though @xmath0 decreases , the increase in @xmath76 and @xmath16 leads to a larger @xmath37 . the larger viscosity allows the gas to dispose of its angular momentum faster , thus allowing a faster inflow of the gas into the black hole . because of angular momentum conservation , the angular momentum of the gas that plummets into the black hole is transferred to larger @xmath17 , where it produces an excess of angular momentum , and thus some matter actually flows to larger @xmath17 , which is seen in panel ( d ) of figure ( [ fig : temp_tau_distr ] ) as positive spikes in the radial velocity distribution . a heating wave is initiated and propagates from the inner disk to larger @xmath17 ( fig . [ fig : temp_tau_distr]a ) . this wave is often referred to as either a ` density wave ' or a ` transition wave ' . as the wave propagates outward , the material on the inside of the wave becomes hot and shifts into the high viscosity state . it then rapidly loses its angular momentum and is transferred into the innermost disk , where the material continuously plummets into the hole . since the innermost stable region is a `` bottle - neck '' for the accreting gas , some material builds up there , which explains the bump in @xmath13 in that region as seen in figure ( [ fig : temp_tau_distr]c ) during the outburst . when the density wave reaches @xmath105 , the instability saturates , since the disk in that region is always gas - pressure dominated ( for the chosen @xmath10 and @xmath67 ) . as the density wave dies away , the outermost unstable regions are cleared of the excess mass and cool down . they are now in the low viscosity state , and thus the rate of mass inflow in the innermost region of the disk drops . an inward propagating cooling wave is initiated and the outburst dies out as the last excess mass in the innermost disk sinks into the hole . finally , a new cycle begins with the accumulation of new excess mass ( since the local accretion rate is smaller than the average rate ) . we now systematically explore the disk behavior with the new viscosity prescription , in order to eventually compare the results with the gross temporal behavior of grs 1915 + 105 . we adopted an energy transport fraction @xmath67 = 0.9 . as with the other parameters , we also tested a wide range of values of @xmath67 , and found that @xmath106 is the most appropriate due to the following considerations . the threshold for the onset of the instability is about @xmath107 0.26 for @xmath108 0.9 , and scales as : @xmath109^{-9/8 } \label{dmo}\ ] ] ( see svensson & zdziarski 1994 and note that their definition of @xmath10 differs from ours by a factor that accounts for the standard disk radiative efficiency , i.e. , @xmath110 ) . it seems clear observationally that @xmath111 should be larger by a factor of at least @xmath112 than the value of @xmath113 corresponding to the transition from the gas- to radiation pressure - dominated regime in the standard theory . indeed , if we assume that the maximum luminosity of @xmath114 erg sec@xmath115 observed in grs 1915 + 105 corresponds to the eddington luminosity for this source , then the instability seems to exist only for @xmath116 . in addition , galactic black hole candidates ( gbhcs ) with a lower persistent luminosity , or weaker transient behavior , have not shown such violent instabilities as grs 1915 + 105 does , and yet many of them are as bright as @xmath117 ( see , e.g. , barret , mcclintok & grindlay 1996 ) . we will discuss the parameter @xmath67 further in [ sect : gross ] . in figure ( [ fig : lcurves ] ) we show the evolution of the integrated disk luminosity as a function of time for the nominal values of the parameters listed above and for @xmath118 @xmath119 , 0.31 , 0.5 , and 1.5 . in agreement with equation ( [ dmo ] ) , we find that for values of @xmath10 below 0.26 the disk is stable , because there is no radiation - pressure dominated zone in the disk . we also find , in general , that as the value of @xmath10 increases , the duty cycle ( i.e. , the fraction of time that the source spends in the high state ) increases , from about 5% at @xmath118 0.26 , to approximately 50% when @xmath118 1.5 . similarly , the ratio of maximum to minimum luminosity through the cycle grows with increasing @xmath10 . the time for the disk to complete one of its cycles ( defined as the cycle time ) also increases with @xmath10 , except for very near the minimum value of @xmath118 0.26 where the instability first sets in ; here the cycle time actually decreases with increasing @xmath10 . the reason is that there seems to exist a `` critical excess mass '' in the inner disk region , such that the instability will appear only when the mass in the disk exceeds this value . the rate for building up the excess mass is @xmath120 , where @xmath111 is the maximum stable accretion rate given by equation ( [ dmo ] ) , and therefore a small increase in @xmath10 above @xmath111 can bring about a substantial decrease in the cycle time . for larger accretion rates , the cycle time starts to increase with increasing accretion rate , because a larger region of the disk is unstable , and it takes longer to clear the excess mass during the outburst . we will now compare our results shown in figure ( [ fig : lcurves ] ) with observations of grs 1915 + 105 . we should acknowledge at the start that it seems a daunting task to try to explain with a single model the diverse , unstable behavior exhibited by this source , but we expect to benefit from a comparison of even the gross properties of our model with the observations . panel ( a ) of figure ( [ fig : lcurves ] ) is similar to figure ( [ fig : data]a ) in that in both figures there exists a long phase of mass accumulation ( the low state ) . furthermore , an examination of the grs 1915 + 105 data reveals that the cyclic limiting behavior disappears when the average count rate is lower than @xmath121 counts / sec , so that there indeed exists a threshold accretion rate below which the instability does not operate . in simulations , the spiky nature of the outburst is explained by the fact that only a very narrow region in @xmath17 within the disk is unstable , and it takes very little time to get rid of the excess mass in that region . the differences with fig . ( 1a ) are mainly due to the fact that an outburst in the data does not seem to completely clear the excess mass , and that more oscillations follow with a gradually declining amplitude , whereas in the simulations the hot state persists until all the excess mass is swallowed by the black hole . panel ( b ) of fig . ( [ fig : data ] ) is also somewhat similar to panels ( b ) and ( c ) of the simulations ( fig . [ fig : lcurves ] ) . the progression of time scales and the average luminosity ( i.e. , roughly speaking , the count rate ) from pattern ( a ) to ( b ) in fig . ( [ fig : data ] ) is similar to that from panel ( a ) to ( b ) in fig . ( [ fig : lcurves ] ) as well . the physical reason for the cycle time getting shorter with increasing @xmath10 is that the excess mass can build up faster for larger accretion rates . next , panel ( c ) in figure ( [ fig : data ] ) is most interesting from the point of view that the high state is long lasting and is clearly stable at least for the first half . panel ( d ) of the simulations can account for some of the properties of this variability pattern . in particular , the duty cycle increased in both the simulations and the data as the accretion rate increased . the cycle time did not increase nearly enough in the simulations , however . it could be made longer by choosing a smaller @xmath69 , but then the time scales in all the other panels would increase as well , which does not appear acceptable . also , a general trend seen in the first three panels ( a - c ) in both the data and the simulations is that the minimum of each light curve shows a slow gradual rise in luminosity before the instability sets in . this is a clear indication of the disk accumulating mass until it reaches a global instability . we note that panel ( d ) of figure ( [ fig : data ] ) seems to be rather different from the other three variability examples . if the high state is again defined as the one with larger count rates , then a peculiarity of panel ( d ) is that the count rate first decreases and then increases by the end of a high state episode . this is the opposite of what is seen for the other states , such as that shown in panel ( c ) of fig . ( 1 ) . from a theoretical point of view , this is a highly significant observation . in all of our simulations , we observed that as the outburst progresses , the amount of mass in the inner disk region builds up because of the greater and greater inflow of mass from larger radii . once in a given ( high or low ) stable state , the local disk luminosity is proportional to the local column density @xmath13 , which can be seen from figure ( [ fig : scurve ] ) . therefore , in an outburst , the luminosity decreases only after the outer disk cools down and the influx of mass stops , leading to a decrease in @xmath13 in the inner disk . thus , the outburst profile in the simulations is such that the second time derivative @xmath122 is always negative , not positive as seen in panel ( d ) of figure 1 . we do not see a clear explanation for this disagreement in the context of the current model , but we will show that there might be a natural cause for this phenomenon if a jet is allowed ( [ sect : jet ] ) . finally , we note that in a number of the panels in figure ( [ fig : lcurves ] ) , the first ( and sometimes also the second ) peak in the light curve is of slightly different intensity or duration than the peaks that come later in the sequence . this is simply a consequence of the disk adjusting to its quasi - steady cyclic pattern after starting from the imposed initial conditions that are given by the shakura - sunyaev solution corresponding to the mean value of @xmath10 . let us now discus the rather extreme value for @xmath67 invoked here , which requires that as much as @xmath123 of the accretion power is carried out of the disk by processes other than the usual radiative diffusion . if this process is magnetic buoyancy or some other mhd process heating the corona , then there appears to be a contradiction with the observations of grs 1915 + 105 . the blackbody component in the spectrum of this source is generally small in the low state , but it can be dominant in the outburst state ( muno , morgan & remillard 1999 ) . one could argue here that as much as @xmath124 of the x - rays emitted by the corona towards the disk may be reprocessed into the soft disk radiation , ( e.g. , magdziarz & zdziarski 1995 ) if the reflection / reprocessing takes place in a _ neutral medium _ ( neutral in the sense of high - z elements like oxygen and iron , that are important for the reflection spectrum ) . thus , in principle , the blackbody flux could exceed that in the non - thermal component by the ratio @xmath125 ( see eqs . 15 in haardt & maraschi 1991 with their parameters @xmath126 and @xmath127 ) . the rather high inferred disk temperature in grs 1915 + 105 , e.g. , @xmath128 kev ( belloni et al . 1997a ) , rules out the possibility of neutral reflection , however . as found by nayakshin & dove ( 1998 ) and nayakshin ( 1998 ) , the integrated albedo of the reflected spectrum in the case of a strongly ionized disk can be much higher than the standard 0.1 - 0.2 of the neutral reflector ( i.e. , @xmath129 in eqs . 15 of haardt & maraschi 1991 ) , and thus it seems difficult to have @xmath67 as large as 0.9 given the large amount of soft power observed from the disk . a possible way out of this dilemma is that the high value of @xmath67 does not necessarily represent the energy flux from the disk into the corona . indeed , equation ( [ fvert ] ) states that the energy transport out of the disk may be @xmath56 times faster than that given by the standard diffusion of radiation . convection of energy in the vertical direction ( e.g. , bisnovatiy - kogan & blinnikov 1977 ; goldman & wandel 1995 ; and references therein ) is one physical mechanism that can speed up the transfer of energy out of the disk . mhd waves dissipating their energy before they reach the corona could be another . in addition , there is no proof that the vertical averaging procedure ( i.e. , a one zone approximation ) used in the standard accretion disk theory does not lead to a substantial underestimation of the vertical radiation flux out of the disk . for example , if @xmath130 , where @xmath131 few , then @xmath111 becomes @xmath132 ( see sz94 ) , and thus one may have @xmath111 as large as that observed in grs 1915 + 105 due to a faster or an additional disk cooling mechanism rather than due to a transfer of most of the disk power into the corona . it is notable that panel ( c ) of fig . ( [ fig : data ] ) shows rapid chaotic oscillations as fast as @xmath1 tens of seconds at the end of the high state , whereas our simulations do not show a similar behavior . furthermore , note that the rise / fall time scales are shorter in the data than they are in our model . it appears that the heating of the unstable region in grs1915 happens on a time scale that is very much shorter than the cycle time . by contrast , the outermost part of the unstable region of the simulated disk becomes unstable only after @xmath133 the duration of the hot phase , that is of the order of the disk viscous time for the high luminosity cases . one explanation here could be that these fast oscillations are failed attempts by the disk to make a state transition from the high to the low state . further , during these fast oscillations the disk is always brighter than it is in the low state , which could be interpreted as an indication that only a part of the disk ( most likely the innermost region ) in grs 1915 + 105 takes part in these rapid oscillations , whereas the rest of the unstable region is still on the upper stable branch of the s - curve . for this to be true , the inner disk must be able to decouple to some extent from the rest of the unstable disk , and be variable on a much shorter time scale than the outer disk . to test this idea , we have carried out several simulations in which the @xmath0-parameter is a decreasing function of radius . our hope is that in this case , since @xmath0 in the inner disk is larger than that in the outer disk , the time scale for oscillations in the inner disk may be much shorter than the overall viscous time , which has the dependence @xmath134 ( e.g. , frank et al . 1992 ) . in figure ( [ fig : variable_alpha ] ) we show one such simulation , in which the viscosity parameter was chosen to be @xmath135 $ ] . this functional form allows the @xmath0-parameter to be roughly @xmath136 in the inner disk region ( @xmath137 ) , and to be @xmath138 for @xmath139 . the presence of these two time scales is obvious just from a perusal of the resulting light curves . oscillations of the inner disk produce the precursor seen before each major outburst in fig . ( [ fig : variable_alpha ] ) . its relative magnitude is small because the amount of excess mass stored in the inner disk is small compared to that in the outer disk . furthermore , in the simulations the inner disk decouples from the outer one only at the beginning of the outburst , not at the end as is seen in the grs 1915 + 105 data . this is due to the fact that once the outer disk makes a transition to the high state , the large mass supply to the inner disk forces the latter to go into the high state as well and remain there . in summary , varying the @xmath0-parameter with radial distance does not appear to be a viable explanation for the rapid chaotic oscillations seen in the data which are superposed on the more regular long time scale disk evolution . an alternative way of explaining the fast oscillations could have been provided if the heating / cooling fronts stall and are reflected back as cooling / heating fronts . this behavior was observed in simulations of the classical thermal ionization disk instability by cannizzo ( 1993 , see text above his equation 5 ) , where it happened to be an unwanted result . however , we have not been able to see such stalled transition fronts ( except for the case presented in fig . [ fig : variable_alpha ] ) , which should not be too surprising , due to the fact that the underlying physics of the ionization instability and the one explored in this paper are vastly different , and one does not expect a direct correspondence between these two instabilities . although we will not present any light curves , we should mention that we have also attempted to allow the fraction @xmath67 to be a function of @xmath85 , since the transition from @xmath140 to @xmath141 means a substantial change in the physical conditions in the disk . our hope was that the freedom in choosing @xmath142 might help to reproduce the disk flickering . however , all our attempts in this regard ( with @xmath67 decreasing or increasing across the transition in @xmath85 ) have been unsuccessful . although our model is able to reproduce a number of the general trends seen in the light curve of grs 1915 + 105 , the rise / fall time scales are always shorter in the data than they are in the simulations . the fast rises and falls are perhaps the reason why the profile of an outburst in some of the actual light curves of grs 1915 + 105 ( e.g. , panel ( c ) of fig . [ fig : data ] ) is reminiscent of a square - wave like shape rather than the rounded shapes that an outburst has in the simulations ( see fig . [ fig : lcurves]d ) . the difference between our simulations and the data appears important enough to us to require the addition of a new feature to our model and test whether it can bring the theory closer to the observations . the following considerations have guided us in choosing the additional ingredient in our model . the speed with which the hot state propagates outward in the accretion disk is equal to the speed of the transition wave , which was studied analytically by , e.g. , meyer ( 1984 ) and numerically by menou , hameury & stehle ( 1998 ) . both of these studies find that the propagation speed of the transition wave is @xmath143 . the time scale for the transition front to traverse an unstable region of size r is @xmath144 , i.e. , the transitions in the disk actually take @xmath145 times longer than the thermal time scale of the outermost part of the unstable disk region . this is the reason why the rises and falls take too long in the simulations compared to the data . however , notice that during the initial stage of the transition from the low to the high state , the luminosity rises quite rapidly in the simulations ( see fig [ fig : lcurves]d ) , at a rate probably as fast as that seen in the data . now , if some physical process were to `` cut '' the simulated light curves at , say , @xmath146 , so that anything emitted by the disk above this value _ is not seen _ by the observer , then the outburst profile would be much more consistent with the data . to find a process able to `` cut '' the light curve , we should recall that powerful plasma ejections are known to occur in grs 1915 + 105 ( e.g. , mirabel and rodriguez 1994 ) . recently , eikenberry et al . ( 1998 ) , mirabel et al . ( 1997 ) and fender & pooley ( 1998 ) have shown that there exists a strong link between the x - ray emission and emission in the infrared and radio frequencies . the radio and infrared emission is attributed to plasma ejection events taking place in the innermost part of the disk . the minimum power deposited at the base of the jet ( that may consist of individual ejection events ) was found ( e.g. , gliozzi , bodo & ghisellini 1999 ) to be as high as @xmath147 few @xmath148 erg sec@xmath115 for the event first discovered by mirabel and rodriguez ( 1994 ) . during other observations ( e.g. , eikenberry et al . 1998 ) , the radio fluxes were a factor of @xmath149 lower than the ones observed by mirabel & rodriguez ( 1994 ) , which still may require @xmath150 few @xmath151 erg sec@xmath115 ejected into the jet , which is as large as the largest x - ray luminosities produced by the source . thus , if one were to model these plasma ejections in the framework of our model , one would have to allow for a significant portion of the accretion _ energy _ to be diverted into the jet . from here on , we will therefore assume that the total accretion disk power @xmath152 consists of two parts : the first is the x - ray luminosity @xmath153 , that is equal to the sum of the thermal disk and non - thermal coronal luminosities ; and the second part @xmath154 is the `` jet power '' , i.e. , the energy ejected into the jet . for simplicity we further assume that the jet power is not seen in x - rays , and appears only in the radio or other non - x - ray wavebands . as before , the locally produced total disk power is given by equation ( [ fvert ] ) , where we will again suppose that @xmath155 . the jet luminosity is then @xmath156 where @xmath157 is the fraction of power that escapes from the disk into the jet . this parameter ( @xmath158 ) may be a constant , or it may be a function of local disk conditions or a function of some global variable , e.g. , the total disk power @xmath152 . since the importance of advection relative to the radiative cooling is roughly given by the ratio @xmath159 , and since according to our discussion in appendix [ sect : geometry ] @xmath160 in this source , the advection of energy is not important in our model . thus , the observed x - ray power @xmath153 will consist from the disk power ( fraction @xmath161 of the total disk power ) and the coronal luminosity which is given by an equation analogous to equation ( [ lj ] ) except with a parameter @xmath162 instead of @xmath158 inside the integral . to eliminate any ambiguities of our approach , we note that the accretion power is divided among the disk , the corona , and the jet as @xmath161 , @xmath163 and @xmath164 , respectively . we should also mention that the amount of mass carried away by the jet is small compared to that accreted into the black hole , so we neglect the former in the mass conservation ( eq . [ eq4 ] ) . the point here is rather simple : if the ultimate source of the jet power is the underlying accretion disk , then to produce one relativistic proton in the jet , many protons in the disk must transfer their gravitational energy ( which is small compared with their rest mass ) to that one jet proton . these protons will therefore have to sink into the black hole in order to send one proton into the jet . more precisely , the mass outflow rate is @xmath165 , where @xmath166 few is the terminal bulk lorenz factor of the jet . this is to be compared with the accretion rate through the disk , that is equal to @xmath167 , where @xmath168 is the `` radiative '' efficiency of the disk and @xmath152 is the total disk power . accordingly , @xmath169 since @xmath170 . this estimate is quite conservative . for example , if the jet power is dominated by magnetic fields or by electron - positron pairs , then the mass carried away in the jet is even smaller . a similar consideration allows us to neglect the angular momentum outflow from the disk into the jet . we point out however that the latter approximation depends on the model of the jet ; if the jet particles actually gain a considerable amount of the angular momentum as they are streaming away from the disk , then the torque exerted by the jet on the disk may become non - negligible . we plan to explore these effects in future work . eikenberry et al . ( 1998 ) found that the ejection events were absent in the low state and started at the onset and during the high state . we thus will accept a parameterization of @xmath158 such that it is close to zero when the disk is in the low state , and it is relatively large ( but still smaller than unity , of course ) when the disk is in the high state . we now explore three different illustrative prescriptions for the function @xmath158 to show the possible effects of the disk power being diverted into a jet . let us assume that the fraction @xmath158 is described by the following ( somewhat arbitrarily chosen ) function : @xmath171 where @xmath172 is the total power of the source in terms of its eddington luminosity . this equation is qualitatively reasonable since one expects particles to be ejected from the disk when the luminosity @xmath102 approaches ( and then exceeds ) the eddington limit . while the radiation may not be the ultimate driving power of the jet , the radiation pressure can cause more matter to be ejected and possibly produce stronger jets . we show in figure ( [ fig : steady1 ] ) the resulting x - ray light curve ( solid line ) and the total disk luminosity @xmath152 ( represented by a dashed curve ; scaled down by the factor of 3 to fit the figure ) for an accretion rate @xmath173 . note that the overall luminosity curve is the same as we would have obtained with our basic model for the same choice of @xmath67 and @xmath10 , but with no jet ejections ( see [ sect : basic ] & [ sect : results ] ) . the shape of the outburst indeed becomes more like a square - wave , and thus the rises and falls appear to be sharper than they were for the model with @xmath174 , thus making it possible to reproduce this feature of the observed light - curves ( in particular , panel ( c ) of fig . [ fig : data ] ) . as a second example , we test the following prescription for the jet power : @xmath175 ( this prescription , in contrast to the one given by eq . [ [ fj1 ] ] , was chosen among several that we tested to reproduce the panels ( c ) and ( d ) of fig . [ [ fig : data ] ] simultaneously as described in [ sect : test3 ] ) . the corresponding x - ray light curve and the total luminosity of the system are shown in figure ( [ fig : steady2 ] ) . one notices that the profile of the outburst in x - rays is now inverted with respect to the actual disk power . when the disk produces most of the energy output , the x - ray light curve actually has a local minimum , since most of the energy is ejected into the jet during that time . this effect may be responsible for the `` strange '' shape of the outburst seen in panel ( d ) of fig ( [ fig : data ] ) . as a final example we present a sequence of light curves produced by our model with a fixed prescription for the jet energy fraction @xmath158 and with the accretion rate being the only parameter that is varied . the functional dependence of @xmath158 is again given by eq . ( [ fj ] ) . further , motivated by the fact that the x - ray emission can fluctuate wildly when it is in the high state ( see fig . 1 ) , whereas it does not fluctuate as much in the low luminosity state , we allow the fraction @xmath67 carried away from the disk other than by the usual radiation diffusion to fluctuate around some mean . numerically , the @xmath67-factor will now contain a _ variable part @xmath176 , such that @xmath177 : @xmath178 where @xmath179 is the constant part fixed at the previously selected value of @xmath180 , @xmath181 is a random number distributed uniformly between -1 and + 1 ( a new value of @xmath181 is randomly chosen every 3 seconds ) , and @xmath182 is a function of radius @xmath183 such that it is unity in the inner disk and it approaches zero for @xmath184 . the latter radial dependence is introduced simply to avoid complications with the outer boundary condition at @xmath185 , where we set the disk structure to be that given by the gas - dominated shakura - sunyaev configuration with a constant @xmath186 ( see [ sect : code ] ) . the shape of the function @xmath187 is almost completely irrelevant for the resulting light curves as long as @xmath188 is large , since most of the disk power is liberated in the inner disk region . finally , although the second term on the right hand side of the equation ( [ ft ] ) appears to be complicated and model - dependent , its presence is not required and serves only to make the light curves look somewhat more random . none of the final conclusions of this paper depend on the particular choice for the random part in @xmath67 . figure ( [ fig : xandj ] ) shows the resulting x - ray light curves for four different accretion rates . it is worthwhile comparing figure ( [ fig : xandj ] ) with figure ( [ fig : lcurves ] ) , in which the plasma ejections and random fluctuations in @xmath67 were absent . the case with the accretion rate just above @xmath111 ( i.e. , panel c in fig . [ fig : lcurves ] ) has not been affected as strongly as that with the higher accretion rate ; this is of course due to the fact that little energy is ejected in the jet in the former case compared with the latter . yet panels ( a ) and ( c ) do look more chaotic , which is entirely due to the fluctuating part in the fraction @xmath67 ( see eq . ( [ ft ] ) . analyzing panels ( b ) and ( d ) of this figure , one notices that it turns out to be possible to reproduce both the sharp square - wave like pattern ( c ) and the `` inverted '' pattern ( d ) of figure 1 with the _ same _ prescription for @xmath158 . it is the same process , i.e. , the plasma ejection , that allows our model to better reproduce the data ( panels [ c ] and [ d ] of fig . 1 ) ; the only difference being that in panel ( d ) the ejected fraction of the disk energy is much greater than in panel ( b ) . for the completeness of presentation , we also plot the jet power @xmath154 with a dotted curve in fig . ( [ fig : jet1 ] ) corresponding to the simulation presented in panel ( b ) of fig . ( [ fig : xandj ] ) . given the relatively good agreement between the model and the observations of grs 1915 + 105 , we believe that we now have a better handle on the key characteristics of accretion in this source . while we feel less confident about the origin of the s - curve in the @xmath189 space , we think that the geometry of the accretion flow in grs 1915 + 105 is that of a geometrically thin and optically thick flow ( see [ sect : geometry ] ) . most of the x - ray luminosity is produced in an optically thin corona , that can cover the entire inner disk or consist of localized transient magnetic flares ( e.g. , haardt , maraschi & ghisellini 1994 ; nayakshin & melia 1997 ; nayakshin 1998 ) . in the latter case , the fraction @xmath67 of the power transported from the disk into the corona and the jet must be thought of as a time averaged variable . further , we believe that plasma ejection events must be an integral part of the accretion process , and are likely to be the result of an excessively large radiation pressure . we will discuss this issue in a broader context in [ sect : discussion ] . belloni et al . ( 1997a , b ) were the first to fit the spectrum in grs 1915 + 105 with a two - component model , consisting of a multi - temperature disk - blackbody with a variable inner disk radius plus a power - law . very importantly , these authors found that most of the complex spectral variability can be explained by a rapid change in the disk inner radius @xmath7 and a corresponding change in the temperature of the disk at @xmath7 . they have also shown a correlation between the disk filling time scale and the inferred radius of the inner disk . our results are qualitatively similar to the framework of the disk instability discussed by belloni et al . for grs 1915 + 105 since in the low state the inner disk is quite dim ( up to a factor of 10 - 30 , see figure [ fig : dim ] ) compared with what it should have been if the disk were stable . thus , the inner part of the disk can be largely unobservable and can be said to be `` missing '' . at the same time , in the high state , the inner disk is brighter than its equilibrium luminosity by typically a factor of a few , which may make the outer part of the disk ( that is stable and so has just the `` normal '' flux ) seem comparatively dim . under these conditions , it may be non - trivial to distinguish observationally between a disk where the temperature is a continuous function of r ( @xmath190 ) and a disk where the temperature is lower than expected in the outer part of the disk , since the outer disk carries a small fraction of the overall luminosity . in addition , the inner disk is never completely empty of mass in our model , although the difference between the inner disk surface density in the high versus that in the low state can be a factor of @xmath1 few to 10 ( see fig . [ fig : temp_tau_distr]c ) . note that @xmath7 as defined by belloni et al . should be identified with the largest radius reached by the heating wave in our simulations . more specifically , during the high state , @xmath7 should be defined as the radius of innermost stable orbit , whereas during the low state it is approximately equal to the largest radius in the disk in which @xmath191 ( i.e. , @xmath192 for fig . [ fig : dim ] ) . our values of @xmath7 are in general much larger than those obtained by belloni et al . the value of @xmath193 inferred from the observations could be larger if one were to fit the grs 1915 + 105 emission with a more complicated ( and likely more realistic ) spectrum than a multi- temperature blackbody appropriate for a shakura - sunyaev disk . moreover , for the high accretion rates observed in this source , the disk may become effectively optically thin and thus radiate as a modified ( rather than a pure ) blackbody ( see discussion in taam et al . since blackbody emission produces the maximum flux for a given temperature , the modified emission would require a larger @xmath7 . finally , the values of @xmath7 obtained by us would presumably be smaller for a kerr black hole . we plan to address this question in the future . belloni et al . ( 1997b ) showed that the outburst duration is proportional to the duration of the preceding quiescent state , but there is no correlation between the former and the quiescent time after the burst . profiles of outbursts in our model are rather regular , so we believe we see a correlation among these three quantities in contrast to the belloni et al . results . on the other hand , in many observations ( other than the one presented in belloni et al . 1997b ) grs 1915 + 105 does not show a good correlation between the duration of the outburst and the preceding quiescent phase ( t. belloni 1999 , private communication ) . we have systematically analyzed the physical principles underlying the behavior of grs 1915 + 105 , and have arrived at several important conclusions regarding the nature of the time - dependent accretion flow in this system . as discussed in appendix [ sect : geometry ] , geometrically thick advection - dominated flows are unacceptable for this source , since they would produce burst - like instabilities with a very small duty cycle compared to those seen in the data , for accretion rates smaller than the eddington value . to produce outbursts as long as @xmath194 seconds , geometrically thick accretion flows must have implausibly small values of @xmath0 . however , if the accretion rate is highly super - eddington , then a thick adaf disk could yield outbursts with a reasonable duty cycle ( @xmath34 ) . but since the viscous time scale for geometrically thick disks is as short as the thermal one , their light curves would be rather smooth which again contradicts the data ( see , e.g. , panels c & d of fig . it is also unclear how the observation ( e.g. , belloni et al . 1997a , b ; muno , morgan & remillard 1999 ) that the disk extends down to the innermost stable orbit during the high state can be reconciled with the structure of a geometrically thick , hot and optically thin flow . for these reasons , the standard cold accretion disk with a modified viscosity law , a corona , and plasma ejections seems to be the only reasonable choice . as we have shown in this paper , this geometry indeed allows one to obtain rather good general agreement between the theory and observations , and to understand particular features of the light curves in terms of fundamental physical processes . nevertheless , a critical reader may question whether the relatively large number of parameters introduced by us in this work allows us to pin - point actual values of these parameters . we believe that the answer to this question is `` yes '' , because spectral constraints , used very sparingly here , may prove to be quite restrictive in future studies . for example , we have shown only the total luminosity light curves in this paper , whereas the data also contain a wealth of information about the spectral evolution in grs 1915 + 105 ( belloni et al . 1997 ; muno et al . 1999 ) . if future observations also provide more examples of the disk - jet connection ( e.g. , eikenberry et al . 1998 ) , then we have a means of constraining the dependence of both @xmath158 and the total fraction @xmath67 on the disk luminosity . further , one should also attempt to explore the fact that other transient and persistent black hole sources do not exhibit instabilities similar to grs 1915 + 105 , at least not to the same extreme degree . thus , putting all these observational and theoretical constraints together may be quite effective in limiting the theoretical possibilities for the instability and the coronal and jet activity . as an example of the need for a detailed study of the spectra and qpos in the context of our model , we point out the following . from the results of muno et al . ( 1999 ) it appears that qpos are present when the power - law component dominates the spectrum , which usually happens in the low state ( `` low '' means the lower count rate ) . further , these authors find that for the burst profiles similar to the one shown in fig . 1(d ) , the qpos are present during the `` high '' state , which is opposite to most of the other cases . if qpos indeed track the presence of a vigorous corona , this would imply that although our panel ( d ) in fig . ( [ fig : xandj ] ) looks similar to fig . 1(d ) , it does not represent the actual situation very well . namely , the data seem to imply that the higher count rate states actually have lower accretion rates through the inner disk than the lower luminosity states ( because excess luminosity escapes into the jet ) . it is thus possible that plasma ejections occur during the sharp dips in fig 1(d ) rather than during the longer phases ( e.g. , from @xmath195 to @xmath1 960 sec in this figure ) . the `` m - shape '' of the outbursts in fig . ( [ fig : data]d ) may then be understood because these would be the low states as tracked by the accretion rate , and the low states of the three other panels ( a - c ) of fig . ( [ fig : data ] ) show similar shapes albeit with lower count rates . although we have not yet made a detailed study of this suggestion , we can probably accommodate it within our model , but only if we allow the parameters in our viscosity law to depend on the disk luminosity . _ note , however , that in either case we must require plasma ejections from the inner disk in order to understand panel ( d ) of fig . ( 1 ) . _ the discussion above of course bears on the particular choice of the viscosity law that we have made in this paper . we emphasize that our model is still rather empirical and we have not identified the actual physics of the viscosity law in grs 1915 + 105 . the real physics might well be yet more complex , and in fact , the existence of the s - curve might be due to other than radiation pressure instabilities , perhaps involving an effect that we do not currently understand . however , based on the results presented in this paper , we feel confident that there is an s - curve in this source , and that this curve can be modeled approximately by the viscosity law given by equation ( [ eq3 ] ) . a clear shortcoming of our work is the use of a non - relativistic disk around a non - spinning black hole . we plan to improve this in our future work . we expect that this will change the `` reasonable '' values of the parameters that we used , such as the fraction @xmath67 , and the inner disk radius , of course , but that it will not affect the general nature of our conclusions . the authors thank t. belloni , r. taam , m. muno , e. morgan , r. remillard , e. vishniac and d. kazanas for useful discussions . this work was supported in part by nasa grants nag5 - 8239 and nag5 - 4057 , under the astrophysics theory program .

in the complex action theory ( cat ) we explicitly examine how the momentum and hamiltonian are defined from the feynman path integral ( fpi ) point of view based on the complex coordinate formalism of our foregoing paper . after reviewing the formalism briefly , we describe in fpi with a lagrangian the time development of a-parametrized wave function , which is a solution to an eigenvalue problem of a momentum operator . solving this eigenvalue problem , we derive the momentum and hamiltonian . oppositely , starting from the hamiltonian we derive the lagrangian in fpi , and we are led to the momentum relation again via the saddle point for . this study confirms that the momentum and hamiltonian in the cat have the same forms as those in the real action theory . we also show the third derivation of the momentum relation via the saddle point for .

1105.1294

quantum theory is described via feynman path integral ( fpi ) . the integrand in fpi has the form of @xmath3 , where @xmath4 is the imaginary unit and @xmath5 is the action . the action is usually taken to be real . however , if we assume that the integrand is more fundamental than the action in quantum theory , then we can consider a theory where the action is taken to be complex , by speculating that since the integrand is complex , the action can be also complex . based on this assumption and other related works in some backward causation developments inspired by general relativity@xcite@xcite and the non - locality explanation of fine - tuning problems@xcite@xcite , the complex action theory ( cat ) has been studied intensively by one of the authors ( h.b.n ) and ninomiya@xcite@xcite . compared to the usual real action theory ( rat ) , the imaginary part of the action is thought to give some falsifiable predictions . indeed , many interesting suggestions have been made for higgs mass@xcite , quantum mechanical philosophy@xcite , some fine - tuning problems@xcite , black holes@xcite , de broglie - bohm particle and a cut - off in loop diagrams@xcite . in ref . we have studied the time - development of some state by a non - hermitian diagonalizable bounded hamiltonian @xmath6 , and shown that we can effectively obtain a hermitian hamiltonian after a long time development by introducing a proper inner product . if the hermitian hamiltonian is given in a local form , a conserved probability current density can be constructed with two kinds of wave functions . we note that the non - hermitian hamiltonian is a generic one , so it does not belong to a class of the pt symmetric non - hermitian hamiltonians , which has been intensively studied recently@xcite@xcite . for reviews see refs . . in ref . , we have proposed the replacement of hermitian operators of coordinate and momentum @xmath7 and @xmath8 and their eigenstates @xmath9 and @xmath10 with non - hermitian operators @xmath11 and @xmath12 , and @xmath13 and @xmath14 with complex eigenvalues @xmath2 and @xmath1 , so that we can deal with complex @xmath2 and @xmath1 obtained at the saddle point . introducing a philosophy to keep the analyticity in parameter variables of fpi and defining a modified set of complex conjugate , real and imaginary parts , hermitian conjugates and bras , we have explicitly constructed them for complex @xmath2 and @xmath1 by squeezing the coherent states of harmonic oscillators . in addition , extending the study of ref . to the complex coordinate formalism , we have investigated a system defined by a diagonalizable non - hermitian bounded hamiltonian , and shown that a hermitian hamiltonian is effectively obtained after a long time development also in the complex coordinate formalism . in addition , as other works related to complex saddle point paths , in refs . @xcite@xcite the complete set of solutions of the differential equations following from the schwinger action principle has been obtained by generalizing the path integral to include sums over various inequivalent contours of integration in the complex plane . in ref . complex langevin equations have been studied . in refs . and a method to examine the complexified solution set has been investigated . the cat has been studied intensively as mentioned above , but there remain many things to be studied . for instance , in the above studies it has been supposed and taken as a matter of course that the momentum and hamiltonian in the cat have the same forms as those in the rat , and it has not been examined explicitly so far . in the rat , we first write the lagrangian @xmath15 in terms of a coordinate @xmath2 and its time derivative @xmath16 , and then define the momentum @xmath1 by the relation @xmath17 . next , we define the hamiltonian @xmath18 via the legendre transformation as @xmath19 . by replacing @xmath2 and @xmath1 with their corresponding operators in the classical hamiltonian @xmath6 , we obtain the quantum hamiltonian . this is a well - known story in the rat , but what happens in the cat ? once we allow the action to be complex , various quantities known in the rat can drastically change , and we can encounter various exotic situations . hence we have to be careful about them . indeed , the momentum relation @xmath17 in the cat is not so trivial . we write @xmath20 as @xmath21 . if we quantum mechanically have in mind that our formalism corresponds to the deformation of the contour of originally real @xmath2 and @xmath1 , then using functional integral fpi and imagining to choose the @xmath22-integrals along the real axis we would intuitively think that both @xmath22 and @xmath23 could be chosen real and thus @xmath16 would be real . this is seemingly a discrepancy for the equation @xmath24 with @xmath25 complex , and also for its naive operator interpretation by expecting that the operators corresponding to these dynamical variables are still hermitian . in fact this seeming discrepancy partly motivated us to construct the complex coordinate formalism in ref . . it is the main subject of the present paper to formulate in such a way that we can quantum mechanically assign at least some meaning to the relation @xmath17 . normally in quantum mechanics @xmath23 has no definite value . rather in fpi the system at time @xmath26 is given in a superposition of essentially all @xmath23-values . to see whether the momentum relation holds in the cat or not , we analyze it explicitly by writing it in the form of the operator acting on appropriate states with a number @xmath0 put as replacement for @xmath23 . see eq.([phat_p_xi2 ] ) . we shall show in this paper that the relation @xmath17 is true in the classical sense along the time development of the saddle points in the integral @xmath27 in @xmath28 . indeed we put forward this saddle point interpretation of @xmath17 as eq.([pjmqj ] ) . since with complex mass @xmath25 the saddle point or classical path will typically be complex , there is nothing strange in @xmath17 of eq.([pjmqj ] ) conceived as a saddle point relation . from the above point of view , in this paper we explicitly examine how the momentum and hamiltonian are defined in the cat based on the complex coordinate formalism of ref . . we replace hermitian operators of coordinate and momentum @xmath7 and @xmath8 and their eigenstates @xmath9 and @xmath10 with non - hermitian operators @xmath11 and @xmath12 , and @xmath13 and @xmath14 with complex eigenvalues @xmath2 and @xmath1 , and utilize the new devices such as a modified set of complex conjugate , hermitian conjugates and bras to realize the philosophy of keeping the analyticity in parameter variables of fpi . in the usual way of deriving and understanding the functional integral , a functional integral corresponding to finite time interval say @xmath29 $ ] with endpoints fixed to specific @xmath30 and @xmath31 values is expressed as @xmath32 , where @xmath33 is a priori taken over real @xmath34 s . we claim that this interpretation is possible even for complex @xmath31 and @xmath30 by following the formalism of ref . . we can deform formally the infinitely many contours as long as the action @xmath5 is an analytical function of all the @xmath34 s . in this way going to complex contours only with restriction to go from @xmath35 to @xmath36 is rather trivial . we , however , stress that after the deformation of the contour in a different way for each moment @xmath37 the contours still start at @xmath35 and ends at @xmath38 , but in between of course they are usually complex . that is to say , appropriate contours are chosen at each time @xmath37 . the present paper has also a role of showing an explicit application of the formalism of ref . . starting from the lagrangian @xmath39 in fpi , we derive the momentum relation by considering an eigenvalue problem of the momentum operator , which includes a parameter @xmath0 . we attempt to split up a wave function @xmath40 into various @xmath0-components , and investigate the time - development of the @xmath0-parametrized wave function in fpi . this study elucidates that the momentum in the cat has the same form as that in the rat . in addition we also derive the hamiltonian @xmath6 in fpi starting from the lagrangian @xmath39 in the cat . next in an opposite way we try to check whether we can reproduce the lagrangian starting from the hamiltonian , and show that the lagrangian is derived from the hamiltonian . this result confirms that the hamiltonian in the cat has the same form as that in the rat . it leads us to the momentum relation again via the saddle point for @xmath1 . we also give the third derivation of the momentum relation via the saddle point for @xmath2 . this paper is organized as follows . in section 2 we briefly review the complex coordinate formalism of ref . . introducing the philosophy of keeping the analyticity in parameter variables of fpi and the new devices to realize it such as a modified set of bra , complex conjugates , hermitian conjugate and hermiticity , we construct non - hermitian operators of coordinate and momentum @xmath11 and @xmath12 , and the hermitian conjugates of their eigenstates @xmath41 and @xmath42 with complex eigenvalues @xmath2 and @xmath1 . in section 3 we derive the momentum relation by solving an eigenvalue problem of the momentum operator and considering the time - development of the wave function solution in fpi . we also derive the free hamiltonian starting from the free lagrangian in fpi . next starting from the lagrangian with a potential term in fpi , we derive the momentum relation and the hamiltonian with the potential term . in section 4 oppositely by using the hamiltonian @xmath6 we derive the lagrangian @xmath39 in fpi . also , we again obtain the momentum relation via the saddle point for @xmath1 . this study reveals that the momentum and hamiltonian in the cat have the same forms as those in the rat . section 5 is devoted to conclusion and outlook . in appendix [ via_saddle_q ] we explain the third derivation of the momentum relation via the saddle point for @xmath2 . in ref . we have proposed the replacement of hermitian operators of coordinate and momentum @xmath7 and @xmath8 and their eigenstates @xmath43 and @xmath10 with non - hermitian operators @xmath11 and @xmath12 , and @xmath13 and @xmath14 with complex eigenvalues @xmath2 and @xmath1 , so that we can deal with complex @xmath2 and @xmath1 obtained at the saddle point . introducing the philosophy of keeping the analyticity in parameter variables of fpi , we have defined the new devices such as a modified set of complex conjugate , real and imaginary parts , hermitian conjugates and bras , by means of which we have explicitly constructed @xmath11 , @xmath12 , and the hermitian conjugates of their eigenstates @xmath41 and @xmath42 with complex @xmath2 and @xmath1 by utilizing the coherent states of harmonic oscillators . in this section we briefly review the complex coordinate formalism of ref . . we begin with the delta function of a complex parameter . in quantum mechanics the delta function is one of essential tools in a theory which has orthonormal bases with continuous parameters , and the parameters are usually real in the rat . on the other hand , parameters in the cat can be complex in general , so we need to define the delta function of complex parameters . the delta function for real @xmath2 is represented as @xmath44 where @xmath45 is real . we assume that @xmath45 is also complex but asymptotic value of @xmath45 is real . in this case we can take an arbitrary path running from @xmath35 to @xmath36 in the complex plane . we call such a path @xmath46 and define @xmath47 for complex @xmath2 by @xmath48 where @xmath49 in the first line of eq.([delta_c_epsilon(q ) ] ) we have introduced a finite small positive real number @xmath50 , and in the second equality we have assumed that @xmath51 goes larger than @xmath52 . eq.([delta_c_epsilon(q ) ] ) is convergent for complex @xmath2 such that @xmath53 since for any analytical test function @xmath54 is a bounded entire function , @xmath55 is constant . so we are considering as @xmath55 an unbounded entire function or a function which is not entire but is holomorphic at least in the region on which the path runs . ] the path @xmath46 of the integral @xmath56 is independent of finite @xmath50 , this @xmath47 satisfies for any @xmath54 @xmath57 as long as we choose a path such that at any @xmath2 its tangent line and a horizontal line form an angle @xmath58 whose absolute value is within @xmath59 to satisfy the inequality ( [ cond_of_q_for_delta ] ) . an example of permitted paths is shown in fig.[fig : contour ] , and the domain of the delta function is drawn in fig.[fig : delta_function ] . at the origin @xmath47 is divergent . in the domain except for the origin , which is painted with inclined lines , @xmath47 takes a vanishing value , while in the blank region the delta function is oscillating and divergent . it is well - defined for @xmath2 such that the condition ( [ cond_of_q_for_delta ] ) is satisfied . to circumvent some difficulties due to the naive extension of @xmath2 to complex numbers , which are explained in ref . , we introduce a philosophy of keeping the analyticity in parameter variables of fpi . if we keep the analyticity in parameter variables of fpi , then we can deform a path , along which the integration is performed . to realize this philosophy we define a modified set of complex conjugate , hermitian conjugates and bras in this subsection . we define a modified complex conjugate for a function of @xmath61-parameters @xmath62 by @xmath63 where @xmath64 denotes the set of indices attached with the parameters in which we keep the analyticity , and on the right - hand side @xmath65 on @xmath55 acts on the coefficients included in @xmath55 . this is in contrast to a usual complex conjugate defined by @xmath66 we show a set of examples of the complex conjugates @xmath65 , @xmath67 , @xmath68 and @xmath69 on a function @xmath70 as follows , @xmath71 where in the first , second , third and fourth relations the analyticity is kept in no parameters , @xmath2 , @xmath1 and both @xmath2 and @xmath1 , respectively . for simplicity we express the modified complex conjugate as @xmath72 . for some state @xmath76 with some complex parameter @xmath77 , we define a modified bra @xmath78 by @xmath79 the modified bra is analytically extended bra with regard to the parameter @xmath77 . it preserves the analyticity in @xmath77 . in the special case of @xmath77 being real it would become a usual bra . next we define a little bit generalized modified bra @xmath80 and a modified hermitian conjugate @xmath75 of a ket , where @xmath81 is a symbolical expression of a set of parameters in which we keep the analyticity . we show some examples , @xmath82 we express the hermitian conjugate @xmath75 of a ket symbolically as @xmath83 also , we see that the hermitian conjugate @xmath75 of a bra can be defined likewise . we express it as @xmath84 for @xmath85 , @xmath86 and some operator @xmath64 , we have the following relations , @xmath87 in eqs.([hermitian_conjugate_rat])([hermitian_conjugate_u])([hermitian_conjugate_v])([hermitian_conjugate_uv ] ) the analyticity is kept in no parameters , the parameter @xmath88 , @xmath89 , and both @xmath88 and @xmath89 , respectively . we can choose one of the four options according to which parameters we keep the analyticity in . we can express the four relations simply as @xmath90 various examples of the usage of the modified complex conjugate , modified bra and modified hermitian conjugate are given in ref . . in this subsection following ref . we summarize how we construct the non - hermitian operators @xmath11 and @xmath12 , and their eigenstates @xmath93 and @xmath94 with complex eigenvalues @xmath2 and @xmath1 , which satisfy the following relations , @xmath95 = i \hbar , \label{commutator_q_p_new } \end{aligned}\ ] ] by utilizing coherent states of harmonic oscillators . here @xmath96 and @xmath97 are modified bras of @xmath41 and @xmath42 , so eqs.([qhatqket = qqket_newrev])([phatpket = ppket_newrev ] ) are equivalent to @xmath98 respectively . a coherent state parametrized with a complex parameter @xmath77 is defined up to a normalization factor by @xmath99 and satisfies the relation @xmath100 in eqs.([lambda_coh_def])([a_lambda_ket = lambda_lambda_ket ] ) @xmath101 and @xmath102 are annihilation and creation operators , which are defined in terms of @xmath7 and @xmath8 by @xmath103 the eigenstates of @xmath7 and @xmath8 are @xmath104 and @xmath105 for real @xmath2 and @xmath1 respectively , and they obey the following relations , @xmath106 = i \hbar . \label{commu_rel_qhatphat}\end{aligned}\ ] ] substituting eq.([annihilation ] ) for eq.([a_lambda_ket = lambda_lambda_ket ] ) , we obtain @xmath107 next we consider another coherent state @xmath108 parametrized with a complex parameter @xmath109 of another harmonic oscillator defined with @xmath110 and @xmath111 . since the annihilation operator of the coherent state is given by replacing @xmath25 and @xmath112 with @xmath110 and @xmath111 in eq.([annihilation ] ) , @xmath113 we obtain @xmath114 eqs.([coh_q])([coh_p ] ) inspire us to define @xmath11 , @xmath12 , @xmath41 and @xmath42 by @xmath115 with the introduction of complex @xmath2 and @xmath1 as @xmath116 these expressions satisfy eqs.([qhatqket = qqket_new])([phatpket = ppket_new])([commutator_q_p_new ] ) , and @xmath41 and @xmath42 are normalized so that they satisfy the following relations , @xmath117 where we have used the expression of eq.([delta_c_epsilon(q ) ] ) , and @xmath118 and @xmath119 are given by @xmath120 for sufficiently large @xmath121 the tamed delta function in eq.([m_qbranew_qketnew ] ) converges for complex @xmath2 and @xmath122 satisfying the condition like eq.([cond_of_q_for_delta ] ) , @xmath123 this condition is satisfied only when @xmath2 and @xmath122 are on the same path . similarly , for sufficiently small @xmath124 the tamed delta function in eq.([m_pbranew_pketnew ] ) converges for complex @xmath1 and @xmath125 satisfying the following condition , @xmath126 thus for sufficiently large @xmath121 and small @xmath127 eqs.([m_qbranew_qketnew])([m_pbranew_pketnew ] ) represent the orthogonality relations for @xmath41 and @xmath42 . in the following we take @xmath121 sufficiently large and @xmath127 sufficiently small . then we have the following relations for complex @xmath2 and @xmath1 , @xmath128 , \label{basis_fourier_transf2 } \\ % & & \langle p ' | q \rangle_{new } = \frac{1 } { \sqrt { 2\pi \hbar } } \exp\left [ - \frac{i}{\hbar } p ' q \right ] \quad \text{for real $ p'$ } , \label{brap'qketnew } \\ % & & \langle q ' | p \rangle_{new } = \frac{1 } { \sqrt { 2\pi \hbar } } \exp\left [ \frac{i}{\hbar } q ' p \right ] \quad \text{for real $ q'$ } , \label{braq'pketnew } \end{aligned}\ ] ] where eqs.([completion_complexq_ket2])([completion_complexp_ket2 ] ) represent the completeness relations for @xmath91 and @xmath92 . thus @xmath11 , @xmath12 , @xmath41 and @xmath42 with complex @xmath2 and @xmath1 obey the same relations as @xmath7 , @xmath8 , @xmath104 and @xmath105 with real @xmath2 and @xmath1 satisfy . for real @xmath122 and @xmath125 @xmath129 and @xmath130 become @xmath131 and @xmath132 respectively , @xmath133 also , @xmath134 and @xmath135 behave like hermitian operators @xmath7 and @xmath8 respectively for @xmath131 and @xmath132 with real @xmath122 and @xmath125 , @xmath136 using @xmath41 we define a wave function @xmath137 with complex @xmath2 by @xmath138 when @xmath2 is real , this wave function becomes a usual one @xmath139 one might feel a bit uneasy about the above replacement of the a priori correct operators @xmath7 and @xmath8 by @xmath11 and @xmath12 . it might help a tiny bit to accept @xmath11 and @xmath12 to have in mind that operators smooth in @xmath7 and @xmath8 like @xmath11 and @xmath12 have generically eigenvalues filling the whole complex plane , while hermitian operators like @xmath7 and @xmath8 have eigenvalues only along a certain curve say on the real axis in the complex plane . the consolation for our replacement is then that for our purpose of having general contours running through eigenvalues we replaced the special operators @xmath7 and @xmath8 by the more generic ones @xmath11 and @xmath12 . the philosophy should be that almost any small disturbance would anyway bring @xmath7 and @xmath8 into operators of the generic type with the whole complex plane as spectrum . the operators @xmath11 and @xmath12 are just concrete examples of such tiny deformation . so we stress that the hermitian operators as @xmath7 and @xmath8 are special by having their eigenvalue spectrum on a curve say on the real axis in the complex plane rather than distributed all over it . had we clung to the believe in curve - spectra , it would have been embarassing for our formalism that under heisenberg time development one could have feared that the curve spectra would be transformed into new curve - spectra which might not match at the free choice of contour from time to time in our scheme . now , however , as already stressed , if we use @xmath11 and @xmath12 , we have already from the beginning gone over to operators having any complex numbers as eigenvalues . so arbitrary deformation of the contour would a priori have no problem . thus we claim that the contours of integration can be chosen freely at each time @xmath37 , so that there is no need for any natural choice , which only has to run from @xmath35 to @xmath36 . in this section we have briefly reviewed the complex coordinate formalism of ref . . see ref.@xcite for the detail of the formalism . for our convenience we show the summary of the comparison of the rat and the cat in table [ tab1:comparison ] . in the next section we will study how the momentum and hamiltonian are defined in the cat based on the complex coordinate formalism . .various quantities in the rat and the cat [ cols="<,<,<",options="header " , ] in this section we study how the momentum and the hamiltonian are defined in the cat . before starting the explicit analysis we mention one of the general aspects of the cat . as a simple case of the cat we consider a non - relativistic particle with position @xmath2 in one dimension with a potential energy @xmath140 , which can now be a complex function of the real variable @xmath2 at first . also the mass @xmath25 of this particle can be complex . the lagrangian is given by @xmath141 and the functional integral takes the form @xmath142 \pi_t \left ( dq_t \frac{m}{2\pi i \hbar \delta t } \right ) , \end{aligned}\ ] ] where discretizing the time direction we have written @xmath16 as @xmath143 and introduced the notation @xmath34 and @xmath144 for @xmath22 and @xmath23 as @xmath145 we regard @xmath34 and @xmath144 as independent variables . we consider a theory such that the asymptotic values of dynamical variables such as @xmath2 or @xmath1 are on the real axis . the path @xmath46 denotes a set of arbitrary paths running from @xmath35 to @xmath36 in the complex plane at each time @xmath37 . since this regularized expression is analytical in all the @xmath34 for all different moments @xmath37 , we can arbitrarily deform the contour for each @xmath34 independent of each other , except for keeping some continuity in @xmath37 . in the lagrangian , since @xmath146 is taken to be a small quantity in the denominator of @xmath16 , the kinetic term @xmath147 tends to dominate over the potential term @xmath140 . therefore a kind of restriction is needed for the kinetic term . in practice we impose on the mass the condition @xmath148 where @xmath149 is the imaginary part of @xmath25 . then we can prevent the kinetic term in the integrand from blowing up for @xmath150 along real axis . we note that @xmath151 can be accommodated with this condition , since @xmath151 is obtained by taking an infinitesimally small limit of positive @xmath149 . therefore in the following we impose on @xmath25 the condition @xmath152 this is the wanted condition for @xmath149 in order that fpi of the cat is meaningful . we first consider the momentum . in classical mechanics of the rat the momentum is defined by the relation @xmath153 can we use eq.([def_p_delldelqdot ] ) as the definition of the momentum also in the cat ? it is not clear whether we can or not from a point of view of quantum mechanics because it includes @xmath144 , a quantity at time @xmath26 . the quantity @xmath144 is somehow unclear unless we define it properly including the fluctuation in the time - development from a quantity at time @xmath37 . hence we attempt to describe @xmath144 via fpi and derive the relation ( [ def_p_delldelqdot ] ) from fpi in the cat . in fpi the time - development of some wave function @xmath154 at time @xmath37 to time @xmath26 is described by @xmath155 for the purpose to derive the relation ( [ def_p_delldelqdot ] ) we take the following approach . first we imagine that we have some wave function @xmath156 with some number @xmath0 which obeys @xmath157 for fixed @xmath0 eq.([phat_p_xi2 ] ) is a differential equation to be obeyed by @xmath156 . this is an eigenvalue problem of the momentum operator . if the set @xmath158 could be an approximately reasonable basis which has at least roughly completeness and orthogonality , @xmath159 we could expand the wave function @xmath154 into a linear combination of @xmath156 as @xmath160 where in the second equality for our convenience we have introduced the notation @xmath161 , which is the contributed part of @xmath154 from @xmath0-component . by solving eq.([phat_p_xi2 ] ) and using the expansion ( [ qbra_atket_exp_lc_xi ] ) we attempt to show that eq.([time_dev_qbraaket ] ) would give @xmath162 because this relation means that only the component with @xmath163 contributes a non - zero value to @xmath164 . the relation ( [ def_p_delldelqdot ] ) can be obtained by showing eq.([prompt_delta ] ) . in the following we take this strategy , and calculate @xmath164 explicitly . by expressing this quantity in terms of @xmath165 and looking at its time - development operator , we can obtain the hamiltonian . before going ahead we mention that the above approach can be presented in a more general prescription . not only the momentum but also other operators including the momentum can be dealt with as follows . we consider relations involving a wave function and a parameter @xmath0 going into the lagrangian at place of @xmath144 as eq.([phat_p_xi2 ] ) . we are thus motivated to think about a family of @xmath0-parameterized operators @xmath166 which operates on a wave function @xmath156 to give a vanishing value as @xmath167 if @xmath168 satisfies the following relation @xmath169 @xmath166 acting on @xmath156 effectively projects out the @xmath170 component . the strategy explained in the previous subsection is an example of this general prescription . indeed if we take @xmath171 , which is the difference between the two operators acting on both sides in eq.([phat_p_xi2 ] ) , eq.([phat_p_xi2 ] ) is reproduced from eq.([a_xi_ket ] ) . for other purpose of studying the hamiltonian , for example , @xmath166 could be taken as @xmath172 . based on the approach explained in the foregoing subsections we shall solve the eigenvalue problem of eq.([phat_p_xi2 ] ) . usually the lagrangian @xmath39 is written as @xmath173 , where @xmath174 is a kinetic term and @xmath175 is a potential term . taylor - expansion of @xmath15 with regard to @xmath16 is written as @xmath176 where in the first equality we have assumed the normal lagrangian such that @xmath177 and in the second equality we have used eq.([dot_q ] ) . we consider the case where @xmath174 is given by @xmath178 then @xmath179 is written as @xmath180 so we have @xmath181 substituting eq.([par_l_par_dotq_xi ] ) for eq.([phat_p_xi2 ] ) , we find that the solution of eq.([phat_p_xi2 ] ) is given by @xmath182 \nonumber \\ % & = & c_a \tilde{\psi}_\xi ( q_t ) , \label{qbra_atket_xi}\end{aligned}\ ] ] where @xmath183 and @xmath184 is a normalization factor . here we note that the above wave function is normalizable for @xmath185 , but not for @xmath186 . namely the wave function diverges except for @xmath151 case when the condition ( [ m_i_positive_cond ] ) , which is generally required for convergence of the cat , is satisfied . it means that the wave function destined to make @xmath144 be @xmath0 by being a solution to eq.([phat_p_xi2 ] ) is not a good wave function , but rather a non - normalizable one . however , we think that it would not make a big problem because if we consider a test function which converges sufficiently , the product of the wave function and the test function in an integrand can converge . we will come back to this point later . next we explain how we normalize the wave function @xmath187 . since we have the following relation , @xmath188 , \label{qbra_atket_xi3}\end{aligned}\ ] ] we calculate the normalization factor according to whether @xmath25 is real or not . in the case of real @xmath25 , namely @xmath189 , we have @xmath190 dq_t \nonumber \\ % & = & \frac { 2 \pi \hbar \delta t } { m } \delta_c ( \xi ' - \xi ) % \exp\left[-\frac{i m ( \xi^2 - \xi'^2)}{2\hbar \delta t } \right ] \nonumber \\ % & = & \frac { 2 \pi \hbar \delta t } { m } \delta_c ( \xi ' - \xi ) , \label{qbra_atket_xi6}\end{aligned}\ ] ] where @xmath191 is well - defined for @xmath0 and @xmath192 such that @xmath193 . therefore we can normalize the wave function by choosing @xmath184 as @xmath194 thus we have @xmath195 and we can show @xmath196 from eqs.([qbra_atket_xi8])([qbra_atket_xi8a ] ) we obtain @xmath197 therefore we see that in the case of real @xmath25 @xmath198 has both of the completeness and orthogonority , and thus @xmath156 can work as a basis . in the case of @xmath199 , we have @xmath200 . \nonumber \\ \label{qbra_atket_xi4}\end{aligned}\ ] ] we note that the integral is divergent for @xmath201 , so we can not perform the gaussian integral and normalize the wave function for @xmath201 . to improve this situation we propose to introduce another special state @xmath202 such that we can have both of the completeness and orthogonality . first we try to find a dual basis , a set @xmath203 such that @xmath204 which is rewritten as @xmath205 from the expression of @xmath206 in eq.([qbra_atket_xi ] ) , we consider the following form of @xmath207 , @xmath208 , \label{bra_antixi_q_ket}\ ] ] from which we have @xmath209 . \label{bra_antixi_q_ket2}\end{aligned}\ ] ] since @xmath210 is written as @xmath211 @xmath212 is normalizable for @xmath186 . substituting eqs.([qbra_atket_xi])([bra_antixi_q_ket ] ) for eq.([anti - bara - xi ] ) , we obtain @xmath213 then we can also show @xmath214 in eq.([c_b*c_a ] ) we have ambiguity or freedom to distribute normalization factor between @xmath203 and @xmath198 , so no unique normalization is required . however , if we choose @xmath184 and @xmath215 as @xmath216 then the wave functions of eqs.([qbra_atket_xi])([bra_antixi_q_ket2 ] ) are explicitly written as @xmath217 , \\ % & & { } _ m\langle_{new}~ q_t | \text{anti } ~\xi \rangle = \sqrt { \frac{m^*}{2\pi \hbar \delta t } } \exp\left [ c_1^ { * } ( q_t - \xi)^2 \right ] , \end{aligned}\ ] ] and we note that with the choice of eqs.([c_acomplexm])([c_bcomplexm ] ) @xmath218 corresponds to @xmath219 when we take @xmath25 real . therefore the normalization of eqs.([c_acomplexm])([c_bcomplexm ] ) seems to be a natural choice . from the above consideration we can say that in the case of complex @xmath25 we have both of the orthogonality and completeness relation for @xmath220 by introducing @xmath203 , and thus @xmath156 can roughly work as a basis . due to the introduction of @xmath218 , eq.([qbra_atket_exp_lc_xi ] ) is replaced with @xmath221 this replacement is applied not only in the complex @xmath25 case but also in the real @xmath25 case since @xmath222 corresponds to @xmath223 in the latter case as mentioned above . in the previous subsection we have obtained the explicit form of @xmath156 , which is the solution to the eigenvalue problem of eq.([phat_p_xi2 ] ) , and showed that @xmath198 has both of the orthogonality and completeness relation . we are thus prepared to calculate @xmath224 explicitly . substituting eqs.([taylor_l])([qbra_atket_xi])([qbra_atket_exp_lc_xi2 ] ) for eq.([time_dev_qbraaket ] ) , we have @xmath225 % dq_t . \label{time_dev_qbraaket2}\end{aligned}\ ] ] we shall evaluate eq.([time_dev_qbraaket2 ] ) first in the free case , and next in the potential case . first we consider the free case , @xmath226 . in this case , since each component of the taylor - expanded @xmath39 in eq.([taylor_l ] ) is given by @xmath227 eq.([time_dev_qbraaket2 ] ) is calculated as @xmath228 dq_t \nonumber \\ % & = & c_a ~{}_m \langle \text{anti } ~\xi | \psi(t ) \rangle 2\pi\hbar \delta_c(\xi - q_{t+dt } ) \frac{dt } { m } \exp\left [ \frac{i m}{2\hbar dt } ( q_{t+dt}^2 - \xi^2 ) \right ] . \label{time_dev_qbraaket_k=0}\end{aligned}\ ] ] the above result means that @xmath229 therefore unless @xmath170 the value of @xmath230 vanishes . i.e. only the component with @xmath170 contributes to the value of @xmath164 . thus we have obtained the momentum relation of eq.([def_p_delldelqdot ] ) from the fpi point of view . next we evaluate @xmath164 . this quantity is calculated as @xmath231 ~{}_m \langle_{new}~ q_t | \psi(t ) \rangle \nonumber \\ % & \simeq & \sqrt { \frac { 2\pi i \hbar \delta t}{m } } % \left\ { { } _ m \langle_{new}~ q_{t+dt } | \psi(t ) \rangle -\frac{i}{\hbar } \left ( -\frac{\hbar^2}{2 m } \right ) \delta t \frac{\partial^2}{\partial q_{t+dt}^2 } ~{}_m \langle_{new}~ q_{t+dt } | \psi(t ) \rangle \right\ } \nonumber \\ % & \simeq & \sqrt { \frac { 2\pi i \hbar \delta t}{m } } { } _ m \langle_{new}~ q_{t+dt } | \exp\left [ -\frac{i}{\hbar } \hat{h_0 } \delta t \right ] |\psi(t ) \rangle , \label{time_dev_qbraaket_k=0_2}\end{aligned}\ ] ] where in the third equality we have inserted @xmath232 and used the following expression , @xmath233 \right)^{*_{q_t , q_{t+dt } } } \nonumber \\ % & = & c_b^ * \exp\left [ \frac{im}{2\hbar \delta t } ( q_t - q_{t+dt})^2 \right ] . \label{comp_conj_exp}\end{aligned}\ ] ] in the fourth equality of eq.([time_dev_qbraaket_k=0_2 ] ) , noting that the exponential converges for @xmath234 , we have used the following relation @xmath235 \simeq \delta_c(x-\beta ) + \frac{1}{4\alpha } \frac{\partial^2}{\partial x^2 } \delta_c(x-\beta ) + \cdots , \label{exp_delta_func}\ ] ] which stands for large positive @xmath236 , and integrated them out with regard to @xmath34 . in the last expression @xmath237 is given by @xmath238 eqs.([time_dev_qbraaket_k=0_2])([h_0 ] ) show that we have obtained the hamiltonian @xmath239 from the lagrangian @xmath39 via the fpi point of view in the free case . next we study the case where the lagrangian includes the potential @xmath175 , which we express as @xmath240 in this case since each component of the taylor - expanded @xmath39 in eq.([taylor_l ] ) is given by @xmath241 eq.([time_dev_qbraaket2 ] ) is calculated as @xmath242 \nonumber \\ % & = & c_a ~{}_m \langle \text{anti } ~\xi | \psi(t ) \rangle % \int_c \exp\left [ \frac{i}{\hbar } ( \xi - q_{t+dt } ) \frac{m}{dt } q_t \right ] d q_t \exp\left [ \frac{i m}{2\hbar dt } ( q_{t+dt}^2 - \xi^2 ) \right ] \nonumber \\ % & & % - \frac{i}{\hbar } dt c_a ~{}_m \langle \text{anti } ~\xi | \psi(t ) \rangle \int_c \sum_{n=2 } b_n q_t^n % \exp\left [ \frac{i}{\hbar } ( \xi - q_{t+dt } ) \frac{m}{dt } q_t \right ] d q_t \exp\left [ \frac{i m}{2\hbar dt } ( q_{t+dt}^2 - \xi^2 ) \right ] . % \label{time_dev_qbraaket_k=0general_a}\end{aligned}\ ] ] in the last expression we note that the first line is the same as eq.([time_dev_qbraaket_k=0 ] ) , and for the second line we shall use the following relation , @xmath243 \frac{m}{dt } q_t^n d q_t = \left ( \frac{\hbar dt}{m } \right)^{n+1 } ( -i)^n 2\pi \frac { \partial^n \delta_c(\xi - q_{t+dt } ) } { \partial ( \xi - q_{t+dt } ) ^n } . \label{general_potential_kzero1}\ ] ] then eq.([time_dev_qbraaket_k=0general_a ] ) is expressed as @xmath244 \nonumber \\ % & & - c_a ~{}_m \langle \text{anti } ~\xi | \psi(t ) \rangle % \sum_{n=2 } \left ( \frac{\hbar dt}{m } \right)^{n+1 } ( -i)^n 2\pi \frac{i dt}{\hbar } b_n \exp\left [ \frac{i m}{2\hbar dt } ( q_{t+dt}^2 - \xi^2 ) \right ] \frac { \partial^n \delta_c(\xi - q_{t+dt } ) } { \partial \xi^n } . \nonumber \\ \label{time_dev_qbraaket_k=0general}\end{aligned}\ ] ] this result shows that @xmath230 is equal to the linear combination of @xmath245 and its derivative . therefore , as shown in the free case , only the component with @xmath170 contributes to the value of @xmath164 . we have thus obtained the momentum relation of eq.([def_p_delldelqdot ] ) in the potential case from the fpi point of view . next we calculate @xmath164 by integrating out eq.([time_dev_qbraaket_k=0general ] ) with regard to @xmath0 . we have already seen that the integration of the first line of eq.([time_dev_qbraaket_k=0general ] ) with regard to @xmath0 gives the last expression of eq.([time_dev_qbraaket_k=0_2 ] ) . the integration of the second line of eq.([time_dev_qbraaket_k=0general ] ) gives @xmath246_{\text{potential part } } \nonumber \\ % & = & - c_a \int_c d\xi ~{}_m \langle \text{anti } ~\xi | \psi(t ) \rangle % \sum_{n=2 } \left ( \frac{\hbar dt}{m } \right)^{n+1 } ( -i)^n 2\pi \frac{i dt}{\hbar } \nonumber \\ % & & \times b_n \exp\left [ \frac{i m}{2\hbar dt } ( q_{t+dt}^2 - \xi^2 ) \right ] \frac { \partial^n \delta_c(\xi - q_{t+dt } ) } { \partial \xi^n } \nonumber \\ % & = & -c_a c_b^ * \frac{i dt}{\hbar } 2\pi \sum_{n=2 } i^n \left ( \frac{\hbar dt}{m } \right)^{n+1 } b_n \nonumber \\ % & & \times % \int_c dq_t \exp\left [ \frac{i m}{2\hbar dt } ( q_t^2 + q_{t+dt}^2 ) \right ] ~{}_m \langle_{new}~ q_t | \psi(t ) \rangle % \int_c d\xi \delta_c ( \xi - q_{t+dt } ) \frac { \partial^n } { \partial \xi^n } \exp\left [ \frac{-i m q_t}{\hbar dt } \xi \right ] \nonumber \\ % & = & - \frac{i dt}{\hbar } \sum_{n=2 } b_n \int_c dq_t q_t^n \exp\left [ \frac{i m}{2\hbar dt } ( q_t - q_{t+dt } ) ^2 \right ] % ~{}_m \langle_{new}~ q_t | \psi(t ) \rangle \nonumber \\ % & \simeq & - \frac{i dt}{\hbar } \sqrt { \frac{2\pi i \hbar dt } { m } } ~{}_m \langle_{new}~ q_{t+dt } | v(\hat{q}_{new } ) | \psi(t ) \rangle , \label{kzero_secondline } % \end{aligned}\ ] ] where in the last relation we have used the following relation , @xmath247 dq_t \simeq \sqrt { \frac{\pi}{a } } q_c^n , \label{qnexpqq - qc2}\ ] ] which stands for large positive @xmath248 . from eqs.([time_dev_qbraaket_k=0_2])([time_dev_qbraaket_k=0general])([kzero_secondline ] ) we obtain the following result , @xmath249 |\psi(t ) \rangle , \label{time_dev_qbraaket_k=0_3}\end{aligned}\ ] ] where @xmath250 is given by @xmath251 therefore also in the potential case we have obtained the hamiltonian @xmath6 from the lagrangian @xmath39 via fpi . in this section we have derived the momentum relation of eq.([def_p_delldelqdot ] ) and the hamiltonian via fpi starting from the lagrangian . the analysis of this section has identified the definitions of the momentum and hamiltonian in the cat , and we have seen that they have the same forms as in the rat . in the next section we make an analysis in the opposite way . we shall attempt to derive the lagrangian starting from the hamiltonian in fpi . in the analysis we shall find again the momentum relation of eq.([def_p_delldelqdot ] ) . in this section starting from the hamiltonian which we obtained in the previous section , we shall derive a lagrangian and see that it coincides with the lagrangian from which we started in the previous section . this study can be performed as done in the rat@xcite , except for the use of the newly introduced devices such as a modified bra . in addition we shall obtain the momentum relation of eq.([def_p_delldelqdot ] ) in a different way from the previous section . we start from the following hamiltonian @xmath252 which we obtained in the previous section . the transition amplitude from an initial state @xmath253 at time @xmath254 to a final state @xmath255 at time @xmath256 is written as @xmath257 where we have divided the time interval @xmath258 into @xmath259 pieces whose interval is @xmath260 using this @xmath261 , we define @xmath262 as @xmath263 to evaluate eq.([braqbqaket ] ) we calculate @xmath264 considering the coordinate ordering implicitly as in the rat as follows , @xmath265 , \label{braqj+1qjket}\end{aligned}\ ] ] where we have used eq.([def_dotq_j ] ) and the following relations , @xmath266 , \\ % & & { } _ m \langle_{new}~ p_j | q_{j } \rangle_{new}~ % = \left ( { } _ m \langle_{new}~ q_{j-1 } | p_j \rangle_{new}~ \right)^{*_{q_{j-1},p_j } } = \frac{1 } { \sqrt { 2\pi\hbar } } \exp\left [ -\frac{i}{\hbar } p_j q_{j } \right ] . \end{aligned}\ ] ] in the last line of eq.([braqj+1qjket ] ) @xmath267 is given by @xmath268 by substituting eq.([braqj+1qjket ] ) for eq.([braqbqaket ] ) @xmath269 is calculated as @xmath270 \nonumber \\ % & = & \int_c { \cal d } p { \cal d } q~ \psi_f(q_f)^{*_{q_f } } \psi_i(q_i ) \exp\left [ \frac{i}{\hbar } \int_{t_i}^{t_f } dt l(p , q , \dot{q } ) \right ] , \end{aligned}\ ] ] where in the second equality we have introduced @xmath271 and @xmath272 . also , we have taken the large @xmath273 limit and introduced @xmath28 and @xmath274 by @xmath275 the integral with regard to @xmath276 is performed as follows , @xmath277 % & = & \int_c \frac{dp_j}{2\pi\hbar } \exp\left [ \frac{i}{\hbar } \delta t \left\ { -\frac{1}{2 m } \left ( p_j - m \dot{q}_j \right)^2 + \frac { 1}{2 } m \dot{q_j}^2-v(q_j ) \right\ } \right ] \nonumber \\ % & = & \sqrt { \frac{m}{2\pi i \hbar \delta t } } \exp\left [ \frac{i}{\hbar } \delta t l(\dot{q}_j , q_j ) \right ] , \end{aligned}\ ] ] where in the last equality based on the condition ( [ m_i_positive_cond ] ) we have performed the gaussian integral around the saddle point for @xmath276 , @xmath278 and in the last expression @xmath279 is given by @xmath280 from eqs.([pjmqj])([lagrangian_dotq_q ] ) we obtain the momentum relation ( [ def_p_delldelqdot ] ) . thus we have obtained @xmath281 , % \end{aligned}\ ] ] where @xmath282 in this section we have derived the lagrangian ( [ lagrangian_dotq_q ] ) starting from the hamiltonian , and have seen that the lagrangian has the same form as that in the previous section . we have thus confirmed that the hamiltonian in the cat has the same form as that in the rat , and justified the classical construction of the hamiltonian @xmath283 via fpi . also , we have obtained the momentum relation ( [ def_p_delldelqdot ] ) via the saddle point for @xmath1 . this is the second derivation of the momentum relation , which is different from that in the previous section . we present the third derivation of the momentum relation in appendix [ via_saddle_q ] . before ending this section , we discuss another possibility to define our hamiltonian . the hamiltonian @xmath6 we have derived is not hermitian , so we can decompose it as @xmath284 where @xmath285 and @xmath286 are the hermitian and anti - hermitian part of @xmath6 respectively , and they are defined by @xmath287 if we adopt @xmath285 as our hamiltonian , then we can define a unitary time - development operator @xmath288 satisfying @xmath289 , while we have @xmath290 . if we define a hermitian lagrangian @xmath291 , we obtain the relation @xmath292 , which has a similar form as a usual one . but using @xmath293 means that we lose the complexity of the action , so this does not match our motivation to study the effect of the complex action . in addition , if @xmath6 gives time - development , @xmath294\neq 0 $ ] . hence the option to use both of @xmath6 and @xmath285 does not seem to work well . therefore we adopt the hamiltonian @xmath6 as our hamiltonian . in other words we prefer the relation @xmath295 to the hermiticity of the hamiltonian . in the complex action theory ( cat ) even a coordinate @xmath2 and a momentum @xmath1 can be complex in general . in ref . , to handle the complex @xmath2 and @xmath1 , we have proposed the replacement of hermitian operators of coordinate and momentum @xmath7 and @xmath8 and their eigenstates @xmath43 and @xmath10 with non - hermitian operators @xmath11 and @xmath12 , and @xmath13 and @xmath14 with complex eigenvalues @xmath2 and @xmath1 . we have introduced a philosophy of keeping the analyticity in parameter variables of feynman path integral ( fpi ) , and defined modified bras @xmath73 and @xmath296 , a modified complex conjugate @xmath60 , and a modified hermitian conjugate @xmath297 to realize the philosophy , by means of which we have explicitly constructed @xmath11 , @xmath12 , and the hermitian conjugates of their eigenstates @xmath41 and @xmath42 by utilizing the coherent states of harmonic oscillators . here @xmath298 denotes a set of parameters which we keep the analyticity in . in this paper , based on the complex coordinate formalism of ref . , we have explicitly examined how the momentum and hamiltonian are defined in the cat from the fpi point of view . in section 2 we have briefly reviewed the complex coordinate formalism of ref . . first we have explained that the delta function can be used also for complex parameters when it satisfies some condition , and introduced the new devices to realize the philosophy of keeping the analyticity in parameter variables of fpi . the philosophy allows us to deform the path of fpi . then we have briefly shown the construction of non - hermitian operators @xmath11 and @xmath12 , and the hermitian conjugates of their eigenstates @xmath41 and @xmath42 with complex eigenvalues @xmath2 and @xmath1 by utilizing the coherent states of harmonic oscillators . in section [ derivation_hamiltonian ] based on the philosophy and using the new devices explained in section [ fundamental ] we have described in fpi with a lagrangian the time development of a @xmath0-parametrized wave function , which is a solution to an eigenvalue problem of a momentum operator . solving the eigenvalue problem and considering the time - development of the wave function solution in fpi we have derived the momentum relation @xmath17 . furthermore we have also studied the form of the hamiltonian in the cat . we have derived the hamiltonian starting from the lagrangian in fpi , first in a free theory case , and then in a potential case . in section [ derivation_lagrangian ] , oppositely starting from the hamiltonian , we have derived the lagrangian in a potential case via fpi , and also the momentum relation again via the saddle point for @xmath1 . these explicit studies have confirmed that both of the momentum and hamiltonian in the cat have the same forms as those in the real action theory ( rat ) . in appendix [ via_saddle_q ] we have explained the third derivation of the momentum relation via the saddle point for @xmath2 . now the momentum and hamiltonian in the cat have been identified clearly , based on the complex coordinate formalism of ref . . as a next step what should we study to develop the cat ? one direction is to study a future - included theory , that is to say , a theory including not only a past time but also a future time as an integration interval of time , which we have not considered in this paper , in the complex coordinate formalism . it is interesting to study such a future - included theory by considering a kind of wave function of universe including the future information , which was discussed in ref . . in addition , it is also important to study an expectation value in the cat . we shall attempt to discuss how the expectation value is defined in the cat generally , and also clarify its relation to the modified hermiticity . furthermore , it is also intriguing to investigate in detail the possible misestimation of the past state by extrapolating back in time with the hermitian hamiltonian . as pointed out in ref . , the misestimation by a historian living in the late time can occur in some fundamental theory which is described by a non - hermitian hamiltonian . we will study them and report the progress in the future . this work was partially funded by danish natural science research council ( fnu , denmark ) , and the work of one of the authors ( k.n . ) was supported in part by grant - in - aid for scientific research ( nos.18740127 and 21740157 ) from the ministry of education , culture , sports , science and technology ( mext , japan ) . k.n . would like to thank all the members and visitors of nbi for their kind hospitality . in this section we give the third derivation of the momentum relation of eq.([def_p_delldelqdot ] ) . taking a momentum eigenstate @xmath299 where @xmath2 and @xmath1 are in general complex , we study the following path integral for an infinitesimal time from @xmath37 to @xmath26 , @xmath300 where @xmath16 is given by eq.([dot_q ] ) . then performing the taylor - expansion of the lagrangian @xmath15 around @xmath301 as in eq.([taylor_l ] ) , we have @xmath302 % dq_t dq_{t+dt } , % \end{aligned}\ ] ] where we note that the gaussian function is very narrow like the delta function since @xmath146 is assumed to be a very small quantity . the saddle point @xmath303 is given by @xmath304 and we have the relation @xmath305 where in the last equality we have used the taylor - expansion of @xmath306 , @xmath307 from eq.([qsaddle_momentum ] ) we have thus derived the momentum relation of eq.([def_p_delldelqdot ] ) by seeing the dominant saddle point in @xmath2 . 0 h. b. nielsen and m. ninomiya , int . j. mod . phys . a * 21 * ( 2006 ) 5151 . h. b. nielsen and m. ninomiya , arxiv : hep - th/0601048 . h. b. nielsen and m. ninomiya , arxiv : hep - th/0602186 . h. b. nielsen and m. ninomiya , int . j. mod . a * 22 * ( 2008 ) 6227 .

the study of codes for powerline communications has garnered much interest over the past decade . various types of codes such as permutation codes , frequency permutation arrays , and constant composition codes have been proposed over the years . in this work we study a type of code called the bounded symbol weight codes which was first introduced by versfeld in 2005 , and a related family of codes that we term constant symbol weight codes . we provide new upper and lower bounds on the size of bounded symbol weight and constant symbol weight codes . we also give direct and recursive constructions of codes for certain parameters . asymptotic bounds , constant composition codes , powerline communications , reed solomon codes , symbol weight codes .

1110.0911

the notion of transmitting data over powerlines has posed an interesting challenge for information and coding theory . the noise characteristics of such a communication channel include permanent narrowband noise , impulse noise and white gaussian noise . communication over this channel also has an additional requirement that the power envelope be as close to constant as possible . vinck @xcite studied this channel and showed that @xmath0-ary frequency shift keying ( @xmath0-fsk ) modulation , in conjunction with the use of permutation codes , provides a constant power envelope , frequency spreading and redundancy to correct errors resulting from the harsh noise pattern . this has since resulted in research on frequency permutation arrays ( fpas ) and constant composition codes ( cccs ) which retain the property of a constant power envelope ( see @xcite , and @xcite for a survey ) . every codeword of an fpa or a ccc has the requirement that the frequency of each symbol is fixed by the parameters of the code . versfeld@xcite introduced the notion of the `` same - symbol weight '' of a code by relaxing the requirement that every symbol must occur a fixed number of times in any codeword . in every codeword of a same - symbol weight code , the frequency of any symbol is bounded . even with this relaxation it is possible to detect permanent narrowband noise . versfeld@xcite used reed - solomon codes to design codes with specified same - symbol weight . in this work we mostly study the asymptotic behavior of codes in the symbol weight space . we use the term _ bounded symbol weight _ ( as opposed to same - symbol weight @xcite ) to denote all words in the hamming space with bounded symbol weight , that is , any symbol in a codeword does not occur more than a fixed number of times , say @xmath1 this terminology is adopted in order to distinguish this space from the _ constant symbol weight space _ , in which every symbol in a word in the hamming space occurs at most @xmath2 times and there exists one symbol which occurs exactly @xmath2 times . the constant symbol weight space is clearly a subset of the bounded symbol weight space . we also use the term symbol weight space to refer to either the bounded symbol weight or constant symbol weight space . the actual space being referred to is made clear from the context and notation . as described in @xcite , the symbol weight determines whether the code can detect and correct narrowband noise in the powerline channel . an fpa or a ccc belongs to some constant symbol weight space . the constant symbol weight space also contains other compositions all of which have the same maximal part , that is , all such codes have the same fixed symbol weight . thus a code in the constant symbol weight space is larger than a ccc of a fixed composition , and is still relevant for correcting narrowband noise . the asymptotic behavior of fpas have been studied in @xcite . in contrast , there are relatively fewer results on the asymptotic behavior of cccs ( see @xcite ) . we consider familiar techniques used to derive classical bounds such as the gv bound , the johnson bound and the singleton bound , on codes in the symbol weight space . however , the derivation of these results are not immediate because of the lack of any reasonable structure in the symbol weight spaces . in particular , even the hamming balls of a fixed radius in these spaces depend on the center of the ball . in section [ sec : lower - bnd - sw ] , we also study non - asymptotic bounds on codes in the symbol weight spaces by expressing them in terms of different cccs . this also raises related combinatorial questions regarding the size and construction of optimal codes , which can be an interesting avenue of future research . in later sections we show that there exists codes , which are subsets of reed - solomon codes , with high rate and relative distance , which are a subset of the constant symbol weight space . throughout this work we are mostly concerned with codes that have positive rate and positive relative distance . hence we do not study codes with very large distances , in the plotkin region . we start with some basic definitions and notations in the following section . we devote section [ sec : asymp - sws ] to deriving the exact and asymptotic size of the symbol weight spaces . the results in section [ sec : asymp - sws ] allow us to determine which constant composition space contained within the symbol weight space contributes the most to the rate of a symbol weight space . these estimates are used in section [ sec : sw - asymp ] to determine upper and lower bounds on bounded symbol weight and constant symbol weight codes . in particular , it is clear that asymptotically some constant composition code determines the rate of a symbol weight code . an upper bound is readily obtained from either the singleton bound or the linear programming bound in the hamming space . in section [ sec : sw - asymp ] we also provide a johnson - type bound on codes in the constant symbol weight space , and use this bound to derive an asymptotic improvement of the singleton bound and the linear programming bound for certain ranges of the minimum distances and the symbol weight . in section [ sec : lower - bnd - sw ] we provide non - asymptotic lower bounds on symbol weight codes . we introduce a new metric on the space of compositions of an integer and use this metric to lower bound the size of symbol weight codes by a sum of sizes of cccs . finally , in section [ sec : constructions ] we provide other constructions of constant symbol weight codes , and in particular show that the asymptotic lower bound presented in section [ sec : asymp - sws ] is tight for certain parameters , for subcodes of reed - solomon codes . let @xmath3 denote an alphabet set of @xmath4 elements . we consider symbol weight codes in the hamming space @xmath5 the _ symbol weight _ of a word is defined as the maximum of the frequencies of occurrences of symbols in the word . for instance , the all-0 word has a symbol weight of @xmath6 the _ bounded symbol weight space _ with symbol weight @xmath2 is the set of all words with symbol weight at most @xmath1 this space is denoted by @xmath7 . the bounded symbol weight space is termed as `` same - symbol weight space '' in the works of versfeld@xcite . we adopt this terminology to distinguish this space from the constant symbol weight space that we define next . in the _ constant symbol weight space _ every word has a symbol weight of exactly @xmath1 this space is denoted as @xmath8 . if every symbol occurs in each codeword we can use the pigeonhole principle to get the lower bound @xmath9 . since any word with this lowest value of symbol weight contains the least repetition of any symbol , these words are considered as ones with the _ optimal _ symbol weight . in this work we study codes in the bounded and constant symbol weight spaces . a bounded ( resp . constant ) symbol weight code is a subset of the bounded ( resp . constant ) symbol weight space . let @xmath10 ( resp . @xmath11 ) denote the maximum size of a bounded ( resp . constant ) symbol weight code with distance @xmath12 in @xmath7 ( resp . @xmath13 ) . we denote a composition of @xmath14 into @xmath4 non - negative parts by @xmath15 $ ] . the constant composition space with composition @xmath15 $ ] is a subset of @xmath16 in every word of which the @xmath17-th symbol occurs exactly @xmath18 times . a constant composition code is a subset of a constant composition space . we use the notation @xmath19 to denote the maximum size of a code in the constant composition space given by the composition @xmath20 and minimum distance at least @xmath12 . we use the notation @xmath21 to denote the maximum size of a code with minimum distance at least @xmath12 in the hamming space . a code @xmath22 of length @xmath23 size @xmath24 distance @xmath25 over @xmath26 is denoted by @xmath27 . if @xmath22 has a constant symbol weight @xmath2 it is denoted by @xmath28 . if @xmath22 is a linear code of dimension @xmath29 over a field @xmath30 it is denoted as @xmath31_q$ ] . an fpa consists of vectors in which every symbol occurs a fixed number , say @xmath32 , of times . hence , an fpa is a ccc with composition @xmath33 $ ] . thus the fpa is a subset of the constant symbol weight space with symbol weight @xmath34 similarly , it can be seen that a ccc with composition @xmath35 $ ] is a subset of the constant symbol weight space with symbol weight @xmath36 . the coded modulation scheme introduced by vinck @xcite for the powerline channel considered @xmath0-fsk modulation along with the use of permutation codes . the demodulator considered is a hard - decision demodulator consisting of an envelope detector with a threshold . at every time instance , the demodulator provides a multivalued output consisting of all the symbols that correspond to frequencies at which the output of the envelope detector exceeds the threshold . a narrowband noise in this context results in the same symbol appearing at all time instances . as explained in versfeld@xcite , a linear code is less effective in this channel . for instance , the all - zero codeword can not be distinguished from a narrowband noise . hence , permutation codes , fpas and cccs are more suitable for communication in this channel . to understand why we study the constant symbol weight space , consider the following example . + consider a ccc in @xmath37 with composition @xmath38 $ ] and minimum distance @xmath39 , that is suitable for correcting narrowband noise in a powerline channel . let @xmath40 be a codeword in this ccc . then the vector @xmath41 is also a suitable vector for correcting narrowband noise . however this vector belongs to a different constant composition space that contains vectors with composition @xmath42 $ ] . both these constant composition spaces are a subset of the constant symbol weight space with symbol weight 3 . in section [ sec : lower - bnd - sw ] , we prove that since the compositions @xmath38 $ ] and @xmath42 $ ] have distance @xmath43 in a specific metric that we define later , any vector from the constant composition space with composition @xmath38 $ ] will be at a distance at least 4 from any vector of the other space with composition @xmath42 $ ] . hence , we can increase the size of the code by including all the codewords from a ccc in the latter space . + it is clear that any constant symbol weight space with symbol weight @xmath2 can be written as the union of different constant composition spaces , each of which contains vectors with the same symbol weight @xmath2 . this relation to the constant composition space is used throughout this work . in the next section , we first determine the size of the constant symbol weight space and the bounded symbol weight space . this size is then used to determine a gv - type bound on the symbol weight spaces . unfortunately , the expression for the size of the symbol weight spaces is unwieldy and gives little insight into the behavior of the lower bounds . the symbol weight spaces are also not ball - homogeneous , that is , the size of a hamming ball in the space depends on the center of the ball . for example , the bounded symbol weight space in @xmath44 with symbol weight @xmath2 at most 2 has 24 vectors . the ball of radius 1 around the vector @xmath45 contains 6 vectors , namely , @xmath46 @xmath47 @xmath48 , @xmath49 @xmath50 @xmath51 . in contrast , the ball of radius 1 around @xmath52 contains 7 vectors , namely , @xmath53 , @xmath54 . this fact makes it difficult to state decent lower bounds . similar comments apply to the computation of the hamming bound . hence , in the following two sections , we instead study the asymptotic behavior of the symbol weight spaces and the rate of the corresponding symbol weight codes . in the following section we determine the asymptotic size of the symbol weight space . this enables us to determine which constant composition space , contained within the symbol weight space , has the largest size . to determine the asymptotic size of the symbol weight space we first state the expression for the non - asymptotic case . each vector in the symbol weight space corresponds to a vector in some constant composition space . hence we introduce some basic definitions below to describe the size of the symbol weight space . let us denote the set of all compositions of @xmath14 into @xmath4 non - negative parts by @xmath55 , that is @xmath56 and define @xmath57 let @xmath58 denote the compositions of @xmath59 into @xmath60 parts , each part bounded between @xmath61 and @xmath62 an expression for the size of @xmath58 is given by @xcite , @xmath63 to be @xmath64 the quantity @xmath65 corresponds to the smallest number of symbols that can occur with frequency exactly @xmath2 in any vector with symbol weight @xmath2 . the size of the set @xmath66 is given by the following lemma . + [ lem : size - cnr ] @xmath67 for some @xmath68 such that @xmath69 . if a vector @xmath70 has a composition @xmath20 such that exactly @xmath29 of the symbols in @xmath26 have composition @xmath2 in @xmath71 , then the rest of the symbols must satisfy the inequality , @xmath72 this inequality , in conjunction with the requirement that at least one symbol must have composition @xmath2 , determines the value of @xmath73 if a composition @xmath20 has exactly @xmath29 symbols with the value @xmath2 , then these @xmath29 symbols can be chosen in @xmath74 ways . the rest of the elements of @xmath20 must correspond to a composition of @xmath75 into @xmath76 parts , each part bounded between @xmath61 and @xmath77 note that @xmath2 must satisfy @xmath78 we have @xmath79 there is exactly one integer @xmath2 which satisfies @xmath80 . this value of @xmath2 is given by @xmath81 , for some @xmath82 such that @xmath83 . the size of the constant symbol weight space @xmath8 can now be determined to be @xmath84 where @xmath85 , and @xmath2 is repeated @xmath29 times in the multinomial coefficient @xmath86 . the size of the bounded symbol weight space is a sum of the sizes of the different constant symbol weight spaces , as shown below : @xmath87 where @xmath85 , @xmath88 , and @xmath89 . the expressions in the equations above can be used to provide gv - type existence bounds on symbol weight codes . a gv bound on the size of a code @xmath90 with minimum distance @xmath12 in a space @xmath91 is given as @xmath92 where @xmath93 is the volume of a ball of radius @xmath94 in the space @xmath91 . although the sizes of the constant and bounded symbol weight spaces are given by the above equations and , respectively , there are several hurdles in applying the gv - type bound directly . first the space itself lacks any suitable structure and is not even ball - homogeneous . even for the special case of an fpa in which all the symbols occur equally often in every vector , the expression for the gv ( and also the hamming bound ) is quite unwieldy because the size of the ball does not have a nice form ; see huczynska ( * ? ? ? * theorem 2.7 ) . secondly , the expressions for the sizes of the spaces are not suitable for expressing the bound in a simple form . we instead study the asymptotic form of this bound in the next section . to determine the asymptotic results , we first need to understand the behavior of the sizes of the symbol weight spaces for large block length @xmath6 the expression for the asymptotic size of the constant symbol weight space is given by the following theorem . a similar expression for the bounded symbol weight space can be readily derived from this theorem , and is mentioned at the end of this section . the following theorem holds for any @xmath4 such that @xmath4 grows at most proportional to @xmath6 note that all the asymptotics are with respect to @xmath14 and so the term @xmath95 below goes to zero as @xmath14 goes to @xmath96 . + [ thm : sw - asymp ] for any @xmath97 such that @xmath98 where @xmath99 is a positive constant and @xmath100 , @xmath101 we first give a brief outline of the proof of this theorem . as mentioned earlier , a constant symbol weight space with symbol weight @xmath2 is a union of several constant composition spaces , each of which contains vectors of symbol weight @xmath2 . we first show in lemma [ lem : nr - size ] that the number of constant composition spaces does not contribute to the rate of the constant symbol weight space . this is not surprising and it is true even for the hamming space , when considered as a union of constant composition spaces . because of this lemma , we now know that there is a constant composition space which dominates the expression for the rate . lemma [ lem : multinomial ] and [ lem : monotone ] below help us determine this dominant term . it turns out that this dominant term comes from the constant composition space which has exactly @xmath65 symbols that occur exactly @xmath2 times in any vector . we continue with the proof of the theorem , by first establishing a sequence of lemmas presented below . let @xmath102 be the @xmath103-ary entropy function defined in the range @xmath104 , as @xmath105 [ lem : nr - size ] @xmath106 the number of terms in the summation over the range @xmath107 is at most @xmath6 hence only one of the terms in the summation dominates in the asymptotics . we note that @xmath108 . also , @xmath109 ( see @xcite ) . for a constant @xmath4 , it shows that @xmath110 grows at most polynomially in @xmath14 and hence @xmath111 for some positive constant @xmath112 . for @xmath113 , and positive constant @xmath99 , we get @xmath114and hence @xmath115 by the above lemma , we can conclude that in the asymptotics of ( [ eq : size - swc ] ) only one term @xmath116 from the inner summation dominates in the asymptotics , and similarly only one term from @xmath117 is present in the asymptotics . the dominant multinomial terms are given by an optimal choice of @xmath118 first , we determine the dominating multinomial term from the inner summation in ( [ eq : size - swc ] ) . we use the following lemma . let @xmath119 denote the gamma function , @xmath120 in particular , for an integer @xmath121 , @xmath122 . + @xcite [ lem : multinomial ] let @xmath123 be non - negative real numbers . then @xmath124 this lemma immediately implies that @xmath125 hence the dominating term in the inner summation in ( [ eq : size - swc ] ) is given by @xmath126 where @xmath127 for @xmath128 times and @xmath129 for @xmath130 times , where @xmath131 denotes the fractional part of a real number @xmath121 . ] for large @xmath14 we obtain the following asymptotic expression for @xmath132 : @xmath133 neglecting the @xmath95 term , the maximum of the expression in equation over @xmath29 yields the rate of the constant symbol weight space . unfortunately , a closed - form expression for the optimizing value of @xmath29 seems difficult to achieve , even if @xmath29 is considered over reals instead of integers . we instead look at how the expression behaves for large @xmath6 the lemma below asserts that the maximum is achieved at @xmath134 as @xmath135 + [ lem : monotone ] let @xmath136 be a family of bounded , strictly monotonic decreasing functions in @xmath121 , defined over the domain @xmath137 $ ] , such that @xmath138 . let @xmath139 be a family of non - negative functions such that @xmath140 where @xmath141 depends only on @xmath14 and @xmath142 . then , @xmath143 } \lim_{n\to\infty } f_n(x)+g_n(x ) = f(x_0)\\ = \lim_{n\to\infty } \max_{x\in [ x_0,x_0 ] } f_n(x ) + g_n(x).\end{gathered}\ ] ] @xmath144_proof : _ the strict monotonicity @xmath145 for any @xmath146 @xmath147 , implies that @xmath148 for all @xmath149 $ ] . now , @xmath150 } f_n(x ) + c_n & \ge \max_{x\in [ x_0,x_0 ] } f_n(x ) + g_n(x ) \ge \max_{x } f_n(x ) \\ \rightarrow f_n(x_0 ) + c_n & \ge \max_{x\in [ x_0,x_0 ] } f_n(x ) + g_n(x ) \ge f_n(x_0 ) \\ \rightarrow & \lim_{n\to\infty } \max_{x\in [ x_0,x_0 ] } f_n(x ) + g_n(x ) = f(x_0 ) . \end{aligned}\ ] ] we also get @xmath151 } \lim_{n\to\infty } f_n(x ) + g_n(x ) = \max_{x\in [ x_0,x_0 ] } f(x ) = f(x_0).\quad\qed\ ] ] this lemma implies that we can determine the asymptotic optimum of @xmath152 by simply taking the limit of the sequence of numbers @xmath153 , which converges to @xmath154 . + we apply lemma [ lem : monotone ] as follows . let @xmath155 be defined over integer @xmath156 $ ] . @xmath157 can be upper bounded by a term independent of @xmath158 @xmath159 and @xmath160 we now note that for every @xmath161 , @xmath162 is strictly monotonically decreasing . to establish this we relax @xmath29 to reals and consider the derivative @xmath163 . we get @xmath164 where the last line follows because of the fact that @xmath165 , and that @xmath166 for @xmath167 , with equality at @xmath168 . we also note that @xmath169 if and only if @xmath170 hence @xmath162 is strictly monotonic decreasing for @xmath170 for @xmath171 , @xmath162 is a constant independent of @xmath29 and @xmath172 , and hence the optimal value of @xmath162 is at @xmath173 . since lemma [ lem : monotone ] is applicable to @xmath174 , we concentrate only on determining the asymptotics of @xmath175 . for @xmath176 we get @xmath177 and @xmath178 for @xmath179 , we have @xmath180 , and @xmath181 this proves theorem [ thm : sw - asymp ] . the exponent of the asymptotic size of the bounded symbol weight space @xmath182 is always @xmath183 since it contains @xmath184 . in this section we provide estimates on the rate of symbol weight codes for all @xmath185 , for any positive constant @xmath186 and for @xmath100 . we considered the asymptotic behavior of the symbol weight spaces because of the difficulty in determining reasonable expressions for fixed @xmath14 . below , we determine upper and lower bounds on the rate of a symbol weight code . first we determine a gv - type bound in theorem [ thm : aqw - rate - q - n - r - const ] below . the singleton and linear programming ( lp ) upper bounds on codes in the hamming space are applicable to the symbol weight codes too . in theorem [ thm : large - r ] , we show that for constant symbol weight codes , the singleton and lp upper bounds can be improved substantially for a specific range of the symbol weight . the following lemma is immediate and it shows that the rate of symbol weight codes can be given in terms of the rate of a ccc . + [ lem : aqw - rate ] @xmath187 note that we clearly have the following upper and lower bounds on @xmath188 : @xmath189 the lemma now follows from an application of lemma [ lem : nr - size ] . the second expression in ( [ eq : aqw - rate ] ) can be determined similarly . we state the lp upper bound on codes in the hamming space from aaltonen @xcite . + @xcite [ thm : aaltonen - lp ] @xmath190 where @xmath191 an upper bound on symbol weight codes is readily obtained by an upper bound on codes in the hamming space , since @xmath192 thus for constant @xmath4 the lp bound is also an upper bound on symbol weight codes . for @xmath4 growing with @xmath14 , the singleton bound is an upper bound on symbol weight codes . below , we provide asymptotic estimates of symbol weight codes . + [ thm : aqw - rate - q - n - r - const ] let @xmath193 , where @xmath194 is a constant , and @xmath100 . let @xmath195 and @xmath196 as @xmath197 , where @xmath198 . then for @xmath4 constant , i.e. , @xmath199 , @xmath200 for @xmath4 increasing with @xmath14 one can use the singleton bound . thus for @xmath201 , we get @xmath202 * remark : * note that for @xmath4 increasing with @xmath14 the following limits can be inferred , @xmath203 we use the following lower bound on the constant symbol weight space , which is actually an elias - type bound on the hamming space ( see levenshtein @xcite ) . this is followed by using the gv bound in the hamming space . @xmath204 where @xmath205 is the @xmath4-ary entropy function and @xmath206 is the volume of the ball of radius @xmath94 in the hamming space . similarly , for the bounded symbol weight space , we obtain @xmath207 for a constant @xmath4 , the asymptotics of these expressions are as given in . for @xmath4 growing with @xmath14 , the upper bound on the symbol weight codes is provided by the singleton bound , @xmath208 using the fact that @xmath209 in the limit as @xmath210 , we get the results as stated in the theorem . in particular for @xmath211 , @xmath212 and @xmath213 . since @xmath10 is greater than @xmath214 , it gives the result stated in ( [ eq : aqw - rate - bd - r ] ) . the lower bound in the theorem may be interpreted as a gv - type bound in the symbol weight space that can be obtained if the volume of a ball of radius @xmath94 in the symbol weight space is upper bounded by the volume of a ball of radius @xmath94 in the hamming space . since , the symbol weight space is not ball - homogeneous , that is , the size of the balls of radius @xmath94 depends on the center , we adopt the above method to derive the gv - type lower bound . , for instance from algebraic geometry codes , can improve on this gv bound on @xmath188 . ] in the following theorem we provide an improvement on the upper bound for a _ constant _ symbol weight code with symbol weight @xmath2 . + [ thm : large - r ] let @xmath215 with @xmath216 let @xmath12 satisfy @xmath217 for @xmath218 let @xmath219 and @xmath220 then , for constant @xmath4 , @xmath221 for @xmath4 growing with @xmath14 , we get @xmath222 the proof of this theorem relies on a johnson - type upper bound , and a lemma given below . we follow some elements of the derivation of the singleton bound in omrani and kumar @xcite . however , our purpose is to improve the singleton bound by using the parameters of the constant symbol weight space . the improvement mainly stems from the following lemma . + [ lem : upper - q ] let @xmath223 then @xmath224 we claim that if there are two codewords @xmath225 both with symbol weight @xmath2 , then the symbol which repeats @xmath2 times must be different in the two codewords . suppose not . then the two codewords @xmath226 must have at least @xmath227 coordinates which contain the same symbol . thus the distance between the codewords is at most @xmath228 which implies @xmath229 , since @xmath230 . this is a contradiction . we get @xmath231 to show the opposite inequality , let @xmath232 be a word with symbol weight @xmath233 then @xmath234 , where @xmath235 is the all - one codeword , are also codewords with symbol weight @xmath1 this establishes that @xmath236 . we next give the johnson - type bound . [ lem : johnson ] @xmath237 consider the code - matrix of the constant symbol weight code with parameters @xmath238 . any row of the code - matrix has at least one symbol of frequency @xmath1 fix one symbol of frequency @xmath2 in each row . there are a total @xmath239 symbols in the code - matrix which do not contribute to the symbol weight in any codeword . the average number of symbols , averaged over the @xmath14 columns , with frequency at most @xmath2 is then @xmath240 the average number per symbol , averaged over @xmath14 columns and @xmath4 symbols is @xmath241 . thus , there exists at least one symbol @xmath242 and at least one column @xmath243 such that the subcode consisting of the symbol @xmath242 in column @xmath243 has size at least @xmath244 discarding the coordinate corresponding to @xmath243 gives us the bound as stated in the lemma . we now proceed to prove the theorem . apply lemma [ lem : johnson ] recursively @xmath245 times to get @xmath246 the recursion stops for @xmath247 such that @xmath248 and for @xmath249 . for this value of @xmath2 and @xmath12 , @xmath250 by lemma [ lem : upper - q ] . the condition @xmath251 implies @xmath252 . constraints on @xmath247 are obtained from the inequalities @xmath253 this gives us the upper bound @xmath254 in the asymptotics as @xmath197 we get @xmath255 . this gives us the upper bounds as stated in the theorem . in the case of @xmath4 growing with @xmath14 , this theorem improves on the singleton bound @xmath256 for @xmath257 the upper bound in theorem [ thm : large - r ] for constant @xmath4 improves on the lp bound for certain range of parameters . the improvements are possible only for large @xmath258 and for @xmath259 for @xmath260 the restrictions @xmath261 and @xmath262 do not leave room for improvement . for constant @xmath4 the upper bound is in fact concave in shape . this can be verified by taking the second derivative with respect to @xmath258 , which results in the negative expression @xmath263 since this bound does not depend on @xmath264 , it seems that further improvements might be possible . an example plot of all the bounds are provided in figures [ fig : constant - q ] and [ fig : variable - q ] . since the improvements are for larger @xmath4 , we show the bounds for @xmath265 in figure [ fig : constant - q ] , the first plot is obtained at a particular value of @xmath264 and the second plot is obtained at a particular value of @xmath258 . the improvements ( over lp ) are obtained in the regions @xmath266 and @xmath267 , respectively . figure [ fig : variable - q ] shows the plots when @xmath4 is increasing with @xmath14 . in this case , we compare against the singleton upper bound . it shows improvements in the region @xmath268 construction of codes which meet this upper bound for any parameters is an open problem . + * remarks : * 1 . for @xmath269 and @xmath270 , dukes @xcite provides a singleton bound , @xmath271 not surprisingly , for @xmath272 this reduces to @xmath256 in the asymptotics . 2 . missing from the list of bounds above is a hamming - type bound on the symbol weight codes . the lack of a simple expression for the size of the ball is the main reason behind this omission . as mentioned in the previous sections , the traditional means of determining the gv - type bounds is not very useful for non - asymptotic block lengths @xmath6 in this section we adopt a different approach to determine lower bounds on the size of symbol weight codes . the lower bounds are obtained by using corresponding constructions and lower bounds on constant composition codes . there is a wide body of literature on cccs that can be used to determine these lower bounds ; see @xcite . to the best of our knowledge , most of the work in the literature on constant composition codes has focused on determining constructions and bounds for either very large distances ( in the region where the johnson bound or plotkin bound is applicable ) or for very small distances such as @xmath273 for fpa and permutation codes , there do exist constructions with distances in the ranges in between ; see @xcite . we note three papers in this connection . sidorenko @xcite provides an asymptotic upper bound on cccs ; this is not useful for this section since we look at lower bounds . in a very recent work , luo and helleseth @xcite construct cccs of almost uniform composition and with relative distances very close to the plotkin limit ( @xmath274 ) . the bound that we provide below requires a relatively large alphabet size so that codes from different constant composition spaces can be combined . hence , the cccs from @xcite are not useful in this context . chu@xcite provide some lower bounds for cccs for large distances . we use some of the constructions from this latter work in this section to provide examples of lower bounds on symbol weight codes . we first describe a method to determine the size of a symbol weight code in terms of cccs . this method may be viewed as a generalization of an elementary bound in bachoc ( * ? ? ? * equation ( 2 ) ) on binary bounded weight codes to @xmath4-ary spaces . let @xmath35 $ ] and @xmath275 $ ] denote two different compositions of @xmath6 the aim here is to lower bound the size of a symbol weight code by the sum of all possible different cccs which have the same symbol weight . thus , we first need to determine the condition on two different compositions @xmath20 and @xmath276 such that any vector @xmath277 with composition @xmath20 is at least distance @xmath12 away from a vector @xmath278 with composition @xmath279 it can be seen that @xmath280 is the maximum number of coordinates in @xmath277 and @xmath278 where the @xmath17-th symbol is common to both . thus , the hamming distance @xmath281 satisfies @xmath282 a sufficient condition for @xmath283 to hold is @xmath284 let @xmath285 then we obtain + @xmath286 is a distance function on @xmath287 @xmath288 is clearly symmetric . to show the triangle inequality , we note that we can rewrite @xmath289 where @xmath290 . also , for any non - negative real numbers @xmath291 , it can be readily verified that @xmath292 thus , we get @xmath293 since @xmath294 , we get that @xmath295 if and only if @xmath296 for all @xmath297 let @xmath298 ( resp . @xmath299 ) be such that for any distinct @xmath300 ( resp . @xmath301 ) we have @xmath302 we can now readily give a lower bound on the size of symbol weight codes in terms of the cccs : @xmath303 for large @xmath14 and @xmath4 the size of the set @xmath66 becomes very large . hence finding all the compositions in @xmath66 which are separated by distance at least @xmath12 is difficult . we instead seek lower bounds on @xmath304 so that the size of the symbol weight codes can be more easily expressed in terms of the sizes of either one or a few cccs . * remark : * note that @xmath286 is a metric on a `` simplex '' which intersects each axis at ( euclidean ) distance @xmath14 from the origin . in particular , the components of @xmath20 need not be restricted to integers for @xmath286 to become a metric . also , @xmath14 need not be restricted to be an integer . in this section we determine lower bounds to the size of @xmath304 . to get these lower bounds , we first obtain a relation between the hamming distance between two compositions and the distance between two compositions as given by ( [ eq : d+ ] ) . + [ lem : n - nprime - dh ] for any two compositions @xmath306 , if @xmath307 then @xmath308 . @xmath144_proof : _ define two sets @xmath309 and @xmath310 . clearly , in the rest of the coordinates , @xmath311 . then we get the following set of equalities . i=0^q-1 n_i & = _ i=0^q-1 n_i + & _ ii^+ n_i + _ ii^- n_i & = _ ii^+ n_i + _ ii^- n_i + & _ ii^+ ( n_i - n_i ) & = _ ii^- ( n_i - n_i ) + & _ ii^+i^- ( n_i - n_i)^+ & = _ ii^+ i^- ( n_i - n_i)^+ . [ eq : n - nprime ] note that the lhs and rhs of the last equation are both equal to @xmath288 . let @xmath312 , then since @xmath307 , we get @xmath313 using the fact that the difference @xmath314 , for @xmath315 and @xmath316 for @xmath317 , we get @xmath318 this lemma immediately allows us to use existing gv bounds in various spaces ( under hamming distance ) to derive lower bounds on @xmath319 let the alphabet set be @xmath321 , that is , @xmath322 in every word of length @xmath323 , let all the @xmath2 symbols occur such that the frequencies of the symbols are in the set @xmath324 and all the frequencies occur . because of this construction , given @xmath2 the values of @xmath325 are restricted as given above . using the gv lower bound on the permutation code with hamming distance at least @xmath326 between two codewords gives us the lower bound @xmath327 is the volume of the ball of radius @xmath328 in @xmath320 . as explained in the proof of theorem [ thm : aqw - rate - q - n - r - const ] , a lower bound on constant symbol weight codes is provided by an elias - type bound in the hamming space . we can obtain another lower bound on constant symbol weight codes by considering lower bounds on @xmath304 . a lower bound on @xmath304 is obtained by letting @xmath29 of the symbols @xmath329 repeat @xmath2 times in every codeword of the constant symbol weight code and the remaining @xmath76 symbols satisfy the condition that there are @xmath130 symbols with composition @xmath330 and @xmath128 symbols with composition @xmath331 . denote this composition by @xmath332 , that is , @xmath333.\ ] ] because of the above choice of the repetitions of each symbol we seek a `` binary constant weight code '' in @xmath304 where @xmath76 coordinates have the value @xmath334 or @xmath335 and the rest @xmath29 coordinates have the value @xmath1 we also want the hamming distance between distinct codewords to be at least @xmath336 denote the maximum size of a binary constant weight code of length @xmath14 , weight @xmath337 and minimum distance @xmath12 by @xmath338 . the gv bound under the above constraints is @xmath339 note that we get @xmath340 in the denominator ( instead of @xmath328 ) since the binary constant weight space affords only even distances . from levenshtein @xcite we know that the lower bound is significant and grows exponentially as @xmath341 , for some constant @xmath342 only when the following conditions are satisfied @xmath343 the above lower bounds on @xmath305 give lower bounds on symbol weight codes as follows . the lower bound on @xmath344 can be stated as follows . [ thm : finite - case - csw ] we get the following results for different compositions . 1 . for @xmath345 $ ] , we get @xmath346 2 . for @xmath332 , we get @xmath347 3 . let @xmath348 , and @xmath349 . then @xmath350 the first two results follow immediately from the lower bounds on @xmath305 . in part 1 , each codeword in the permutation code corresponds to a rearrangement of the composition @xmath351.$ ] in part 2 , each codeword in the binary constant weight code corresponds to a rearrangement of the composition in @xmath332 . for the third result , we include a larger range of cccs . the expression is obtained by taking constant composition codes from separate constant composition spaces @xmath352 , whose compositions are separated by a hamming distance of at least @xmath326 . two different compositions @xmath352 and @xmath353 correspond to taking binary constant weight codes with weights separated by @xmath336 note that this choice of separate compositions corresponds to a binary bounded weight code , as studied in @xcite . there is a trade - off between the size of the constant composition space with composition @xmath354 and the size of the constant weight code in @xmath355 . the size of the constant weight code in @xmath355 is substantial only for large @xmath29 around @xmath356 on the other hand , the size of the constant composition space is large for small @xmath29 , thus potentially allowing for a larger ccc . the lower bound in equation is in fact useful in the case of a bounded symbol weight code with symbol weight at most @xmath2 . it is unclear how to combine codes of different symbol weights @xmath357 , where @xmath358 , such that we can obtain a computable expression . we instead use equation and optimize over the different symbol weights @xmath357 and the smallest weight @xmath359 . note that for a given symbol weight @xmath360 the quantity @xmath359 corresponds to the minimum number of symbols with frequency @xmath357 that we include in our estimate . + [ thm : finite - case - sw ] @xmath361 where @xmath362 and @xmath363 . we consider some numerical examples in order to show how the expressions in the previous section can be used to obtain lower bounds on symbol weight codes . we adopt the exponential notation of chu@xcite to denote a composition in a compact form . the notation @xmath364 is used to denote the composition @xmath365.\ ] ] we also recall the notion of a refinement of a composition from the same work . the composition @xmath366 $ ] is called a _ refinement _ of a composition @xmath367 $ ] if there is a partition @xmath368 of @xmath329 such that @xmath369 for every @xmath370 we write @xmath371 if @xmath20 is a refinement of @xmath372 this notion is important because of the following inequality ( see @xcite ) : @xmath373 , d ) \ge a_p([m_0,\dots , m_{p-1 } ] , d).\ ] ] below , we use lower bounds on fpas , where the lower bound is taken from @xcite . using equation , lower bounds on cccs are obtained from the lower bounds on the fpas . the lower bound on fpa mentioned below rely on the existence of certain ( generalized ) distance preserving mappings from @xmath16 to the permutation space @xmath374 ( see @xcite ) . the distances between compositions used in this section are all taken in the @xmath286 metric , unless mentioned otherwise . + [ ex:1 ] in this example we show how theorem [ thm : finite - case - csw ] and equation can be used . we know from ( * ? ? ? * example 3.7 ) that @xmath375 . since @xmath376 we immediately obtain that @xmath377 . in this case , @xmath378 and @xmath379 and the number of symbols occurring with frequency @xmath380 is @xmath381 . hence theorem [ thm : finite - case - csw ] is not applicable . but equation can be applied directly . for instance the compositions @xmath382 and @xmath383 satisfy @xmath384 which is greater than 7 . thus , @xmath385 in fact , the compositions @xmath386 and @xmath387 have the special property that if a symbol @xmath17 has different compositions @xmath388 then @xmath389 we can exploit this property to get a variant of lemma [ lem : n - nprime - dh ] below . + [ lem : n - nprime - dh - special ] for two compositions @xmath390 let @xmath391 for @xmath392 , if either @xmath393 or it is zero , then @xmath394 @xmath144_proof : _ the proof is very similar to the proof of lemma [ lem : n - nprime - dh ] . let @xmath395 be as defined in the proof of that lemma , and let @xmath396 finally , use equation to get @xmath397 _ example [ ex:1 ] _ ( contd . ) : returning to this example , we know that if two compositions differ in a symbol , then the difference is 4 , and so @xmath398 we will have the distance between two compositions at least 7 if we ensure ( using lemma [ lem : n - nprime - dh - special ] ) that @xmath399 , that is , @xmath400 using @xmath401 , we get that @xmath402 . the size of the binary code is obtained from agrell@xcite : @xmath403 this gives the much improved lower bound @xmath404 + [ ex:12 ] we continue with the previous example and a finer refinement @xmath405 we show in this example that directly using equation can lead to a better bound , compared to theorem [ thm : finite - case - csw ] . in this case , we apply theorem [ thm : finite - case - csw ] with @xmath406 @xmath407 @xmath408 and @xmath409 from the table of constant weight codes in @xcite , we get @xmath410 this gives @xmath411 however , this bound can be improved by using equation . a greedy search through the compositions with the maximum value of each part being 2 , show that the compositions @xmath412 @xmath413 @xmath414 @xmath415 and @xmath416 are mutually at distance at least 7 from one another . since each of them is a refinement of @xmath417 we get that @xmath418 . + finally , we look at an example which considers the bounded symbol weight and demonstrate the use of theorem [ thm : finite - case - sw ] . consider the refinement @xmath419 we can consider a code in this space to be embedded in the constant composition space with 16 symbols . thus , we get @xmath420 . we use the fact that if we consider all compositions containing 8 symbols occurring with frequency @xmath421 , then the difference of frequency between two symbols from different compositions is either 0 or 3 . thus , @xmath422 , @xmath423 and @xmath265 using lemma [ lem : n - nprime - dh - special ] , we need to ensure that @xmath424 we get the lower bound @xmath425 . from @xcite we have @xmath426 hence , @xmath427 in this particular example , using the lower bound in equation on the constant symbol weight codes with @xmath428 does not yield a better bound , primarily due to the absence of a known good lower bound on the corresponding ccc . the improvement mainly stems from the fact that we use a large binary constant weight code with weight @xmath429 . + as is evident from the above examples , theorems [ thm : finite - case - csw ] and [ thm : finite - case - sw ] help in actual computation of the bounds . in example [ ex:12 ] , the number of ordered partitions of @xmath430 into @xmath431 parts with each part taking values between 0 and 2 , inclusive , is @xmath432 a non - exhaustive greedy search could only find 5 compositions . it is computationally difficult to search exhaustively in such a large space . similarly , for larger lengths and alphabet sizes , finding the size of , and compositions in , @xmath304 is difficult . instead , by relying on known bounds on binary constant weight codes and cccs , we compute the sizes of the symbol weight codes more easily . in this section we determine constructions of symbol weight codes . versfeld@xcite provided constructions of bounded symbol weight codes from reed - solomon codes . we seek to obtain constant symbol weight codes with positive rate and positive relative distance . the following two constructions provide us with such codes with positive rate and positive relative distance , given that we already have a constant symbol weight code with positive rate and positive relative distance . * @xmath433 construction : * let @xmath22 be a constant symbol weight code with parameters @xmath434 over @xmath26 . let @xmath435 be an fpa over @xmath26 with parameters @xmath436 . then the @xmath433 construction results in a code @xmath437 . it is obtained by taking all codewords as follows : @xmath438 the code @xmath437 has parameters @xmath439 over @xmath26 . in particular if the code @xmath22 had the minimum symbol weight @xmath440 then so does the code @xmath437 , that is , @xmath441 . * concatenated construction : * let @xmath22 be a code over @xmath26 with parameters @xmath442 . let @xmath435 be an fpa with parameters @xmath443 over @xmath444 , such that @xmath445 the concatenated code @xmath437 with @xmath435 as the inner code and @xmath22 as the outer code has parameters @xmath446 . it is obtained by replacing every @xmath4-ary symbol of @xmath22 with a codeword from @xmath435 . in particular , the resulting code @xmath437 has the minimum symbol weight @xmath447 in this section , we use the symbol @xmath29 to denote the dimension of the reed - solomon code . let @xmath448_q$ ] be a reed - solomon code over a finite field @xmath30 with @xmath449 and @xmath450 versfeld@xcite showed that aside from the reed - solomon codewords which correspond to a constant polynomial , the reed - solomon code has maximum symbol weight of @xmath451 it is also established in the same works that there exists a coset of the reed - solomon code such that the maximum symbol weight of any codeword in the code is at most @xmath452 these codes belong to the bounded symbol weight space @xmath7 . by the singleton bound , these are optimal codes . in this section we establish several results which show that subsets of reed - solomon codes or their cosets can achieve the gv - type lower bound in theorem [ thm : aqw - rate - q - n - r - const ] . first , lemma [ lem : rs - coset ] below shows that for any _ constant _ symbol weight @xmath2 , there exists a coset of the reed - solomon code that attains the gv bound asymptotically . in theorem [ thm : rs - code ] , we provide a more explicit description of a subset of the reed - solomon code itself that has the constant symbol weight @xmath2 for @xmath453 , such that it attains the gv bound asymptotically . since @xmath454 , this also means that this latter result holds only for reed - solomon codes with rate more than @xmath455 we first show by an averaging argument that the gv bound can be achieved by subcodes of cosets of reed - solomon codes . [ lem : rs - coset ] let @xmath31_q$ ] be a family of reed - solomon codes . for @xmath197 let @xmath456 and @xmath220 then there exists a family of subcodes @xmath435 which is a subset of some coset of the code @xmath22 such that @xmath457 let @xmath458 denote the cosets of the reed - solomon code . since the cosets of the reed - solomon code are disjoint and they partition the hamming space , we have @xmath459 thus , the average size of the intersection of a coset with the space @xmath8 is @xmath460 . hence , there exists at least one coset whose intersection with @xmath8 has size at least this average . in the asymptotics for @xmath461 , we get the result stated in the lemma . in the remaining part of this section , we give a more explicit description of a subcode of the reed - solomon code with rate equal to the gv - type bound . the derivation of this result uses a lemma and a proposition stated below . the proposition [ prop : sw - le - wt - distr ] below states that asymptotically the rate of the constant symbol weight code with symbol weight @xmath2 that is a subset of the reed - solomon code , can not exceed the rate of the subcode formed by all the codewords of weight @xmath462 lemma [ lem : wt - distr ] gives an upper bound on the size of the number of codewords of weight @xmath462 the combination of this proposition and the lemma imply that the rate of the constant symbol weight code , which is a subset of the reed - solomon code , with symbol weight @xmath463 , can not exceed the gv - type lower bound @xmath464 that we obtained in theorem [ thm : aqw - rate - q - n - r - const ] , _ for any _ @xmath465 theorem [ thm : rs - code ] below shows that this rate can be attained for any @xmath258 satisfying @xmath466 we state the proposition and the lemma first , and defer their proofs to after the proof of the theorem . + [ prop : sw - le - wt - distr ] let @xmath448_q$ ] be a family of reed - solomon codes with parameters @xmath467 . let @xmath468 denote the set of vectors with symbol weight exactly @xmath469 for @xmath470 , and let @xmath471 denote the number of vectors of weight @xmath462 then , @xmath472 [ lem : wt - distr ] the weight distribution @xmath473 of a linear maximum distance separable ( mds ) code with parameters @xmath474_q$ ] satisfies : @xmath475 for @xmath476 . + the main theorem in this section is now stated below . [ thm : rs - code ] let @xmath31_q$ ] denote the family of reed - solomon codes with @xmath477 and @xmath449 . let @xmath478 . for @xmath197 , let @xmath196 and @xmath220 there exists a family of subcodes @xmath435 of @xmath22 of symbol weight exactly @xmath2 such that @xmath479 @xmath144_proof : _ every codeword of the reed - solomon code consists of coordinates which are the evaluations at all the non - zero points of @xmath30 , of a polynomial of degree at most @xmath480 . let @xmath481 be a polynomial in @xmath482 let @xmath483 if @xmath484 has symbol weight @xmath2 then it implies that @xmath485 for some @xmath486 and for @xmath2 different values of @xmath121 in @xmath487 in other words , @xmath488 has exactly @xmath2 distinct roots . note that @xmath2 is restricted to be @xmath489 since the polynomials can not have more than @xmath480 roots . let @xmath484 be a polynomial of degree @xmath480 such that it has exactly @xmath2 nonzero distinct roots @xmath490 in @xmath491 . then @xmath484 can be written as @xmath492 where @xmath493 and @xmath494 is a product of monic irreducible polynomials , each of degree at least @xmath495 the total degree of @xmath494 is @xmath496 since @xmath453 , the polynomial @xmath484 can not attain the value @xmath242 , where @xmath497 , at more than @xmath2 different points @xmath498 since there are @xmath499 points at which the function is nonzero . hence , the symbol weight of the codeword represented by @xmath484 is exactly @xmath1 we seek the asymptotic exponent of the number of such polynomials @xmath484 . this number is dominated by the number of possible monic irreducible polynomials @xmath494 . to describe this number , we recall the definition of the mbius function @xmath500 , @xmath501 the number of monic irreducible polynomials of degree @xmath502 is given by the sum @xmath503 ( see lidl and niederreiter ( * ? ? ? * theorem 3.25 ) ) . in particular , for large @xmath502 , this sum is dominated by just the first term @xmath504 as a consequence , asymptotically the number of polynomials @xmath484 is described by the number of monic irreducible polynomials @xmath494 of degree @xmath496 the asymptotic exponent of this count is approximately @xmath505 . thus the subset @xmath506 of @xmath90 consists of all the codewords obtained from at least these polynomials . hence the count above provides a lower bound on the rate of @xmath506 : @xmath507 the upper bound on the rate of @xmath506 is obtained by applying both lemma [ lem : wt - distr ] and proposition [ prop : sw - le - wt - distr ] . we get @xmath508 and , @xmath509 let @xmath510 be a codeword in the reed - solomon code . then @xmath277 is the image of a polynomial @xmath511 evaluated at all points of @xmath512 . we write @xmath513 if @xmath277 resulting from @xmath511 has symbol weight exactly @xmath2 then so do the codewords obtained from the polynomials @xmath514 for @xmath515 consider the subset @xmath516 of @xmath468 that is obtained by retaining exactly one monic polynomial from the set @xmath517 for any polynomial @xmath511 . thus , the size of @xmath516 satisfies @xmath518 . we claim that @xmath519 . to show this , we claim that there exists an injection mapping from @xmath516 to the set of all vectors of weight @xmath520 . since any @xmath511 in @xmath516 has symbol weight exactly @xmath2 , there exists a @xmath521 such that @xmath522 has exactly @xmath2 distinct roots . thus , the codeword @xmath523 has hamming weight exactly @xmath462 this is the only such vector . if there exists @xmath524 and @xmath486 such that @xmath525 , then the two polynomials @xmath511 and @xmath526 must satisfy the relation @xmath527 since they are the same on @xmath477 points and their degrees are at most @xmath528 . thus , @xmath529 , which is not possible since @xmath516 contains exactly one polynomial of this form . the expression for the weight distribution satisfies ( see ( * ? ? ? * chapter 11)):@xmath530 retaining only the first term gives the required upper bound on @xmath471 . the above expression can be rewritten as @xmath531 - \\ { i}(2\nmid k - r-1)\binom{n - r}{k - r-1}(q-1)\bigg\},\end{gathered}\ ] ] where @xmath532 is an indicator function that is 1 if @xmath533 does not divide @xmath534 and 0 otherwise . we show that each of the terms in the summation above is positive and hence we can upper bound @xmath471 by only the first term . this is proved via the following sequence of inequalities . for any @xmath535 , we have crl & j ( q^k - r - j-1 ) & > ( q^k - r - j-1 - 1 ) + & q^k - r - j - 1 & > ( q^k - r - j-1 - 1 ) + & q & > . we use the inequality @xmath536 . this expression is greater than the rhs of the above because of the inequalities @xmath537 , @xmath538 , and @xmath539 . this proves the lemma . for correcting narrowband noise in the powerline channel , it is desirable that the symbol weight be close to the minimum possible value of @xmath540 . it remains open to determine a large subset of the reed - solomon code for the case @xmath541 . it follows from proposition [ prop : sw - le - wt - distr ] that the rate of this subset can not exceed the gv - type bound @xmath542 . the more interesting question is whether this bound can be achieved , especially in the cases where @xmath2 is small . the work of konyagin and pappalardi @xcite gives an affirmative answer to this question in the case of @xmath543 and for low relative distance @xmath544 they show that the number of permutation polynomials of degree at most @xmath545 is approximately @xmath546 for @xmath547 . this is asymptotically , @xmath548 where @xmath549 . for the other ranges of @xmath2 , when @xmath2 is growing with @xmath23 we believe that it should be possible to attain the gv bound . for instance , it would be interesting to prove that for large @xmath550 , and for any irreducible polynomial @xmath494 of degree @xmath534 , there exists @xmath2 distinct and nonzero points @xmath551 in @xmath491 such that the polynomial @xmath552 has symbol weight exactly @xmath1 since most of the count of polynomials attaining constant symbol weight @xmath2 comes from the count of irreducible polynomials , proving this will show that the rate attains the gv - type bound . we have obtained no counterexample on performing an exhaustive computer search over all such irreducible polynomials @xmath494 of degree @xmath553 , with @xmath554 in all finite fields up to @xmath555 . we are unable to verify for larger fields because the computations become prohibitive . we believe that the conjecture is not true for very small @xmath2 ( when @xmath2 does not grow with @xmath14 ) . for instance for @xmath270 the polynomial @xmath556 , where @xmath557 is an irreducible polynomial in @xmath558 , has symbol weight greater than 1 for every choice of @xmath559 . we derive the asymptotic estimates of the sizes of symbol weight codes . we also provide means of obtaining lower bounds on such codes and show that it is possible to provide symbol weight codes with the minimal possible symbol weight via recursive constructions , given we start with a known such code . finally , we provided constructions of asymptotically good constant symbol weight codes . it remains open to determine families of codes with positive rate and positive relative distance with symbol weights that are optimal or close to the optimal value of @xmath540 . we thank the associate editor navin kashyap for pointing out an error in an earlier version of the manuscript , and for his comments , which helped us improve the presentation of this article . y. m. chee , g. ge , and a. c. h. ling , `` group divisible codes and their application in the construction of optimal constant - composition codes of weight three '' , _ ieee trans . theory , _ vol . 54 , pp . 35523564 , august 2008 . y. m. chee , s. h. dau , a. c. h. ling , and s. ling , `` linear size optimal @xmath4-ary constant - weight codes and constant - composition codes '' , _ ieee trans . theory , _ vol . 56 , pp . 140151 , january 2010 . d. j. j. versfeld , a. j. h. vinck , and h. c. ferreira , `` reed - solomon coding to enhance the reliability of m -fsk in a power line environment '' , in _ proc . . power line commun . and its applications _ , vancouver , bc , canada , april 2005 , pp . 100104 . d. j. j. versfeld , a. j. h. vinck , j. n. ridley , and h. c. ferreira , `` constructing coset codes with optimal same - symbol weight for detecting narrowband interference in @xmath0-fsk systems '' , _ ieee trans . theory _ , vol . 56 , pp . 63476353 , december 2010 .

we define a heegaard - scharlemann - thompson ( hst ) splitting of a 3-manifold to be a sequence of pairwise - disjoint , embedded surfaces , , such that for each odd value of , is a heegaard splitting of the submanifold of cobounded by and . our main result is the following : suppose ( or ) is an irreducible submanifold of a triangulated 3-manifold , bounded by a normal or almost normal surface , and containing at most one maximal normal 2-sphere . if is a strongly irreducible hst splitting of then we may isotope it so that for each even value of the surface is normal and for each odd value of the surface is almost normal . we then show how various theorems of rubinstein , thompson , stocking and schleimer follow from this result . we also show how our results imply the following : ( 1 ) a manifold that contains a non - separating surface contains an almost normal one , and ( 2 ) if a manifold contains a normal heegaard surface then it contains two almost normal ones that are topologically parallel to it . keywords : heegaard splitting , normal surface , almost normal surface

math0309049

since rubinstein s announcement of an algorithm to recognize the 3-sphere in 1993 @xcite , almost normal surfaces , the main technical innovation of that work , have received wide attention . in 1994 thompson @xcite reworked rubinstein s proof of the existence of an almost normal 2-sphere in @xmath7 using gabai s language of _ thin position _ @xcite . also in 1993 rubinstein announced that any strongly irreducible heegaard spliting could be made almost normal . this result , combined with his later work with william jaco @xcite , implies that any non - haken 3-manifold admits at most finitely many heegaard splittings of a given genus . in 1996 stocking @xcite provided a new proof of the fact that strongly irreducible heegaard splittings can be made almost normal , by following thompson s use of thin position type arguments . more recently the author @xcite and , independently , rieck and sedgwick @xcite used almost normal surfaces to prove that for any orientable 3-manifold with torus boundary , @xmath8 , in all but finitely many dehn fillings of @xmath8 the core of the attached solid torus can be isotoped onto every strongly irreducible heegaard splitting . saul schleimer @xcite has used almost normal surfaces to show that the set of distances of the heegaard splittings of a given 3-manifold is finite . the main result of this paper is that one may isotope a _ heegaard - scharlemann - thompson _ ( hst ) splitting into an appropriate normal form . roughly speaking , an hst splitting of a 3-manifold is a sequence of surfaces such that each consecutive pair cobounds a compression body , and if @xmath9 is such compression body then the subindex of @xmath10 is odd ( a more precise definition is given in section [ s : hst ] ) . for each odd @xmath2 the surface @xmath3 is referred to as a _ thick level _ , and for each even @xmath2 , @xmath3 is a _ thin level_. just as for classical heegaard splittings , the concept of _ strong irreducibility _ is a useful notion of non - triviality for an hst splitting . the precise statement of our main result is the following : * theorem [ t : main ] * _ suppose @xmath0 ( @xmath6 or @xmath7 ) is an irreducible submanifold of a triangulated 3-manifold which is bounded by a normal or almost normal surface , and contains at most one maximal normal 2-sphere . if @xmath1 is a strongly irreducible hst splitting of @xmath0 then each thin level is isotopic to a normal surface , and each thick level is isotopic to an almost normal surface . _ consequences of theorem [ t : main ] include many of aforementioned results on the existence of almost normal surfaces . we also show that several new results follow that may be of independent interest . * theorem [ nonsep ] * _ if a triangulated 3-manifold contains a non - separating surface then it contains a non - separating almost normal surface . _ * theorem [ compressible ] * _ suppose @xmath0 is a submanifold of a triangulated irreducible 3-manifold which is bounded by a connected normal surface , and contains at most one maximal normal 2-sphere . if @xmath11 is compressible in @xmath0 then there is an almost normal surface in @xmath0 which is topologically parallel to @xmath11 , and normalizes monotonically to @xmath11 . _ a result similar to theorem [ compressible ] has been independently obtained by culler , dunfield , jaco and shalen in their program to solve the weak lopez conjecture . an immediate corollary is the following : * corollary [ normalheegaard ] * _ if @xmath12 is a heegaard surface for a 3-manifold which is isotopic to a normal surface in the complement of a maximal normal 2-sphere then @xmath12 is isotopic to two , topologically parallel , almost normal surfaces . _ we conclude the paper by proving a result similar to theorem [ t : main ] , which gives an exact characterization of handlebodies in terms of the existence of certain normal and almost normal surfaces . we hope to use this result in subsequent work to give a new algorithm to determine when an almost normal surface is a heegaard surface . the techniques used to prove theorem [ t : main ] may be of independent interest . while in this paper we make crucial use of ideas from such earlier papers as @xcite , our arguments differ from those of previous authors in several key respects . we outline these differences here . * for the existence of almost normal strongly irreducible heegaard splittings both rubinstein and stocking produce a sequence of surfaces which eventually converge to the desired one . for example , in rubinstein s iterated " sweepout approach he shows that almost normal surfaces occur at the maximum of a sweepout which is in some way minimal " . but the first sweepout which he chooses may not have the desired surface as a maximum . so a sequence of sweepouts of smaller and smaller submanifolds is chosen in which the maximum is eventually the desired surface . in the approach here we show that if the right concept of minimal " is chosen for the initial sweepout then all _ local _ maxima are almost normal , and the desired surface must be one of these . * several results are used here to streamline the argument which were not available to rubinstein or stocking . for example , we appeal to scharlemann s _ local detection of strongly irreducible heegaard splittings _ @xcite to classify the intersection of certain surfaces with the interior of a 3-simplex . * rather than using sweepouts ( as does rubinstein ) or thin position ( as do thompson and stocking ) we use the general framework of _ relative _ hst splittings ( see section [ s : relhst ] ) . we find that this general framework provides a much cleaner approach and further streamlines some of the arguments . * we deal with 2-sphere components that may be pinched off " in the formation of an almost normal surface in an entirely different way than previous authors . the author thanks saul schleimer for helpful conversations , especially regarding lemma [ l : untang ] , during the preparation of this paper . the author also thanks peter shalen . in this section , we give some of the standard definitions that will be used throughout the paper . the expert in 3-manifold theory may still want to skim this , as one or two new terms are introduced . a 2-sphere in a 3-manifold which does not bound a 3-ball on either side is called _ essential_. if a manifold does not contain an essential 2-sphere , then it is referred to as _ irreducible_. a loop on a surface is called _ essential _ if it does not bound a disk in the surface . given a surface , @xmath13 , in a 3-manifold , @xmath0 , a _ compressing disk _ for @xmath13 is a disk , @xmath14 , such that @xmath15 , and such that @xmath16 is essential on @xmath13 . if we let @xmath17 denote a thickening of @xmath18 in @xmath0 , then to _ compress @xmath13 along @xmath18 _ is to remove @xmath19 from @xmath13 , and replace it with @xmath20 . a _ compression body _ is a 3-manifold which can be obtained by starting with some surface , @xmath13 ( not necessarily connected ) , forming the product , @xmath21 , attaching some number of 2-handles to @xmath22 , and capping off all remaining 2-sphere boundary components with 3-balls . the boundary component , @xmath23 , is often referred to as @xmath24 . the other boundary component is referred to as @xmath25 . if @xmath26 , then we say the compression body is a _ a compression body is _ non - trivial _ if it is not a product . a surface , @xmath13 , in a 3-manifold , @xmath0 , is a _ heegaard surface for m _ if @xmath13 separates @xmath0 into two compression bodies , @xmath9 , and @xmath27 , such that @xmath28 . such a splitting is _ non - trivial _ if both @xmath9 and @xmath27 are non - trivial , and _ doubly - trivial _ if both @xmath9 and @xmath27 are trivial . a separating surface , @xmath13 , in a 3-manifold , @xmath0 , is _ strongly irreducible _ if every compressing disk for @xmath13 on one side intersects every compressing disk for @xmath13 on the other . [ l : haken ] if @xmath0 contains a non - trivial or doubly - trivial strongly irreducible heegaard surface , then @xmath11 is incompressible in @xmath0 . lemma [ l : haken ] follows directly from a lemma of haken @xcite . a _ heegaard - scharlemann - thompson ( hst ) splitting _ of a manifold , @xmath0 , is a sequence of closed , embedded , pairwise disjoint surfaces , @xmath29 , such that 1 . for each odd @xmath2 strictly less than @xmath30 , @xmath3 is a non - trivial or doubly - trivial heegaard splitting of the submanifold of @xmath0 cobounded by @xmath4 and @xmath5 . if @xmath30 is odd , then there is a non - trivial compression body , @xmath9 , such that @xmath31 , and @xmath32 . 3 . @xmath33 . for each odd @xmath2 the surface @xmath3 is referred to as a _ thick level _ , and for each even @xmath2 , @xmath3 is a _ thin level_. if @xmath9 is a compression body , then @xmath34 is an hst splitting of @xmath9 . the only thick level of this hst splitting is @xmath10 , and the only thin level is @xmath35 . if @xmath0 is a closed 3-manifold , and @xmath13 is a heegaard surface for @xmath0 , then an hst splitting of @xmath0 is @xmath36 , with a unique thick level , @xmath13 . if @xmath37 , then @xmath38 is an hst splitting of @xmath0 . an hst splitting , @xmath29 , is _ strongly irreducible _ if for each odd @xmath2 strictly less than @xmath30 , the thick level , @xmath3 , is strongly irreducible in the submanifold of @xmath0 cobounded by @xmath4 and @xmath5 . [ d : order ] for any surface , @xmath13 , let @xmath39 , where @xmath40 are the components of @xmath13 . if @xmath41 and @xmath42 denote compact , embedded surfaces in a 3-manifold , @xmath0 , then we say @xmath43 if @xmath44 . note that this ordering is defined so that if @xmath41 is obtained from @xmath42 by a compression , then @xmath43 . for any hst splitting , @xmath45 , let @xmath46 denote the ordered set @xmath47 is a thick level@xmath48 , where repeated integers are included , and the ordering is non - increasing . if @xmath49 and @xmath50 denote hst splittings of some 3-manifold , @xmath0 , then we say @xmath51 if @xmath52 , where the comparison is made lexicographically . every 3-manifold admits a minimal hst splitting . choose any hst splitting of @xmath0 . now consider any decreasing sequence of hst splittings of @xmath0 that begins with the chosen one . this sequence must be finite , since any decreasing sequence of sets of non - negative integers terminates under the lexicographical ordering . the following two theorems of scharlemann and thompson originally appear in @xcite , with slightly different terminology . one can find proofs of both that are more consistent with the language presented here in @xcite . [ t : st ] [ scharlmann - thompson @xcite ] a minimal hst splitting is strongly irreducible . [ t : stinceven ] [ scharlmann - thompson @xcite ] each thin level of a strongly irreducible hst splitting is incompressible . to proceed , we must first understand arcs in a compression body . if @xmath9 is a compression body , recall that @xmath9 can be built by starting with a product , @xmath21 , and attaching 2- and 3-handles to @xmath22 . anything that remains of @xmath22 after the attachment becomes part of @xmath35 . we say a properly embedded arc , @xmath54 , is _ straight _ in @xmath9 if @xmath55 , where @xmath56 is a point such that @xmath57 . we are now ready to generalize the definition of a compression body : [ d : kcompbod ] a _ @xmath58-compression body _ , @xmath59 , is 1 . a 3-manifold , @xmath9 , which can be obtained by starting with some surface , @xmath13 ( not necessarily connected ) , forming the product , @xmath21 , attaching some number of 2-handles to @xmath22 , and capping some ( but not necessarily all ) remaining 2-sphere boundary components with 3-balls . the boundary component , @xmath23 , is @xmath60 , and the other boundary component is @xmath35 . 2 . a 1-manifold , @xmath61 , such that 1 . @xmath58 is a disjoint union of embedded arcs 2 . each arc of @xmath58 has at least one endpoint on @xmath10 3 . if @xmath54 is an arc of @xmath58 with @xmath62 , then there is a disk , @xmath63 , with @xmath64 , @xmath65 , and @xmath66 . if @xmath54 is an arc of @xmath58 with one endpoint on @xmath10 , then @xmath54 is straight . 5 . for each 2-sphere component , @xmath67 , of @xmath68 , @xmath69 . if @xmath67 is a 2-sphere component of @xmath10 then either @xmath67 does not bound a 3-ball component of @xmath9 or @xmath70 . a @xmath58-compression body , @xmath59 , is _ non - trivial _ if either @xmath9 is not a product , or at least one arc of @xmath58 is not straight . if @xmath58 is a 1-manifold which is properly embedded in a 3-manifold , @xmath0 , then a _ heegaard splitting of the pair _ @xmath53 is an expression of @xmath0 as a union of @xmath71-compression bodies , @xmath72 and @xmath73 , such that @xmath74 , and @xmath75 . such a splitting is _ non - trivial _ if both @xmath72 and @xmath73 are non - trivial . when the context is clear , we will refer to the surface , @xmath60 , as a _ heegaard surface _ of @xmath53 . if @xmath58 is a 1-manifold which is properly embedded in a 3-manifold , @xmath0 , then a _ heegaard - scharlemann - thompson ( hst ) splitting of the pair _ @xmath53 is a sequence of closed , embedded , pairwise disjoint surfaces , @xmath76 , such that 1 . for each odd @xmath77 between 1 and @xmath30 , there is a non - trivial @xmath78-compression body , @xmath79 , such that @xmath80 , and @xmath81 . 2 . for each even @xmath77 between 2 and @xmath30 , there is a non - trivial @xmath82-compression body , @xmath83 , such that @xmath84 , and @xmath85 . 3 . @xmath86 4 . @xmath87 . for each odd @xmath77 the surface @xmath88 is referred to as a _ thick level _ , and for each even @xmath77 , @xmath88 is a _ thin level_. [ e : thin ] suppose that @xmath89 is an arbitrary knot or link with no trivial components , and @xmath90 is some standard height function on @xmath7 ( so that for each @xmath91 , @xmath92 is a 2-sphere ) , which is a morse function when restricted to @xmath58 . let @xmath93 denote the critical values of @xmath90 restricted to @xmath58 , and let @xmath94 be some point in the interval @xmath95 . the following terminology is standard in _ thin position _ arguments ( see @xcite ) . if @xmath77 is such that @xmath96 and @xmath97 , then we say the surface @xmath98 is a _ thick level _ of @xmath58 . similarly , if @xmath99 and @xmath100 , then we say the surface @xmath98 is a _ thin level _ of @xmath58 . suppose there are @xmath30 thick levels for @xmath58 . let @xmath101 , @xmath102 denote the set of thick levels of @xmath58 , and @xmath103 denote the set of thin levels . then @xmath104 is an hst splitting of @xmath105 , and the thick and thin levels of this hst splitting are precisely the thick and thin levels of @xmath58 . if @xmath106 is some properly embedded 1-manifold then let @xmath107 denote @xmath0 with a regular neighborhood of @xmath58 removed . if @xmath8 is any subset of @xmath0 then let @xmath108 . suppose @xmath13 is a surface in a 3-manifold @xmath0 , and @xmath58 is some properly embedded 1-manifold in @xmath0 , transverse to @xmath13 . a _ relative compressing disk _ for @xmath13 is a disk , @xmath18 , such that @xmath109 , where @xmath110 , and @xmath111 . if there are no relative compressions for some surface , then we say that it is _ relatively incompressible_. to _ perform a relative compression _ is to use a relative compressing disk to guide an isotopy , thereby reducing the number of intersections with @xmath58 by exactly two . [ d : weakreduce ] suppose @xmath13 is a separating surface in a 3-manifold , @xmath0 , and @xmath58 is some properly embedded 1-manifold in @xmath0 , transverse to @xmath13 . then @xmath13 is _ weakly reducible with respect to @xmath58 _ if there exist disks , @xmath18 and @xmath112 , on opposite sides of @xmath13 , such that either 1 . @xmath18 and @xmath112 are disjoint compressing disks for @xmath113 in @xmath107 2 . @xmath18 and @xmath112 are relative compressing disks for @xmath13 that are either disjoint , or meet in a point of @xmath58 3 . @xmath18 is a relative compressing disk for @xmath13 , and @xmath112 is a disjoint compressing disk for @xmath113 in @xmath107 4 . @xmath18 is a compressing disk for @xmath113 in @xmath107 , and @xmath112 is a disjoint relative compressing disk for @xmath13 . we say @xmath13 is _ strongly irreducible with respect to @xmath58 _ if it is not weakly reducible with respect to @xmath58 . [ e : thin3 ] let @xmath58 , @xmath90 , and @xmath114 be as in example [ e : thin ] . the _ width _ of @xmath58 is defined to be the quantity @xmath115 . a knot is said to be in _ thin position _ if @xmath90 is chosen so that the width of @xmath58 is minimal ( see @xcite ) . recall from example [ e : thin ] that any choice of @xmath90 induces an hst splitting , @xmath116 , of @xmath105 . if , when @xmath90 is chosen so as to minimize the width of @xmath58 , it turns out that @xmath117 ( _ i.e. _ @xmath58 is also in _ bridge position _ with respect to @xmath90 ) , then we claim that @xmath118 is a strongly irreducible heegaard splitting of @xmath105 . suppose this were not the case . then @xmath118 is weakly reducible with respect to @xmath58 . let @xmath18 and @xmath112 be a pair of disks , as in the above definition of weak reducibility . we now examine each of the possible cases in that definition : 1 . _ @xmath18 and @xmath112 are relative compressing disks for @xmath118 that are either disjoint , or meet in a point of @xmath58 . _ then there is an isotopy of @xmath58 which reduces its width , as depicted in figure [ f : thinmove ] . + = 2 in 2 . _ @xmath18 is a relative compressing disk for @xmath118 , and @xmath112 is a disjoint compressing disk for @xmath119 in @xmath107 . _ then @xmath120 bounds disks , @xmath121 and @xmath122 , on @xmath118 ( since @xmath118 is a sphere ) . one of these disks ( @xmath121 , say ) is disjoint from @xmath18 . as @xmath112 is a compressing disk for @xmath119 , it must be that there is a relative compressing disk inside the sphere bounded by @xmath123 . however , this relative compressing disk is disjoint from @xmath18 , so we are reduced to case 1 . @xmath18 is a compressing disk for @xmath119 in @xmath107 , and @xmath112 is a disjoint relative compressing disk for @xmath13 . _ this case and the previous one are completely symmetric . @xmath18 and @xmath112 are disjoint compressing disks for @xmath119 in @xmath107 . _ as in case 2 , @xmath120 bounds disks , @xmath121 and @xmath122 on @xmath118 . also , @xmath16 bounds disks , @xmath124 and @xmath125 on @xmath118 . since @xmath126 , one of the disks , @xmath121 or @xmath122 is disjoint from one of the disks , @xmath124 or @xmath125 . say @xmath127 . as in case 2 , there is a relative compression , @xmath128 , inside the ball bounded by @xmath123 , and a relative compression , @xmath12 , inside the ball bounded by @xmath129 . but @xmath130 , so we are again reduced to case 1 above . in lemma [ l : untang ] we will present a generalization of this example for graphs in 3-manifolds , as opposed to 1-manifolds in @xmath7 . proofs of the next two lemmas originally appeared in @xcite with different terminology . one can find proofs of both that are more consistent with the language presented here in @xcite . [ l : essential ] if @xmath13 is a strongly irreducible heegaard surface of @xmath53 and @xmath131 is an essential surface then @xmath13 can be isotoped rel @xmath58 so that no arc of @xmath132 cobounds a relative compressing disk for @xmath13 , and so that no loop of @xmath132 bounds a compressing disk for @xmath113 in @xmath107 . [ l : relhaken ] if @xmath13 is a strongly irreducible heegaard surface of @xmath53 then @xmath133 is incompressible in @xmath107 , and @xmath11 is relatively incompressible in @xmath0 . an hst splitting , @xmath76 of a pair , @xmath53 , is _ strongly irreducible _ if for each odd @xmath77 strictly less than @xmath30 , the thick level @xmath88 is strongly irreducible with respect to @xmath58 in the submanifold of @xmath0 cobounded by @xmath134 and @xmath135 . we now present one further generalization of the concept of a hst splitting . if @xmath136 is any graph then let @xmath137 and @xmath138 denote the sets of vertices and edges of @xmath136 . if we are discussing a particular graph , @xmath136 , embedded in a 3-manifold , @xmath0 , then @xmath139 will denote @xmath0 with a regular neighborhood of @xmath137 removed . let @xmath136 be a properly embedded graph in a 3-manifold , @xmath0 . we define an _ hst splitting of @xmath140 _ to be an hst splitting of @xmath141 . if @xmath136 is a graph properly embedded in an irreducible 3-manifold , @xmath0 , and we are given an hst splitting of @xmath140 , then we can define an hst splitting of @xmath0 by forgetting " @xmath136 . [ d : underlyinghst ] if @xmath76 is an hst splitting of @xmath140 , where @xmath0 is an irreducible 3-manifold other than @xmath142 or @xmath7 , then we define its _ underlying hst splitting _ , @xmath143 $ ] , as follows : 1 . let @xmath144 be the onto , monotone function , such that @xmath145 iff the submanifold of @xmath0 co - bounded by the non-@xmath146 components of @xmath147 and @xmath88 is a product . 2 . for each @xmath2 between 0 and @xmath148 , choose some @xmath149 , and let @xmath3 denote the non-@xmath146 components of @xmath88 . 3 . let @xmath150 be the maximal subset of @xmath151 such that @xmath152 is an hst splitting of @xmath0 . 4 . define @xmath143=\{f_i\}_{i \in \sigma}$ ] . we leave it to the reader to check that @xmath143 $ ] is well defined up to isotopy . let @xmath136 be a properly embedded graph in a 3-manifold , @xmath0 . for any surface , @xmath153 , let @xmath154 , where @xmath40 are the components of @xmath13 , and @xmath155 denotes @xmath156 with a neighborhood of @xmath136 removed . if @xmath41 and @xmath42 denote compact , embedded surfaces in @xmath0 , then we say @xmath157 if @xmath158 . in the next lemma we relate the thick and thin levels of an hst splitting of a pair to those of its underlying hst splitting . this will be an important step in the eventual production of almost normal surfaces . [ l : underlyingodd ] let @xmath136 be a properly embedded graph in an irreducible 3-manifold @xmath0 other than @xmath7 or @xmath142 . let @xmath116 be an hst splitting of @xmath140 , and suppose @xmath159 $ ] . then for each thick ( thin ) level , @xmath160 , of @xmath1 there is a thick ( thin ) level of @xmath116 which becomes parallel to @xmath160 when all @xmath146 components are removed . let @xmath161 and @xmath150 be as given in definition [ d : underlyinghst ] . let @xmath162 be the @xmath163 element of @xmath150 , for some odd number , @xmath54 . @xmath164 is parallel to the non-@xmath146 components of @xmath88 , for all @xmath165 . since @xmath161 is monotone , @xmath166 either contains an odd number , or contains only one element . assume the latter is true , and let @xmath167 . the assumption that @xmath166 has only one element implies the non-@xmath146 components of @xmath168 , @xmath169 , and @xmath170 ( which we denote @xmath171 , @xmath172 , and @xmath173 ) are non - parallel . now , let @xmath174 denote the predecessor of @xmath162 in @xmath150 . since @xmath54 is odd , @xmath175 . since @xmath176 is even , @xmath177 . as @xmath178 , we know that @xmath172 and @xmath171 are not parallel , so it must be that @xmath179 . but @xmath178 also implies that @xmath180 is parallel to @xmath171 . since @xmath174 is the predecessor of @xmath162 in @xmath150 , @xmath181 . since @xmath150 is maximal , @xmath175 implies @xmath182 . this is a contradiction , as @xmath183 , @xmath180 is parallel to @xmath171 , and @xmath179 . the proof in the case that @xmath54 is even is completely analogous . in this section we examine the properties of strongly irreducible hst splittings of pairs and prove an important theorem about their existence . we begin with a result which is completely analogous to theorem [ t : stinceven ] . its proof is similar to that of theorem 7.4 . of @xcite . [ t : minwrtk ] let @xmath58 be a properly embedded 1-manifold in a 3-manifold , @xmath0 . each thin level of a strongly irreducible hst splitting of @xmath53 is incompressible in @xmath107 , and relatively incompressible in @xmath0 . let @xmath184 be a strongly irreducible hst splitting of @xmath53 . for each odd @xmath77 , let @xmath185 denote the submanifold of @xmath0 cobounded by @xmath134 and @xmath135 . suppose @xmath186 is a loop that bounds a compressing disk , or an arc which cobounds a relative compressing disk , for some surface , @xmath187 , where @xmath188 is even . if @xmath186 is a loop , then let @xmath189 denote a disk in @xmath0 , such that @xmath190 . if @xmath186 is an arc , then let @xmath189 be a disk in @xmath0 such that @xmath191 , where @xmath192 . in either case , choose @xmath189 so that @xmath193 is minimal . it follows from lemma [ l : relhaken ] that for each odd @xmath77 , @xmath194 is incompressible and relatively incompressible in @xmath185 . hence , @xmath189 can not lie entirely in @xmath195 or @xmath196 . we conclude then that there is some loop or arc of intersection of the interior of @xmath189 with @xmath197 . let @xmath198 denote an innermost such loop . let @xmath199 denote the subdisk of @xmath189 bounded by @xmath198 . @xmath199 lies in @xmath200 , for some odd number , @xmath201 . as @xmath202 is incompressible in @xmath200 , @xmath198 must bound a disk , @xmath203 , on @xmath202 . now , let @xmath204 be an innermost loop of @xmath205 , and let @xmath206 be the subdisk of @xmath203 bounded by @xmath204 . then we can use @xmath206 to surger @xmath189 , and thereby obtain a new disk , with the same boundary as @xmath189 , contradicting our minimality assumption . we conclude then that @xmath186 can not be a loop , and if @xmath186 is an arc , then @xmath189 contains no loops of intersection with @xmath197 . let @xmath207 then denote an arc of intersection which is outermost on @xmath189 . @xmath207 and a subarc of @xmath198 cobound a subdisk , @xmath208 , of @xmath189 . @xmath208 is then a relative compressing disk for @xmath194 , for some @xmath77 , contradicting lemma [ l : relhaken ] . if @xmath136 is a properly embedded graph in an irreducible 3-manifold , @xmath0 , then each thin level of a strongly irreducible hst splitting of @xmath140 is incompressible in @xmath209 , and relatively incompressible in @xmath139 . for any hst splitting , @xmath45 , of @xmath140 , let @xmath210 denote the ordered set @xmath211 is a thick level of @xmath212 , where repeated integers are included , and the ordering is non - increasing . if @xmath49 and @xmath50 denote hst splittings of @xmath140 , then we say @xmath213 if @xmath214 , where the comparison is made lexicographically . [ l : untang ] let @xmath136 be a properly embedded graph in an irreducible 3-manifold , @xmath0 . if @xmath1 is a strongly irreducible hst splitting of @xmath0 then there exists a strongly irreducible hst splitting , @xmath116 , of @xmath140 such that @xmath143=\{f_i\}$ ] . among all hst splittings of @xmath140 whose underlying hst splitting is @xmath1 , choose one , @xmath184 , that is smallest . we would like to show that @xmath116 is strongly irreducible . for each odd @xmath77 strictly less than @xmath148 , let @xmath185 denote the submanifold of @xmath0 cobounded by @xmath134 and @xmath135 . let @xmath215 be a thick level , so that @xmath201 is odd . we need to show that @xmath215 is strongly irreducible with respect to @xmath216 in @xmath200 . if not , then there are disks , @xmath18 and @xmath112 , in @xmath200 , as in definition [ d : weakreduce ] . note that we may assume that @xmath16 does not bound a disk , @xmath124 , on @xmath215 such that @xmath217 bounds a ball containing a single unknotted arc of @xmath138 . if this were the case then there is a relative compressing disk , @xmath218 , inside the ball bounded by @xmath219 . as @xmath220 , we may use @xmath218 instead of @xmath18 . we make a similar choice for @xmath112 . our reason for such choices for @xmath18 and @xmath112 will be made clear shortly . let @xmath221 , @xmath222 , and @xmath223 be the surfaces obtained from @xmath215 by ( relative ) compression along @xmath18 , ( relative ) compression along @xmath112 , and ( relative ) compression along both @xmath18 and @xmath112 . by our choice of @xmath18 from the previous paragraph we know that if some component @xmath67 of @xmath221 bounds a ball in @xmath200 then @xmath224 . this a necessary condition for @xmath221 and @xmath225 to cobound a @xmath138-compression body . ( see condition 2(f ) of definition [ d : kcompbod ] . ) we now produce a new hst splitting from @xmath116 . there are four cases : 1 . @xmath226 , @xmath227 . + remove @xmath215 from @xmath116 . in its place , insert @xmath228 and reindex . 2 . @xmath229 , @xmath227 . + replace @xmath230 with @xmath231 . @xmath226 , @xmath232 . + replace @xmath233 with @xmath234 . 4 . @xmath229 , @xmath232 . + replace @xmath235 with @xmath223 and reindex . in all cases it is a routine matter to check that we have defined a new hst splitting of @xmath140 which is smaller than @xmath116 . what remains to be checked is that the underlying hst splitting is still @xmath1 . we claim that the only way in which this can fail is if the non-@xmath146 components of both @xmath221 and @xmath222 are not parallel to the non-@xmath50 components of @xmath215 . the proof of this assertion is similar in all four cases . we present a proof for the first ( and most difficult ) case and leave the remaining cases as exercises . let @xmath150 be as in definition [ d : underlyinghst ] . if the non-@xmath146 components of two surfaces , @xmath203 and @xmath236 , are parallel then we write @xmath237 . suppose that @xmath238 . then @xmath18 is either a relative compressing disk , or @xmath16 bounds a disk on @xmath215 that meets @xmath136 ( and hence @xmath16 is essential on @xmath239 , but inessential on @xmath215 ) . since @xmath223 is obtained from @xmath222 by ( relative ) compression along @xmath18 , we conclude that @xmath240 . in case 1 above , the new hst splitting that we construct looks like @xmath241 as @xmath242 , and @xmath243 , the underlying hst splitting of this is the same as the underlying hst splitting of the following sequence : @xmath244 since @xmath112 is a disk in the submanifold of @xmath0 cobounded by @xmath215 and @xmath245 we will not include the index of @xmath222 in @xmath150 when forming the underlying hst splitting of this sequence . hence , the underlying hst splitting of this last sequence is the same as that of @xmath116 . what we have established is that if the underlying hst splitting of the new hst splitting of @xmath246 constructed above is not equal to @xmath1 then both @xmath18 and @xmath112 were compressing disks for @xmath215 . as @xmath215 is parallel to some thick level of @xmath1 ( by lemma [ l : underlyingodd ] ) , and @xmath18 and @xmath112 are disjoint , this contradicts our assumption that @xmath1 was strongly irreducible . in this section , we discuss the necessary background material on normal surfaces . a _ normal curve _ on the boundary of a tetrahedron is a simple loop which is transverse to the 1-skeleton , made up of arcs which connect distinct edges of the 1-skeleton . the _ length _ of such a curve is simply the number of times it crosses the 1-skeleton . a _ normal disk _ in a tetrahedron is any embedded disk , whose boundary is a normal curve of length three or four , and whose interior is contained in the interior of the tetrahedron , as in figure [ f : normal ] . = 3.25 in a _ normal surface _ in @xmath0 is the image of an embedding , @xmath201 , of some closed surface , @xmath13 , into @xmath0 , such that @xmath247 is a union of normal disks . in addition , we say @xmath247 is an _ almost normal surface _ if it consists of all normal disks , plus one additional piece in one tetrahedron . this piece can be either a disk with normal boundary of length 8 ( depicted in figure [ f : octagon ] ) , or two normal disks connected by a single unknotted tube ( as in figure [ f:1tube ] ) . almost normal surfaces were first explored by rubinstein in @xcite , and later used by thompson @xcite and stocking @xcite . in @xcite , we produce new applications of almost normal surfaces by generalizing them to include surfaces with non - empty boundary . one application of the results we have discussed thus far comes about when we let @xmath136 be the 1-skeleton of a triangulation of a 3-manifold , @xmath0 . if @xmath249 is such a triangulation then we denote the @xmath30-skeleton of @xmath249 by @xmath250 . we now focus on strongly irreducible hst splittings of @xmath248 . [ bubble ] suppose @xmath13 is an embedded surface in @xmath0 , equipped with a triangulation , @xmath249 . a _ bubble _ for @xmath13 is a ball , @xmath236 , such that @xmath251 , where @xmath252 and @xmath253 are disks , @xmath252 is contained in a single tetrahedron , @xmath254 , @xmath255 , and @xmath256 . [ l : nobubbles ] suppose @xmath0 is a submanifold of a 3-manifold , @xmath257 , with triangulation , @xmath249 , which is bounded by a normal or almost normal surface . let @xmath116 be a strongly irreducible hst splitting of @xmath248 . then we may isotope @xmath116 so that for all @xmath77 , @xmath88 does not contain any bubbles . suppose @xmath236 is a bubble for @xmath88 , for some @xmath77 , where @xmath258 , as in definition [ bubble ] . we can use @xmath236 to guide an isotopy from @xmath253 to @xmath252 . this may push other surfaces of @xmath116 which had non - empty intersection with @xmath259 , but it can only destroy bubbles for those surfaces , too . the isotopy is supported on a neighborhood of @xmath236 , which is disjoint from @xmath260 . hence , if @xmath116 was strongly irreducible before the isotopy then it will remain so . since there are a finite number of surfaces in @xmath116 , and a finite number of bubbles for each , we eventually arrive at an hst splitting with the desired properties . [ t : normal ] suppose @xmath0 is a submanifold of a 3-manifold , @xmath257 , with triangulation , @xmath249 , which is bounded by a normal or almost normal surface . let @xmath116 be a strongly irreducible hst splitting of @xmath248 . then we may isotope @xmath116 so that each thin level is a normal surface . we begin by using lemma [ l : nobubbles ] to isotope @xmath116 , removing all bubbles . now , let @xmath77 be some even number , let @xmath261 be some tetrahedron in @xmath249 , and let @xmath262 be a face of @xmath261 . first , we examine the possibilities for @xmath263 . let @xmath264 be an innermost simple closed curve , bounding a disk , @xmath252 in @xmath262 . by theorem [ t : minwrtk ] , @xmath264 must bound a disk , @xmath253 , in @xmath88 . @xmath265 is irreducible ( it s a handlebody ) , so @xmath266 bounds a bubble for @xmath88 . this is a contradiction , so we see no simple closed curves in any face . if there are any curves which run from one edge of @xmath262 to itself , then there is an outermost such one . let @xmath18 denote the sub - disk it cuts off in @xmath262 . then @xmath18 is a relative compressing disk for @xmath88 , also contradicting theorem [ t : minwrtk ] . we conclude that @xmath263 is a collection of normal arcs . we now consider the possibilities for @xmath267 . it is easy to show that the only possibilities for normal loops are curves of length 3 , or 4n ( see , for example , @xcite ) . if there are any curves of length greater than 4 , then there must be a disk , @xmath18 , such that @xmath109 , where @xmath268 , and @xmath269 ( see @xcite ) . this is a relative compressing disk for @xmath88 , which is again a contradiction . we conclude that @xmath267 consists of normal loops of length 3 and 4 . finally , it follows from theorem [ t : minwrtk ] that every loop of @xmath267 bounds a disk on @xmath88 . since we have already ruled out simple closed curves in faces of @xmath261 , such disks must lie entirely inside @xmath261 . we conclude @xmath88 is a normal surface . our goal now is to show that once bubbles are removed from the thick levels of @xmath116 , they become almost normal in @xmath0 . first , we shall need a few lemmas . [ l : nlarcs ] suppose @xmath0 is a submanifold of a 3-manifold , @xmath257 , with triangulation , @xmath249 , which is bounded by a normal or almost normal surface . let @xmath184 be a strongly irreducible hst splitting of @xmath248 , where @xmath11 normal implies @xmath148 is even , and @xmath11 almost normal implies @xmath148 is odd . then we may isotope @xmath116 , so that each thin level is a normal surface , and so that for each thick level , @xmath88 , @xmath263 is a collection of normal arcs , for every 2-simplex , @xmath262 , of @xmath249 . we begin by appealing to lemma [ l : nobubbles ] and theorem [ t : normal ] to isotope @xmath116 so that there are no bubbles for any of its surfaces , and so that each thin level is normal . we now produce a further isotopy of @xmath116 , which fixes all thin levels , and which results in an hst splitting with all of the desired properties . the isotopy will be defined individually for each odd value of @xmath77 ( _ i.e. _ , for each thick level ) . let @xmath77 be some odd number , and let @xmath185 denote the submanifold of @xmath0 cobounded by @xmath134 and @xmath135 . the strong irreducibility of @xmath116 implies that @xmath88 is a strongly irreducible heegaard surface for @xmath270 . lemma [ l : essential ] now implies that a standard innermost disk / outermost arc argument can be used to isotope @xmath88 so that no arc of @xmath271 cobounds a relative compressing disk for @xmath88 in @xmath185 , and so that no loop of @xmath271 bounds a compressing disk for @xmath272 in @xmath273 . now , suppose there are non - normal components of @xmath263 , for some 2-simplex , @xmath262 . if there are loops of @xmath263 , then let @xmath264 be a loop which is innermost on @xmath262 . then @xmath264 bounds a subdisk , @xmath18 , of @xmath262 . but , as no loop of @xmath271 bounds a compressing disk for @xmath272 in @xmath273 , it must be that @xmath264 also bounds a disk , @xmath124 , on @xmath272 . but then @xmath129 bounds a bubble for @xmath88 . as all such bubbles have been removed , this is a contradiction . we conclude then that if there are non - normal components of @xmath263 , then they must be arcs which run from some edge of @xmath260 back to itself . but then if @xmath198 is such an arc which is outermost on @xmath262 , then @xmath198 cobounds a subdisk of @xmath262 which is a relative compressing disk for @xmath88 , a contradiction . [ 348 ] let @xmath274 and @xmath116 be as given by the conclusion of lemma [ l : nlarcs ] . then for each odd @xmath77 , @xmath88 meets the boundary of every tetrahedron in normal curves of length 3 , 4 , and at most one curve on at most one tetrahedron of length 8 . this lemma is taken straight from @xcite . we refer the reader to this paper for its proof . the necessary assumptions are that @xmath88 meets every tetrahedron in normal arcs , and that there is no pair of disks which form a weak reduction for @xmath88 . a connected surface in a ball , @xmath236 , is _ unknotted _ if it is isotopic to a neighborhood of the graph obtained by coning @xmath30 points on @xmath275 to a point in the interior of @xmath236 . for the proof of theorem [ t : an ] we will need the following result : [ t : sch ] [ scharlemann @xcite ] suppose @xmath13 is a strongly irreducible heegaard surface for a 3-manifold , @xmath0 , and @xmath236 is a ball in @xmath0 . if @xmath276 is incompressible in @xmath277 , then @xmath278 is connected and unknotted . to prove theorem [ t : an ] , we will need to restate the previous result for heegaard surfaces of @xmath53 : [ t : relsch ] suppose @xmath13 is a strongly irreducible heegaard surface of @xmath279 and @xmath236 is a ball in @xmath107 . if @xmath276 is incompressible in @xmath280 , then @xmath278 is connected and unknotted . the proof of this theorem is exactly the same as scharlemann s proof of theorem [ t : sch ] . the point is that there is very little difference between a heegaard surface for @xmath0 and a heegaard surface for @xmath53 , as viewed from inside a ball that lies entirely in the complement of @xmath58 . [ t : an ] suppose @xmath0 is a submanifold of a 3-manifold , @xmath257 , with triangulation , @xmath249 , which is bounded by a normal or almost normal surface . let @xmath184 be a strongly irreducible hst splitting of @xmath248 , where @xmath11 normal implies @xmath148 is even , and @xmath11 almost normal implies @xmath148 is odd . then we may isotope @xmath116 , so that each thin level is a normal surface and each thick level is almost normal . isotope @xmath116 to satisfy the conclusion of lemma [ l : nlarcs ] . let @xmath77 be some odd number , so that @xmath88 is a thick level . let @xmath261 be some tetrahedron in @xmath249 . let @xmath67 be a copy of @xmath281 , pushed slightly into @xmath261 . we first claim that @xmath282 is incompressible in @xmath265 , to the outside of @xmath67 ( the side which is not contained in @xmath261 ) . suppose not , and let @xmath18 denote a compressing disk which meets @xmath281 minimally . as @xmath18 lies in @xmath265 , it must be that @xmath283 is a ( possibly empty ) collection of simple closed curves , for each 2-simplex , @xmath262 , of @xmath281 . let @xmath264 denote an innermost such simple closed curve . then @xmath264 bounds a subdisk of @xmath262 which is disjoint from @xmath88 ( since @xmath88 meets @xmath262 only in normal arcs ) . hence , we can use this subdisk to surger @xmath18 , thereby obtaining a compressing disk for @xmath282 which meets @xmath281 in one fewer curve . our conclusion , then , is that @xmath284 . but then @xmath18 lies between @xmath67 and @xmath281 . as these two surfaces are parallel , this is impossible . now , choose a complete collection of compressing disks for @xmath282 inside @xmath285 , and surger @xmath67 along this collection . we obtain in this way a collection of spheres , @xmath286 . @xmath287 bounds a ball , @xmath288 , in @xmath261 , and by definition , @xmath289 is incompressible in the complement of @xmath290 in @xmath291 . these are the conditions necessary to apply theorem [ t : relsch ] . the conclusion is that inside each @xmath288 , @xmath88 is a connected surface , which looks like the neighborhood of a graph which is the cone on some collection of points in @xmath292 . so , in particular , if @xmath293 is a single curve , then it bounds a disk in @xmath288 , and hence so does the corresponding curve in @xmath281 . = 5 in suppose there is some @xmath2 such that @xmath293 consists of three or more curves , of length 3 or 4 . the only ways this can happen are shown in figure [ f:2tubes ] . in all cases we see a compressing disk on one side of @xmath88 which is disjoint from a relative compressing disk on the other side ( see figure [ f : hlcomp ] ) . = 3.25 in now suppose that for some @xmath2 , @xmath293 consists of two normal curves , of length 3 or 4 . theorem [ t : relsch ] tells us that the picture must be two normal disks , tubed together by a single unknotted tube , as in figure [ f:1tube ] . note that in this situation , we see a relative compressing disk on one side , and a compressing disk on the other . hence , there can not be more than one place where we see this picture . otherwise , we d see either two disjoint compressing disks on opposite sides , or a compressing disk on one side disjoint from a relative compressing disk on the other . neither of these situations can happen for a surface which is strongly irreducible with respect to @xmath260 . = 3.75 in furthermore , suppose @xmath267 contains a curve of length 8 . then we see relative compressing disks on both sides as in figure [ f : octagon ] , and hence , there can not be a tube anywhere else ( including attached to this disk ! ) . =3 in we conclude that @xmath88 is made up of all normal disks , with the exception of either a single disk with a boundary curve of length 8 , or a single place where there are two normal disks tubed together by an unknotted tube . this is the precise definition of an almost normal surface . our proof is complete by noting that there must be an octagonal disk or a tube _ somewhere _ , because @xmath88 is a thick level , and hence there is at least one compressing or relative compressing disk on both sides . if there were no tubes or octagons , then we would not have this . the next theorem is an application of the results of this section . this result is an important step in the proof of rubinstein and thompson s recognition algorithm for @xmath7 . [ s3 ] [ rubinstein @xcite , thompson @xcite ] let @xmath249 be a triangulation of some 3-manifold , @xmath0 . if @xmath236 is a ball in @xmath0 such that @xmath275 is the unique normal 2-sphere in @xmath236 , and such that @xmath236 does not contain any of the vertices of @xmath294 , then @xmath236 contains an almost normal 2-sphere with an octagon as its exceptional piece . inside @xmath236 the 1-skeleton is simply a 1-manifold . choose an hst splitting , @xmath116 , of @xmath295 which is minimal . we first claim that @xmath116 is strongly irreducible . if not , then we can perform one of the four operations described in the proof of lemma [ l : untang ] to produce an hst splitting which is smaller than @xmath116 . we now call upon theorem [ t : an ] to isotope @xmath116 so that all thick levels are almost normal . suppose @xmath296 is such a thick level . let @xmath67 be the sphere component of @xmath296 which contains the exceptional piece , which is either an octagon or a tube . in the latter case @xmath297 is obtained from @xmath296 by compressing the tube , yielding two normal 2-spheres . as the only such normal 2-spheres are normally parallel to @xmath275 , the sequence @xmath298 is also an hst splitting of @xmath140 , contradicting the minimality of @xmath116 . we conclude then that the exceptional piece of @xmath67 was an octagon . in this section we tie together the main results of this paper to show how one can prove that various topologically interesting surfaces can be made almost normal . if @xmath0 is an irreducible submanfold of a triangulated 3-manifold then @xmath299 is a _ maximal normal 2-sphere _ in @xmath0 if there is no other normal 2-sphere which bounds a ball in @xmath0 which contains @xmath67 . if @xmath0 is not homeomorphic to @xmath142 or @xmath7 , and has a unique maximal normal 2-sphere , @xmath67 , then we recall two basic facts about @xmath67 from @xcite or @xcite . first , @xmath67 bounds a ball , @xmath236 , in @xmath0 which contains all of the vertices of @xmath300 . second , there are no almost normal 2-spheres in @xmath0 that are disjoint from @xmath236 . [ t : main ] suppose @xmath0 ( @xmath6 or @xmath7 ) is an irreducible submanifold of a triangulated 3-manifold which is bounded by a normal or almost normal surface , and contains at most one maximal normal 2-sphere . if @xmath1 is a strongly irreducible hst splitting of @xmath0 then each thin level is isotopic to a normal surface , and each thick level is isotopic to an almost normal surface . let @xmath67 denote a maximal normal 2-sphere in @xmath0 . @xmath67 bounds a ball , @xmath236 which contains all of the vertices of @xmath300 . collapsing @xmath236 to a point , @xmath301 , then turns @xmath260 into a properly embedded graph , @xmath136 , whose unique vertex is @xmath301 . note that any surface which is disjoint from @xmath301 after the collapse can be identified with a surface which is disjoint from @xmath236 before the collapse . let @xmath1 be a strongly irreducible hst splitting of @xmath0 . lemma [ l : untang ] implies that there is a strongly irreducible hst splitting , @xmath116 , of @xmath140 such that @xmath143=\{f_i\}$ ] . by lemma [ l : underlyingodd ] , for each thick ( thin ) level of @xmath1 there is a thick ( thin ) level of @xmath116 whose non-@xmath146 components are parallel to it . note that as @xmath116 is a hst splitting of @xmath140 , which , by definition , is a hst splitting of @xmath302 , we can also think of @xmath116 as a strongly irreducible hst splitting of @xmath303 . by theorem [ t : an ] we may isotope @xmath116 so that each thin level is normal and each thick level is almost normal . note that for each almost normal thick level , the exceptional piece is not contained in any 2-sphere component , because there are no almost normal 2-spheres in @xmath304 . hence , every thick level of @xmath1 is isotopic to an almost normal surface . [ hs ] [ rubinstein @xcite , stocking @xcite ] in any triangulation of a closed , irreducible 3-manifold other than @xmath7 , any strongly irreducible heegaard surface is isotopic to an almost normal surface . let @xmath13 be a strongly irreducible heegaard surface for a closed , irreducible 3-manifold , @xmath0 . then @xmath36 is a strongly irreducible hst splitting of @xmath0 with @xmath13 as a thick level . by the results of @xcite or @xcite , @xmath0 contains a unique maximal normal 2-sphere . hence , theorem [ t : main ] implies that @xmath13 can be made almost normal . [ sch ] [ schleimer @xcite ] suppose @xmath0 is a submanifold of a 3-manifold with triangulation , @xmath249 , such that @xmath0 is bounded by two topologically , but not combinatorially , parallel normal surfaces . if , in addition , @xmath0 does not contain a normal 2-sphere then @xmath0 contains a boundary parallel almost normal surface . by assumption , @xmath0 is homeomorphic to @xmath21 , for some subsurface , @xmath13 , of @xmath11 . the sequence , @xmath38 , is then a strongly irreducible hst splitting of @xmath0 , with @xmath305 as a thick level . theorem [ t : main ] now implies that @xmath305 can be made almost normal . let @xmath67 denote a maximal normal 2-sphere in an irreducible 3-manifold , @xmath306 . as in the proof of theorem [ hs ] , @xmath67 bounds a ball which we collapse to a point , transforming @xmath260 to a 1-vertex graph , @xmath136 . it is well known that if a 3-manifold contains a non - separating surface then it contains an essential one . we may also assume that we have such a surface , @xmath13 , that is disjoint from @xmath67 , and that this surface can be normalized in @xmath307 . let @xmath308 and @xmath309 denote two parallel copies of @xmath13 . finally , let @xmath0 denote the submanifold of @xmath306 cobounded by @xmath308 and @xmath309 which contains @xmath67 . as @xmath0 is cobounded by surfaces which are essential in @xmath0 there is a strongly irreducible hst splitting , @xmath29 , of @xmath0 such that @xmath310 and @xmath311 . suppose @xmath160 is a thick level of this splitting . then @xmath160 separates @xmath308 from @xmath309 in @xmath0 , but is non - separating in @xmath306 . theorem [ t : main ] may now be applied to @xmath0 to show that @xmath160 can be made almost normal . suppose @xmath118 is a normal surface in a triangulated 3-manifold , and @xmath13 is any surface disjoint from @xmath118 . then we say @xmath13 _ normalizes monotonically to _ @xmath118 if there is a @xmath58-compression body , @xmath59 , such that @xmath312 , @xmath313 , and @xmath314 . the following theorem characterizes when a normal surface compresses in terms of the existence of certain almost normal surfaces . in @xcite we use this to give a new algorithm to recognize when a given normal surface is compressible . [ compressible ] suppose @xmath0 is a submanifold of a triangulated irreducible 3-manifold which is bounded by a connected normal surface , and contains at most one maximal normal 2-sphere . if @xmath11 is compressible in @xmath0 then there is an almost normal surface in @xmath0 which is topologically parallel to @xmath11 , and normalizes monotonically to @xmath11 . let @xmath315 denote a maximal collection of compressing disks for @xmath11 in @xmath0 . let @xmath316 denote the surface obtained from @xmath11 by compressing along all disks in @xmath315 . let @xmath317 denote side of @xmath316 opposite @xmath11 in @xmath0 . alter @xmath316 by throwing away any 2-sphere component which bounds a ball in @xmath317 , and continue to denote this new surface as @xmath316 . if @xmath318 , then @xmath0 is a handlebody , and @xmath319 is a strongly irreducible hst splitting of @xmath0 . otherwise @xmath316 is essential in @xmath317 , and we can find a strongly irreducible hst splitting , @xmath320 , of @xmath317 , where @xmath321 and @xmath322 . in this case , let @xmath323 , so that @xmath324 is a strongly irreducible hst splitting of @xmath0 . in any case , @xmath323 is a thick level . hence , we may apply theorem [ t : main ] to show that it is isotopic to an almost normal surface in @xmath0 . [ normalheegaard ] if @xmath12 is a heegaard surface for a 3-manifold which is isotopic to a normal surface in the complement of a maximal normal 2-sphere then @xmath12 is isotopic to two , topologically parallel , almost normal surfaces . let @xmath325 denote a normal representative of @xmath12 . then @xmath325 bounds two compression bodies , @xmath9 and @xmath27 . both of these satisfy the hypotheses of theorem [ compressible ] , and hence , both contain almost normal surfaces which are topologically parallel to @xmath325 . the following result is important because it gives an exact characterization of handlebodies in terms of the existence of normal and almost normal surfaces . in @xcite we use this to give a new algorithm to recognize when a given almost normal surface is a heegaard surface . [ t : handlebody ] suppose @xmath9 is a submanifold of a triangulated 3-manifold such that @xmath326 is almost normal , and such that @xmath9 does not contain a normal 2-sphere . then @xmath9 is a handlebody if and only if there is a sequence of pairwise disjoint surfaces , @xmath327 , such that 1 . @xmath328 , 2 . @xmath329 , 3 . for each even @xmath77 , @xmath88 is normal , 4 . for each odd @xmath77 , @xmath88 is almost normal , and normalizes monotonically to @xmath134 and @xmath135 , and 5 . @xmath330 for all @xmath331 ( with respect to the ordering given in definition [ d : order ] ) . the proof that the existence of such a sequence implies that @xmath9 is a handlebody is left as an exercise to the reader . the proof of the converse is similar to that of theorem [ t : main ] . suppose that @xmath9 is a handlebody that satisfies the hypotheses . then @xmath332 is a strongly irreducible hst splitting of @xmath9 . lemma [ l : untang ] implies that that there is a strongly irreducible hst splitting , @xmath184 , of @xmath333 such that @xmath143=\{\emptyset , \partial w\}$ ] . this immediately implies that @xmath329 , @xmath334 , and that for all @xmath331 , @xmath330 . by theorem [ t : an ] we may isotope @xmath116 so that each surface of even index is normal , and each surface of odd index is almost normal . furthermore , by definition there is a @xmath260-compression body between consecutive thick and thin level , so each almost normal thick level normalizes monotonically to the next ( and previous ) normal thin level . as @xmath326 is assumed to be almost normal , we conclude that @xmath335 , for some @xmath30 .

we prove asymptotic formulas for the number of rational points of bounded height on smooth equivariant compactifications of the affine space . @equation@subsection @equation@subsection @tabletocline03pt plus2pt0pt _ _ [ subsection]proposition [ subsection]theorem [ subsection]corollary [ subsection]condition [ subsection]lemma [ subsection]assumption [ subsection]definition [ subsection]problem [ subsection]question [ subsection]conjecture [ subsection]remark [ subsection]remarks [ subsection]example [ subsection]exercise p h # 1 # 1 # 1#2#1 pdfoutput = 9 nous tablissons un dveloppement asymptotique du nombre de points rationnels de hauteur borne sur les compactifications quivariantes lisses de lespace affine .

math0005015

a theorem of northcott asserts that for any real number @xmath0 there are only finitely many rational points in the projective space @xmath1 with _ height _ smaller than @xmath0 . an asymptotic formula for this number ( as @xmath0 tends to infinity ) has been proved by schanuel @xcite . naturally , it is interesting to consider more general projective varieties and there are indeed a number of results in this direction . the techniques employed can be grouped in three main classes : * the classical circle method in analytic number theory permits to treat complete intersections of small degree in projective spaces of large dimension ( cf . for example @xcite ) ; * harmonic analysis on adelic points of reductive groups leads to results for toric varieties @xcite , flag varieties @xcite and horospherical varieties @xcite ; * elementary ( but nontrivial ) methods for del pezzo surfaces of degree 4 or 5 cf . de la bretche @xcite , and for cubic surfaces ( salberger , swinnerton - dyer , heath - brown , cf . @xcite and the references therein ; this research has been stimulated by a conjecture put forward by batyrev and manin . they proposed in @xcite an interpretation of the growth rate in terms of the mutual positions of the class of the line bundle giving the projective embedding , the anticanonical class and the cone of effective divisors in the picard group of the variety . peyre refined this conjecture in @xcite by introducing an adelic tamagawa - type number which appears as the leading constant in the expected asymptotic formula for the anticanonical embedding . batyrev and the second author proposed an interpretation of the leading constant for arbitrary ample line bundles , see @xcite . in this paper we consider a new class of varieties , namely _ equivariant compactifications of vector groups_. on the one hand , we can make use of harmonic analysis on the adelic points of the group . on the other hand , such varieties have a rich geometry . in particular , in contrast to flag varieties and toric varieties , they admit geometric deformations , and in contrast to the complete intersections treated by the circle method their picard group can have arbitrarily high rank . their geometric classification is a difficult open problem already in dimension 3 ( see @xcite ) . a basic example of such algebraic varieties is of course @xmath1 endowed with the action of @xmath2 by translations . a class of examples is provided by the following geometric construction : take @xmath3 endowed with the translation action of @xmath4 . let @xmath5 be a smooth subscheme of @xmath6 which is contained in the hyperplane at infinity . then , the blow - up @xmath7 contains the isomorphic preimage of @xmath4 and the action of @xmath4 lifts to @xmath8 . the rank of @xmath9 is equal to the number of irreducible components of @xmath5 . using equivariant resolutions of singularities , one can produce even more complicated examples , although less explicitly . our first steps towards this paper are detailed in @xcite and @xcite . there we studied the cases @xmath10 with @xmath5 a finite union of @xmath11-rational points and @xmath12 with @xmath5 a smooth hypersurface contained in the hyperplane at infinity . we now describe the main theorem of this article . _ let @xmath8 be an equivariant compactification of the additive group @xmath4 over a number field @xmath13 . unless explicitly stated , we shall always _ assume that @xmath8 is smooth and projective . _ the boundary divisor @xmath14 is a sum of irreducible components @xmath15 ( @xmath16 ) . we do not assume that they are geometrically irreducible . the picard group of @xmath8 is free and has a canonical basis given by the classes of @xmath17 . the cone of effective divisors consists of the divisors @xmath18 with @xmath19 for all @xmath20 . denote by @xmath21 the anticanonical line bundle on @xmath8 and by @xmath22 its class in @xmath9 . let @xmath23 be a class contained in the interior of the cone of effective divisors @xmath24 and @xmath25 the corresponding line bundle , equipped with a smooth adelic metric ( see [ defi.adelic ] for the definition ) . with the above notations , we have @xmath26 for all @xmath20 . denote by @xmath27 the associated exponential height on @xmath28 . let @xmath29 and let @xmath30 be the cardinality of @xmath31 put @xmath32 for example , one has @xmath33 and @xmath34 . we denote by @xmath35 the tamagawa number as defined by peyre in @xcite . [ theo.main ] the series @xmath36 converges absolutely and uniformly for @xmath37 and has a meromorphic continuation to @xmath38 for some @xmath39 , with a unique pole at @xmath40 of order @xmath30 . moreover , @xmath41 has polynomial growth in vertical strips in this domain . there exist positive real numbers @xmath42 and a polynomial @xmath43 $ ] of degree @xmath44 with leading coefficient @xmath45 such that the number @xmath46 of @xmath13-rational points in @xmath4 with height @xmath27 smaller than @xmath0 satisfies @xmath47 for @xmath48 . for @xmath49 we have @xmath50 . granted the smoothness assumption on the adelic metric , note that the normalization of the height in theorem [ theo.main ] can be arbitrary . by a theorem of peyre @xcite , this means that the points of bounded anticanonical height are equidistributed with respect to the tamagawa measure ( compatible with the choice of the height function ) in the adelic space @xmath51 . let us explain this briefly . fix a smooth adelic metric on the anticanonical line bundle , this defines a height function @xmath52 . let @xmath53 be the renormalized tamagawa measure on @xmath51 . a smooth positive function @xmath54 on @xmath51 determines another height function , namely @xmath55 . applied to such @xmath56 , the main theorem implies that @xmath57 theorem [ theo.main ] implies that the open subset @xmath4 does not contain any accumulating subvarieties . the proof proceeds as follows . first , we extend the height function to the adelic space @xmath58 . next we apply the additive poisson formula and find a representation of @xmath59 as a sum over the characters of @xmath60 of the fourier transforms of @xmath52 . the fourier transforms of @xmath52 decompose as products over all places @xmath61 of `` global '' integrals on @xmath62 which are reminiscent of igusa zeta functions . the most technical part of the paper is devoted to the evaluation of these products : meromorphic continuation and control of their growth in vertical strips . for this , we need to consider the special case where @xmath63 has strict normal crossings : this means that over an algebraic closure of @xmath13 , @xmath63 is a sum of smooth irreducible divisors meetings transversally . then , at almost all nonarchimedean places , we compute explicitly the local fourier transforms in terms of the reduction of @xmath8 modulo the corresponding prime . the obtained formulas resemble denef s formula in @xcite for igusa s local zeta function . for the remaining nonarchimedean places we find estimates . this leads to a proof of the meromorphic continuation of the fourier transforms at each character . to consider the general cas , where @xmath63 is not assumed to have strict normal crossings , we then introduce a proper modification @xmath64 which is a composition of equivariant blow - ups with smooth centers lying on @xmath63 such that @xmath65 is a smooth projective equivariant compactification of @xmath4 whose boundary divisor @xmath66 has strict normal crossings : over an algebraic closure of @xmath13 , @xmath66 is a sum of smooth irreducible divisors meeting transversally . considering @xmath67 , all the computations shall be made on @xmath65 . we also prove in lemma [ lemma.singularities ] that @xmath68 and @xmath69 . in the poisson formula , invariance properties of the height reduce the summation over @xmath70 to one over a lattice . the meromorphic continuation in part a ) of the theorem follows then from additional estimates for the fourier transforms at the infinite places . at this stage one has a meromorphic continuation of @xmath59 to the domain @xmath38 for some @xmath71 , with a single pole at @xmath40 whose order is less or equal than @xmath30 . it remains to check that the order of the pole is exactly @xmath30 ; we need to prove that the limit @xmath72 is strictly positive . for @xmath73 this is more or less straightforward : the main term is given by the summand corresponding to the trivial character and the tamagawa number defined by peyre appears naturally in the limit . for other @xmath74 we use the poisson formula again and relate the limit to an integral of the height over some subspace which is shown to be strictly positive . part b ) follows by a tauberian theorem . the work of the second author was partially supported by the nsa . let @xmath8 be a smooth projective equivariant compactification of the additive group @xmath4 of dimension @xmath75 over a field @xmath13 : @xmath8 is a ( smooth , projective ) algebraic variety endowed with an action of @xmath4 with a dense orbit isomorphic to @xmath4 . we will always assume that @xmath8 is smooth and projective . let @xmath14 denote the boundary divisor . we have a decomposition in irreducible components @xmath76 where @xmath77 is a finite set and for each @xmath78 , @xmath15 is an integral divisor in @xmath8 . we do not assume that the @xmath15 are geometrically irreducible , nor that they are smooth . there is however a similar description over any extension of @xmath13 . in particular , let @xmath79 be a separable closure of @xmath13 and let @xmath80 be the decomposition of @xmath81 in irreducible components . the natural action of @xmath82 on @xmath83 induces an action of @xmath84 on @xmath85 such that for any @xmath86 , @xmath87 . [ prop.picard ] one has natural isomorphisms of @xmath84-modules ( resp . @xmath84-monoids ) @xmath88 where @xmath89 is the monoid of classes of effective divisors of @xmath8 . subsequently , we identify the divisors @xmath17 ( and the corresponding line bundles ) with their classes in the picard group . these maps are equivariant under the action of @xmath84 . hence , it remains to show injectivity and surjectivity . for this we may assume @xmath90 . let @xmath91 be a line bundle on @xmath8 . as @xmath8 is smooth and @xmath92 , there is a divisor @xmath63 in @xmath8 not meeting @xmath4 such that @xmath93 . such a divisor @xmath63 is a sum @xmath94 . moreover , such a @xmath63 is necessarily unique . if @xmath95 for @xmath96 and @xmath97 , then the canonical rational section @xmath98 of @xmath99 is a rational function on @xmath8 without zeroes nor poles on @xmath4 . lemma ( see lemma [ lemm.rosenlicht ] below ) implies that @xmath98 is constant , so that @xmath100 . we remark that any effective cycle @xmath101 on @xmath8 is rationally equivalent to a cycle that does not meet @xmath4 . indeed , if @xmath102 is the parameter of a subgroup of @xmath4 isomorphic to @xmath103 , we can consider the specialization of the cycles @xmath104 when @xmath105 . [ lemm.rosenlicht ] if @xmath106 has neither zeroes nor poles on @xmath4 , then @xmath107 . \1 ) the @xmath84-module @xmath108 is a permutation module . in particular , plus3cm-30 plus-3 cm @xmath109 . \2 ) @xmath110 is a free @xmath111-module of finite rank equal to the number of @xmath84-orbits in @xmath85 . for any @xmath16 , we shall denote by by @xmath112 the algebraic closure of @xmath13 in the function field of @xmath15 . over @xmath112 , the irreducible components of @xmath15 are geometrically irreducible . we also denote by @xmath113 the dedekind zeta function of the number field @xmath112 . let @xmath54 be a nonzero linear form on @xmath4 viewed as an element of @xmath114 . its divisor can be written as @xmath115 where @xmath116 is the unique irreducible component of @xmath117 that meets @xmath4 and @xmath118 are integers . ( the divisor @xmath116 can also be seen as the closure in @xmath8 of the hypersurface of @xmath4 defined by @xmath54 . ) since @xmath116 is rationally equivalent to @xmath119 , the preceding proposition implies the following lemma : for any nonzero linear form @xmath54 and any @xmath16 , one has @xmath120 . if @xmath121 , let @xmath122 be the associated linear form and define @xmath123 [ prop.invsection ] let @xmath8 be a normal equivariant compactification of @xmath4 over @xmath13 . every line bundle @xmath91 on @xmath8 admits a unique @xmath4-linearization . if @xmath91 is effective then @xmath124 has a unique line of @xmath4-invariant sections . since @xmath4 has no nontrivial characters , it follows from proposition 1.4 and from the proof of proposition 1.5 in @xcite , chapter 1 , that any line bundle on @xmath8 admits a unique @xmath4-linearization . assume that @xmath125 and consider the induced action of @xmath4 on the projectivization @xmath126 . borel s fixed point theorem ( @xcite , theorem 10.4 ) implies that there exists a nonzero section @xmath127 such that the line @xmath128 is fixed under this action . as @xmath4 has no nontrivial characters , @xmath129 itself is fixed . the divisor @xmath130 is @xmath4-invariant ; therefore , @xmath130 does not meet @xmath4 and is necessarily a sum @xmath131 such that @xmath132 . because of proposition [ prop.picard ] , every other such section will be proportional to @xmath129 . we observe that if @xmath133 for integers @xmath134 , then the canonical section @xmath135 of @xmath136 is @xmath4-invariant . we now recall some facts concerning vector fields on equivariant compactifications of algebraic groups . let @xmath137 be a connected algebraic group over @xmath13 and @xmath138 its lie algebra of invariant vector fields . let @xmath8 be a smooth equivariant compactification of @xmath137 . denote by @xmath139 the boundary . we assume that @xmath63 is a divisor with strict normal crossings . let @xmath140 be the tangent bundle of @xmath8 . evaluating a vector field at the neutral element @xmath141 of @xmath137 induces a `` restriction map '' @xmath142 conversely , given @xmath143 , there is a unique vector field @xmath144 such that for any open subset @xmath145 of @xmath8 and any @xmath146 , @xmath147 . the map @xmath148 is a section of the restriction map . [ lemm.vectorfields ] for @xmath149 and for any @xmath143 the restriction @xmath150 is invariant under @xmath137 . ( this is of course peculiar to the additive group . for a general algebraic group @xmath137 , vector fields on @xmath137 of the form @xmath150 need not be invariant . ) [ prop.logvanishing ] let @xmath151 and fix a local equation @xmath152 of @xmath63 in a neighborhood @xmath145 of @xmath153 . then , for any @xmath143 , @xmath154 is a regular function in @xmath145 . by purity , it is sufficient to prove this when @xmath63 is smooth in a neighborhood of @xmath153 ( since @xmath8 is smooth , a function which is regular in the complement to a codimension 2 subscheme is regular everywhere ) . then , we can choose ( tale ) local coordinates @xmath155 in @xmath145 ( i.e. , elements of @xmath156 , @xmath157 is free over @xmath158 with basis @xmath159 ) such that @xmath160 . moreover , we can write uniquely @xmath161 for some functions let @xmath163 be the irreducible component of @xmath63 containing @xmath153 . necessarily , @xmath137 stabilizes @xmath163 so that the function @xmath164 is identically @xmath165 in a zariski neighborhood of @xmath141 in @xmath166 and a fortiori , @xmath167 vanishes on @xmath145 . by definition @xmath168 hence @xmath169 is a multiple of @xmath170 : there exists a unique @xmath171 such that @xmath172 and @xmath173 the lemma is proved . for @xmath174 $ ] or @xmath175 $ ] , the lie algebra @xmath138 has a canonical basis @xmath176 , given by the local parameter @xmath153 at the neutral element ( respectively @xmath165 and @xmath177 ) of @xmath137 . if we embed @xmath137 in @xmath178 equivariantly , we get @xmath179 for @xmath180 and @xmath181 for @xmath182 . we see that it vanishes at infinity ( being @xmath183 or @xmath184 , accordingly ) . this is a general fact , as the following lemma shows . [ lemm.rho ] there exist integers @xmath185 such that @xmath186 if @xmath149 , then for each @xmath20 , @xmath187 . we only prove that @xmath185 , because the sharper bound which is valid in the case of @xmath149 wo nt be used below . we refer to @xcite , theorem 2.7 for its proof . let @xmath188 be a basis of @xmath138 . then , @xmath189 is a global section of the line bundle @xmath190 . moreover , @xmath191 does not vanish on @xmath137 . therefore , we can write @xmath192 for nonnegative integers @xmath193 . necessarily , @xmath194 , hence we have to prove that these integers are positive . fix any @xmath153 in the smooth part of @xmath195 , so that there exists a unique @xmath16 such that @xmath196 . pick local coordinates @xmath155 in a neighborhood @xmath145 of @xmath153 in such a way that in @xmath145 , @xmath63 is defined by the equation @xmath197 . write @xmath198 , with @xmath199 . we have seen in proposition [ prop.logvanishing ] that for any @xmath200 , @xmath201 . hence , @xmath202 . let @xmath13 be a number field and @xmath203 its ring of integers . denote by @xmath204 the completion of @xmath13 at a place @xmath61 , by @xmath205 the ring of integers in @xmath204 if @xmath61 is nonarchimedean , by @xmath206 the ring of adeles of @xmath13 and by @xmath207 the ( restricted ) product of @xmath204 over the nonarchimedean places @xmath61 . the valuation in @xmath204 is normalized in such a way that for any haar measure @xmath208 on @xmath204 and any measurable subset @xmath209 , @xmath210 . in particular , it is the usual absolute value for @xmath211 , its square for @xmath212 , it satisfies @xmath213 if @xmath214 and if @xmath215 is the norm map , @xmath216 . let @xmath8 be an equivariant compactification of @xmath4 as above and @xmath91 a line bundle on @xmath8 , endowed with its canonical linearization . [ defi.adelic ] a _ smooth adelic metric _ on @xmath91 is a family of @xmath61-adic norms @xmath217 on @xmath91 for all places @xmath61 of @xmath13 satisfying the following properties : 1 . if @xmath61 is archimedean , then @xmath217 is @xmath218 ; 2 . if @xmath61 is nonarchimedean , then @xmath217 is locally constant ; + _ ( i.e. , the norm of any local nonvanishing section is @xmath218 , resp . locally constant ) _ 3 . there exists an open dense subset @xmath219 , a flat projective @xmath145-scheme @xmath220 extending @xmath8 together with an action of @xmath221 extending the action of @xmath4 on @xmath8 and a linearized line bundle @xmath91 on @xmath220 extending the linearized line bundle on @xmath8 , such that for any place @xmath61 lying over @xmath145 , the @xmath61-adic metric on @xmath91 is given by the integral model . let @xmath61 be a nonarchimedean valuation of @xmath13 and @xmath217 a locally constant @xmath61-adic norm on @xmath91 . then the stabilizer of @xmath222 , i.e. , the set of @xmath223 which act isometrically on @xmath222 , is a compact open subgroup of @xmath224 . first we assume that @xmath91 is effective . by proposition [ prop.invsection ] , we have a nonzero invariant global section @xmath129 . if @xmath225 and @xmath226 denote the action and the second projection @xmath227 , respectively , endow the trivial line bundle @xmath228 on @xmath229 with the tensor - product metric . this is a locally constant metric on the trivial line bundle . the function on @xmath230 given by @xmath231 is the norm of the canonical basis @xmath177 and therefore extends to a locally constant function on @xmath232 . its restriction to the compact subset @xmath233 is uniformly continuous . since it is locally constant and equal to @xmath177 on @xmath234 , there exists a neighborhood of @xmath234 on which it equals @xmath177 . such a neighborhood contains a neighborhood of the form @xmath235 , where @xmath236 is a compact open subgroup of @xmath224 . this proves the lemma in the effective case . in the general case , we write @xmath237 for two effective line bundles ( each having a nonzero global section ) . we can endow @xmath238 with any locally constant @xmath61-adic metric and @xmath239 with the unique ( necessarily locally constant ) @xmath61-adic metric such that the isomorphism @xmath240 is an isometry . by the previous case , there exist two compact open subgroups @xmath241 and @xmath242 contained in @xmath224 which act isometrically on @xmath239 and @xmath238 , respectively . their intersection acts isometrically on @xmath91 . [ prop.stab ] if @xmath91 is endowed with a smooth adelic metric then for all but finitely many places @xmath61 of @xmath13 the stabilizer of @xmath222 is equal to @xmath224 . therefore , their product over all finite places of @xmath13 is a compact open subgroup of @xmath243 . it suffices to note that if @xmath61 lies over the open subset @xmath219 given by the definition of an adelic metric then the stabilizer of @xmath222 equals @xmath224 . from now on we choose adelic metrics on the line bundles @xmath91 in such a way that the tensor product of two line bundles is endowed with the product of the metrics ( this can be done by fixing smooth adelic metrics on a @xmath111-basis of @xmath9 and extending by linearity ) . we shall denote by @xmath244 the compact open subgroup of @xmath243 stabilizing all these metrized line bundles on @xmath8 . we also fix an open dense @xmath245 , a flat and projective model @xmath220 over @xmath145 and models of the line bundles @xmath91 such that the chosen @xmath61-adic metrics for all line bundles on @xmath8 are given by these integral models . let @xmath8 be an algebraic variety over @xmath13 and @xmath246 an adelically metrized line bundle on @xmath8 . the associated _ height function _ is defined as @xmath247 where @xmath248 is any @xmath13-rational section of @xmath249 not vanishing at @xmath153 . the product formula ensures that @xmath250 does not depend on the choice of the @xmath13-rational section @xmath129 ( though the local heights @xmath251 do ) . in section [ section.metrizations ] we have defined simultaneous metrizations of line bundles on equivariant compactifications of @xmath4 . this allows to define compatible systems of heights @xmath252 fix for any @xmath16 some non zero @xmath4-invariant section @xmath253 of @xmath254 . we can then extend the height pairing @xmath52 to a pairing @xmath255 by mapping @xmath256 recall that @xmath244 denotes the compact open subgroup of @xmath58 stabilizing all line bundles on @xmath8 together with their chosen metrization . by construction , we have the following proposition . [ prop.heights ] the height pairing @xmath52 defined in is @xmath244-invariant in the first component and ( exponentially ) linear in the second component . [ prop.est ] assume that @xmath91 is in the interior of the effective cone of @xmath9 ( i.e. , @xmath132 for some positive integers @xmath257 ) . then , for any real @xmath0 , there are only finitely many @xmath258 such that @xmath259 . in view of lemma [ lemm.rho ] , this applies to the anticanonical line bundle @xmath21 . let @xmath260 be an ample line bundle on @xmath8 and let @xmath261 be a sufficiently large integer such that @xmath262 is effective . it follows from the preceding section that @xmath263 has a section @xmath129 which does not vanish on @xmath4 . this implies that the function @xmath264 is bounded from above on @xmath70 . therefore , there exists a constant @xmath265 such that for any @xmath258 , @xmath266 we may now apply northcott s theorem and obtain the desired finiteness . the same argument shows that the rational map given by the sections of a sufficiently high power of @xmath91 is an embedding on @xmath4 . the main tool in the study of asymptotics for the number of points of bounded height is the _ height zeta function _ @xmath267 [ prop.abs-conv ] there exists an non - empty open subset @xmath268 such that @xmath269 converges absolutely to a bounded holomorphic function in the tube domain @xmath270 in the complex vector space @xmath271 . fix a basis @xmath272 of @xmath9 consisting of ( classes of ) line bundles lying in the interior of the effective cone of @xmath9 . let @xmath273 denote the open set of all linear combinations @xmath274 such that for some @xmath275 , @xmath276 . fix some ample line bundle @xmath260 . it is well known that the height zeta function of @xmath8 relative to @xmath260 converges for @xmath277 big enough , say @xmath278 . in the proof of proposition [ prop.est ] we may choose some @xmath261 which works for any @xmath279 , so that for any @xmath275 @xmath280 since @xmath281 is bounded from below on @xmath70 it follows that for any @xmath282 @xmath283 therefore , the height zeta function converges absolutely and uniformly on the tube domain @xmath284 . the following sections are devoted to the study of analytic properties of @xmath285 . we recall basic facts concerning harmonic analysis on the group @xmath286 over the adeles @xmath287 ( cf . , for example , @xcite ) . for any prime number @xmath288 , we can view @xmath289 as the @xmath288-sylow subgroup of @xmath290 and we can define a local character @xmath291 of @xmath292 by setting @xmath293 at the infinite place of @xmath294 we put @xmath295 ( here @xmath296 is viewed as an element in @xmath297 ) . the product of local characters gives a character @xmath298 of @xmath299 and , by composition with the trace , a character of @xmath300 . if @xmath301 , recall that @xmath122 is the corresponding linear form on @xmath58 and let @xmath302 . this defines a map @xmath303 . it is well known that this map is a pontryagin duality . the subgroup @xmath304 is discrete , cocompact and we have an induced pontryagin duality @xmath305 we fix selfdual haar measures @xmath306 on @xmath307 for all @xmath61 . we refer to @xcite for an explicit normalization of these measures . we will use the fact that for all but finitely many @xmath61 the volume of @xmath308 with respect to @xmath306 is equal to 1 . thus we have an induced selfdual haar measure @xmath309 on @xmath310 and the product measure @xmath311 on @xmath58 . the fourier transform ( in the adelic component ) of the height pairing on @xmath312 is defined by @xmath313 we will use the poisson formula in the following form ( cf . @xcite , p. 280 ) . [ theo.poisson ] let @xmath314 be a continuous function on @xmath315 such that both @xmath314 and its fourier transform @xmath316 are integrable and such that the series @xmath317 converges absolutely and uniformly when @xmath318 belongs to @xmath319 . then , @xmath320 the following lemma ( a slight strengthening of proposition [ prop.abs-conv ] ) verifies the two hypotheses of the poisson formula [ theo.poisson ] concerning @xmath52 . let @xmath8 be a smooth projective equivariant compactification of @xmath4 and @xmath52 the height pairing defined in section [ section.heights ] . there exists a nonempty open subset @xmath321 such that for any @xmath322 the series @xmath323 converges absolutely , uniformly in @xmath324 and locally uniformly in @xmath325 . moreover , for such @xmath325 , the function @xmath326 is integrable over @xmath58 . since the natural action of @xmath4 on @xmath9 is trivial , for any @xmath327 the function @xmath328 is a height function for @xmath91 , induced by a `` twisted '' adelic metric . when @xmath318 belongs to some compact subset @xmath329 of @xmath58 there exists a constant @xmath330 such that for any @xmath331 and any @xmath332 , @xmath333 indeed , the quotient @xmath334 defines a bounded continuous function of @xmath335 . ( only at a finite number of places do they differ , and at these places @xmath61 , a comparison is provided by the compactness of @xmath62 . ) as @xmath60 is compact , one may take for @xmath329 any compact set containing a fundamental domain . this implies the uniform convergence of the series once @xmath325 belongs to the tube domain of proposition [ prop.abs-conv ] . the integrability follows at once : for such @xmath325 , @xmath336 the following proposition follows from the invariance of the height under the action of the compact @xmath244 . [ prop.vanishing ] for all characters @xmath337 of @xmath58 which are nontrivial on the compact subgroup @xmath244 of @xmath243 and all @xmath325 such that @xmath338 is integrable , we have @xmath339 consequently , the set @xmath340 of all @xmath341 such that @xmath337 is trivial on @xmath342 is is a sub-@xmath203-module of @xmath70 , commensurable with @xmath343 . consequently , provided the poisson formula applies and we have a formal identity for the height zeta function : @xmath344 the last four sections of the paper are concerned with analytic arguments leading to the evaluation of @xmath345 and to the fact that equation holds for all @xmath325 in some tube domain . it will be necessary to assume that the boundary divisor @xmath63 has strict normal crossings , which means that over the separable closure @xmath79 of @xmath13 , @xmath63 is a union of smooth irreducible components meeting transversally . we sum up these sections in the following proposition . recall that for @xmath121 , @xmath346 is the set of @xmath16 such that @xmath347 . in particular , @xmath348 . also , for any @xmath349 , we denote by @xmath273 the open tube domain in @xmath350 defined by the inequalities @xmath351 . finally , let @xmath352 denote any norm on the real vector space @xmath353 . [ prop.main-calc ] assume that @xmath63 has strict normal crossings . for any @xmath354 , @xmath355 is integrable on @xmath58 . for any @xmath356 and @xmath357 there exists a holomorphic bounded function @xmath358 on @xmath359 such that for any @xmath360 @xmath361 for any @xmath362 there exist constants @xmath363 and @xmath364 such that for any @xmath365 and any @xmath366 one has the estimate @xmath367 for the proof of our main theorem , we need to extend the estimates of proposition [ prop.main-calc ] when @xmath63 is not assumed to have strict normal crossings . [ lemma.singularities ] proposition [ prop.main-calc ] remains true without the hypothesis that @xmath63 has strict normal crossings . equivariant resolution of singularities ( in char . @xmath165 , see @xcite or @xcite ) shows that there exists a composition of equivariant blow - up with smooth centers @xmath368 such that * @xmath65 is a smooth projective equivariant compactification of @xmath4 ; * @xmath369 is equivariant ( hence induces an isomorphism on the @xmath4 ) ; * the boundary divisor @xmath370 has strict normal crossings . the map @xmath369 induces a map from the set @xmath371 of irreducible components of @xmath66 to @xmath77 . moreover , @xmath369 has a canonical section obtained by associating to a component of @xmath63 its strict transform in @xmath65 . this allows to view @xmath77 as a subset of @xmath371 , the complementary subset @xmath372 consisting of exceptional divisors . via the identifications of @xmath9 with @xmath373 and @xmath374 with @xmath375 , the map @xmath376 on line bundles induces a linear map @xmath377 which is a section of the first projection @xmath378 . ( we also extend @xmath200 by linearity to the vector spaces of complexified line bundles . ) concerning metrizations , it is possible to endow line bundles on @xmath65 with adelic metrics which extend the metrics coming from @xmath8 via @xmath376 . remark that with such a choice , the height function on @xmath58 and its fourier transform on @xmath65 extend those on @xmath8 . for instance , for any @xmath379 and any @xmath380 , one has @xmath381 as @xmath200 maps @xmath382 into @xmath383 , it follows in particular that @xmath355 is integrable for @xmath354 and that for any @xmath301 and any @xmath354 , one has @xmath384 unless @xmath369 is an isomorphism , @xmath369 is not smooth and @xmath385 has a nonzero coefficient at each exceptional divisor . for any effective @xmath74 , this implies that @xmath386 lies on the boundary of @xmath387 . ( recall that the classes of @xmath388 and @xmath389 in @xmath373 and @xmath375 are denoted @xmath390 and @xmath391 . ) it follows that @xmath392 and that @xmath393 can be identified with @xmath394 . consequently , any `` exceptional '' factor @xmath395 for @xmath396 extends to a holomorphic bounded function on @xmath397 this shows that proposition [ prop.main-calc ] still stands true without the hypothesis that @xmath63 has strict normal crossings . one has also the following result which shall allow for the determination of the constant in the asymptotics . [ prop.tamagawa ] denote by @xmath398 the tamagawa number of @xmath8 defined by the adelic metrization on @xmath399 . one then has the equality @xmath400 in particular , it is a strictly positive real number . recall first peyre s definition @xcite of the tamagawa measure in this context , which extends weil s definition in the context of linear algebraic groups @xcite . for any place @xmath61 of @xmath13 and any local non - vanishing differential form @xmath401 on @xmath62 , the self - dual measure on @xmath402 induces a local measure @xmath403 on @xmath62 . these local measures glue to define a global measure @xmath404 on @xmath62 . recall that artin s @xmath405-function of @xmath406 is equal to @xmath407 . using deligne s theorem on weil s conjectures ( but it will also follow from our calculations ) , peyre has shown that the euler product @xmath408 converges absolutely and he defined the tamagawa number of @xmath8 as @xmath409 ( note however that peyre apparently does nt use the selfdual measure for his definition in _ loc . _ , but inserts the appropriate correcting factor @xmath410 . ) on the open subset @xmath402 , we can define the local measure with the differential form @xmath411 induced by the canonical coordinates of @xmath4 . let @xmath412 be such that @xmath413 . on @xmath402 , we thus have @xmath414 so that , the boundary subset being of measure @xmath165 , @xmath415 moreover , it follows from these results and the absolute convergence of the euler products of dedekind zeta functions for @xmath416 that for any @xmath417 , @xmath418 is integrable on @xmath51 and that one has @xmath419 the estimates of proposition [ prop.bbb ] below allow us to interchange the limit @xmath420 and the infinite product @xmath421 , so that @xmath422 because of the product formula . it follows from these results that the poisson formula applies for any @xmath354 and that @xmath423 hence , @xmath424 this last series is a sum of holomorphic functions on @xmath425 and taking @xmath426 $ ] , the estimate in proposition [ prop.main-calc ] implies that the series converges locally uniformly in that domain . therefore , we have shown that the function @xmath427 has a holomorphic continuation to @xmath425 with polynomial growth in vertical strips . the restriction of @xmath269 to a line @xmath428 ( with @xmath26 for all @xmath20 ) gives @xmath429 where @xmath430 is a holomorphic function for @xmath431 such that @xmath432 providing a meromorphic continuation of @xmath433 to this domain . the rightmost pole is at @xmath434 and its order is less or equal than @xmath435 . let @xmath8 be a smooth projective equivariant compactification of @xmath4 over a number field @xmath13 . for any strictly effective @xmath436 , the height zeta function @xmath433 converges absolutely for @xmath437 , has a meromorphic extension to the left of @xmath438 , with a single pole at @xmath438 of order @xmath439 . when @xmath73 , @xmath440 , @xmath441 and proposition [ prop.tamagawa ] implies the equality @xmath442 hence the order of the pole is exactly @xmath443 . using a standard tauberian theorem , we deduce an asymptotic expansion for the number of rational points in @xmath4 of bounded anticanonical height . [ theo.rho ] let @xmath8 be a smooth projective equivariant compactification of @xmath4 over a number field @xmath13 . there exists a real number @xmath39 and a polynomial @xmath444 $ ] of degree @xmath445 such that the number of @xmath13-rational points on @xmath446 of anticanonical height @xmath447 satisfies @xmath448 the leading coefficient of @xmath449 is given by @xmath450 let @xmath451 be an effective class and @xmath25 the corresponding line bundle on @xmath8 equipped with a smooth adelic metric as in section [ section.metrizations ] . the function @xmath452 is holomorphic for @xmath453 and admits a meromorphic continuation to the left of @xmath454 . recall from the previous section that it has a pole of order @xmath455 at @xmath454 , where @xmath30 is the cardinality of @xmath456 ( geometrically , the integer @xmath30 is the codimension of the face of the effective cone @xmath457 containing the class @xmath458 . ) in this section we prove that the order of the pole of the height zeta function @xmath433 at @xmath459 is exactly @xmath435 and derive our main theorem . denote by @xmath460 the sub - series @xmath461 it follows from the calculations above that @xmath462 since all other terms converge to zero when @xmath463 , with uniform convergence of the series . the set @xmath464 of @xmath465 such that @xmath346 contains @xmath466 is a sub - vector space . let @xmath467 be the @xmath13-sub - vector space defined by the corresponding linear forms @xmath468 . then , the autoduality on @xmath58 identifies @xmath464 with the dual of @xmath469 . we apply the poisson summation formula and obtain @xmath470 ( the justification of poisson summation formula is as in section [ section.poisson ] . ) [ prop.tamlambda ] for each @xmath471 there exists a strictly positive real @xmath472 such that @xmath473 it is sufficient to prove the proposition when @xmath474 as the integrals for different values of @xmath153 are comparable . the zariski closure @xmath5 of @xmath475 in @xmath8 need not be smooth . nevertheless , we may introduce an equivariant proper modification @xmath64 such that the zariski closure @xmath476 of @xmath475 is a smooth equivariant compactification whose boundary is a divisor with strict normal crossings obtained by intersecting the components of the boundary of @xmath65 with @xmath477 . the arguments of lemma [ lemma.singularities ] show that we do not lose any generality by assuming this on @xmath8 itself . there exist integers @xmath478 for @xmath16 such that @xmath479 moreover , @xmath480 if and only if @xmath481 . we prove this by induction on the codimension of @xmath5 in @xmath8 . if @xmath482 , the adjunction formula shows that @xmath483 for @xmath484 the nonnegative integer @xmath485 is equal to @xmath165 if and only if @xmath486 . the lemma is proved . let @xmath487 be the ( nonrenormalized ) tamagawa measure on @xmath488 : by definition , on @xmath489 , @xmath490 . we have to estimate @xmath491 the integral in the limit is of the type studied in prop [ prop.main-calc ] , but on the subgroup @xmath5 . denote therefore by primes @xmath492 restrictions of objects from @xmath8 to @xmath5 . thanks to the previous lemma , one has @xmath493 , @xmath494 and and @xmath495 . therefore , with notations from proposition [ prop.main-calc ] , @xmath496 the proposition is proved . we now can conclude as in the case of the anticanonical line bundle : [ theo.lambda ] let @xmath497 be the class of a line bundle contained in the interior of the effective cone @xmath89 , and equip the corresponding line bundle @xmath25 with a smooth adelic metric as in section [ section.metrizations ] . there exist a polynomial @xmath498 $ ] of degree @xmath44 and a real number @xmath499 such that the number @xmath500 of @xmath13-rational points on @xmath4 of @xmath25-height @xmath447 satisfies @xmath501 the leading term of @xmath502 is equal to @xmath503 it is compatible with the description by batyrev and tschinkel in @xcite . in the remaining sections we have to prove prop . [ prop.main-calc ] . first we define a set of bad places @xmath504 : as in section [ section.metrizations ] , fix a good model @xmath220 ( flat projective @xmath145-scheme ) over a dense open subset @xmath245 . in particular , we assume that @xmath220 extends @xmath8 as an equivariant compactificatation of @xmath4 whose boundary consists of the schematic closures of the @xmath15 . moreover , we can assume that for any @xmath20 , the chosen section @xmath253 of @xmath505 extends to @xmath220 and that its norm coincides with the one that can be defined with this model at all finite places dominating @xmath145 . in particular , at such places @xmath61 , the local height functions restricted to @xmath402 are invariant under @xmath224 . we can also assume that for any finite place @xmath61 dominating @xmath145 , the ( nonrenormalized ) tamagawa measure @xmath404 is given on @xmath402 by @xmath506 . let @xmath504 be the set of places @xmath61 of @xmath13 such that either * @xmath61 is archimedean , or * @xmath61 does nt dominate @xmath145 , or * the residual characteristic of @xmath61 is @xmath507 or @xmath508 , or * the volume of @xmath224 with respect to @xmath509 is not equal to 1 ( equivalently , @xmath13 is ramified at @xmath61 ) , or * over the local ring @xmath205 , the union @xmath510 is not a transverse union of smooth relative divisors over @xmath145 . for @xmath366 , let @xmath511 be the corresponding linear form on @xmath4 , considered as an element of @xmath114 or of @xmath512 and @xmath513 its divisor . we denote by @xmath514 the set of all valuations @xmath61 such that either * @xmath61 is contained in @xmath504 , or * @xmath513 is not flat over @xmath145 , _ i.e. _ if @xmath515 denotes the maximal ideal of @xmath61 in @xmath203 , @xmath516 belongs to @xmath517 . finally , recall that for any @xmath349 , @xmath273 is the tube domain in @xmath350 consisting of classes @xmath518 such that for any @xmath20 , @xmath519 . in section [ section.estimates ] we prove estimates for finite products @xmath520 . in sections [ section.trivchar ] and [ section.otherchars ] we compute explicitly the local fourier transforms @xmath521 for all @xmath522 . _ in the following four sections we temporarily assume that the irreducible components of the boundary of @xmath8 are geometrically irreducible . _ in section [ section.nonsplit ] , we shall explain how our results extend to the general case , ( still assuming that @xmath63 has strict normal crossings ) . [ prop.estim ] let @xmath523 be any finite set of places of @xmath13 containing the archimedean places . for every @xmath524 and any @xmath362 there exists a constant @xmath525 such that for all @xmath526 and all @xmath527 one has the estimate @xmath528}}{\prod_{v|\infty}(1+\norm{\mathbf a}_v)^n},\ ] ] where @xmath529 . we subdivide the proof of this proposition into a sequence of lemmas . [ prop.integrability ] the function @xmath530 is integrable on @xmath402 if and only if @xmath531 ( _ i.e. _ , @xmath532 for all @xmath16 ) . moreover , for all @xmath524 and all nonarchimedean places @xmath61 , there exists a constant @xmath533 such that for all @xmath534 and all @xmath535 one has the estimate @xmath536 without loss of generality , we can assume that @xmath537 . now , using an analytic partition of unity on @xmath62 and the assumption that the boundary @xmath538 is a divisor with strict normal crossings , we see that it suffices to compute the integral over a relatively compact neighborhood of the origin in @xmath539 ( denoted by @xmath540 ) on which we have a set coordinates @xmath155 so that in @xmath540 the divisor @xmath63 is defined by the equations @xmath541 for some @xmath542 . in @xmath540 there exist continuous bounded functions @xmath543 ( for @xmath78 ) such that @xmath544 the integral over @xmath545 is comparable to an integral of the same type with functions @xmath546 replaced by @xmath165 . the lemma is now a consequence of the following well - known lemma . let @xmath329 be a local field . the function @xmath547 is integrable on the unit ball in @xmath329 if and only if @xmath548 . we may assume @xmath549 . choose @xmath5500;1\right[$ ] such that the annulus @xmath551 has positive measure in @xmath329 and let @xmath552 . then , we have @xmath553 it follows that the integral over @xmath329 converges if and only if the geometric series @xmath554 converges , that is if @xmath555 . let @xmath61 be an archimedean place of @xmath13 and let @xmath556 $ ] . for any @xmath362 and any @xmath499 , there exists a constant @xmath557 such that for any @xmath558 and all @xmath366 ( and @xmath559 ) , we have @xmath560 let us assume for the moment that @xmath211 . we shall write @xmath561 , @xmath562 and @xmath563 . then , in the domain @xmath564 one has @xmath565 now integrate by parts : for any @xmath566 we have @xmath567 and by induction @xmath568 for any @xmath20 , let us define @xmath569 ; it is a @xmath218 function on @xmath570 . let @xmath153 be a point of @xmath571 and let @xmath572 be the set of @xmath16 such that @xmath196 . if @xmath573 is a local equation of @xmath15 in a neighborhood @xmath145 of some @xmath574-point of @xmath15 , then there is a @xmath218 function @xmath575 on @xmath145 such that for @xmath576 we have @xmath577 it then follows from propositions [ lemm.vectorfields ] and [ prop.logvanishing ] that for each @xmath20 , @xmath578 extends to a @xmath218 function on @xmath571 . from the equality @xmath579 we deduce by induction the existence of an isobaric polynomial @xmath580 $ ] of degree @xmath581 ( with the convention that each @xmath582 has weight @xmath288 ) and such that @xmath583 this implies that there exists a constant @xmath584 such that @xmath585 choosing @xmath275 such that @xmath586 is maximal gives @xmath587 , hence an upper bound @xmath588 where @xmath589 is the trivial character and @xmath590 some positive constant . to conclude the proof it suffices to remark that for any @xmath499 , @xmath591 is bounded on the set @xmath592 ( but the bound depends on @xmath593 ) . the case @xmath212 is treated using a similar integration by parts . ( the exponent @xmath507 on the numerator comes from the fact that for a complex place @xmath61 , @xmath217 is the square of a norm . ) in this section we consider only @xmath594 . let @xmath595 be the maximal ideal , @xmath596 and @xmath597 . recall that we have fixed a good model @xmath220 over some @xmath598 . to simplify notations we will write @xmath599 , @xmath600 etc . the following formula is an analogue of denef s formula in ( * ? ? ? * thm 3.1 ) for igusa s local zeta function . [ theo.trivcalc ] for all @xmath594 and all @xmath601 we have @xmath602 for @xmath603 we get @xmath604 , the expected local density at @xmath61 . we split the integral along residue classes mod @xmath515 . let @xmath605 and @xmath606 so that @xmath607 . we can introduce local ( tale ) coordinates @xmath608 ( @xmath609 ) and @xmath610 ( @xmath611 , @xmath612 ) around @xmath613 such that locally , the divisor @xmath614 is defined by the vanishing of @xmath608 . then , the local tamagawa measure identifies with the measure @xmath615 on @xmath616 . if @xmath617 denotes the fixed measure on @xmath402 , one has the equality of measures on @xmath618 : @xmath619 consequently , @xmath620 where the last equality follows from @xmath621 summing these equalities for @xmath622 gives the desired formula . to estimate further @xmath623 and the product over all places , we need to give an estimate for the number of @xmath624-points in @xmath17 which is uniform in @xmath61 . [ lemm.uniform ] there exists a constant @xmath625 such that for all @xmath594 and all @xmath626 we have the estimates : * if @xmath627 , @xmath628 ; * if @xmath629 , @xmath630 . we use the fact that for any projective variety @xmath5 of dimension @xmath631 and degree @xmath632 the number of @xmath624-points is estimated as @xmath633 ( see , e.g. , lemma 3.9 in @xcite ) . since @xmath8 is projective , all @xmath634 can be realized as subvarieties in some projective space . this proves the second part . to prove the first part , we apply lang - weil s estimate @xcite to the geometrically irreducible smooth @xmath145-scheme @xmath614 . [ prop.bbb ] for all @xmath499 there exists a constant @xmath635 such that for any @xmath636 and any finite place @xmath637 , @xmath638 using the uniform estimates from lemma [ lemm.uniform ] we see that in the formula for @xmath591 , each term with @xmath639 is @xmath640 , with uniform constants in @xmath641 . turning to the remaining terms , we get @xmath642 finally , we have the desired estimate . for @xmath8 as above and @xmath643 , the ( multi - variable ) artin @xmath405-function is given by @xmath644 from standard properties of dedekind zeta functions , we conclude that @xmath645 admits a meromorphic continuation and that it has polynomial growth in vertical strips . for @xmath646 we get the usual artin @xmath405-function of @xmath9here a power of the dedekind zeta function which has been used in the regularization of the tamagawa measure in prop . [ prop.tamagawa ] . [ coro.trivtamagawa ] for all @xmath356 there exists a holomorphic bounded function @xmath647 on @xmath648 such that for any @xmath649 one has @xmath650 moreover , there exist constants @xmath362 and @xmath635 such that for any @xmath325 in @xmath359 , @xmath651 for any place @xmath61 of @xmath13 , let @xmath652 the previous proposition shows that in @xmath653 , the euler product @xmath654 converges absoluetely to a holomorphic and bounded function @xmath54 . for any @xmath354 , one has @xmath655 it suffices to define @xmath656 the polynomial growth of @xmath647 in vertical strips follows from the boundedness of @xmath54 in @xmath359 and from the fact that dedekind zeta functions have a polynomial growth in such vertical strips . our aim here is to compute as explicitely as possible the fourier transforms of local heights with character @xmath337 . in general , this will be only possible up to some error term . the calculations in the preceding section allow us to strengthen proposition [ prop.estim ] . for @xmath121 , we denote by @xmath657 a norm of @xmath516 in the real vector space defined by extension of scalars via the diagonal embedding @xmath658 . for any @xmath499 , there exist an integer @xmath659 and for any @xmath660 , a constant @xmath661 such that for any @xmath662 and all @xmath535 we have @xmath663 for @xmath664 the local integrals converge absolutely in the domain under consideration and are bounded as in proposition [ prop.estim ] . for @xmath637 , we have shown that there exists a constant @xmath665 such that for all @xmath666 and all @xmath667 one has the estimate @xmath668 this implies that @xmath669 is bounded independently of @xmath516 , @xmath670 and @xmath671 . for any nonzero @xmath672 , there is a trivial estimate @xmath673 which implies that for some constant @xmath674 , @xmath675 using proposition [ prop.estim ] , we have for all @xmath676 , @xmath677 } \prod_{v|\infty } ( 1+\norm{\mathbf a}_v)^{\kappa - n } \\ & \leq c'(s,\eps , n ) ( 1+\norm{s})^{n[f:\q ] } ( 1+\norm{\mathbf a}_\infty)^{(\kappa - n)[f:\q ] } .\end{aligned}\ ] ] replacing @xmath581 by @xmath678 $ ] and @xmath674 by @xmath679 $ ] concludes the proof of the proposition . for the explicit calculation at good places , let us recall some notations : @xmath680 is the maximal ideal , @xmath681 a uniformizing element , @xmath624 the residue field , @xmath682 , @xmath683 , @xmath599 , @xmath684 , @xmath600 . as in section [ section.geometry ] , let @xmath685 be a linear form on @xmath4 and write @xmath686 . we have also defined @xmath687 and @xmath688 . [ prop.estimatewithcharacters ] there exists a constant @xmath635 independent of @xmath526 such that for any @xmath689 , @xmath690 similarly to the proof of proposition [ prop.bbb ] in the preceding section , this proposition is proved by computing the integral on residue classes . let @xmath691 and @xmath692 . we now consider three cases : _ case 1 . _ the sum of all these contribution is equal to the integral over @xmath224 : @xmath694 _ case 2 . @xmath695 and @xmath696 . _ we can introduce analytic coordinates @xmath608 and @xmath610 around @xmath613 such that locally @xmath697 with @xmath698 . then , we compute the integral of @xmath699 as @xmath700 [ lemm.cancel ] for all integers @xmath701 and @xmath702 and all @xmath698 , @xmath703 if @xmath704 , the computation runs as follows @xmath705 for @xmath706 , let @xmath707 . since we assumed @xmath204 to be unramified over @xmath708 , @xmath709 . we will integrate over disks @xmath710 for @xmath711 suitably chosen . indeed , if @xmath712 and @xmath713 , @xmath714 hence , if @xmath715 is chosen such that @xmath716 @xmath717 we can find such an @xmath718 if and only if @xmath719 , _ i.e. _ @xmath720 . if @xmath721 , this is true since @xmath706 . if @xmath722 , one has @xmath723 since we assumed @xmath724 this lemma implies the following trichotomy : @xmath725 _ case 3 . @xmath639 or @xmath726 and @xmath727 . _ under some transversality assumption , we could compute explicitely the integral as before . we shall however content ourselves with the estimate obtained by replacing @xmath298 by 1 in the integral . the total contribution of these points will therefore be smaller than @xmath728 finally , the fourier transform is estimated as follows : @xmath729 with an `` error term '' @xmath730 smaller than the expression in . it is now a simple matter to rewrite this estimate as in the statement of [ prop.estimatewithcharacters ] . we deduce from these estimates that @xmath731 has a meromorphic continuation : [ coro.otherchars ] for any @xmath356 and @xmath732 there exists a holomorphic bounded function @xmath358 on @xmath359 such that for any @xmath360 @xmath733 moreover , for any @xmath362 there exist constants @xmath363 and @xmath364 such that for any @xmath365 , one has the estimate @xmath734 in this section we extend the previous calculations of the fourier transform to the nonsplit case , _ i.e. _ when the geometric irreducible components of @xmath735 are no longer assumed to be defined over @xmath13 . let @xmath736 and @xmath737 . fix for any orbit @xmath738 some element @xmath20 and let @xmath112 be the field of definition of @xmath15 . then there exist an open neighborhood @xmath145 of @xmath153 , tale coordinates around @xmath153 over @xmath79 , @xmath739 ( @xmath609 ) and @xmath740 such that @xmath739 is a local equation of @xmath15 and such that the induced morphism @xmath741 descends to an tale morphism @xmath742 chose some element @xmath20 in each orbit @xmath744 and fix a local equation @xmath739 for the corresponding @xmath15 which is defined over the number field @xmath112 . then if @xmath745 ( for some @xmath746 ) is another element in the orbit of @xmath20 , set @xmath747 . this is well defined since we assumed that the equation @xmath739 is invariant under @xmath748 . let @xmath61 be a place of @xmath13 . the above lemma allows us to identify a neighborhood of @xmath153 in @xmath62 ( for the analytic topology ) with a neighborhood of @xmath165 in the product @xmath751 , the local heights @xmath752 being replaced by @xmath753 where @xmath754 is a smooth function . the assertions of section [ section.estimates ] still hold in this more general case and the proofs require only minor modifications . however the calculations of sections [ section.trivchar ] and [ section.otherchars ] have to be redone . in the general case , we introduce the multi - variable artin @xmath405-function of @xmath108 as @xmath761 its restriction to the line @xmath762 is the usual artin @xmath405-function of @xmath108 . it has a pole of order @xmath763 at @xmath764 .

we study the radio / x - ray correlation in cyg x-3 . it has been known that the soft and hard x - ray fluxes in the hard spectral state are correlated positively and negatively , respectively , with the radio flux . we show that this implies that the observed kev flux ( which is a fair approximation to the bolometric flux ) is completely uncorrelated with the radio flux . we can recover a positive correlation ( seen in other sources and expected theoretically ) if the soft x - rays are strongly absorbed by a local medium . then , however , the intrinsic x - ray spectrum of cyg x-3 in its hard state becomes relatively soft , similar to that of an intermediate spectral state of black - hole binaries , but not to their true hard state . we also find the radio spectra in the hard state of cyg x-3 are hard on average , and the flux distributions of the radio emission and soft x - rays can be described by sums of two log - normal functions . we compare cyg x-3 with other x - ray binaries using colour - colour , colour - eddington ratio and eddington ratio - radio flux diagrams . we find cyg x-3 to be spectrally most similar to grs 1915 + 105 , except that cyg x-3 is substantially more radio loud , which appears to be due to its jet emission enhanced by interaction with the powerful stellar wind from the wolf - rayet donor . [ firstpage ] radiation mechanisms : non - thermal radio continuum : stars stars : individual : cyg x-3 stars : winds , outflows x - rays : binaries .

1508.02049

0.4 cm cyg x-3 is an binary system possessing a number of unique and puzzling characteristics . it is the brightest radio source among x - ray binaries @xcite , showing extremely strong radio outbursts and resolved jets ( e.g. , @xcite ) . it is the only x - ray binary that is certainly powered by accretion for which high - energy @xmath1-ray emission has been detected at a high statistical significance @xcite . it is the only known binary in the galaxy containing both a compact object and a wolf - rayet star @xcite . its orbital period of @xmath2 d is unusually short for a high - mass binary , which indicates a past spiral - in episode during a common - envelope evolutionary stage . given the value of @xmath3 and estimates of the masses and mass - loss rate , the compact object orbits within the donor photosphere . in spite of its discovery as early as in 1966 @xcite , the nature of its compact object remains uncertain , due to the lack of fully reliable determination of the mass functions and inclination . based on the currently available information , @xcite found the range of the compact - object mass of @xmath4 , allowing either a neutron star ( ns ) or a low - mass black hole ( bh ) . the nature of the compact object is of significant interest , in particular because cyg x-3 is a certain progenitor of a double compact system , after its wolf - rayet star explodes as a supernova @xcite . the presence of a bh over an ns is favoured by considering the x - ray and radio emission ( @xcite , hereafter szm08 ; @xcite ) . @xcite have shown that the differences between the shape of the x - ray spectra of cyg x-3 in its hard spectral state from those of confirmed bh binaries could be accounted for by compton scattering in the strong stellar wind from the donor . that model also could account for the lack of high frequencies @xcite in the power spectra of cyg x-3 . the distance to cyg x-3 has been estimated as @xmath59 kpc @xcite : at this @xmath6 , its observed ( i.e. , not corrected for intrinsic absorption ) bolometric x - ray luminosity is @xmath7 erg s@xmath8 . its x - ray spectra have been classified into six states by @xcite , based on _ rossi x - ray timing explorer _ ( _ rxte _ ) pointed observations . here , we adopt a simpler classification into three states , hard , intermediate and soft , based on the relationship between the 35 kev and 1450 kev fluxes . an unresolved issue is the value of the absorbing column within the stellar wind of the donor . for example , @xcite found that the hard - state spectral data from _ integral _ can be fitted with either a high or low absorbing column . the two resulting models then significantly differ in both the spectral slope of the soft x - rays and in the bolometric flux . in the former and latter models , the intrinsic spectrum is soft , with the photon index of @xmath9 , and hard , with @xmath10 ( defined by the photon spectrum , @xmath11 ) . only the latter behaviour corresponds to the standard definition of the hard state , e.g. , @xcite . correlated radio x - ray behaviour of cyg x-3 was discovered in the soft ( 16 kev ) and hard ( 20100 kev ) x - rays by @xcite and @xcite , respectively . the correlations were then studied by @xcite , @xcite and szm08 . curiously , the soft x - rays are positively correlated with the radio flux in the hard state whereas the hard x - rays are anticorrelated . this differs , e.g. , from another accreting high - mass binary , cyg x-1 , in which both the soft and hard x - rays are positively correlated with the radio emission in the hard state ( @xcite ; hereafter zspl11 ) . this raises the issue of how the bolometric flux ( dominated by the x - rays , emitted most likely by the accretion flow ) is correlated with the radio emission , originating in the jets of this system . theoretically , a positive correlation is expected @xcite in a steady accretion - flow jet system . generally , this is because both the accretion luminosity and the jet power should be positively correlated with the accretion rate , and the radio flux is positively correlated with the jet power @xcite . in this work , we study in detail the relationship between the radio emission and x - rays . in particular , we estimate the bolometric flux of the system , its relation to the radio flux , and the radio spectra . we then consider implications of our findings for the nature of this accreting source , and compare cyg x-3 to other x - ray binaries . we study here radio monitoring data at 15 ghz from the ryle telescope , which cover mjd 4923153905 , and the arcminute microkelvin imager ( ami ) , mjd 5461257211 . the combined data set contains 76776 measurements . the ami large array is the re - built and reconfigured ryle telescope . @xcite describe the normal operating mode for the ryle telescope in the monitoring observations ; the observing scheme for the ami large array is very similar . the new correlator has a useful bandwidth of about 4 ghz ( compared with 0.35 ghz for the ryle ) , but the effective centre frequency is similar . the data are subject to variations in the flux calibration of @xmath12 per cent from one day to another ; we hereafter assume a 10 per cent fractional error . also , we use the 2.25 ghz and 8.3 ghz monitoring data @xcite from the green bank interferometer ( gbi ) . the data are for mjd 5040951823 and contain 11022 measurements . we average all of the data used here over 1 day ( except for the batse , see below ) . we have found that 1 day is a time scale which both strongly reduces the measurement errors and averages over the orbital modulation ( with the orbital period of 1/5 d ) , see also @xcite . at the same time , it retains most of the correlated radio and x - ray variability . for soft and medium x - rays , we use daily - averaged monitoring data from the all - sky monitor ( asm ; @xcite ) on board _ rxte _ , giving count rates in three channels at energies of 1.33 kev , 35 kev and 512.2 kev , spanning mjd 5008755915 . we also use the data from the monitor of all - sky x - ray image ( maxi ; @xcite ) on board of _ international space station _ ( _ iss _ ) . we use daily averaged data during mjd 5505857401 . the data are given as photon fluxes in three channels , 24 , 410 and 1020kev . these are converted from the count rates by scaling to crab assuming its photon spectrum of the form of @xmath13 where the h column density is @xmath14 @xmath15 , @xmath16 is the bound - free cross section per h atom averaged over the cosmic composition ( calculated using wabs in xspec , @xcite ) , @xmath17 @xmath15 s@xmath8 , and @xmath18 . we convert the maxi data to the energy fluxes using the above . for other data sets used here , we use the above form of the crab spectrum except for taking into account a break to @xmath19 at @xmath20 kev . for hard x - rays , we use daily - averaged data from the _ swift _ burst alert telescope ( bat ; @xcite ) survey data . they were retrieved from the heasarc public archive and processed using the bat_imager software @xcite . they form a 14195 kev 8-channel count - rate light curve , similar to that used in @xcite , but covering now a substantially longer interval of mjd 5334857279 . the energy fluxes for the asm and bat data have been obtained by using scaling to the crab spectrum as given above . we note that @xcite have estimated that , if the photon index of the actual source differs from the crab index by 1 or the column density is up to 10 times larger , the asm energy - range flux differs by at most 30 per cent . we thus expect similarly minor effects on the scaling of the maxi and bat data . we also use data from the burst and transient source experiment ( batse ; @xcite ) on board of _ compton gamma ray observatory _ ( _ cgro _ ) . these monitoring data have been obtained using the earth occultation analysis technique @xcite . we use the same data as used by @xcite and szm08 , which are given as 3-day average energy fluxes in the 2050 , 50100 and 100230 kev bands for the interval of mjd 4836351686 . in appendix [ lc ] , we present and discuss the histograms of the flux distributions and the light curves for the x - ray monitors and for the ryle / ami . 0.4 cm fig . [ states ] defines the spectral states of cyg x-3 used in this paper via the relationship between the soft and hard x - rays , similarly to @xcite . the horizontal dashed line in both panels delineates the upper boundary of the soft state , with @xmath21@xmath22 kev @xmath15 s@xmath8 . the vertical dotted lines delineate the upper boundary of the hard state , with @xmath23@xmath24 kev @xmath15 s@xmath8 for the asm data , and @xmath25@xmath26 kev @xmath15 s@xmath8 for the maxi data . the intermediate state is defined for the fluxes above both boundaries . [ rx4 ] shows the energy dependence of the 15-ghz / x - ray relationship based on the asm / bat ( top ) and maxi / bat ( bottom ) data . each data point corresponds to a daily average of the x - ray flux and the corresponding average radio flux . in figs . [ rx4](a c ) and ( e g ) , the data points showing a strong positive correlation seen on the left correspond to the hard state . the dashed lines in figs . [ rx4](b , e ) show the upper flux limits of the hard state in the 35 and 24 kev bands , respectively , defined above . as found by szm08 , as the soft x - ray flux increases above the upper hard - state limit , the radio flux becomes highly variable , dropping to very low values and increasing to very high values , corresponding to major radio flares . [ rx4 ] does not show the temporal behaviour ( shown in szm08 ; see also appendix [ lc ] ) , but we do see the radio flux varying by 4 orders of magnitude while the soft x - ray flux changes only within a factor of @xmath02 . as we also see in fig . [ rx4 ] , the slope of the radio / x - ray correlation changes with the energy , becoming increasingly steep with increasing energy up to @xmath010 kev , and then changing its sign to negative in the 1450 and 2075 kev bands . the vertical dashed lines in figs . [ rx4](d , h ) show the lowest flux in those bands corresponding to the hard state . the change of the sign does happen between the asm and bat energy ranges , as we have checked by looking at the 1420 kev band , which already shows a negative correlation ( though with more scatter ) . since that dependence is similar to those in the 1450 and 2075 kev bands , we do not show it here . we have found the steepening of the correlation with the energy in the hard state in the simultaneous asm / gbi data , at radio frequencies of 2.25 and 8.3 ghz , to be very similar to those shown in figs . [ rx4](a c , d g ) . also , we have correlated the batse and 15 ghz data . we have found an anticorrelation at 20100 kev range in the hard state similar to those shown in figs . [ rx4](d , h ) , but with much higher statistical scatter . the 100230 kev batse data have then very limited statistics , given the spectrum of cyg x-3 being strongly cut off at those energies . given the similarities of those results to those using the asm , bat and maxi and the 15-ghz data , we do not show them here . 0.1 cm the existence of the anticorrelation between radio and hard x - rays in the hard state raises the issue of how the bolometric luminosity is correlated with the radio flux . the observed @xmath27 spectra in the hard state peak at @xmath020 kev ( e.g. , szm08 ) , and the spectral slope in soft x - rays is @xmath28 , which gives rise to the classification of this state as ` hard ' . thus , the anticorrelation takes place in the spectral region in which at least @xmath01/2 of the observed bolometric flux is emitted . figs . [ rx_bol](a , c ) show the relationship between the 1.3100 and 2100 kev flux , respectively , and the radio flux . as it can be seen in the spectra of , e.g. , szm08 , these x - ray energy ranges contain most ( though not all ) of the observed x - ray energy and thus they give fair approximations to the bolometric flux . ( for the asm / bat data , we have neglected here a minor contribution from the 12.214 kev band , not covered by either asm or bat . ) we see that there is hardly any correlation , with the data just forming a long vertical strip . obviously , there is substantial absorption in cyg x-3 , which needs to be taken into account . its amount is , however , rather uncertain . @xcite found that their hard - state data , from _ integral _ , can be fitted by either weak absorption , with the resulting intrinsic slope below 10 kev of @xmath10 , or strong one , with @xmath9 . @xcite found that their hard - state _ bepposax _ data can be fitted with a weak absorption ( @xmath10 ) . as we find here , only relatively strong absorption , corresponding to @xmath29 , gives reasonable indices of the hard - state correlation of the intrinsic bolometric flux , @xmath30 , with the radio flux , @xmath31 , @xmath32 . thus , the hard state of cyg x-3 would then be similar to an intermediate state of bh binaries . the correlation index for most bh low - mass x - ray binaries ( lmxbs ) in the hard state for the 39 kev x - ray range is @xmath330.8 @xcite . then , some lmxbs have been found to have @xmath34 , in particular the bh lmxb h1743 - 322 @xcite , with the two branches intersecting at @xmath35@xmath36 erg s@xmath8 , see fig . 9 in @xcite . as discussed in @xcite , the two branches of the correlation correspond to the jet being powered by either radiatively inefficient ( @xmath37 ) or radiatively efficient ( @xmath34 ) accretion . [ rx_bol](b , d ) show the correlation between the radio flux and the intrinsic , unabsorbed , broad - band x - ray flux for a choice of the intrinsic absorber yielding @xmath38 . we use the ionized absorber model zxipcf of @xcite in xspec , which is based on the treatment of ionized absorption by xstar @xcite . it is characterized by a column density , @xmath39 , and an ionization parameter , @xmath40 , where @xmath41 is the medium density and @xmath42 is the distance from the ionizing source to the medium . we assumed @xmath43 @xmath15 and @xmath44 erg cm s@xmath8 , which values are similar to those fitted in @xcite to the average _ rxte _ x - ray spectra for the orbital phases around the superior and inferior conjunctions . in addition , we took into account interstellar absorption with @xmath45 @xmath15 @xcite . we have then created average x - ray spectra in the three spectral states , defined as in fig . [ states ] , based on the simultaneous asm / bat and maxi / bat data . the spectra as observed and the intrinsic ones , i.e. , corrected for absorption , are shown in fig . [ spectra ] . we see that the intrinsic spectra in the hard state are almost flat in @xmath46 below @xmath010 kev , with @xmath47 . the slope of the observed radio / x - ray correlation corresponding to the hard - state asm / bat spectrum shown in fig . [ spectra](b ) , and fitted as described in zspl11 , is @xmath48 , which we show by the dotted line in fig . [ rx_bol](b ) . given the overlap of the points in different states , the value of @xmath49 also bears a substantial systematic error . this slope is very similar to that found in the lower - branch ( radiatively efficient ) radio / x - ray correlation of lmxbs , see above . the corresponding slope for the maxi / bat data , shown in fig . [ rx_bol](d ) , is somewhat higher , which appears to be due to the lower boundary of the maxi data being 2 kev compared to about 1.3 kev for the asm data and the steepening of the correlation with the increasing energy up to @xmath010 kev , found here . in order to get more insight into the nature of the hard state , we consider the radio spectra based on three measurements , at 2.25 , 8.3 and 15 ghz . we calculate the spectral index , @xmath50 , between two frequencies , defined by @xmath51 . we show the results in fig . [ alpha ] . we identify the hard state using the simultaneous asm data , by our standard criterion of @xmath235 kev@xmath52 kev @xmath15 s@xmath8 , which we show by the black symbols . note that some points not identified in this way due to the lack of simultaneous asm data also belong to that state . we see that the hard state corresponds to the range of the 2.25-ghz fluxes of @xmath5335150 mjy and @xmath54 to @xmath55 , with the average of @xmath56 . these radio spectra are harder than those of cyg x-1 , with @xmath57 @xcite . however , they are similar to those of those exhibited by some bh binaries . for example , the lmxb gx 3394 has shown @xmath50 in the hard state during the 201011 outburst saturated at @xmath58 ( @xcite , see their fig . 2 ) . ) based on two radio frequencies vs. the flux of the lower frequency for 1-d average gbi and ryle data . ( a ) 2.258.3 ghz ; ( b ) 2.2515 ghz . the hard state points identified by the simultaneous asm data is shown by black symbols . only the 2.25 , 8.3 ghz and 35 kev flux measurements with the significance of @xmath59 have been taken into account . , width=219 ] ) based on two radio frequencies vs. the flux of the lower frequency for 1-d average gbi and ryle data . ( a ) 2.258.3 ghz ; ( b ) 2.2515 ghz . the hard state points identified by the simultaneous asm data is shown by black symbols . only the 2.25 , 8.3 ghz and 35 kev flux measurements with the significance of @xmath59 have been taken into account . , width=219 ] thus , the hard state in cyg x-3 is characterized by hard radio spectra , similar to the case of bh binaries . such spectra are emitted by jets with the partially synchrotron self - absorbed emission @xcite . on the other hand , the soft state shows a wide range of @xmath50 , with majority of them with @xmath50 between @xmath60 and 0 . the emission with @xmath61 is likely optically thin synchrotron . however , the presence of spectra with @xmath62 during the intermediate and soft states , see fig . [ alpha ] , indicates that synchrotron self - absorption is also taking place in those states . we have found , based on the asm , maxi , bat and ryle / ami data , that the cyg x-3 radio flux in the hard state is highly variable and completely uncorrelated with the measured bolometric flux , as shown in figs . [ rx_bol](a , c ) . however , a positive correlation is recovered when taking into account absorption of soft x - rays provided it is strong enough , as shown in figs . [ rx_bol](b , d ) . we have found that an absorber similar to that fitted to the _ rxte _ pointed data in @xcite gives a reasonable slope of the @xmath30-@xmath63 correlation , @xmath64 , similar to those found in the lower branch of the lxmb radio / x - ray correlation @xcite and in cyg x-1 in the hard state @xcite . however , this index can be reduced due to an effect following from the physics of a continuous jet ( e.g. , @xcite ) coupled with the free - free absorption in the stellar wind . namely , the height above the binary plane of a jet at which the bulk of partially synchrotron self - absorbed emission at a given @xmath65 emerges increases with the increasing @xmath66 . this owes to the associated increase of the density of the jet relativistic electrons , which in turn increases the synchrotron self - absorption optical depth within the jet . for the standard model of @xcite and assuming that the radio emission is partially synchrotron self - absorbed ( which leads to @xmath67 ) , the predicted correlation between the height and the flux was derived in @xcite , @xmath68 , where @xmath69 . here @xmath70 is the index of the power - law electron distribution in the jet , and @xmath71 is the emitted radio flux . as we have shown in section [ radio ] , the radio spectrum of cyg x-3 in the hard state has indeed @xmath67 , and the above formalism approximately applies . then , at a higher flux , and thus a greater height , the photon path through the wind becomes shorter , and the orbit - averaged optical depth , @xmath72 , for free - free absorption becomes lower . a further effect is an increase of the temperature of the wind in response to the increased x - ray flux , which also reduces @xmath72 . the presence of these effects leads to the intrinsic correlation index , @xmath73 , becoming lower than the one observed , @xmath74 where @xmath31 is the observed radio flux ( zspl11 ) . the value of @xmath42 can be estimated based on the dependence of the depth of an orbital modulation of the radio flux on @xmath65 . in the case of cyg x-1 , @xcite estimated @xmath75 , leading to @xmath76 in that binary . in the case of cyg x-3 , the orbital modulation of the hard - state radio emission is very weak , with the full depth of the modulation of 15 ghz emission of @xmath77 per cent ( zdziarski et al . , in preparation ) . this is much weaker than that in the hard state of cyg x-1 , where it is @xmath78 per cent @xcite . this weakness is most likely due to two effects . first , the separation between the binary components in cyg x-3 is about an order of magnitude lower than that of cyg x-1 ( given the orbital period of the former is much shorter than of the latter ) . second , the height along the jet at which a given radio frequency is emitted is most likely higher in cyg x-3 than in cyg x-1 , due to the accretion rate in the former being much higher than in the latter . therefore , the asymmetry of the paths of 15-ghz photons through the wind between the superior and inferior conjunction in cyg x-3 is much lower than in cyg x-1 . this is then likely to lead to the value of @xmath72 at 15 ghz being much lower in cyg x-3 than in cyg x-1 , in spite of the wind mass loss rate being about an order of magnitude higher . given that , we expect the value of @xmath42 in cyg x-3 to be much closer to unity than that in cyg x-1 . thus , the above effect of the intrinsic correlation index being lower than the observed one is probably minor in cyg x-3 . therefore , it is unlikely that the value of @xmath49 , which we have estimated correcting for the x - ray absorption , is significantly higher than our value of @xmath531.5 . the anticorrelation between hard and soft x - rays within the hard state of cyg x-3 is a consequence of the soft spectrum becoming softer when brighter . this leads to a pivoting in the @xmath531020 kev energy range . then , a flux increase in soft x - rays corresponds to a decrease in hard x - rays , and vice versa . this pivoting pattern is seen , e.g. , in fig . 2 of szm08 , see their spectra 1 and 2 . on the other hand , cyg x-1 has only moderate spectral variability in the hard state . based on the energy dependence of the fractional variability , see , e.g. , fig . 12 in @xcite , the hard - state pivot is at @xmath0100 kev . consequently , there is no observed anticorrelation between the soft and hard x - rays in the hard state , as found by zspl11 . as calculated in @xcite , the x - ray variability patterns in the hard state are well explained by originating from thermal comptonization . if the dominant driver of the variability is in a changing flux of seed photons , the spectrum softens with the increase of the total flux , leading to a pivoting behaviour . the lower value of the pivot in cyg x-3 than in cyg x-1 is most likely related to the hard state in the former being in fact intermediate . in cyg x-1 , the transition from the hard to the intermediate state is indeed characterized by pivoting around @xmath01020 kev , see fig . 13 in @xcite . given our results in section [ rx ] and the discussion in section [ intrinsic ] , the absorption in cyg x-3 has to be strong , and the intrinsic soft x - ray spectrum in the hard state has to be rather soft , @xmath47 . thus , the hard state of cyg x-3 corresponds indeed to the intermediate state of bh binaries . as found by zspl11 ( see their figs . 8a , b ) , the positive correlation between the radio and bolometric x - ray fluxes in cyg x-1 is present in both the hard and intermediate states , and breaks down only in the soft state . we thus expect a similar situation in cyg x-3 , which can explains the observed correlation even if its spectral state is intermediate . if the apparent hard state in cyg x-3 is intrinsically intermediate , this can explain the difference in its spectral shape with respect to bh binaries in the hard state without scattering in the stellar wind , invoked by @xcite . the apparent peak at @xmath010 kev in the hard - state @xmath46 spectrum is then an artefact of strong absorption , and the spectrum at higher energies is due to non - thermal , rather than thermal , comptonization @xcite . also , the hard state of cyg x-3 appears then to be similar to the hardest spectral state of the bh lmxb grs 1915 + 105 , see , e.g. , fig . 4 in @xcite , where the hardest spectra are similar to the intrinsic hard - state spectra of cyg x-3 shown in figs . [ spectra](b , d ) . in order to determine the relationship of cyg x-3 to other accreting x - ray binaries , we compare quantitatively their overall x - ray and radio properties . we first compare the positions on the x - ray colour - colour and colour - luminosity diagrams . we use the compilations of the data from @xcite , @xcite , @xcite and @xcite for galactic bh and weakly - magnetized ns binaries . those authors used the soft and hard colours defined by the energy flux ratio of the bands 46.4 kev to 34 kev and 9.716 kev to 6.49.7 kev , and the bolometric luminosity in the eddington units . we have updated the values of the distances and bh masses according to the current best estimates , in particular those of @xcite and @xcite . we calculate the colours and eddington ratios , @xmath79 , for cyg x-3 assuming @xmath80 kpc , @xmath81 @xcite and using the asm - bat absorption - corrected spectra shown in fig . [ spectra](b ) . here @xmath82 is the eddington luminosity , calculated for pure he ( @xmath83 ) for cyg x-3 , and for the cosmic composition ( @xmath84 ) for other binaries . to calculate the fluxes within a part of a bin , we assume the spectral shape within a bin to be flat in @xmath46 . the spectra at @xmath85 kev are most likely dominated by disc blackbody emission , and we thus add a contribution from those energies assuming @xmath86 . we obtain the bolometric fluxes of 2.3 , 5.5 and @xmath87 erg @xmath15 s@xmath8 in the hard , intermediate and soft state , respectively , which correspond to the bolometric ( isotropic ) luminosities of 1.8 , 4.2 , @xmath88 erg s@xmath8 , and the eddington ratios of 0.18 , 0.42 and @xmath89 , respectively . we can see in fig . [ lcd ] that indeed the colours and eddington ratios of the spectral states of cyg x-3 ( at @xmath90 ) are within the range occupied by grs 1915 + 105 ( using the data of @xcite ) , on both the colour - colour and colour-@xmath79 diagrams . on the other hand , @xcite used similar diagrams and argued that the most likely bh mass is @xmath91 . in fact , that mass may also be compatible with our values of colour-@xmath79 , and it would correspond to moving the points on fig . [ lcd](b ) down to a region populated by other bh binaries , but not grs 1915 + 105 . however , the cyg x-3 points on the colour - colour diagram would still be within those of grs 1915 + 105 , which represents an inconsistency . also , such a high bh mass is not found in other bh binaries ( see @xcite for the lack of dynamical evidence for a high mass bh in ic10 x-1 ) , as well as it is in contradiction with the later results of @xcite . compared to @xcite , the absorber in cyg x-3 adopted by us is stronger , which would move the values of the colours in that paper to lower values , making them more compatible with those of grs 1915 + 105 . we also note that the distance to grs 1915 + 105 has been revised down from 12.5 kpc to @xmath92 kpc @xcite , which has decreased its eddington ratios by a factor of 2.1 ( at the best - fit distance ) . this results in our cyg x-3 points for @xmath81 to be entirely within the region occupied by grs 1915 + 105 . we have also compared the colours and the eddington ratios of cyg x-3 to those of weakly magnetized ns binaries , assuming now @xmath93 . we have found the results of this comparison compatible with cyg x-3 containing an ns , with its colours and eddington ratios similar to those of z sources . however , the strength of the radio emission relative to the x - ray in cyg x-3 exceeds that of all of bh binaries , see below , and it is much much higher than for ns ones @xcite . ; black error bars ) compared to that of cyg x-1 ( 1.86 kpc , @xmath94 ; the red triangles , green squares and blue circles for the hard , intermediate and soft state , respectively , from zspl11 ) , gx 3394 ( 8 kpc , @xmath95 ; green squares fitted by the green dashed line , from @xcite , but multiplied by 1.25 to account for those measurements being at 8.6 ghz ) and grs 1915 + 105 ( 8.6 kpc , @xmath96 ; cyan squares with error bars ) , with the magenta squares corresponding to the plateau @xmath97 state using the criteria described in section [ xray ] . ] we then compare the radio vs. the bolometric luminosity relationship of cyg x-3 with those of selected bh binaries , see fig . [ rbol ] , where the @xmath98 axis shows @xmath99 at 15 ghz . the black error bars shows the points for cyg x-3 at @xmath81 and @xmath80 kpc , where the shown values of @xmath100 have been corrected with respect to @xmath101@xmath102 in the same way as described above for fig . [ lcd](b ) . we compare first cyg x-3 with gx 3394 , for which the radio / x - ray correlation is the best studied one @xcite . unfortunately , the published correlations are given for narrow bands of either 39 kev or 110 kev only . as we have argued above , a narrow - band x - ray flux is likely to be significantly affected by spectral variability ( as it is very significantly the case in cyg x-3 , and significantly in cyg x-1 , see zspl11 ) . the only radio/@xmath100 correlation available for an lmxb appears to be that of @xcite for gx 3394 , which we show in green points . the magenta line shows the best fit power - law dependence , with @xmath103 . the radio fluxes are at 8.6 ghz @xcite , which we multiply by a factor of 1.25 in order to estimate the 15 ghz flux using the average @xmath104 observed in the hard state of gx 3394 @xcite . on the other hand , we have checked that the 8.315 ghz index of cyg x-3 is @xmath105 in its hard state , cf . [ alpha](a b ) , and it is @xmath105 for cyg x-1 in the hard state @xcite . thus , the effect of the radio band on the presented correlation appears minor . we see cyg x-3 in its hard state is significantly more radio - loud than gx 3394 , by a factor of @xmath534 on average ( though there is a systematic uncertainty concerning the correction for the intrinsic x - ray absorption in cyg x-3 ) . as we have shown above , cyg x-3 appears relatively similar spectrally in x - rays to grs 1915 + 105 . in order to them in the radio band , we use the intense ryle telescope monitoring of grs 1915 + 105 conducted during mjd 4985653898 , and the asm data , during mjd 5008855859 . given the fast variability of that object , we have tested the effect of various averaging methods . we have finally chosen to average the radio over observation periods containing gaps shorter than 0.2 d , which resulted in the average length of an averaged segment of 0.09 d ( usually one per day ) with @xmath106 d. we then correlated those data with the asm dwell data which were within @xmath107 d from the average time of a given radio segment . we have corrected the averaged asm data for line - of - sight absorption assuming @xmath108 @xmath15 @xcite and the intrinsic spectra with @xmath109 within each asm bin . we have further added the contribution from @xmath110 kev in the same way as for cyg x-3 , i.e. , assuming @xmath86 , see above . we have also considered the available batse data for this object . we have found that addition of the 20160 kev fluxes to those of the asm was possible only for about 40 per cent of the asm points , and an upper part of the correlation was not covered by the asm+batse points . the average 20160 kev to 1.312.2 kev flux ratio equals 0.11 . given that the @xmath27 spectra of grs 1915 + 105 in almost all states are decreasing above several kev @xcite , we estimate the flux contribution from 1220 kev to be similar to that from @xmath111 kev . we have thus applied an approximate bolometric correction of a factor of 1.2 to the fluxes from the asm . we show the results by the cyan error bars in fig . we see grs 1915 + 105 does not show a distinct radio / x - ray correlation . on the other hand , @xcite claimed a strong correlation within the plateau ( steady and hard ) state of grs 1915 + 105 . when using the same time intervals as those given in @xcite as used for their correlation , we have found a rather scattered diagram , with a steep positive slope at high radio fluxes and a negative one at low radio fluxes . when we used the same criteria as given in that paper , i.e. , the ratio of the count rates in the asm third to second channels of @xmath112 and the total asm rate of 3050 s@xmath8 , we found a similar discrepancy . we have recovered a correlation only after imposing an additional criterion of the radio flux of @xmath113 mjy ( which , however , implies that the x - ray state by itself does not set the radio behaviour , in contrast to cyg x-3 ) . those points are shown in fig . [ rbol ] by green crosses . we do see some correlation , with the slope of @xmath114 , somewhat steeper than @xmath115 given in @xcite . we see that grs 1915 + 105 occupies a similar range of @xmath79 as cyg x-3 ( as also seen in fig . [ lcd]b ) , but it is both less radio loud and with a much less clear radio / x - ray correlation . the likely reason for that is the difference in the jet environments . cyg x-3 is a binary with a high - mass donor , and interaction between its wind and the jet leads to presence of shocks , in particular of recollimation shocks @xcite . those shocks are likely to efficiently accelerate electrons , leading to enhanced radio emission . however , no such interaction takes place in grs 1915 + 105 , an lxmb . on the other hand , cyg x-1 , which is also a high - mass x - ray binary , has the hard - state radio fluxes , shown by the red triangles in fig . [ rbol ] , an order of magnitude below those of the gx 3394 correlation . an explanation for that may be provided by the wind ram pressure around the jet in cyg x-1 being much lower than that in cyg x-3 , owing to both the lower mass - loss rate and the larger separation . the jet lateral pressure in cyg x-1 may then be always larger than that of the wind , provided the jet power is above some critical value @xcite . then no recollimation shock would form . furthermore , as discussed in section [ intrinsic ] , the radio flux in cyg x-1 is significantly free - free absorbed in the wind from its donor , see zspl11 and @xcite . in the model of the latter , the slope of the intrinsic hard - state correlation in cyg x-1 is close to that of the lower branch of the radio / x - ray correlation , first found for h1743322 @xcite . however , the correction for the wind absorption in cyg x-1 is relatively uncertain . more detailed knowledge of the wind structure in cyg x-1 is needed to determine the degree of free - free absorption . our results point out to the importance of taking into account an estimate of the bolometric flux when correlating x - rays with radio , reinforcing the similar conclusion of zspl11 . due to the pivoting spectral variability in x - rays the hard x - rays are often a very bad predictor of the bolometric flux , and can be either uncorrelated or negatively correlated with it . this effect appears to be present also in seyfert galaxies , in which a common variability pattern in soft x - rays is the steepening of their soft x - ray spectrum with the increasing flux . this then leads to a pivot in hard x - rays . this effect can explain the discrepancy between the result of @xcite , who found the radio and soft x - ray emission of their sample of seyferts to be strongly correlated , and the result of @xcite , who found a lack of such correlation for hard x - rays , using _ swift_/bat data . the latter authors doubted then the reality of the soft x - ray correlation . however , this discrepancy can be resolved by a pivoting spectral variability of x - rays . in cyg x-3 , it leads to the 1020 kev flux to be completely uncorrelated with radio , see fig . [ rx4](g ) , in spite of the correlation in soft x - rays , fig . [ rx4](e ) , being very strong . we have studied implications of the presence of the positive and negative correlation with the radio flux for the soft and hard x - ray fluxes , respectively , in the hard spectral state of cyg x-3 . we have found that the radio flux is then completely uncorrelated with the observed , absorbed , broad - band , 1.3100 kev x - ray flux , which appears to be a good approximation to the absorbed bolometric flux . an intrinsic lack of correlation is unlikely on theoretical grounds , as well as it is unlikely given the presence of the strong narrow x - ray band correlations . however , we can recover a clear positive correlation if the absorption by the stellar wind of the donor in cyg x-3 is strong enough . this implies that the intrinsic soft x - ray spectrum in the hard state is relatively soft , similar to those of intermediate states of bh binaries . this nature of the apparent hard state of cyg x-3 is also consistent with the luminosity of the hard state being much higher in this object than in the hard state of the high - mass bh binary , cyg x-1 . we have found the radio spectra in the hard state to be hard on average , with @xmath67 . such spectra are characteristic to bh x - ray binaries in the hard state . we have compared cyg x-3 with other x - ray binaries . on the x - ray colour - colour and colour - eddington ratio diagrams , cyg x-3 , after correcting its spectra for absorption and for the bh mass of @xmath116 , appears similar to grs 1915 + 105 , with similar colours and @xmath1171 . on the other hand , cyg x-3 , at @xmath118 is also similar on such diagrams to ns z - type binaries . however , the radio emission of those sources is much weaker than that of cyg x-3 , which is both the most radio - luminous and , in its hard state , the most radio - loud ( i.e. , with the highest ratio of the radio to bolometric luminosities ) source among all known x - ray binaries . still , its radio loudness is much closer to those of bh binaries than to those of ns ones . this represents a strong argument for the presence of a bh in cyg x-3 . we interpret the radio loudness and radio fluxes of cyg x-3 being much higher than those of grs 1915 + 105 as due to an enhancement of the jet emission in the former due to interaction of the jet with the very strong stellar wind from the donor . we have found ( appendix [ lc ] ) that the flux distributions of the radio emission and soft x - rays can be described by two log - normal functions , one corresponding to the hard state and one to soft state . this indicates a multiplicative and correlated character of variability of cyg x-3 in both the hard and soft state and in both the accretion flow and the jet . finally , we have presented ( appendix [ lc ] ) the long - time light curves of cyg x-3 using the available data , and compared them to the occurrences of high - energy @xmath1-ray detection . we have found that a low hard x - ray flux , @xmath119 , is both a necessary and sufficient criterion for detectable @xmath1-rays . we have found that this criterion has not been satisfied since mjd 55650 ( 2011 march 30 ) until now , mjd 57333 . we thank patryk pjanka for help with the data analysis , antonino dai for valuable comments , and chris done and marek gierliski for providing us with the colour - flux data on bh binaries . this research has been supported in part by the polish ncn grants 2012/04/m / st9/00780 and 2013/10/m / st9/00729 . the ryle telescope and the ami arrays are supported by stfc and the university of cambridge . this research has also made use of the maxi data provided by riken , jaxa and the maxi team , and of the quick - look results provided by the _ asm team . arnaud k. a. , 1996 , in jacoby g. h. , barnes j. , eds . , astronomical data analysis software and systems v , asp conf . series vol . 101 , san francisco , p. 17 axelsson m. , larsson s. , hjalmarsdotter l. , 2009 , mnras , 394 , 1544 barthelmy s. d. et al . , 2005 , space sci . , 120 , 143 bautista m. a. , kallman t. r. , 2001 , apjs , 134 , 139 belczyski k. , bulik t. , mandel i. , sathyaprakash b. s. , zdziarski a. a. , mikoajewska j. , 2013 , apj , 764 , 96 blandford r. d. , konigl a. , 1979 , apj , 232 , 34 bradt h. v. , rothschild r. e. , swank j. h. , 1993 , a&as , 97 , 355 bulgarelli a. , et al . , 2012 , a&a , 538 , a63 burlon d. , ghirlanda g. , murphy t. , chhetri r. , sadler e. , ajello m. , 2013 , mnras , 431 , 2471 casares j. , jonker p. g. , 2014 , ssrv , 183 , 223 chapuis c. , corbel s. , 2004 , a&a , 414 , 659 chu k. w. , bieging j. h. , 1973 , apj , 179 , 21 corbel s. , fender r. p. , tzioumis a. k. , nowak m. , mcintyre v. , durouchoux p. , sood r. , 2000 , a&a , 359 , 251 corbel s. , nowak m. a. , fender r. p. , tzioumis a. k. , markoff s. , 2003 , a&a , 400 , 1007 corbel s. , fender r. p. , tomsick j. a. , tzioumis a. k. , tingay s. , 2004 , apj , 617 , 1272 corbel s. , koerding e. , kaaret p. , 2008 , mnras , 389 , 1697 corbel s. , et al . , 2012 , mnras , 421 , 2947 corbel s. , coriat m. , brocksopp c. , tzioumis a. k. , fender r. p. , tomsick j. a. , buxton m. m. , bailyn c. d. , 2013a , mnras , 428 , 2500 corbel s. , et al . , 2013b , mnras , 431 , l107 coriat m. , et al . , 2011 , mnras , 414 , 677 corral - santana j. m. , casares j. , munoz - darias t. , bauer f. e. , martinez - pais i. g. , russell d. m. , 2016 , a&a , in press , arxiv:1510.08869 dickey j. m. , 1983 , apj , 273 , l71 dickey j. m. , lockman f. j. , 1990 , ara&a , 28 , 215 done c. , gierliski m. , 2003 , mnras , 342 , 1041 done c. , wardziski g. , gierliski m. , 2004 , mnras , 349 , 393 done c. , gierliski m. , kubota a. , 2007 , a&arv , 15 , 1 ebisawa k. , 1998 , iaus , 188 , 392 fender r. p. , hanson m. m. , pooley g. g. , 1999 , mnras , 308 , 473 fender , r. p. , pooley , g. g. , durouchoux , p. , tilanus , r. p. j. , brocksopp , c. , 2000 , mnras , 312 , 853 fermi lat collaboration , 2009 , sci , 326 , 1512 gallo e. , fender r. p. , pooley g. g. , 2003 , mnras , 344 , 60 giacconi r. , gorenstein p. , gursky h. , waters j. r. , 1967 , apj , 148 , l119 gladstone j. , done c. , gierliski m. , 2007 , mnras , 378 , 13 harmon b. a. et al . , 2002 , apjs , 138 , 149 heinz s. , sunyaev r. a. , 2003 , mnras , 343 , l59 hjalmarsdotter l. , zdziarski a. a. , larsson s. , beckmann v. , mccollough m. , hannikainen d. c. , vilhu o. , 2008 , mnras , 384 , 278 hjalmarsdotter l. , zdziarski a. a. , szostek a. , hannikainen d. c. , 2009 , mnras , 392 , 251 int zand j. j. m. , jonker p. g. , markwardt c. b. , 2007 , a&a , 465 , 953 koljonen k. i. i. , hannikainen d. c. , mccollough m. l. , pooley g. g. , trushkin s. a. , 2010 , mnras , 406 , 307 krimm h. a. , et al . , 2013 , apjs , 209 , 14 lauqu r. , lequeux j. , nguyen - quang - rieu , 1972 , nat . phys . , 239 , 119 laycock , s. g. t. , maccarone t. j. , christodoulou d. m. , 2015 , mnras , 452 , l31 levine a. m. , bradt h. , cui w. , jernigan j. g. , morgan e. h. , remillard r. , shirey r. e. , smith d. a. , 1996 , apj , 469 , l33 ling z. , zhang s. & tang s. , 2009 , apj , 695 , 1111 loh a. , corbel s. , dubus g. , corbet r. , 2016 , atel , 8591 markwardt c. b. , tueller j. , skinner g. k. , gehrels n. , barthelmy s. d. , mushotzky r. f. , 2005 , apj , 633 , l77 mart j. , paredes j. m. , peracaula m. , 2001 , a&a , 375 , 476 matsuoka , m. et al . , 2009 , pasj , 61 , 999 mccollough m. l. , et al . 1999 , apj , 517 , 951 merloni a. , heinz s. , di matteo t. , 2003 , mnras , 345 , 1057 migliari s. , fender r. p. , 2006 , mnras , 366 , 79 mioduszewski a. j. , rupen m. p. , hjellming r. m. , pooley g. g. , waltman e. b. , 2001 , apj , 553 , 766 pooley g. g. , fender r. p. , 1997 , mnras , 292 , 925 poutanen j. , zdziarski a. a. , ibragimov a. , 2008 , mnras , 389 , 1427 predehl p. , burwitz v. , paerels f. , trmper j. , 2000 , a&a , 357 , l25 reeves j. , done c. , pounds k. , terashima y. , hayashida k. , anabuki n. , uchino m. , turner m. , 2008 , mnras , 385 , l108 reid m. j. , mcclintock j. e. , steiner j. f. , steeghs d. , remillard r. a. , dhawan v. , narayan r. , 2014 , apj , 796 , 2 remillard r. a. , mcclintock j. e. , 2006 , ara&a , 44 , 49 rushton a. , spencer r. , fender r. , pooley g. , 2010 , a&a , 524 , a29 salvato m. , greiner j. , kuhlbrodt b. , 2004 , apj , 600 , l31 segreto a. , cusumano g. , ferrigno c. , la parola v. , mangano v. , mineo t. , romano p. , 2010 , a&a , 510 , a47 szostek a. , zdziarski a. a. , 2008 , mnras , 386 , 593 szostek a. , zdziarski a. a. , mccollough m. l. , 2008 , mnras , 388 , 1001 ( szm08 ) tavani m. , et al . , 2009 , nature , 462 , 620 tavani m. , et al . , 2016 , atel , 8597 uttley p. , mchardy i. m. , vaughan s. , 2005 , mnras , 359 , 345 van kerkwijk m. h. , et al . , 1992 , nature , 355 , 703 van kerkwijk m. h. , 1993 , a&a , 276 , l9 van kerkwijk m. h. , geballe t. r. , king d. l. , van der klis m. , van paradijs j. , 1996 , a&a , 314 , 521 waltman e. b. , fiedler r. l. , johnston k. j. , ghigo f. d. , 1994 , aj , 108 , 179 waltman e. b. , ghigo f. d. , johnston k. j. , foster r. s. , fiedler r. l. , spencer j. h. , 1995 , aj , 110 , 290 waltman e. b. , foster r. s. , pooley g. g. , fender r. p. , ghigo f. d. , 1996 , aj , 112 , 2690 watanabe h. , kitamoto s. , miyamoto s. , fielder r. l. , waltman e. b. , johnston k. j. , ghigo f. d. , 1994 , apj , 433 , 350 williams p. k. g. , et al . , 2011 , apj , 733 , l20 yoon d. , heinz s. , 2015 , apj , 801 , 55 yoon d. , zdziarski a. a. , heinz s. , 2016 , mnras , 456 , 3638 zdziarski a. a. , 2012 , mnras , 422 , 1750 zdziarski a. a. , poutanen j. , paciesas w. s. , wen l. , 2002 , apj , 578 , 357 zdziarski a. a. , gierliski m. , mikoajewska j. , wardziski g. , smith d. m. , harmon b. a. , kitamoto s. , 2004 , mnras , 351 , 791 zdziarski a. a. , misra r. , gierliski m. , 2010 , mnras , 402 , 767 zdziarski a. a. , skinner g. k. , pooley g. g. , lubiski p. , 2011 , mnras , 416 , 1324 ( zspl11 ) zdziarski a. a. , maitra c. , frankowski a. , skinner g. k. , misra r. , 2012 , mnras , 426 , 1031 zdziarski a. a. , mikoajewska j. , belczyski k. , 2013 , mnras , 429 , l104 we show here distributions of the fluxes and the light curves for the x - ray monitors used here and for the ryle / ami . fig . [ hist ] shows the histograms of the 15 ghz , asm , maxi and bat fluxes and count rates ( for simplicity , we use the symbol @xmath120 for both ) . the histogram for the batse flux is shown in @xcite . the rates for 15 ghz , and asm and maxi are fair estimators of the jet power in the former case and the accretion power in the latter , see section [ rx ] . on the other hand , the hard x - rays are anticorrelated with the accretion luminosity in the hard / intermediate state due to the spectral pivoting . thus , their distribution does not reflect the accretion power . consequently , we find the 15 ghz , asm and maxi histograms can be fitted by a sum of two log - normal distributions , one present in the hard / intermediate state and one in the soft state , while it is not the case for the bat histogram . the values of the mean and standard deviation for the hard / intermediate - state and soft - state log - normal distributions for the 15 ghz , asm and maxi data are @xmath121 , @xmath122 , @xmath123 , respectively ( in @xmath124 units ) . we see that the width of the flux distribution in each of the state in the asm data is twice as large as that of the maxi data . this is due to the maxi 10 - 20 kev channel containing the pivot of the spectral variability , see fig . [ rx4](g ) , which reduces the range of the flux variability . on the other hand , the asm 512.2 kev channel is still below the pivot , see fig . [ rx4](c ) . all of the asm and maxi distributions are narrower than the corresponding 15-ghz ones , especially in the soft state . this reflects the dramatic variability of the radio emission observed in that state . the presence of a log - normal distribution shows that the underlying physical process has to be multiplicative , with variations coupled together on all time scales @xcite . interestingly , we find the log - normal variability to be present in the both main states of cyg x-3 , hard and soft . on the other hand , @xcite and @xcite have found it in the hard state of cyg x-1 on @xmath00.110 s ( for soft x - rays ) and @xmath125 d ( for soft x - rays and radio ) time scales , respectively . however , it appears that a log - normal flux distribution has not been found before in the soft state of an x - ray binary . [ lc901 ] shows the light curves . since the variability in all energy bands spans a large ratio and it is log - normal for some detectors , we normalize each light curve to its geometric unweighted average over the respective observation time span , i.e. , we calculate @xmath126 . for that , we have to select only points with the measured fluxes of @xmath127 . this will lead to a certain overestimate of the overall average , but the effect is minor for the light curves presented below . for notational simplicity , we denote @xmath128 . also , in order to reduce a contribution from the measurements of poor quality , we include only data with the fractional error of @xmath129 , where @xmath130 is the flux error . we show the 1-d average rates except for the batse , where we show the 3-d average . in the case of bat , we use the public 1550 kev data @xcite . that light curve is very similar to our 1450 kev one . for the maxi , we show the 210 kev light curve instead of the full 220 kev one in order to have as much correspondence as possible to the asm 1.312.2 kev light curve . the average rates and standard deviations are given in table a1 . the standard deviations correspond to the excess variance , i.e. , the variance due to the measurement errors is subtracted , @xmath131 . \label{sigma}\ ] ] the lower value of @xmath132 for the batse data than that for the bat is probably due to the former data being averaged over 3-d rather than 1-d intervals . the large values of @xmath132 for all the radio data is the consequence of the dramatic radio variability in this object , showing both very bright flares and very weak states , see fig . [ lc901 ] . for the 15 ghz data , we have used the estimate of @xmath133 ( section [ data ] ) . for clarity , we do not show the gbi light curves , which are relatively similar to the ryle / ami one . we also mark the intervals corresponding to the detected high - energy @xmath1-ray emission in fig . [ lc901 ] . these are mjd 5456654647 , 5482154850 , 5532455326 @xcite , 5475054820 , 5499055045 @xcite , 5534355345 @xcite , 5558655610 , 55642 - 55644 @xcite and mjd 5739857412 @xcite . all those periods correspond to the soft state as defined in our paper , and , with exception for the last occurence , to periods with both the lowest 1550 kev flux detected , with @xmath134 , and occurrences of radio flares , with the 1-d averaged 15 ghz flux of @xmath135 ( as noted , e.g. , by @xcite ; @xcite ) . however , the former of these conditions , i.e. , a low hard x - ray flux , appears to be both necessary and sufficient . we note that this condition has not been satisfied during mjd 5565057397 , see fig . [ lc901 ] . indeed , no high - energy @xmath1-ray detection from cyg x-3 has been reported during that period .

direction of arrival ( doa ) estimation in array processing using uniform / sparse linear arrays is concerned in this paper . while sparse methods via _ approximate _ parameter discretization have been popular in the past decade , the discretization may cause problems , e.g. , modeling error and increased computations due to dense sampling . in this paper , an _ exact _ discretization - free method , named as sparse and parametric approach ( spa ) , is proposed for uniform and sparse linear arrays . spa carries out parameter estimation in the continuous range based on well - established covariance fitting criteria and convex optimization . it guarantees to produce a sparse parameter estimate without discretization required by existing sparse methods . theoretical analysis shows that the spa parameter estimator is a large - snapshot realization of the maximum likelihood estimator and is statistically consistent ( in the number of snapshots ) under uncorrelated sources . other merits of spa include improved resolution , applicability to arbitrary number of snapshots , robustness to correlation of the sources and no requirement of user - parameters . numerical simulations are carried out to verify our analysis and demonstrate advantages of spa compared to existing methods . array processing , doa estimation , sparse and parametric approach ( spa ) , continuous parameter estimation , compressed sensing .

1312.7695

farfield narrowband source localization based on observed snapshots of a sensor array is a major problem in array signal processing @xcite , known also as direction of arrival ( doa ) estimation . the difficulty of the problem arises from the fact that the observed snapshots are nonlinear functions of the directions of interest . according to estimation schemes adopted , existing methods for the doa estimation can be classified into three categories : parametric , nonparametric and semiparametric , which are described as follows . _ parametric _ methods explicitly carry out parameter estimation using optimization or other methods . a prominent example is the nonlinear least squares ( nls ) method ( see , e.g. , @xcite ) , which adopts the least squares criterion and has a strong statistical motivation . however , nls admits the following two shortcomings : 1 ) it requires the knowledge of the source number which is typically unavailable , and 2 ) its global optimum can not be guaranteed with a practically efficient algorithm due to nonconvexity caused by the nonlinearity nature . either of them may greatly degrade the parameter estimation performance . music @xcite does the parameter estimation by studying the subspace of the data covariance matrix and is a large - snapshot realization of the maximum likelihood ( ml ) method in the case of uncorrelated sources . but it also requires the source number and does not have reliable performance when the number of snapshots is small or correlation exists between the sources . other subspace - based methods include esprit and their variants ( see @xcite for a complete review ) . in contrast , a _ nonparametric _ method typically produces a _ dense _ spectrum whose peaks are interpreted as source directions . so an alternative name _ dense _ is also used . examples include conventional beamformer and mvdr ( or capon s method ) ( see e.g. , @xcite ) and recently introduced iterative adaptive approach ( iaa ) @xcite which eliminates to a large extent leakage of the beamformer and is robust to correlation of the sources . but it will be shown in this paper that iaa suffers from resolution limit , especially in the moderate / low snr regime . while the classification above is consistent with @xcite , music is sometimes categorized in the literature as a nonparametric / dense method by interpreting the plot of its objective function as its power spectrum whose peaks are the optima of the directions . _ semiparametric _ methods have been popular in the past decade which remove the direction variables in the observation model by parameter discretization and transform the nonlinear parameter estimation problem into a sparse signal recovery problem under a linear model , followed by some sparse signal recovery technique and support detection of the sparse solution . in particular , the continuous direction range is approximated by a set of discrete grid points under the assumption that the grid is fine enough such that any of the true sources lies on ( practically , close to ) some grid point . after that , the knowledge is exploited that each of the expanded vector of source signals [ composed of ( virtual ) source signals from candidate directions on the grid ] is a sparse signal since the grid size is greatly larger than the source number , and a sparse solution is sought after . finally , the directions are retrieved from the support of the sparse solution . consequently , the semiparametric methods are also named as _ sparse _ methods in contrast to _ dense_. in principle , a source will be detected once the estimated source signal / source power from some direction is nonzero . following from the literature of sparse signal representation ( ssr ) and later developed compressed sensing ( cs ) @xcite , @xmath0 norm minimization and other sparse signal recovery techniques have been widely used in the semiparametric methods ( see , e.g. , @xcite ) . however , their theoretical support based on ssr and cs can not be applied due to dense discretization . moreover , the sparse recovery techniques usually require one or more practically unknown parameters , e.g. , the noise statistics , the source number or regularization parameters , etc . the semiparametric iterative covariance - based estimation ( spice ) method @xcite is a breakthrough of the semiparametric methods , in which covariance fitting criteria are adopted with sound statistical motivation , the source power and noise variance(s ) are estimated in a natural manner , no user - parameters are required and a connection to the @xmath0 norm minimization is shown . in principle , the semiparametric methods are approximation methods due to the discretization scheme adopted . furthermore , grid selection remains a major problem since 1 ) a coarse grid leads to a high modeling error and 2 ) too dense a grid is computationally prohibitive and might result in computational instability ( see , e.g. , @xcite ) . to alleviate the drawbacks of the discretization , preliminary results have been obtained in @xcite . overall , because of the inherent discretization scheme , the performance of semiparametric methods is dependent on the trade off between the discretization grid size and the computational workload . this paper aims at developing discretization - free techniques for doa estimation with an affordable computational workload in a common scenario of linear arrays , more specifically , uniform linear arrays ( ulas ) and sparse linear arrays ( slas ) . we consider stochastic source signals and the same covariance fitting criteria as in spice . by exploiting the hermitian toeplitz structure in the data covariance matrix , the covariance fitting problem is cast as semidefinite programming ( sdp ) and solved using off - the - shelf sdp solvers , e.g. , sdpt3 @xcite , in a polynomial time . a postprocessing technique is presented in this paper to retrieve , from the data covariance estimate , the parameters of interest including source locations , source powers and noise variance(s ) . the proposed method is a _ sparse _ method because it utilizes the same covariance fitting criteria of the semiparametric spice method and guarantees to produce a sparse parameter estimate . at the same time , it is a _ parametric _ method since it is proven to be equivalently solving a covariance fitting problem parameterized by the aforementioned parameters . therefore , the proposed method is named as sparse and parametric approach ( spa ) . the spa method differs from the existing sparse / semiparametric methods by no need of discretization . unlike the existing parametric methods , spa is based on convex optimization and does not require the source number . theoretical analysis shows that spa is a large - snapshot realization of the ml estimation and is statistically consistent ( in the number of snapshots ) under uncorrelated sources . other merits of spa include improved resolution , applicability to arbitrary number of snapshots , robustness to correlation of the sources and no requirement of user - parameters . numerical simulations are carried out to verify our theoretical results and demonstrate the superior performance of spa compared to existing methods . though most , if not all , of the existing sparse methods can be applied to sensor arrays with arbitrary geometry , problems may occur when applied to the ula and sla cases as studied in this paper . in particular , their parameter estimates might suffer from some identifiability problem and not be truly sparse as demonstrated in this paper , which conflicts with the name _ sparse _ and decreases their resolution considerably . as a byproduct , a modified spice method , named as spice - pp , is presented to rectify the problem by incorporating the postprocessing technique presented in this paper . in the single - snapshot case , the parameter estimation problem studied in this paper is mathematically equivalent to spectral analysis @xcite , for which discretization - free methods have been developed recently in @xcite . it is noted that the results of @xcite are based on very different techniques . furthermore , @xcite and @xcite are mainly focused on the noise - free case ( the former on the ula case and the latter on the sla case in the language of this paper ) . @xcite studies the ula case in the presence of i.i.d . gaussian noise with known noise statistics . the spa method proposed in this paper can deal with all of the above scenarios without the knowledge of noise statistics . while this paper is focused on the array processing applications in which the multisnapshot case is of the main interest , it is still not clear whether the techniques in @xcite specialized for the single - snapshot case can be extended to other cases . notations used in this paper are as follows . @xmath1 , @xmath2 and @xmath3 denote the sets of real numbers , nonnegative real numbers and complex numbers , respectively . boldface letters are reserved for vectors and matrices . for an integer @xmath4 , @xmath5 $ ] is defined as the set @xmath6 . @xmath7 denotes the absolute value of a scalar or cardinality of a set . @xmath8 , @xmath9 and @xmath10 denote the @xmath0 , @xmath11 and frobenius norms , respectively . @xmath12 , @xmath13 and @xmath14 are the matrix transpose , conjugate transpose and complex conjugate of @xmath15 , respectively . @xmath16 is the @xmath17th entry of a vector @xmath15 and @xmath18 is the @xmath19th entry of a matrix @xmath20 . unless otherwise stated , @xmath21 is a subvector of @xmath15 with the index set @xmath22 . for a vector @xmath15 , @xmath23 is a diagonal matrix with @xmath15 being its diagonal . @xmath24 means @xmath25 for all @xmath17 . @xmath26 denotes the trace of a matrix @xmath20 . for positive semidefinite matrices @xmath20 and @xmath27 , @xmath28 means that @xmath29 is positive semidefinite . @xmath30 denotes expectation and @xmath31 is an estimator of @xmath32 . for notational simplicity , a random variable and its numerical value will not be distinguished . the rest of this paper is organized as follows . section [ sec : problem ] describes the problem of array processing with linear arrays . section [ sec : spa_ula ] presents the proposed spa method in the ula case followed by section [ sec : spa_sla ] on the sla case . section [ sec : property ] introduces theoretical properties of spa . section [ sec : connectionprior ] discusses its connections to existing methods . section [ sec : simulation ] presents our numerical simulations . section [ sec : conclusion ] concludes this paper . consider @xmath33 narrowband farfield sources @xmath34 , @xmath35 , impinging on a linear array of omnidirectional sensors from directions @xmath36 , @xmath35 . we are interested in estimation of the direction vector @xmath37 , known as the direction of arrival ( doa ) estimation problem @xcite . denote @xmath38 , @xmath35 . then we can equivalently estimate @xmath39 since the relation @xmath40 is one - to - one . @xmath41 is called the frequency parameter and its meaning will be clarified later . in this paper , we concern only with the estimation of @xmath41 for convenience . according to @xcite , the time delays at different sensors can be represented by simple phase shifts , leading to the observation model : where @xmath42 indexes the snapshot and @xmath4 is the snapshot number , @xmath43 , @xmath44 and @xmath45 denote the observed snapshot , the vector of source signals and the vector of measurement noise at snapshot @xmath42 , respectively , and @xmath46 is the number of sensors ( some notations will be redefined in the sla case ) . @xmath47 is the so - called array manifold matrix and @xmath48 is the steering vector of the @xmath49th source which is determined by the geometry of the sensor array and will be given later . more compactly , ( [ formu : observation_model1 ] ) can be written into where @xmath50 , and @xmath51 and @xmath52 are similarly defined . the number of sources @xmath33 is assumed unknown in this paper . so , the objective of the array processing is to estimate the unknown parameter @xmath41 ( or equivalently , the vector of doas @xmath53 ) given the sensor measurements @xmath54 and the mapping @xmath55 . we introduce some standard assumptions for the problem formulation and solution . @xmath56 , @xmath57 , are assumed to be spatially and temporarily white , i.e. , where @xmath58 is the noise variance parameter and @xmath59 is a delta function that equals @xmath60 if @xmath61 or @xmath62 otherwise . the source signals and the noise are assumed to be uncorrelated with each other . moreover , assume that the source signals are uncorrelated spatially and temporarily , i.e. , where @xmath63 denotes the source power parameter . under the assumptions above , the data snapshots @xmath64 are uncorrelated with each other and have the covariance matrix it is worthy noting that the spatial uncorrelatedness of the sources may not be satisfied in practice , i.e. , two sources @xmath65 and @xmath66 can be correlated or even coherent ( i.e. , completely correlated ) . however , the method proposed in this paper will be shown to be robust to this assumption . both uniform and sparse linear arrays will be studied in this paper . in the ula case , the sensors are uniformly spaced with a spacing of @xmath67 , where @xmath68 denotes the wavelength of the sources . for an @xmath46-element ula , the steering vector @xmath48 of source @xmath49 has the following form with @xmath69 : it is clear that @xmath70 is the frequency of the uniformly sampled complex sinusoid @xmath48 , a reason for why @xmath41 is called the frequency parameter . a well - known result is that up to @xmath71 sources can be detected using an @xmath46-element ula ( see , e.g. , @xcite ) . an sla can be considered as a ula with `` missing '' sensors , i.e. , an sla takes only a subset , say @xmath72 , of the sensors of a ula . thus an sla can be represented by its sensor index set @xmath73 . without loss of generality , we assume that @xmath72 is sorted ascendingly with @xmath74 and @xmath75 , where @xmath76 denotes the array size ( otherwise , we can redefine @xmath72 , say @xmath77 , such that @xmath78 , @xmath79 $ ] , and let @xmath80 ) . then the steering vector of the sla @xmath72 for source @xmath49 , denoted by @xmath81 , is denote @xmath82 a selection matrix such that the @xmath17th row of @xmath83 contains all @xmath62s but a single @xmath60 at the @xmath84th position . it is clear that let an sla defined by @xmath85 is called the coarray of @xmath72 . @xmath72 is called a redundancy array if @xmath85 defines a ula , i.e. , @xmath86 . it is obvious that a ula is a redundancy array . the maximum number of sources detectable using the array @xmath72 is determined by its coarray @xmath85 @xcite . in particular , a redundancy array can detect up to @xmath71 sources like a ula . for a non - redundancy sla , the maximum number of sources detectable will be less than @xmath71 . it is noted that redundancy slas are quite common and there generally exists such an array satisfying that @xmath87 according to @xcite . the sla @xmath88 is a redundancy array , where @xmath89 and @xmath90 . so maximally @xmath91 sources can be detected using this @xmath92-element array . since a ula can detect up to @xmath71 sources , the condition @xmath93 can be considered as _ a priori _ knowledge while the exact value of @xmath33 is unavailable . the covariance of the data snapshots @xmath54 is given in ( [ formu : ycovariance ] ) under the assumptions specified in subsection [ sec : assmptions ] . denote by @xmath94 the sample covariance . when @xmath95 and @xmath96 are both invertible we consider the following covariance fitting criterion for the purpose of parameter estimation ( see @xcite and the references therein ) : @xmath97 exists in the presence of noise , i.e. , @xmath98 for @xmath99 . then @xmath100 exists with probability one if @xmath101 . according to @xcite , the minimization of the criterion in ( [ formu : criterion1 ] ) is a large - snapshot realization of the ml estimator . in the case where @xmath95 is singular ( it happens when @xmath102 ) we consider an alternative criterion : which has been studied in @xcite . we defer the latter case to subsection [ sec : tlessm ] . the covariance fitting criteria above exploit the assumption that the sources are uncorrelated which results in the expression of @xmath96 in ( [ formu : ycovariance ] ) . however , a theoretical explanation is provided in @xcite to show that the criteria are robust to correlations of the sources . moreover , according to @xcite they are connected to the @xmath0 norm minimization which is known to be robust to the correlations . consequently , such robustness property is maintained in the method proposed in this paper which utilizes the same criteria . a simple calculation shows that it is challenging to minimize @xmath103 with respect to the unknown parameters @xmath41 , @xmath104 and @xmath105 due to their nonlinear relation to @xmath96 by ( [ formu : ycovariance ] ) . we propose to estimate @xmath96 firstly by reparameterization and then determine the parameters of interest . let it is easy to see that @xmath106 and @xmath107 . moreover , @xmath108 is thus a ( hermitian ) toeplitz matrix which is determined by @xmath46 complex numbers and can be written as @xmath109 for some @xmath110 , where it follows from ( [ formu : toeplitz ] ) that the characterization captures the structure of @xmath96 under the constraints @xmath112 and @xmath113 . further observations reveal that @xmath96 is inherently determined by @xmath114 numbers ( @xmath46 for the diagonal and the other @xmath71 for the off diagonal ) . however , @xmath115 numbers ( @xmath46 in @xmath116 and the other @xmath46 in @xmath105 ) have been used in the above expression of @xmath96 ( @xmath117 for the diagonal and @xmath71 for the off diagonal ) . as a result , redundancy exists along the diagonal of @xmath96 . the effect of the redundancy is twofold . on one hand , it will cause an identifiability problem , which will be tackled in subsection [ sec : postproc ] , that in general @xmath116 , @xmath105 can not be uniquely identified from @xmath96 without accounting for additional information except @xmath112 and @xmath113 . it means that the solution @xmath118 of the sdps to be presented can not be directly used as the final estimate of the noise variance . on the other hand , the adoption of one redundant variable enables us not to impose the nonconvex rank - deficiency constraint on @xmath119 to characterize @xmath96 ( it will be clarified in subsection [ sec : postproc ] ) . however , the redundancy problem will not affect the estimation of @xmath96 which is our current focus . with the characterization of @xmath96 in ( [ formu : r_in_u ] ) the minimization of @xmath103 is equivalent to then we can show the following equivalences : so the problem in ( [ formu : problem_tlarge ] ) can be formulated as an sdp and thus is convex . as the result , @xmath96 can be estimated by solving the sdp with its estimate given by @xmath120 , where @xmath121 is the solution of the sdp . when @xmath102 , @xmath95 ceases to be nonsingular and the criterion in ( [ formu : criterion2 ] ) is used which can be written into hence , an sdp similar to ( [ formu : sdp_nlarge ] ) can be formulated as follows : it is reasonable to assume that the noise variances @xmath122 , @xmath99 , are equal in some scenarios . though the formulations in the last subsection can be applied to such a case by imposing the constraint @xmath123 , simpler formulations in fact exist . in such a case , @xmath124 with @xmath125 . therefore , the covariance @xmath96 has a toeplitz structure and the characterization captures its structure without redundancy under the constraint @xmath112 for some @xmath110 . note that the @xmath119 here is the same as that in ( [ formu : r_in_u ] ) off the diagonal but different on the diagonal , i.e. , the two @xmath116 s are the same except the first entry . following from the similar procedures as in the last subsection , similar sdps can be formulated . in particular , when @xmath101 we have when @xmath102 , note that the constraint @xmath112 has been implicitly included in the constraints in ( [ formu : sdp_nlarge_identical ] ) and ( [ formu : sdp_nsmall_identical ] ) . the simplified sdp formulations will lead to faster computations in practice . by solving one of the sdps , @xmath126 is obtained as @xmath127 given the solution @xmath128 . for notational convenience , we say @xmath129 in this case . in fact , note that ( [ formu : r_equalvar ] ) , ( [ formu : sdp_nlarge_identical ] ) and ( [ formu : sdp_nsmall_identical ] ) can be obtained from ( [ formu : r_in_u ] ) , ( [ formu : sdp_nlarge ] ) and ( [ formu : sdp_nsmall ] ) , respectively , by setting @xmath130 . after obtaining @xmath126 , the following task is to estimate the parameters @xmath41 , @xmath104 and @xmath105 by writing it back into the form of ( [ formu : ycovariance ] ) . to do this , we decompose @xmath126 into where @xmath131 is the estimate of @xmath132 in ( [ formu : c ] ) and @xmath133 is the noise covariance estimate . such @xmath134 always exists according to the way @xmath126 is given , however , it is generally not unique . in particular , for any @xmath135 satisfying that @xmath136 , @xmath137 leads to one realization of the decomposition , which in fact also enumerates all possible realizations . we utilize the prior knowledge that @xmath93 to make the decomposition unique as follows . it follows from @xmath93 that @xmath107 . therefore , it is natural to impose that @xmath138 , i.e. , @xmath139 is rank - deficient . a direct result is that @xmath140 is an eigenvalue of @xmath141 . then by @xmath136 we see that and thus the decomposition is unique , where @xmath142 denotes the minimum eigenvalue . ) can be done if we have only @xmath126 rather than @xmath128 and @xmath118 . to see this , let @xmath143 be the transpose of the first row of @xmath126 . then @xmath144 and @xmath145 is obtained as the diagonal of @xmath126 minus @xmath146 . ] the postprocessing is used to separate sources and noise in the estimated covariance matrix such that the source part can be represented by as few sources as possible based on the minimum description length principle @xcite ( see the next subsection for clarity ) . in fact , the idea of postprocessing has been studied in the literature for a special case where @xmath126 is a toeplitz matrix as in the case of equal @xmath111 ( see , e.g. , ( * ? ? ? * section 4.9.2 ) ) . it is noted that the postprocessing is very important in spa , without which the final parameter estimate is generally not unique and does not have the statistical properties that will be shown in section [ sec : property ] . even worse , the frequency estimate of spa can cease to be sparse . to see this , suppose that @xmath141 has full rank ( this is generally the case in the presence of noise ) . choose @xmath147 arbitrarily and let @xmath148 . it follows that the residue @xmath149 which still has the toeplitz structure . then we can choose @xmath150 and @xmath151 similarly based on the residue . the process can be repeated infinitely many times and results in infinitely long vectors @xmath41 and @xmath104 . [ rem : postproc ] the remaining task is to retrieve the frequency estimate @xmath152 and the power estimate @xmath153 given @xmath154 , which is based on the following classical vandermonde decomposition lemma for positive semidefinite toeplitz matrices ( see , e.g. , @xcite ) . any positive semidefinite toeplitz matrix @xmath155 can be represented as where @xmath156 , @xmath157 for @xmath158 $ ] , and @xmath159 . moreover , the representation is unique up to permutation of elements of @xmath41 and @xmath104 if @xmath160 . @xmath161 [ lem : toeplitz ] it follows from lemma [ lem : toeplitz ] that @xmath152 and @xmath153 can be uniquely determined given @xmath154 since @xmath138 . in practice , @xmath152 and @xmath153 can be obtained as follows . given @xmath162 , it is easy to show that since @xmath163 , where @xmath164 takes all but the first rows of the matrix @xmath165 which denotes the complex conjugate of @xmath166 . so we build a system of @xmath114 equations that is linear in @xmath153 whose length is maximally @xmath71 , where each column of the coefficient matrix corresponds to a uniformly sampled sinusoid ( after permutation of rows ) . according to @xcite , prony s method can be applied to this type of systems to efficiently solve @xmath152 and @xmath153 . in particular , @xmath152 is firstly obtained from zeros of a polynomial which is obtained by solving a linear system involving only @xmath167 . after that , @xmath153 is solved from ( [ formu : linsys ] ) . readers are referred to @xcite for the detailed procedure . the proposed spa algorithm for the array processing with a ula is presented in algorithm [ alg : spa ] . the covariance matrix @xmath96 is firstly estimated by solving an sdp . then the postprocessing procedure is carried out for @xmath126 to resolve an identifiability problem that exists in the solution of the sdp ( the noise variance is estimated at this step ) . finally , the frequency and the power are solved using prony s method . to carry out the parameter estimation , spa requires only the data snapshots @xmath54 of the sensor array without any other user parameters . more properties of spa will be presented in section [ sec : property ] . input : observed snapshots @xmath54 . 1 . estimate @xmath96 by solving the solution @xmath121 of an sdp ; 2 . postprocess @xmath126 to obtain @xmath167 and @xmath145 via @xmath168 and @xmath169 ; 3 . solve @xmath152 and @xmath153 from ( [ formu : linsys ] ) using prony s method . output : parameter estimator @xmath170 . [ alg : spa ] in general , spa can not correctly determine the true number of sources @xmath33 but provides additionally spurious sources . this is because that we do not assume in spa any knowledge of the source number or the noise variance(s ) . consequently , spa solves the problem as in the worst case where up to @xmath71 sources can be present . in fact , the source number estimation problem itself is very difficult which is known as model order selection @xcite , and lacking guarantees on the estimated source number seems to be a common feature of sparse methods , especially when no prior knowledge mentioned above is available ( see , e.g. , @xcite ) . a positive side of spa is that it has strong statistical properties which will be shown in section [ sec : property ] , and theoretically the powers of the spurious sources can be very close to zero under sufficient snapshots or appropriately low snr , which enables the spurious sources to be distinguished from the real ones . currently , methods have been proposed to remove spurious sources by modifying the solution of a sparse method . for example , intuitive thresholding is used in @xcite and information criteria - based methods are presented in @xcite . in future studies , we may seek to incorporate model order selection in spa such that the source number can be estimated automatically together with the parameters but it is beyond the scope of this paper . in this section we extend spa to the sla case . with respect to an sla @xmath72 , denote the steering matrix by @xmath171 , the data snapshots by @xmath172 , the covariance matrix by @xmath173 and the sample covariance by @xmath174 , where @xmath175 denotes the noise variance parameter . it follows from ( [ formu : steervector_sla ] ) that @xmath176 and then @xmath177 , where the toeplitz matrix @xmath178 for some @xmath110 . a careful study of @xmath179 reveals that i.e. , the entries of @xmath179 are specified by the coarray @xmath85 defined in ( [ formu : coarray ] ) . in this paper we are mainly interested in the case where the sla @xmath72 is a redundancy array . as in the ula case , @xmath93 is considered as _ a priori _ knowledge according to subsection [ sec : lineararray ] . the matrix @xmath179 contains all elements of @xmath116 explicitly by ( [ formu : entry_sla ] ) . it implies that the relation @xmath180 is one - to - one . given a redundancy sla @xmath88 , we have @xmath181 and @xmath182 , where all elements of @xmath116 are contained . it follows that the covariance matrix @xmath183 can be characterized as under the constraints @xmath112 and @xmath184 . consider similar covariance fitting criteria as in the ula case , i.e. , the minimization of @xmath185 when @xmath186 and @xmath187 otherwise . then spa can be extended to the sla case . in particular , when @xmath186 we obtain the following sdp : when @xmath188 , it is in the case of equal noise variances , @xmath183 can be characterized as @xmath189 . then similar sdps can be formulated as in ( [ formu : sdp_nlarge_miss ] ) and ( [ formu : sdp_nsmall_miss ] ) by simply setting @xmath190 . note that the problem dimension in such a case can not be further reduced as in the ula case since @xmath191 does not imply @xmath112 . note that the same identifiability problem exists in the solution @xmath192 of the sdps above . to resolve the problem , the same procedures can be carried out as in the ula case by exploiting that @xmath138 . then the parameter estimate @xmath193 can be obtained . when the sla @xmath72 is not a redundancy array , i.e. , the coarray @xmath85 is not a ula , spa presented above can be applied straightforwardly . in such a case , it is natural to require that @xmath194 , where @xmath195 denotes the maximum number of sources detectable using the non - redundancy sla . however , the introduced implementation of spa can only use the information up to @xmath93 , which should be a waste of knowledge . a thorough study on exploitation of the full information should be investigated in the future but is beyond the scope of this paper . as its name suggests , spa is a _ sparse _ and _ parametric _ method . it carries out the optimization by reparameterization and thus is a parametric method . the length of its frequency estimator @xmath152 is maximally @xmath71 . since this is an important result of this paper , we state it formally in the following theorem . the length of the frequency estimator @xmath152 of spa is maximally @xmath71 . @xmath161 we consider ulas and redundancy slas in this subsection , or collectively , redundancy arrays . notice that dimensions of the parameter estimators @xmath152 and @xmath153 may be different from the true dimension @xmath33 . to discuss statistical properties of the spa estimator , the dimension problem should be resolved first . notice that the proposed spa is equivalently assuming @xmath196 ( the worst case ) with the knowledge @xmath93 . then we expand both the true parameter and its estimator to the same dimension . define for a vector @xmath197 of dimension no more than @xmath71 , where @xmath198 is arbitrary . it is easy to see that @xmath199 and @xmath200 are physically equivalent since all of the added virtual sources have zero powers . then we have the following results . assume that @xmath201 and @xmath202 are uncorrelated , @xmath203 and @xmath204 for @xmath205 . moreover , assume that the sensor array is a redundancy array and @xmath93 . then the parameter estimator @xmath170 of spa is statistically consistent ( in @xmath4 ) . and @xmath105 denote respectively @xmath206 and @xmath207 in the sla case , or @xmath208 and @xmath209 in the case of equal noise variances . it is the same for theorems [ thm : aml ] and [ thm : efficiency ] . ] [ thm : consistency ] without loss of generality , we consider only the ula case . the proof is based on the observation that the parameter estimator @xmath170 can be uniquely determined given the covariance estimator @xmath126 . suppose that the true parameter value is @xmath210 . as @xmath211 , @xmath95 approaches the true covariance @xmath212 . then by ( [ formu : criterion1 ] ) the sdp of spa admits a unique minimizer @xmath213 of @xmath96 , which determines the unique parameter estimate @xmath210 given @xmath93 . assume that @xmath201 and @xmath202 are uncorrelated and both are i.i.d . circular gaussian with means zero and covariance matrices @xmath214 and @xmath215 , respectively . moreover , assume that the sensor array is a redundancy array and @xmath93 . then @xmath216 is asymptotically an ml estimator of @xmath217 . [ thm : aml ] consider first the ula case . since @xmath170 can be uniquely determined given @xmath126 , the spa estimator is equivalent to solving the following ( nonconvex ) optimization problem : are removed in spa as well as the corresponding entries of the solution of @xmath41 . ] on the other hand , the data snapshots @xmath218 are i.i.d . gaussian with mean zero and covariance @xmath96 under the assumptions . according to the extended invariance principle ( exip ) @xcite and the derivations in @xcite , the global minimizer of ( [ formu : spa_nonconvex ] ) is a large - snapshot realization of the ml estimator of @xmath217 . similarly , the spa estimator in the redundancy sla case is the global minimizer of the following problem : then the same result follows . theorem [ thm : aml ] states that @xmath216 is asymptotically an ml estimator of @xmath217 under some technical assumptions . theorem [ thm : consistency ] shows that this estimator is also consistent . since , asymptotically , the power estimator @xmath219 may lie on the boundary of @xmath220 and thus the asymptotic normality of @xmath216 does not hold directly . however , it indeed holds in the case of @xmath196 where the true parameter @xmath221 is an interior point of @xmath220 . so we have the following theorem . under the assumptions of theorem [ thm : aml ] and @xmath196 with @xmath222 and @xmath223 , the spa estimator @xmath170 is asymptotically normal with asymptotic unbiased mean @xmath210 and covariance @xmath224 , where @xmath225 denotes the fisher information matrix of the parameter . it follows that @xmath170 is statistically asymptotically efficient . [ thm : efficiency ] the result follows from theorems [ thm : consistency ] and [ thm : aml ] and the properties of consistent ml estimators . the asymptotic covariance matrix @xmath224 is usually referred to as the crammer - rao lower bound ( crlb ) , which can be computed following from @xcite and will be omitted in this paper . alternatively , its numerical results will be presented in section [ sec : simulation ] . the spa method proposed in this paper is closely connected to spice in @xcite . the two methods adopt the same covariance fitting criteria . roughly speaking , spice can be considered as a discretized version of spa when applied to ulas and slas . however , the following two main differences make their performances very different . one is that spa is a parametric method while spice is a semiparametric method . in particular , the covariance @xmath96 is approximated in spice by discretizing the continuous range @xmath226 of the frequency . consider the ula case as an example . denote @xmath227 the discretized sampling grid of the frequency and @xmath228 the corresponding power vector , where @xmath229 denotes the grid size . then @xmath96 is expressed as which is a linear function of @xmath230 . with ( [ formu : rgrid ] ) , the covariance fitting criterion ( [ formu : criterion1 ] ) or ( [ formu : criterion2 ] ) becomes a convex function of @xmath230 and is optimized in @xcite via an iterative algorithm , named as spice . note that the characterization of @xmath96 in ( [ formu : rgrid ] ) is only an approximation since there is no guarantee that the true frequencies lie on the grid @xmath231 . thus the approximation error ( or modeling error ) , which depends on the grid density , is one potential reason causing estimation inaccuracy of spice . the other difference is that the parameter estimate of spice is obtained directly from the solution @xmath232 of the covariance fitting optimization problem while an additional postprocessing procedure is carried out in spa . in particular , a source at @xmath233 is detected in spice once @xmath234 and the frequency estimate is constrained on the grid , which will be referred to as the on - grid issue hereafter and becomes a second potential reason causing inaccuracy of spice . furthermore , the parameter estimate without the postprocessing is generally not unique according to remark [ rem : postproc ] due to an identifiability problem which also exists in the solution @xmath232 of spice . this can be a third potential reason causing inaccuracy of spice . in fact , the spice estimate might not be _ sparse _ since the support of @xmath235 can be as large as its dimension @xmath229 according to remark [ rem : postproc ] , which will be numerically verified in section [ sec : simulation ] . in summary , there are three potential reasons that may cause inaccuracy to the parameter estimation of spice , including the modeling error , the on - grid issue and the identifiability problem , rendering that spice does not possess the sparse and the statistical properties of spa presented in section [ sec : property ] . the first reason corresponds to the first difference and is introduced by the discretization , which can be alleviated by adopting a dense sampling grid but at the cost of more expensive computations . the last two are related to the second difference . inspired by spa , we present spice - pp to resolve them using the postprocessing technique in the next subsection . before proceeding to spice - pp , we compare computational costs of spice and spa . we consider only the dominant part in the following comparison , i.e. , the computation of @xmath236 or @xmath237 is excluded for both spice and spa which takes @xmath238 or @xmath239 flops , respectively , and the postprocessing and parameter solving are also excluded for spa which take @xmath240 flops . spice is an iterative algorithm whose computational complexity equals the complexity per iteration times the number of iterations , i.e. , @xmath241 in the ula case and @xmath242 in the sla case , where @xmath22 denotes the number of iterations which is hard to quantify and empirically observed to vary in different scenarios . for spa we adopt an off - the - shelf sdp solver , sdpt3 @xcite , where the interior - point method is implemented to solve the sdp . denote by @xmath243 and @xmath244 the variable size and dimension of the positive semidefinite matrix in the semidefinite constraint of an sdp , respectively . then the sdp can be solved in @xmath245 flops in the worst case according to @xcite . in the ula case , @xmath243 is on the order of @xmath246 and @xmath247 is proportional to @xmath46 . it follows that the complexity of spa is @xmath248 . in the sla case , @xmath243 and @xmath247 are on the order of @xmath249 and @xmath250 , respectively . then the complexity is @xmath251 since @xmath252 . without surprise , the order on @xmath46 or @xmath253 for spa is higher than that for spice . but the positive side is that the complexity of spa does not depend on the grid size @xmath229 which is typically much greater than @xmath46 in array processing . as a result , spa can be possibly faster than spice if a dense sampling grid ( large @xmath229 ) is adopted in spice for obtaining high accuracy , and vise versa . note that the order on @xmath46 or @xmath253 might be decreased in the future for both spa and spice if there are more sophisticated algorithms but the linear dependence on @xmath229 for spice probably can not due to the discretization ( similar results hold for other discretization - based methods ) . like spa , we modify spice and obtain the parameter estimate within three steps . firstly , @xmath96 is estimated as given the solution @xmath232 of the original spice . then the postprocessing is applied to decompose @xmath126 into @xmath254 by exploiting @xmath93 . finally , the parameter estimate is obtained via the vandermonde decomposition of @xmath154 . the modified spice algorithm is named as spice with the postprocessing , abbreviated as spice - pp . note that the frequency estimate of spice - pp is no long constrained on the grid . it is also noted that this postprocessing technique can be possibly applied to other covariance - based methods . in the limiting single - snapshot case , i.e. , when @xmath255 , the array processing problem is mathematically equivalent to spectral analysis @xcite . for the latter topic , discretization - free techniques have been recently proposed in @xcite based on atomic norm minimization ( see @xcite ) which , also called total variation norm in @xcite , is a continuous version of the @xmath0 norm . it is noted that the spa method proposed in this paper can also be applied to the single - snapshot case . in fact , the sdps in ( [ formu : sdp_nsmall ] ) and ( [ formu : sdp_nsmall_identical ] ) can be further simplified in such case . given ( [ formu : sdp_nsmall_identical ] ) as an example . it follows from @xmath256 that @xmath257 . so , an alternative formulation of ( [ formu : sdp_nsmall_identical ] ) will be it is interesting to note that , though obtained from a very different technique , ( [ formu : sdp_n1 ] ) is quite similar to the sdp formulations in @xcite . a detailed investigation of the relation is beyond the scope of this paper and will be posed as a future work since this paper is focused on the array processing applications in which the multisnapshot case is of the main interest . in this section we illustrate the performance of the proposed spa method and compare it with existing methods via numerical simulations . the methods that we consider include spice @xcite , spice - pp , iaa @xcite , music and ogsbi - svd @xcite . spice is a semiparametric method which can be roughly considered as a discretized version of spa as described in subsection [ sec : connectionspice ] . spice - pp is a modified version of spice by incorporating the postprocessing technique presented in this paper . iaa is an enhanced nonparametric method . music is a classical subspace - based parametric method . ogsbi - svd is a semiparametric method for off - grid doa estimation . the information of source number , @xmath33 , is required in music but not in the other methods . since spa and spice operate in different manners in the cases of equal / different noise variances , ` + ' is used to indicate the case when the equal noise variances assumption is imposed . for example , spa+ refers to spa with the assumption . without ambiguity , `` spa '' can refer to either the collectively called spa technique or the spa method with different noise variances ( in contrast to spa+ ) hereafter ( similarly for the use of `` spice '' ) . some setups of the algorithms above are as follows . the sdps of spa are implemented using cvx with the sdpt3 solver @xcite . spice is implemented as in @xcite and terminated when the relative change of the objective function value in two consecutive iterations falls below @xmath258 or the maximum number of iterations , set to 500 , is reached . iaa is terminated if the relative change of the @xmath11 norm of the power vector in two consecutive iterations falls below @xmath258 or the maximum number of iterations , set to 500 , is reached . ogsbi - svd is implemented as in @xcite except that the source number is unknown ( see details in @xcite ) . we compare spectra of the aforementioned methods in this subsection . in our simulation , we consider @xmath259 uncorrelated / coherent sources with power @xmath260 from directions specified by the frequency vector @xmath261 . a ula with @xmath262 is used to receive the signals . each of the source signals is randomly generated with constant amplitude and random phase , which is usually the situation in communications applications @xcite . zero - mean white circular gaussian noise is added with the noise variance @xmath263 . the signal to noise ratio ( snr ) is defined as the ratio of the minimum source power to the noise variance ( in db ) , i.e. , @xmath264 . the grid number is set to 500 for spice and iaa . our simulation results of uncorrelated sources are presented in fig . [ fig : spectra ] with respect to different settings of @xmath265 , where some curves are omitted for better visual effects . in the case of large snapshots , e.g. , @xmath266 as shown in figs . [ fig : spectra1 ] and [ fig : spectra2 ] , we observe the following phenomena . iaa produces a dense spectrum and exhibits significant resolution degradation in the moderate / low snr regime . it can not separate the first two sources in fig . [ fig : spectra2 ] . spice+ produces a sparse spectrum which contains only a few spikes in the high snr regime , e.g. , @xmath267db . however , it produces a dense spectrum due to the identifiability problem as addressed in subsection [ sec : connectionspice ] in the moderate / low snr regime , e.g. , @xmath268db , where it returns a zero estimate of the noise variance . unlike spice+ , spice always produces a dense spectrum whenever the snr is ( its spectra are omitted in fig . [ fig : spectra ] ) . without surprise , the spa method proposed in this paper always produces a sparse spectrum . small spurious spikes exhibit due to the presence of noise and absence of the knowledge of source number and the noise level . by using the postprocessing technique presented in this paper both spice+-pp and spice - pp produce sparse spectra , and their differences from spa+ ( or spa ) are caused by the modeling error of spice as described in subsection [ sec : connectionspice ] . moreover , spice+ and spice+-pp have the same spectrum in the high snr regime since the postprocessing does not alter the spectrum when spice+ has already produced a sparse spectrum . this point will be revisited later . the limiting case @xmath269 with a very low snr ( @xmath270db ) is studied in fig . [ fig : spectra3 ] , where the true covariance matrix is adopted to implement spa , spice and music . iaa is not considered in this scenario . it is shown that spa ( similarly for music ) can exactly localize the three sources , which verifies the conclusion of theorem [ thm : consistency ] that spa is statistically consistent . spice+ has low resolution in the low snr regime due to the identifiability problem and can not separate the first two sources . due to the modeling error of spice , spice+-pp is not consistent as well . though this paper is mainly focused on the case of moderate / large snapshots , it is noted that spa can be applied to the case of small or even a single snapshot as shown in fig . [ fig : spectra4 ] where music fails . [ fig : spectra_coh ] presents simulation results of coherent sources , where source 3 is exactly a replica of source 1 . it is shown that the proposed spa method has consistently good performance in the presence of complete correlation due to the adopted covariance fitting criterion as shown in @xcite . in contrast , iaa has low resolution as in the case of uncorrelated sources . music typically misses coherent sources . spice tends to miss coherent sources as well in the moderate / low snr regime . finally , note that the power estimates of the coherent sources are attenuated in spa , the reason of which should be investigated in the future . quantitative comparisons will be carried out in this subsection to demonstrate advantages of our discretization - free spa method compared to spice . readers are referred to @xcite for performance comparisons of spice with iaa and music . unlike spa , spice might produce a dense spectrum as the nonparametric methods . the frequency estimate of spice is obtained using the peaks of the spectrum following from @xcite . to illustrate effects of the discretization adopted in spice , three discretization levels are considered with the grid size @xmath271 , respectively . for convenience , the spice algorithm adopting the three discretization schemes will be referred to as spice1 , spice2 and spice3 , respectively . the ` + ' symbol will be used as before . metrics recorded include mean squared error ( mse ) and cpu time usage . the mse of the frequency estimation is computed as @xmath272 and then averaged over a number of monte carlo runs , where @xmath273 denotes the frequency estimate which is obtained by keeping the associated largest @xmath33 entries of the power estimate @xmath153 . the crlb is commonly used as a benchmark when evaluating the performance of various estimators though it is a lower bound for only unbiased estimators . note also that to compute the crlb requires the knowledge of @xmath33 which is not used in spa and spice . so , there might exist a gap between the crlb and the performance of spa or spice that we study . the simulations study both the ula and sla cases and are focused on uncorrelated sources . _ experiment 1 _ studies performance variation with respect to the snr . we consider a ula with @xmath262 . without loss of generality , @xmath274 uncorrelated sources impinge on the array with ( off - grid ) frequencies @xmath275 and @xmath276 and unit powers . notice that each frequency is a third grid interval away from the nearest grid point for spice . since the best frequency estimate for a given source is the nearest grid point , the mse of the frequency estimation of spice is lower bounded by @xmath277 regardless of the snr . the number of snapshots is set to @xmath266 and the snr varies in @xmath278db . 200 monte carlo runs are used for each algorithm to obtain the metrics , where the source signals and the noise are both i.i.d . [ fig : mse_ula ] plots mses of the simulation results , where the two cases of equal and different noise variances are separately presented to provide a better illustration . spice has a better performance with a finer discretization but lower bounded by some constant as mentioned above . to the contrary , the mses of the spa methods improve constantly with the snr and gradually approach the crlbs . both the spa methods and the spice methods have similar performance trends in the two cases of equal and different noise variances . however , spice - pps perform differently . in the former case , spice+-pps and spice+s coincide in the high snr regime since they produce the same sparse spectra as shown in subsection [ sec : spectra ] . in the latter case , spice - pps outperform the associated spices when the snr is larger than some threshold thanks to the postprocessing technique presented in this paper . but when the snr is sufficiently high , the modeling error caused by the discretization dominates the total uncertainties and further performance improvement is impossible . on the other hand , in the moderate / low snr regime where the measurement noise dominates the uncertainties , spice - pp coincides with spa as expected . note that the bad performances of spice2 + -pp and spice3 + -pp at @xmath279db are caused by very few outliers ( 1 and 4 trials , respectively , out of 200 ) , where spice+ might not converge within 500 iterations and result in a less accurate @xmath126 that is used for parameter estimation . finally , notice that spice can possibly outperform spa in the snr range @xmath280db ( depending on the discretization level ) , which will be discussed in subsection [ sec : discussion ] . , @xmath274 uncorrelated sources with @xmath281 and @xmath282 , and @xmath266 . the horizontal dashed lines are lower bounds of the spices due to the discretization.,title="fig:",width=336 ] , @xmath274 uncorrelated sources with @xmath281 and @xmath282 , and @xmath266 . the horizontal dashed lines are lower bounds of the spices due to the discretization.,title="fig:",width=336 ] fig . [ fig : time_ula ] presents cpu times of _ experiment 1_. since the postprocessing can be applied efficiently , the time usage of spice - pp is slightly longer than spice and is omitted . both spice+s and spices have similar performance trends with different discretization levels because the number of iterations used is approximately the same . as a result , the time usage of spice is proportional to the grid size @xmath229 while the proposed discretization - free spa does not depend on @xmath229 . [ fig : time_ula ] shows that when @xmath283 spice+ and spice are constantly slower than spa+ and spa , respectively . when @xmath284 , spice+ is also slower than spa+ sometimes . uncorrelated sources . some settings include : ula with @xmath262 , @xmath281 , @xmath282 and @xmath266.,width=336 ] _ experiment 2 _ studies the performance with respect to the array length @xmath46 . we repeat _ experiment 1 _ but set @xmath285db and vary @xmath46 in @xmath286 . moreover , we consider only the case @xmath283 for spice . only the case where @xmath287 is considered for spa ( excluding spa+ ) due to time consideration . the simulation results are presented in fig . [ fig : ula_varym ] . since crlb+ and crlb are slightly different and almost undistinguishable , only crlb+ is plotted . as before , spice3 + , spice3 and spice3 + -pp share the same lower bound due to discretization and the property of the spice+ estimator , while spa+ , spa and spice3-pp can outperform this bound . as @xmath46 increases , e.g. , when @xmath288 , the performances of spa and spice3-pp hardly improve and the gaps between the algorithms and crlb+ become larger . two possible reasons are as follows : 1 ) unlike crlb+ , spa does not use the knowledge of @xmath33 but @xmath93 , which becomes rougher when @xmath46 increases and @xmath33 keeps unaltered , and 2 ) the modeling error of spice increases as @xmath46 increases . the figure at the bottom indicates that the speed of spa+ scales well with @xmath46 in our considered scenario and is faster than spice3 + as @xmath289 and constantly faster than spice3 . the number of iterations of spice3 + is empirically observed to decrease with increasing array length in our considered scenario , leading to a seeming strange result that spice3 + gets faster when @xmath46 increases . -element ula compared with spice . some settings include : @xmath274 uncorrelated sources with @xmath290 and @xmath282 , @xmath291 and @xmath266.,title="fig:",width=336 ] -element ula compared with spice . some settings include : @xmath274 uncorrelated sources with @xmath290 and @xmath282 , @xmath291 and @xmath266.,title="fig:",width=336 ] _ experiment 3 _ studies the sla case where a 4-element redundant array @xmath88 is considered . we try to verify theorem [ thm : efficiency ] and attempt to locate maximally @xmath292 uncorrelated sources with the frequency vector @xmath293 and power vector @xmath294 . moreover , we set @xmath285db and vary the number of snapshots @xmath4 in @xmath295 . 1000 monte carlo runs are used to obtain the metrics for each @xmath4 . to evaluate the performance , we calculate the ratio @xmath296 for each algorithm at each @xmath4 . for an unbiased estimator , the ratio is called its efficiency . the larger the ratio is ( @xmath297 for an unbiased estimator ) , the more accurate the estimator will be . [ table : sla_frequency ] presents the simulation results . remarkably , spa+ ( or spa ) can outperform the crlb+ ( or crlb ) when @xmath298 since the ratios are larger than 1 ( or 0.9281 ) . though spice+ ( or spice ) can outperform the crlb+ ( or crlb ) as well at @xmath266 , its gap to the crlb+ ( or crlb ) becomes larger as @xmath4 increases . moreover , spice+-pp ( or spice - pp ) is consistently better than spice+ ( or spice ) and worse than spa+ ( or spa ) . notice that as @xmath4 gets larger , the gap between spa+ ( or spa ) and 1 ( or 0.9281 ) becomes smaller , which is consistent with the conclusions of theorem [ thm : efficiency ] that @xmath299 , as @xmath300 . it is also noted that the source power and noise variance estimates of the proposed spa method have similar performances and their metrics are omitted . the frequency and power estimates of spa+ at @xmath266 are plotted in fig . [ fig : sla_n200 ] . .mses of frequency estimates of spa compared with spice and crlb . each of the presented values is the ratio @xmath301 ( the larger the better ) . [ cols="<,>,>,>,>",options="header " , ] [ table : sla_frequency ] uncorrelated sources with a 4-element sla @xmath88 when @xmath266 and @xmath291 ( 1000 monte carlo runs ) . black circles indicate the true frequencies and corresponding powers.,width=336 ] we compare spa with the off - grid method ogsbi - svd in @xcite in this subsection . ogsbi - svd utilizes the sparse bayesian learning ( sbl ) technique @xcite , which mimics the ml estimation according to @xcite , and is based on an off - grid observation model which is a first - order approximation of the exact model with continuous frequencies and has a reduced modeling error ( recall that the observation model of spice and many others is a zeroth - order approximation while spa relies on the exact model ) . in ogsbi - svd the grid offset ( the distance from a true frequency location to its nearest grid point ) is estimated jointly with the sparse signal and singular value decomposition ( svd ) is used for reducing dimension of the observed data and faster convergence . since the covariance @xmath96 is also involved in ogsbi - svd as in spice which can not be appropriately separated into the source and noise parts , ogsbi - svd usually produces a dense spectrum like spice . in _ experiment 4 _ , we repeat _ experiment 1 _ for spa+ and ogsbi - svd and consider two discretization levels , @xmath302 , for ogsbi - svd ( denoted by ` 1 ' , ` 2 ' respectively ) . the mses are plotted in fig . [ fig : mse_ogsbi ] . as expected , ogsbi - svd can exceed the lower bounds of on - grid methods ( horizontal dashed lines ) . as snr increases , e.g. , at @xmath303db , the modeling error of ogsbi - svd becomes nonnegligible . then spa is more accurate than ogsbi - svd . note also that spa is more robust to noise . for example , spa can produce satisfactory results at @xmath304db while ogsbi - svd can not . in computational speed , spa is slightly slower than ogsbi - svd with @xmath305 and about 3 times faster with @xmath306 . uncorrelated sources using spa compared with ogsbi - svd and the crlb , with @xmath266 . the horizontal dashed lines refer to lower bounds of on - grid methods with @xmath302.,width=336 ] in _ experiment 5 _ , we repeat the simulation by fixing @xmath291 and varying the number of snapshots @xmath4 from 1 to 200 at a step of 3 . simulation results presented in fig . [ fig : mse_ogsbi_varyn ] show that spa has consistently satisfactory performance though it is slightly worse than ogsbi - svd ( the mses of spa are less than @xmath307 times those of ogsbi - svd in most of the scenarios ) . since ogsbi - svd mimics the ml estimation and spa is a large - snapshot realization of the ml , the performance gap between them vanishes as @xmath4 increases . uncorrelated sources using spa compared with ogsbi - svd and the crlb , with @xmath291 . the horizontal dashed lines refer to lower bounds of on - grid methods with @xmath302.,width=336 ] if an infinitely fine discretization scheme is allowed in spice ( to eliminate the modeling error ) , then spice and spa will produce the same covariance matrix estimate @xmath126 . however , spa always has a sparse spectrum while spice may produce a dense one . the two spectra can be considered as two different decomposition schemes applied to @xmath126 to estimate the source powers . the sparse decomposition of spa has good statistical properties as presented in subsection [ sec : statproperty ] while the dense decomposition of spice does not . moreover , as shown in figs . [ fig : spectra3 ] and [ fig : spectra_coh2 ] , the dense decomposition of spice has potentially inferior resolution . however , the simulation results of fig . [ fig : mse_ula ] show that spice can outperform spa in terms of mse in the middle range of the snr ( similar results are presented in figs . [ fig : mse_ogsbi ] and [ fig : mse_ogsbi_varyn ] for ogsbi - svd ) . it is because that 1 ) spa is empirically observed to produce a more heavy - tailed frequency estimator than spice since the frequency of a source is determined by a single point unlike spice for which the frequency estimate is given by the peaks of the spectrum , and 2 ) the mse metric is sensitive to heavy - tailed estimators ( in fact , a careful study reveals that spa+ is more accurate than spice3 + in over half of the monte carlo runs in the snr range @xmath308db in fig . [ fig : mse_ula ] ) . in this paper , the linear array signal processing problem was studied and a discretization - free technique named as spa was proposed . the new method adopts the covariance fitting criteria of spice and was formulated as an sdp followed by a postprocessing technique . spa is a parametric method and guarantees to produce a sparse parameter estimate and in the mean time enjoys several other merits . its asymptotic statistical properties were analyzed and practical performance was demonstrated via simulations compared to existing methods . the following directions should be investigated in the future , some of which have been mentioned in the main context of this paper : 1 ) connection of the proposed spa method to the atomic norm - based discretization - free methods in @xcite for spectral analysis , 2 ) postprocessing techniques for general but not redundancy slas to utilize the full information of the range of the source number , 3 ) performance analysis of spa in the finite - snapshot case while this paper is mainly on its asymptotic statistical properties , 4 ) fast implementations of the spa method via developing more computationally efficient algorithms for solving the sdps involved , 5 ) modified spa methods with automatic source number estimation , and 6 ) discretization - free methods for array processing with general array geometries . t. yardibi , j. li , p. stoica , m. xue , and a. b. baggeroer , `` source localization and sensing : a nonparametric iterative adaptive approach based on weighted least squares , '' _ ieee transactions on aerospace and electronic systems _ , vol . 46 , no . 1 , pp . 425443 , 2010 . d. malioutov , m. cetin , and a. willsky , `` a sparse signal reconstruction perspective for source localization with sensor arrays , '' _ ieee transactions on signal processing _ 53 , no . 8 , pp . 30103022 , 2005 . a. c. gurbuz , v. cevher , and j. h. mcclellan , `` bearing estimation via spatial sparsity using compressive sensing , '' _ ieee transactions on aerospace and electronic systems _ 48 , no . 2 , pp . 13581369 , 2012 . h. abeida , q. zhang , j. li , and n. merabtine , `` iterative sparse asymptotic minimum variance based approaches for array processing , '' _ ieee transactions on signal processing _ , 61 , no . 4 , pp . 933944 , 2013 . p. stoica , p. babu , and j. li , `` new method of sparse parameter estimation in separable models and its use for spectral analysis of irregularly sampled data , '' _ ieee transactions on signal processing _ , 59 , no . 1 , pp . 3547 , 2011 . c. d. austin , r. l. moses , j. n. ash , and e. ertin , `` on the relation between sparse reconstruction and parameter estimation with model order selection , '' _ selected topics in signal processing , ieee journal of _ , vol . 4 , no . 3 , pp . 560570 , 2010 . z. yang , c. zhang , and l. xie , `` robustly stable signal recovery in compressed sensing with structured matrix perturbation , '' _ ieee transactions on signal processing _ 60 , no . 9 , pp . 46584671 , 2012 . l. hu , z. shi , j. zhou , and q. fu , `` compressed sensing of complex sinusoids : an approach based on dictionary refinement , '' _ ieee transactions on signal processing _ , vol . 60 , no . 7 , 38093822 , 2012 . p. stoica and a. nehorai , `` performance study of conditional and unconditional direction - of - arrival estimation , '' _ ieee transactions on acoustics , speech and signal processing _ , vol . 38 , no . 10 , pp . 17831795 , 1990 . z. yang , `` analysis , algorithms and applications of compressed sensing , '' ph.d . dissertation , nanyang technological university , available online at https://dl.dropboxusercontent.com/u/34897711/thesis_yangzai.pdf , 2013 .

the paradigm of layered networks is used to describe many real - world systems from biological networks , to social organizations and transportation systems . while recently there has been much progress in understanding the general properties of multilayer networks , our understanding of how to control such systems remains limited . one fundamental aspect that makes this endeavor challenging is that each layer can operate at a different timescale , thus we can not directly apply standard ideas from structural control theory of individual networks . here we address the problem of controlling multilayer and multi - timescale networks focusing on two - layer multiplex networks with one - to - one interlayer coupling . we investigate the practically relevant case when the control signal is applied to the nodes of one layer . we develop a theory based on disjoint path covers to determine the minimum number of inputs ( ) necessary for full control . we show that if both layers operate on the same timescale then the network structure of both layers equally affect controllability . in the presence of timescale separation , controllability is enhanced if the controller interacts with the faster layer : decreases as the timescale difference increases up to a critical timescale difference , above which remains constant and is completely determined by the faster layer . we show that the critical timescale difference is large if layer i is easy and layer ii is hard to control in isolation . in contrast , control becomes increasingly difficult if the controller interacts with the layer operating on the slower timescale and increasing timescale separation leads to increased , again up to a critical value , above which still depends on the structure of both layers . this critical value is largely determined by the longest path in the faster layer that does not involve cycles . by identifying the underlying mechanisms that connect timescale difference and controllability for a simplified model , we provide crucial insight into disentangling how our ability to control real interacting complex systems is affected by a variety of sources of complexity .

1608.03604

over the past two decades , the theory of networks proved to be a powerful tool for understanding individual complex systems @xcite . however , it is now increasingly appreciated that complex systems do not exist in isolation , but interact with each other @xcite . indeed , an array of phenomena from cascading failures @xcite to diffusion @xcite can only be fully understood if these interactions are taken into account . traditional network theory is not sufficient to describe the structure of such systems , so in response to this challenge , the paradigm of multilayer networks is being actively developed . here we study a fundamental , yet overlooked aspect of multilayer networks : each individual layer can operate at a different timescale . particularly , we address the problem of controlling multilayer , multi - timescale systems focusing on two - layer multiplex networks . recently significant efforts have been made to uncover how the underlying network structure of a system affects our ability to influence its behavior @xcite . however , despite the appearance of coupled systems from infrastructure to biology , the existing literature with a few notable exceptions @xcite has focused on control of networks in isolation , and the role of timescales remains unexplored . control of multilayer networks is important for many applications . for example , consider a ceo aiming to lead a company consisting of employees and management . studying the network of managers or the network of employees in isolation does not take into account important interactions between the different levels of hierarchy of the company . on the other hand , treating the system as one large network ignores important differences between the dynamics of the different levels , e.g. management may meet weekly , while employees are in daily interaction . in general , the interaction of timescales plays an important role in organization theory @xcite . or consider gene regulation in a living cell . external stimuli activate signaling pathways which through a web of protein - protein interactions affect transcription factors responsible for gene expression . the activation of a signaling pathway happens on the timescale of seconds , while gene expression typically takes hours @xcite . as a third example , consider an operator of an online social network who wants to enhance the spread of certain information by interacting with its users . however , a user may subscribe to multiple social networking services and may opt to share news encountered in one network through a different one out of reach of the operator . the dynamics of user interaction on different websites can be very different depending on user habits and the services offered @xcite . for example the url shortening service bit.ly reports that the half - life of shared links depends on the social networking platform used : half the clicks on a link happened within 2.8 hours after posting on twitter , within 3.2 hours on facebook and within 7.4 hours on youtube @xcite . common features of these examples are that ( i ) each interacting subsystem is described by a separate complex network ; ( ii ) the dynamics of each subsystem operate on a different , but often comparable timescale and ( iii ) the external controller directly interacts with only one of the subsystems . here we study the control properties of a model that incorporates these common features , yet remains tractable . more specifically , we study discrete - time linear dynamics on two - layer multiplex networks , meaning that we assume one - to - one coupling between the nodes of the two layers . this choice ensures both analytical tractability and the isolation of the role of timescales from the effect of more complex multilayer network structure . identifying the underlying mechanisms that govern the controllability of this simple model provides crucial insight into disentangling how our ability to control real interacting complex systems is affected by a variety of sources of complexity . so far only limited work investigated controllability of multilayer networks . menichetti et al . investigated the controllability of two - layer multiplex networks governed by linear dynamics such that the dynamics of the two layers are not coupled , but the input signals in the two layers are applied to the same set of nodes @xcite . yuan et al . identified the minimum number of inputs necessary for full control of diffusion dynamics , allowing the controller to interact with any layer @xcite . zhang et al . investigated the controllable subspace of multilayer networks with linear dynamics without timescale separation if the controller is limited to interact with only one layer ; showing that it is more efficient to directly control peripheral nodes than central ones @xcite . here we also limit the controller to one layer , yet by exploring the minimum input problem , we offer a direct metric which allows us to compare our findings to previous results for single - layer networks @xcite . more so , the key innovation of our work is that we take into account the timescale of the dynamics of each layer , a mostly overlooked aspect of multilayer networks . it is worth mentioning the recent work investigating the related , but distinct problem of controllability of networks with time - delayed linear dynamics @xcite . the key difference between time - delay and timescale difference is that for time - delayed dynamics the state of a node will depend on some previous state of its neighbors ; however , the typical time to change the state of a node remains the same throughout the system . while in case of timescale difference , the typical time needed for changes to happen can be different in different parts of the system . in the next section , we introduce a simple model that captures some common properties of multilayer networks and we describe the problem setup . in sec . [ sec : min_input_prob ] , we develop a theory to determine the minimum number of inputs required for controlling multiplex , multi - timescale networks with discrete - time linear dynamics relying on graph combinatorial methods . in sec . [ sec : results ] , we use networks with tunable degree distribution to systematically uncover the role of network structure and timescale separation . we study three scenarions : no timescale separation , layer i operates faster and layer ii operates faster . finally , in sec . [ sec : conclusions ] we provide a discussion of our results and we outline open questions . we aim to study the controllability of coupled complex dynamical systems with the following properties : ( i ) each subsystem ( layer ) is described by a complex network ; ( ii ) the operation of each layer is characterized by a different timescale and ( iii ) the controller only interacts directly with one of the layers . we propose a model that satisfies these requirements and yet is simple enough to remain tractable . we focus on two - layer multiplex systems , meaning that there is a one - to - one correspondence between the nodes of the two layers . the model is defined by a weighted directed two - layer multiplex network @xmath1 which consists of two networks @xmath2 and @xmath3 called layers and a set of links @xmath4 connecting the nodes of the different layers . each layer @xmath5 ( where @xmath6 ) consists of a set of nodes @xmath7 and a set of links @xmath8 , where a directed link @xmath9 is an ordered node pair and a weight representing that node @xmath10 influences node @xmath11 with strength @xmath12 . the two layers are connected by link set @xmath13 , in other words , there is directed one - to - one coupling from layer i to layer ii ( fig . [ fig:3examples]a ) . although the links are weighted , the exact values of the weights do not have to be known for our purposes . our goal is to control the system by only interacting directly with layer i. we study linear discrete - time dynamics @xmath14 where @xmath15 and @xmath16 represent the state of nodes in layer i and ii ; the matrices @xmath17 and @xmath18 are the transposed weighted adjacency matrices of layer i and ii , capturing their internal dynamics . the weighted diagonal matrix @xmath19 captures how layer i affects layer ii . vector @xmath20 provides the set of independent inputs and the matrix @xmath21 defines how the inputs are coupled to the system . to differentiate between the function @xmath22 and an instance of the function at a given time step , we refer to a component @xmath23 of vector @xmath22 as an independent input , and we call its value at time step @xmath24 , @xmath25 , a signal . finally , @xmath26 are the timescale parameters of each subsystem , meaning that the state of layer i is updated according to eq . ( [ eq : dynamics ] ) every @xmath27th time step ; and layer ii is updated every @xmath28th time step . and @xmath29 is the kronecker comb , meaning that layer i directly impacts the dynamics of layer ii if the two layers simultaneously update . we investigate three scenarios : ( i ) the subsystems operate on the same timescale , i.e. @xmath30 ; ( ii ) layer i updates faster , i.e. @xmath31 and @xmath32 ; and ( iii ) layer ii updates faster @xmath33 and @xmath34 . we seek full control of the system as defined by kalman @xcite , meaning that with the proper choice of @xmath22 , we can steer the system from any initial state to any final state in finite time . to characterize controllability , we aim to design a matrix @xmath35 such that the system is controllable and the number of independent control inputs , @xmath36 , is minimized . the minimum number of inputs , @xmath0 , serves as our measure of how difficult it is to control the system . to find a robust and efficient algorithm to determine @xmath0 , we rely on the framework of structural controllability @xcite . we say that a matrix @xmath37 has the same structure as @xmath38 , if the zero / nonzero elements of @xmath38 and @xmath37 are in the same position , and only the value of the nonzero entries can be different , in other words , in the corresponding network the links connect the same nodes , only the link weights can differ . a linear system of eq . ( [ eq : dynamics ] ) defined by matrices @xmath39 is structurally controllable if there exists matrices with the same structure @xmath40 such that the dynamics defined by @xmath40 are controllable according to the definition of kalman . note that ultimately we are interested in controllability and not structural controllability . yet , structural controllability is a useful tool because ( i ) if a linear system is structurally controllable , it is controllable for almost all link weight combinations @xcite and ( ii ) determining structural controllability can be mapped to a graph combinatorial problem allowing for efficient and numerically robust algorithms . before addressing the minimum input problem of multiplex networks , we revisit the case of single - layer networks by providing an alternative explanation of the minimum input theorem of liu et al . this new approach readily lends itself to be extended to multiplex , multi - timescale networks . thus providing the basis for sec . [ sec : multiplex_net ] , in which we develop an algorithm to determine @xmath0 for two - layer multiplex networks . the linear discrete - time dynamics associated to a single - layer weighted directed network @xmath41 are formulated as @xmath42 where @xmath43 , @xmath38 , @xmath35 and @xmath22 are defined similarly as in eq . ( [ eq : dynamics ] ) ( fig . [ fig : singlenet]a ) . to obtain a graph combinatorial condition for structural controllability we rely on the dynamic graph @xmath44 , which represents the time evolution of a system from @xmath45 to @xmath46 @xcite . each node @xmath47 in @xmath41 is split into @xmath48 copies @xmath49 , each copy @xmath50 represents the state of node @xmath47 at time step @xmath51 . we add a directed link @xmath52 for @xmath53 if they are connected by a directed link @xmath54 in the original network , representing that the state of node @xmath55 at time @xmath56 depends on the state of its in - neighbours at the previous time step . to account for the controller , for each independent input we create @xmath57 nodes @xmath58 ( @xmath59 ; @xmath53 ) each representing a control signal ( i.e. the value of the @xmath60th input at time step @xmath51 ) . we draw a directed link @xmath61 for @xmath53 if @xmath62 , where @xmath63 is an element of matrix @xmath35 . according to theorem 15.1 of ref . @xcite , a linear system @xmath64 is structurally controllable if and only if in the associated dynamic graph @xmath65 node sets @xmath66 ( green nodes in fig . [ fig : singlenet]b ) and @xmath67 ( blue nodes ) are connected by @xmath68 disjoint paths ( red links ) , i.e. there exists a set of disjoint paths @xmath69 such that @xmath70 contains the set of starting points and @xmath71 is the set of endpoints . a path @xmath72 of length @xmath73 between node @xmath74 and @xmath75 is a sequence of @xmath73 consecutive links @xmath76 $ ] such that each node is traversed only once . node @xmath74 is the starting point and @xmath75 is the endpoint of @xmath72 . two paths @xmath77 and @xmath78 are disjoint if no node is traversed by both @xmath77 and @xmath78 , a set of paths is disjoint if all paths in the set are pairwise disjoint . a possible interpretation of this result is that if a @xmath79 path has starting point @xmath80 and endpoint @xmath81 , we say that the signal @xmath82 is assigned to set @xmath83 , the state of node @xmath84 at time @xmath85 , through path @xmath79 . therefore we refer to path @xmath79 as a control path . the clear meaning of the dynamic graph and the control paths makes this condition useful to formulate proofs and to interpret results . however , it is rarely implemented to test controllability of large networks , because the size of the dynamical graph grows as @xmath86 , rendering such algorithms too slow . in the following , we provide a condition that only requires the dynamic graph @xmath87 as input ; therefore it is more suitable for practical purposes . it was shown in refs . @xcite that a linear system @xmath64 is structurally controllable if and only if ( i ) in @xmath87 we can connect nodes @xmath88 ( green nodes in fig . [ fig : singlenet]c ) and nodes @xmath89 ( blue nodes ) via @xmath68 disjoint paths ( red links ) and ( ii ) all nodes are accessible from the inputs . this result can be understood as a self - consistent version of the previous condition involving @xmath65 : instead of keeping track of the entire control paths as we previously did , we concentrate on a single time step . consider the dynamic graph @xmath87 representing the time evolution of the system from @xmath45 to @xmath90 , and assume that the system is controllable . by definition we can set the state of each node independently at @xmath45 ; therefore we can treat them as control signals to control the system at a later time step . now let us aim to control the system at @xmath90 , according to our previous condition , it is necessary that @xmath68 disjoint paths exist between nodes @xmath88 and nodes @xmath89 . this is exactly requirement ( i ) , together with the accessibility requirement ( ii ) it is a sufficient and necessary condition . note that @xmath91 is a bipartite network ( each link is connected to exactly one node in @xmath92 and one node in @xmath93 ) and each disjoint path in @xmath87 is a single link . the minimum input problem aims to identify the minimum number of inputs that guarantee controllability for a given network , in other words , the goal is to design a @xmath94 for a given @xmath38 such that @xmath36 is minimized . for this we consider the dynamic graph @xmath87 without nodes representing control signals . we then find a maximum cardinality matching , where a matching is a set of links that do not share an endpoint . the matching is a set of disjoint paths connecting node sets @xmath95 and @xmath93 . controllability requires @xmath68 disjoint paths between @xmath92 and @xmath93 ; therefore @xmath96 , where @xmath97 is the size of the maximum matching ( if @xmath98 , @xmath99 ) . allowing the inputs to be connected to multiple nodes we can guarantee that all nodes are accessible from the inputs . thus we recovered the minimum input theorem of liu et al . @xcite . in summary , by relying on a self - consistent condition for structural controllability we re - derived the known result that identifying @xmath0 is equivalent to finding a maximum matching in @xmath87 . in the next section we show that this new self - consistent approach lends itself to be extended to the multiplex , multi - timescale model defined by eq . ( [ eq : dynamics ] ) , allowing us to derive analogous method to identify @xmath0 . to find the minimum number of inputs @xmath0 for multiplex , multi - timescale networks , we first extend the definition of the dynamic graph . we define the dynamic graph @xmath100 such that it captures the time evolution of a multiplex system defined by @xmath39 and eq . ( [ eq : dynamics ] ) from @xmath45 to @xmath101 . for sake of brevity , we assume that @xmath102 and @xmath103 , the case of @xmath104 and @xmath105 is treated similarly ( fig . [ fig:3examples]d ) . each node @xmath106 in layer i is split into @xmath107 copies @xmath108 ; each node @xmath109 in layer ii is split into two copies @xmath110 , because layer ii does not update during the intermediate time steps . we draw a link from @xmath111 to @xmath112 ( @xmath113 ) if they are connected in layer i by a directed link @xmath114 , and similarly we connect @xmath115 to @xmath116 if they are connected in layer ii . in addition we draw a link between each pair @xmath117 and @xmath118 to account for the interconnectedness . as a natural extension of self - consistent approach introduced in sec . [ sec : single - layer_net ] , assume that the system is controllable . if the system is controllable , we can set the state of each node independently at @xmath45 . to control the system at @xmath101 , all nodes at @xmath101 in @xmath100 ( blue nodes in fig . [ fig:3examples ] ) have to be connected to a node at @xmath45 or to a control signal ( green nodes ) via a disjoint path ( red links ) . in other words , a linear two - layer system @xmath39 is structurally controllable only if there exists @xmath119 disjoint paths in the dynamic graph connecting node set @xmath120 and node set @xmath121 . in other words , a linear two - layer system @xmath39 is structurally controllable only if there exists @xmath119 disjoint paths in the dynamic graph connecting node set @xmath122 and node set @xmath121 . to test whether the system is controllable by @xmath36 independent inputs , we need to find a @xmath123 such that the system is controllable . we do not have to check all possibilities , because if such @xmath35 exists , then the system is also controllable for @xmath124 where @xmath125 has no zero elements ; therefore , we only check the case when each input is connected to each node in layer i. given matrices @xmath126 , we now have to count the number of disjoint paths connecting @xmath92 and @xmath127 in the corresponding dynamic graph @xmath128 . we find these paths using maximum flow : we set the capacity of each link and each node to 1 , we then find the maximum flow connecting source node set @xmath92 to target node set @xmath127 using any maximum flow algorithm of choice . if the system is structurally controllable , the maximum flow equals to @xmath119 ; if it is less than @xmath119 , additional inputs are needed . we can now identify the minimum number of inputs @xmath0 by systematically scanning possible values of @xmath36 . a simple approach is to first set @xmath129 , and test if the system is controllable . if not , increase @xmath36 by one . repeat this until the smallest @xmath36 yielding full control is found . significant increase in speed is possible if we find the minimum value of @xmath36 using bisection . we initially know that @xmath130 . we set @xmath131 , and test if the system is controllable . if yes , we set @xmath132 ; if no , we set @xmath133 . we repeat this until @xmath134 , which provides @xmath0 . for implementation , we used google or - tools and igraph python packages @xcite . the one - to - one coupling between layer i and layer ii guarantees that full control is possible with at most @xmath68 independent inputs ; therefore we often normalize @xmath0 by @xmath68 , i.e. @xmath135 . note that in the above argument we rely on the test of structural controllability based on the dynamic graph , which was originally introduced for single - timescale networks @xcite . the sufficiency of the condition relies on the fact that the zero is the only degenerate eigenvalue of a matrix @xmath38 if the nonzero elements of @xmath38 are uncorrelated . however , this might not remain true for the spectrum of @xmath136 , where @xmath137 , due to correlations arising in the nonzero elements of @xmath136 . if a @xmath138 eigenvalue has larger geometric multiplicity than the multiplicity of @xmath139 , @xmath0 would be larger than predicted by the dynamic graph ; if a @xmath138 eigenvalue has larger geometric multiplicity than @xmath140 but smaller than the multiplicity of zero , it does not affect @xmath0 , but may require connecting an input to multiple nodes @xcite . in the @xmath141 and @xmath34 case , a control signal is only injected into layer ii every @xmath27 time step ( fig . [ fig:3examples]d ) ; therefore , the spectrum of @xmath142 becomes relevant . however , we are interested in large and sparse complex networks whose spectra is dominated by the zero eigenvalue @xcite . therefore it is reasonable to expect that the spectrum of @xmath136 will be dominated by zero eigenvalues as well . meaning that the minimum number of inputs is correctly given by this graph combinatorial condition . furthermore the one - to - one coupling between the layers guarantees that control is possible by only interacting with layer i directly . so far , we developed a method to characterize controllability of a multiplex , multi - timescale system based on the underlying network structure and the timescale of each of its layers . in the next section , we rely on these tools to systematic study how network characteristics and timescales affect @xmath0 . in this section we investigate how different timescales and the degree distribution of each layer affect controllability . for timescales , we consider three scenarios : ( i ) the subsystems operate on the same timescale , i.e. @xmath30 ; ( ii ) layer i updates faster , i.e. @xmath31 and @xmath32 ; and ( iii ) layer ii updates faster @xmath33 and @xmath34 . to uncover the effect of degree distribution , we consider layers with poisson ( er ) or scale - free ( sf ) degree distribution , the latter meaning that the distribution has a power - law tail . we generate scale - free layers using the static model @xcite : we start with @xmath68 unconnected nodes . each node @xmath47 is assigned two hidden parameters @xmath143 and @xmath144 , where @xmath145 . the weights are then shuffled to eliminate any correlations of the in- and out - degree of individual nodes and between layers . we then randomly place @xmath146 directed links by choosing the start- and endpoint of the link with probability proportional to @xmath147 and @xmath148 , respectively . for large @xmath68 this yields the degree distribution @xmath149}{\zeta_\text{in / out}}\frac{\gamma(k-1/\zeta_\text{in / out},c[1-\zeta_\text{in / out}])}{\gamma(k+1)},\ ] ] where @xmath150 is equal to the average degree , and @xmath151 is the upper incomplete gamma function . for large @xmath152 , @xmath153 , where @xmath154 is the exponent characterizing the tail of the distribution . to reduce the number of parameters we only study layers with symmetric degree distribution , e.g. @xmath155 ; however , the in- and out - degree of a specific node can be different . in the special case when both layers operate on the same timescale , i.e. @xmath30 ( fig . [ fig:3examples]b ) , there is no qualitative difference between the dynamics of the layers . the reason why the system can not be treated as a single large network is that we are only allowed to directly interact with layer i. recently iudice et al . developed methodology to identify @xmath0 if the control signals can only be connected to a subset of nodes @xcite . however , the one - to - one coupling between the layers enables us to find @xmath0 using a simpler approach . finding @xmath0 for a single - layer network is equivalent to finding a maximum matching of the network @xcite . a matching is a set of directed links that do not share starting or end points , and a node is unmatched if there is no link in the matching pointing at it . liu et al . showed that full control of a network is possible if each unmatched node is controlled directly by an independent input ; therefore @xmath0 is provided by the minimum number of unmatched nodes . to determine @xmath0 for a two - layer network , we first find a maximum matching of the combined network of layer i and layer ii . if there are no unmatched nodes in layer ii , we only have to interact with layer i ; therefore we are done . if a node @xmath156 is unmatched in layer ii , @xmath157 is necessarily matched by some node @xmath158 , otherwise the size of the matching could be increased by adding @xmath159 . by taking out the link @xmath160 from the matching and including @xmath161 the size of the maximum matching does not change , and we moved the unmatched node from layer ii to layer i. we repeat this for all unmatched nodes in layer ii . ( note that it may be necessary to connect inputs to additional nodes so that all nodes are reached by the control signals . due to the one - to - one coupling between the layers this too can be accomplished by interacting only with layer i. ) this simplified method allows faster identification of @xmath0 using the hopcroft - karp algorithm @xcite and analytically solving @xmath162 for random networks based on calculating the fraction of always matched nodes as described in appendix [ app : sec : analytical ] @xcite . first , we measure @xmath163 while fixing the average degree of layer ii ( @xmath164 ) and varying the average degree of layer i ( @xmath165 ) . for both er - er and sf - sf networks , we find that @xmath163 decreases for increasing values of @xmath165 and converges to @xmath166 , the normalized number of inputs needed to control layer ii in isolation ( fig . [ fig : seciiia]a ) . the latter observation is easily understood : @xmath163 is determined by the fraction of unmatched nodes in the combined network of the two layers ; if @xmath165 is high enough , layer i is perfectly matched ; therefore all unmatched nodes are in layer ii . based on the same argument , @xmath167 also serves as a lower bound for @xmath163 . varying both @xmath165 and @xmath164 for er - er and both @xmath168 and @xmath169 for sf - sf with constant average degrees @xmath170 , we find that dense networks require less inputs than sparse networks ( fig . [ fig : seciiia]b ) and degree heterogeneity makes control increasingly difficult ( fig . [ fig : seciiia]c ) in line with results for single - layer networks @xcite . we also observe that @xmath163 is invariant to exchanging layer i and layer ii . this is explained by the fact that the size of the maximum matching is invariant to flipping the direction of all links , and on the ensemble level this is the same as swapping the two layers for networks with @xmath171 . in summary , for no timescale separation controllability is equally affected by the network structure of both layers , and @xmath163 is greater or equal to the number of inputs necessary to control any of its layers in isolation . similarly to single - layer networks , networks with low average degree and high degree heterogeneity require more independent inputs than sparse homogeneous networks . in the previous section we found that the network structure of the two layers equally affect @xmath163 if @xmath30 . this is not the case if the timescales are different , for example if layer i updates faster than layer ii , we expect that we need fewer inputs than in the same timescale case by the virtue of having more opportunity to interact with the faster system ( fig . [ fig:3examples]b ) . in this section we systematically study this effect using the algorithm described in sec . [ sec : multiplex_net ] and analytical arguments . by measuring @xmath163 for er - er and sf - sf networks as a function of @xmath28 , we find that @xmath163 monotonically decreases with increasing @xmath28 ( fig . [ fig : seciiib]a ) , confirming our expectations . for both er - er and sf - sf networks @xmath172 converges to @xmath173 which is the normalized number of inputs needed to control layer i in isolation . this can be understood by the following argument : suppose that @xmath174 , the maximum number of time steps needed to impose control on any network with @xmath68 nodes @xcite . we use the state of layer i at @xmath45 to set the state of layer ii at @xmath175 , and we have @xmath68 time steps to impose control on layer i as if it was just by itself . for a given network we define the critical timescale parameter @xmath176 as the minimum value of @xmath28 for which @xmath177 . above the critical timescale separation , layer i completely determines @xmath172 independent of the structure of layer ii , in other words , the multiplex nature of the system no longer plays a role in determining @xmath163 . measuring @xmath178 we find that for both er - er and sf - sf networks @xmath178 monotonically increases with increasing @xmath165 for fixed @xmath164 , and decreases with increasing @xmath164 for fixed @xmath165 ( fig [ fig : seciiib]b ) . that is @xmath178 is the highest if layer i is dense and layer ii is sparse . sf - sf networks have significantly lower @xmath178 than er - er networks with the same average degree . to understand the observed pattern we provide an approximation to calculate @xmath178 . we call a node @xmath157 externally controlled if in the dynamic graph @xmath179 is connected to an external signal @xmath180 via a disjoint control path ( e.g. nodes @xmath181 and @xmath182 in fig . [ fig:3examples]c ) , and the number of such nodes is denoted by @xmath183 . we have previously shown that we require @xmath184 independent inputs at @xmath178 . for each independent input and each time step , we have one control signal @xmath58 ; therefore we need timescale parameter @xmath185 to insert enough signals required by the @xmath186 externally controlled nodes , where @xmath187 is the ceiling function . equation ( [ eq : tauiic ] ) is not yet useful as it contains @xmath178 on both side . observing that @xmath183 is a monotonically increasing function of @xmath28 and @xmath188 , we can write @xmath189 in the special case when layer ii is fully connected , @xmath190 and @xmath191 . in the case when layer ii is entirely disconnected , i.e. is composed of isolated nodes , @xmath192 . these two opposite limiting cases suggest that it is reasonable to approximate @xmath186 by its lower bound : @xmath193 which entirely depends on quantities that we can easily measure or analytically compute . we find that eq . ( [ eq : tauiic_approx ] ) preforms remarkably well : figure [ fig : seciiib]b compares direct measurements of @xmath178 to approximations obtained by using measurements and analytically computed values of @xmath194 and @xmath184 . the approximation based on measurements out performs the analytical calculations , because the analytical results provide the expectation value of the numerator and denominator for er and sf network ensembles ; and therefore the ceiling function is applied to the fraction of averages , instead of averaging after applying the ceiling function . to further test the eq . ( [ eq : tauiic_approx ] ) , we fix @xmath195 and @xmath196 and we analytically calculate @xmath165 and @xmath164 for sf - sf networks with varying degree exponent @xmath197 using the framework developed in appendix [ app : sec : analytical ] . then we generate sf - sf networks and measure @xmath178 as a function of @xmath198 . the approximation predicts that @xmath178 remains constant , in line with our observations ( fig . [ fig : seciiib]c ) . the good performance of eq . ( [ eq : tauiic_approx ] ) is partly due to the role of the ceiling function , as it is insensitive to changes in the numerator that are small compared to @xmath184 . indeed , errors are more pronounced if @xmath199 is close to an integer ( e.g. data point @xmath200 and @xmath201 in fig . [ fig : seciiib]b for er - er ) , or @xmath202 ( e.g. data points @xmath203 in fig . [ fig : seciiib]c ) . what we learn from this approximation is that @xmath178 depends only indirectly on the degree distribution of layer i and layer ii through the control properties of the system without timescale separation @xmath194 and @xmath204 . in sec . [ sec : noseparation ] , we showed that @xmath205 , therefore @xmath178 is expected to be large if layer i is easy to control ( e.g. it is dense and has homogeneous degree distribution ) and layer ii is hard to control ( e.g. it is sparse and has heterogeneous degree distribution ) . in summary , if layer i updates faster , timescale separation enhances controllability up to a critical timescale parameter @xmath178 , above which @xmath206 and is completely determined by layer i. the critical timescale parameter @xmath178 largely depends on the controllability of the system without timescale separation , it is expected to be large if layer i is easy and layer ii is hard to control . finally we investigate the case when layer ii operates faster than layer i , i.e. @xmath33 and @xmath34 ( fig . [ fig:3examples]d ) . measurements show that @xmath163 monotonically increases in function of @xmath27 for both er - er and sf - sf networks , and @xmath163 remains constant if @xmath207 , where @xmath208 is defined for a single network ( fig . [ fig : seciiic]a ) . to understand these results consider the following argument : some nodes of layer ii are internally controlled , meaning that the state of these nodes at @xmath209 is set by the state of nodes within layer ii at @xmath45 connected to them via disjoint control paths ( node @xmath210 in fig . [ fig:3examples]d ) ; while the rest of the nodes of layer ii have to be controlled by nodes of layer i. the maximum number of internally controlled nodes is set by the number of disjoint paths of length @xmath27 . a directed open path traversing @xmath73 links in layer ii yields a path in the dynamic graph of at most length @xmath73 ; therefore if @xmath211 the path can no longer be used for control . for example , in fig . [ fig:3examples]a path @xmath212 consists of a single link ; therefore , we can use it for control if @xmath31 ( fig . [ fig:3examples]b ) and it is no longer useful if @xmath33 ( fig . [ fig:3examples]d ) . however , a cycle can support a path in the dynamic graph of any length , e.g. the self - loop @xmath213 in fig . [ fig:3examples ] . this predicts that @xmath214 where @xmath215 is the maximum fraction of nodes that can be covered with cycles in layer ii . furthermore , it also means that @xmath216 where @xmath217 is the maximum length of a control path that does not involve cycles , a quantity that only depends on the structure of layer ii . we provide the formal definition @xmath217 and algorithms to measure @xmath218 and @xmath217 in appendix [ app : sec : algorithms ] . both @xmath217 and @xmath218 only depend on layer ii , furthermore both strongly depend on whether layer ii contains a strongly connected component ( scc ) or not . uncorrelated random directed networks both er and sf undergo a percolation transition at @xmath219 @xcite . if @xmath220 , the network is composed of small tree components , meaning the @xmath221 and @xmath217 is equal to the diameter @xmath222 of the network . if the system is in the critical point @xmath219 , the size of the largest component @xmath223 diverges as @xmath224 , but the relative size @xmath225 remains zero . the largest component contains a small number of cycles ; therefore @xmath222 is only approximately equal to @xmath217 . if @xmath226 , a unique giant scc emerges which contains cycles ; therefore @xmath227 and @xmath217 is no longer directly connected to the diameter . rigorous mathematical results show that the diameter of the er model scales as @xmath228 for @xmath229 , and @xmath230 for @xmath219 , the latter corresponding to percolation transition point @xcite , suggesting that the critical timescale parameter @xmath231 also depends on @xmath68 . indeed , figure [ fig : seciiic_n ] shows that @xmath231 monotonically increases with @xmath68 for both er - er and sf - sf networks . we now scan possible values of @xmath165 while keeping @xmath164 and @xmath68 fixed , we find that @xmath232 and @xmath233 quickly converges to its respective lower and upper bound provided by eqs . ( [ eq : taui_ni_upperbound ] ) and ( [ eq : tauic_upperbound ] ) ( fig . [ fig : seciiic]b - c ) . varying @xmath164 and keeping @xmath165 fixed shows more intricate behavior : @xmath234 increases , peaks and decreases again ( fig . [ fig : seciiic]d ) . this is explained by changes in the structure of layer ii : for small @xmath164 the network is composed of small components with tree structure , increasing @xmath164 agglomerates these components , thus increasing @xmath217 . for large @xmath164 , a giant scc exists supporting many cycles , as @xmath164 increases more and more nodes can be covered with cycles reducing @xmath217 . at the critical point @xmath235 the giant scc emerges , and the largest component consists of @xmath236 nodes ( @xmath237 ) with only few cycles , providing the peak of @xmath234 . although @xmath235 for both er and sf networks in the @xmath238 limit , finite size effects delay the peak of @xmath231 for sf - sf networks . below the transition point , @xmath231 is smaller for er - er networks than for sf - sf networks with the same average degree . in contrast , above the transition point sf - sf networks have larger @xmath231 . a likely explanation is that the cycle cover of sf networks is smaller than the cycle cover of er networks with the same average degree , thus more nodes can potentially participate in the longest control path that does not involve cycles . the number of inputs above the critical timescale parameter @xmath239 is also affected by the cycle cover of layer ii ( fig . [ fig : seciiic]e ) : for @xmath240 , layer ii does not contain cycles yielding @xmath241 ; for large @xmath164 , layer ii can be completely covered with cycles , and @xmath239 is determined by @xmath242 , the number of inputs needed to control layer i in isolation . in summary , if layer ii updates faster , timescale separation reduces controllability up to a critical timescale parameter @xmath208 . for the model networks , the value of @xmath208 depends on whether layer ii has a giant scc ; @xmath208 has the highest value at the percolation threshold of layer ii . if layer ii does not contain a giant scc , degree heterogeneity decreases @xmath208 ; above the percolation threshold homogeneous networks have lower @xmath208 . for all timescale parameters , it remains true that er - er networks require less independent inputs than sf - sf networks with the same average degree . here we explored controllability of interconnected complex systems with a model that incorporates common properties of these systems : ( i ) it consists of two layers each described by a complex network ; ( ii ) the operation of each layer is characterized by a different , but often comparable timescale and ( iii ) the external controller only interacts with one layer directly . we focused on two - layer multiplex networks , meaning that we assume one - to - one coupling between the nodes of the two layers . our motivation for this choice was to ensure analytical tractability and to isolate the specific role of timescales from the effect of more complex multilayer network structure . results obtained for more general multilayer networks will ultimately be shaped by a variety of features such as complex interconnectivity structure , correlations in network structure and details of dynamics . however , even by studying multiplex networks , we uncovered nontrivial phenomena , attesting that without understanding each individual effect , it is impossible to fully understand a system as a whole . using structural controllability we were able to solve the model , thereby directly linking controllability to a graph combinatorial problem . we investigated the effect of network structure and timescales by measuring the minimum number of independent inputs needed for control , @xmath0 . overall we found that dense networks with homogeneous degree distribution require less inputs than sparse heterogeneous networks , in line with previous results for single - layer networks @xcite . we showed that if we control the faster layer directly , @xmath0 decreases with increasing timescale difference , but only up to a critical value . above the critical timescale difference , @xmath0 is completely determined by the faster layer and we do not have to take into account the multiplex structure of the system . this critical timescale separation is expected to be large if the faster layer would be easy to control and the slower layer would be hard to control in isolation . if we interact with the slower layer , control is increasingly difficult for increasing timescale difference , again up to a critical value , above which @xmath0 still depends on the structure of both layers . in this case the critical timescale difference largely depends on the longest control path that does not involve cycles in the faster layer . although our model offers only a stylized description of real systems , it is a tractable first step towards understanding the role of timescales in control of interconnected networks . by identifying the network characteristics that affect important measures of controllability , such as minimum number of inputs needed for control and critical timescale difference , our results serve as a starting point for future work that aims to relax some of the model s assumptions . some of these extensions are relatively straightforward using the tool set developed here , for example , the effect of higher order network structures can be studied by adding correlations to the underlying networks . other extensions are more challenging , e.g. if the interconnection between the layers is incomplete or the layers contain different number of nodes , the minimum input problem is computationally more difficult ; therefore investigating such systems would require development of efficient approximation schemes . structural control theory does not take the link weights into account ; therefore answering questions that depend on the specific strength of the connections require the development of different tools . for example , for continuous - time systems the timescales are encoded in the strength of the interactions ; or the minimum control energy also depends on value of the link weights . we thank yang - yu liu , philipp hvel and zsfia pnzvlt for useful discussions . we gratefully acknowledge support from the us army research office cooperative agreement no . w911nf-09 - 2 - 0053 and muri award no . w911nf-13 - 1 - 0340 , and the defense threat reduction agency basic research awards hdtra1 - 10 - 1 - 0088 and hdtra1 - 10 - 1 - 00100 . in this section we derive an analytical solution of @xmath243 in case of @xmath30 for two - layer random networks with predefined degree distribution as defined in sec . [ sec : results ] . this network model is treelike in the @xmath238 limit ; therefore it lends itself to the generating function formalism . the approach described here is based on calculating the fraction of nodes that are matched in all possible maximum matchings @xcite . this solution is substantially simpler than the one described in ref . @xcite ; however , it only applies to bipartite networks ( or to bipartite representations of directed networks ) , and can not be generalized to unipartite networks . we aim to calculate the expected size of the maximum matching of the following undirected bipartite network @xmath244 . layer i @xmath245 and layer ii @xmath246 are generated independently either using the er or the sf model ; @xmath247 and @xmath248 are the node and link sets of @xmath245 and @xmath249 and @xmath250 are the node and link sets of @xmath246 . each node in @xmath251 is split into two copies @xmath252 and @xmath253 , we draw a link @xmath254 if there exists a link @xmath114 in @xmath245 . we treat @xmath246 similarly . we then add links @xmath255 for all @xmath60 . that is all links in @xmath244 connect exactly one node in @xmath256 to one node in @xmath257 . nodes in @xmath258 belong to layer i , and nodes in @xmath259 belong to layer ii . the network @xmath244 is the undirected version of the dynamical graph @xmath87 without control signals . in general , multiple possible maximum matchings may exist in a network . we first calculate the fraction of nodes that are matched in all possible maximum matchings . it was shown in ref . @xcite that in any network @xmath260 a node @xmath261 is always matched if and only if at least one of its neighbors is not always matched in @xmath262 , where @xmath262 is the network obtained by removing node @xmath261 from @xmath260 . we translate this rule to a set of self - consistent equations to calculate the expected fraction of always matched nodes in our random network model in the @xmath238 limit . we provide comments on the issues of applying the rule proven for finite networks to infinite ones at the end of this section . to proceed we define a few probabilities . we randomly select a link @xmath263 connecting two nodes @xmath252 and @xmath264 . let @xmath265 be the probability that @xmath117 is always matched in @xmath266 , and @xmath267 be the probability that @xmath268 is always matched in @xmath266 . similarly we randomly select a link @xmath263 connecting a node @xmath252 with a node @xmath269 . let @xmath270 be the probability that node @xmath117 is always matched in @xmath266 , and @xmath271 be the probability that node @xmath272 is always matched in @xmath266 . the probabilities @xmath273 and @xmath274 are defined similarly . according to the rule described above these quantities can be determined by the following set of equations : @xmath275 where @xmath276 are the generating functions of the degree distributions and @xmath277 are the generating functions of the excess degree distributions . if we remove a node @xmath261 which is not always matched , the size of the maximum matching does not decrease . however , if @xmath261 is matched in all maximum matchings , the number of matched nodes will decrease by two . therefore to count the size of the maximum matching , we first count the number of nodes that are always matched . by doing so , we have double counted the case when an always matched node is matched by another always matched one . this case occurs for each link @xmath263 that connects two nodes that are not always matched in @xmath278 . combining these two contributions , the expected number of links in the matching is @xmath279 + n[1 - g^{\textrm{i}}(\theta^{\textrm{ii}}_{0 } ) ] + n[1 - g^{\textrm{ii}}(\theta^{\textrm{i}}_{1 } ) ] + n[1 - g^{\textrm{ii}}(\theta^{\textrm{ii}}_{0})\theta^{\textrm{i , ii}}_{0 } ] -\\ - & c_{\textrm{i}}n(1-\theta^{\textrm{i}}_{0})(1-\theta^{\textrm{i}}_{1 } ) - n(1-\theta^{\textrm{i , ii}}_{0})(1-\theta^{\textrm{i , ii}}_{1 } ) - c_{\textrm{ii}}n(1-\theta^{\textrm{ii}}_{0})(1-\theta^{\textrm{ii}}_{1 } ) , \end{split}\ ] ] where the first four terms count the number of nodes that are always matched in @xmath280 , @xmath281,@xmath282 and @xmath283 , respectively ; and the last three terms correct the double counting . the expected number of independent inputs needed is determined by the number of unmatched nodes in @xmath281 and @xmath283 : @xmath284 due to the links between layer i and layer ii , the size of the maximum matching is at least @xmath68 , meaning that @xmath285 . therefore we normalize @xmath0 by @xmath68 , yielding @xmath286 the method we described to calculate the expected size of the maximum matching does not work for unipartite er or sf networks generally . the reason for this is that above a critical average degree @xmath287 a densely connected subgraph forms , which is referred to as the core of the network ( sometimes leaf removal core or computational core ) @xcite . to derive eq . ( [ eq : app : theta ] ) , we assume that the neighbors of a randomly selected node @xmath261 are independent of each other in @xmath288 and removing a single node does not influence macroscopic properties , e.g. @xmath289 . the effect of the core is that these assumptions no longer hold and removing just a few nodes may drastically change the number of always matched nodes . possible way of circumventing this problem is to introduce a new category of nodes : in addition to keeping track of nodes that are sometimes matched and always matched , we separately account for nodes that are almost always matched @xcite . the reason why the calculation works for bipartite networks is that a core in the bipartite network will have two sides : all nodes on one side will be always matched and all nodes on other will be some times matched @xcite . if the expected size of the core on the two sides is different , finite removal of nodes will not change macroscopic properties . if the expected size of the two sides of the core is the same , removal of finite nodes may change which side is always matched and which side is sometimes matched @xcite . however , this does not change expected fraction of matched nodes ; therefore does not interfere with the calculations . to find the maximum cycle cover of a directed network @xmath41 , we assign weight @xmath139 to each link in @xmath41 ; and we add a self - loop with weight @xmath140 to each node that does not already have a self - loop . then we find the minimum weight maximum directed matching in @xmath41 augmented with self - loops by converting the problem to a minimum cost maximum flow problem . the maximum matching is guaranteed to be perfect , because each node has a self - loop . the minimum weight perfect matching in the directed network corresponds to a perfect cycle cover where the number of self - loops with weight @xmath140 is minimized . therefore the maximum cycle cover in @xmath41 without extra self - loops is @xmath291 where @xmath292 is the sum of the weights of the links in the minimum weight perfect matching . in this section we provide the algorithm to measure the longest control path not involving cycles @xmath217 of layer ii of a two - layer network for the case @xmath293 and @xmath34 . the algorithm itself serves as the precise definition of @xmath217 . given a two - layer directed network @xmath1 , let @xmath290 be the maximum number of nodes that can be covered by node disjoint cycles in layer ii . to measure @xmath217 , first we construct the dynamical graph @xmath294 representing the time evolution of the layer ii between time @xmath45 and @xmath295 as if it would be isolated as defined in sec . [ sec : single - layer_net ] . we search for disjoint control paths connecting nodes at time step @xmath45 with nodes at time step @xmath295 , e.g. each control path connects a node @xmath115 with @xmath296 . the maximum number of such paths @xmath297 provides the maximum number of internally controlled nodes if @xmath298 . to determine @xmath297 we convert the problem to a maximum flow problem : we set the capacity of each link and each node in @xmath294 to 1 . we then find the maximum flow connecting source node set @xmath299 to target node set @xmath300 using a maximum flow algorithm of choice . the maximum flow provides @xmath297 . and @xmath217 is defined as one less than the smallest value of @xmath73 such that @xmath301 47ifxundefined [ 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 * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * ( ) @noop * * ( ) @noop * * ( ) @noop * * ( ) @noop * * ( ) @noop * * , ( ) @noop * * ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ , ed . ( , ) @noop * * , ( ) in @noop _ _ ( , ) pp . in @noop _ _ ( , ) pp . @noop `` '' ( ) , @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) in @noop _ _ , vol . ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) http://igraph.org [ ( ) ] @noop ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * ( ) @noop * * ( ) _ _ , @noop ph.d . thesis , ( ) @noop * * , ( ) @noop * * , ( ) @noop ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) , we construct the dynamic graph representing the time evolution of the system from @xmath302 to @xmath303 . the system is controllable only if all nodes at @xmath85 ( blue ) are connected to nodes at @xmath304 or nodes representing control signals ( green ) via disjoint paths ( red ) . * ( b ) * in case of no timescale separation ( @xmath30 ) , each disjoint control path consists of a single link , yielding @xmath305 . * ( c ) * if layer i updates twice as frequently as layer ii ( @xmath31 , @xmath306 ) , we are allowed to inject control signals at time steps @xmath45 and @xmath140 , reducing the number of inputs to @xmath99 . * ( d ) * on the other hand , if layer ii is faster ( @xmath307 , @xmath34 ) , layer ii needs to support longer control paths , yielding @xmath308 . ] and @xmath309 . * ( b ) * the dynamic graph @xmath65 representing the time evolution of the dynamics from @xmath45 to @xmath175 . the system is controllable , because we can connect the set of nodes representing control signals ( green ) to the set of nodes at @xmath175 ( blue ) via disjoint paths ( red ) . * ( c ) * the dynamic graph @xmath87 representing the time evolution of the dynamics from @xmath45 to @xmath90 . the system is controllable , because we can connect the control signals and nodes at @xmath45 ( green ) to the set of nodes at @xmath90 ( blue ) via disjoint paths ( red ) , and all nodes are accessible from control signals . ] in function of @xmath165 for er - er and sf - sf ( @xmath310 ) networks . the circles represent simulations , the continuous line is the analytical solution , and the dashed line is the analytical solution of @xmath311 , the number of independent inputs necessary to control layer ii in isolation @xcite . * ( b ) * @xmath163 for er - er networks with varying average degrees @xmath165 and @xmath164 . in both layers @xmath312 , therefore the heatmap is symmetric with respect to the diagonal . increasing @xmath313 in either layer enhances controllability . * ( c ) * @xmath163 for sf - sf networks with @xmath314 and varying degree exponents @xmath168 and @xmath169 . increasing degree heterogeneity in either layer increases @xmath163 . each data point is the average over @xmath315 randomly generated networks with @xmath316 . the standard deviation of the measurements remains below 0.01 . ] for single er - er and sf - sf ( @xmath310 ) networks with @xmath316 and varying timescale parameter @xmath28 . the number of inputs @xmath163 monotonically decreases with increasing @xmath28 , and for @xmath317 , @xmath318 . * ( b ) * the critical timescale parameter @xmath178 for er - er and sf - sf ( @xmath310 ) networks with varying average degree @xmath165 and @xmath164 . the crosses represent direct measurements of @xmath178 ; the squares represent the approximation obtained by applying eq . ( [ eq : tauiic_approx ] ) to measurements of @xmath195 and @xmath196 ; and the dashed line is an approximation obtained using analytically calculated expectation values of @xmath195 and @xmath196 . * ( c ) * we measure @xmath178 for sf - sf networks with the same @xmath195 and @xmath196 as a function of @xmath197 . equation ( [ eq : tauiic_approx ] ) predicts that @xmath178 remains constant ( dashed line ) , in line with our observations . for ( b - c ) , each data point is the average over @xmath315 randomly generated networks with @xmath316 and error bars represent the standard deviation . ] for single er - er and sf - sf ( @xmath310 ) networks with @xmath316 and varying timescale parameter @xmath27 . the number of inputs @xmath163 monotonically increases with increasing @xmath27 ; for @xmath319 , @xmath320 . * ( b ) * @xmath231 as a function of @xmath165 . for @xmath321 , @xmath231 quickly reaches its upper bound ; for @xmath322 , the convergence is somewhat delayed . * ( c ) * @xmath239 as a function of @xmath165 . increasing @xmath165 facilitates control , until @xmath163 reaches its lower bound . * ( d ) * @xmath231 as a function of @xmath164 with fix @xmath323 . the peak of @xmath231 corresponds to the critical point where the giant strongly connected component in layer ii emerges . * ( e ) * @xmath239 as a function of @xmath164 with fix @xmath323 . for @xmath240 , layer ii does not contain cycles , therefore @xmath241 ; for large @xmath164 , layer ii can be completely covered with cycles , and @xmath239 is determined by @xmath167 . for ( b - e ) , each data point is the average over @xmath315 randomly generated networks with @xmath316 and error bars represent the standard deviation . ( a ) * layer ii has no giant strongly connected component ( @xmath324 ) , @xmath217 equals the diameter @xmath222 of layer ii which scales as @xmath325 for er networks , and the diameter of sf networks is smaller than the diameter of er networks with the same average degree . the fact that @xmath326 suggest that @xmath327 . * ( b ) * at the critical point @xmath328 the diameter of er networks scales as @xmath230 , suggesting that @xmath231 scales as a powerlaw of @xmath68 . * ( c ) * above the critical point ( @xmath329 ) there is no direct connection between @xmath222 and @xmath231 , nonetheless observations suggest @xmath330 . in contrast with the @xmath331 case , @xmath231 increases more rapidly for sf - sf networks than for er - er networks . each data point is the average over @xmath332 randomly generated networks with @xmath323 and error bars represent the standard deviation . ] * example 1 . * * ( a ) * a directed network with tree structure ; therefore not containing cycles . the diameter @xmath333 is the length of the longest path . * ( b ) * we count the maximum number of disjoint control paths @xmath297 which connect nodes at time step @xmath139 with nodes at time step @xmath73 . we find that @xmath334 is the smallest value of @xmath73 such that @xmath335 ; therefore @xmath336 . there are no cycles ; therefore @xmath337 . ] * example 2 . * * ( a ) * a directed network containing a cycle . the size of the maximum cycle cover is @xmath338 . * ( b ) * we count the maximum number of disjoint control paths @xmath297 which connect nodes at time step @xmath139 with nodes at time step @xmath73 . we find that @xmath339 is the smallest value of @xmath73 such that @xmath340 ; therefore @xmath341 . @xmath297 remains non - zero for @xmath342 , showing that cycles can support control paths of any length . ]

we prove a bosonic formula for the generating function of level - restricted paths for the infinite families of affine kac - moody algebras . in affine type a this yields an expression for the level - restricted generalized kostka polynomials . supported by the `` stichting fundamenteel onderzoek der materie '' . ] partially supported by nsf grant dms-9800941 . ]

math9812106

let @xmath0 be an affine kac - moody algebra , @xmath1 a @xmath2-submodule of a finite direct sum @xmath3 of irreducible integrable highest weight @xmath4-modules , and @xmath5 the limit of the demazure operator for an element @xmath6 of the weyl group as @xmath7 . the main theorem of this paper gives sufficient conditions on @xmath1 so that the formula @xmath8 holds , where @xmath9 is the character of @xmath1 . when @xmath1 is the one - dimensional @xmath2-module generated by the dominant integral weight @xmath10 then is the weyl - kac character formula . the above result is well - known when @xmath1 is a union of demazure modules for any kac - moody algebra @xmath0 . let @xmath11 be the derived subalgebra of @xmath0 . consider the @xmath11-module @xmath1 given by a tensor product of finite - dimensional @xmath12-modules that admit a crystal of level at most @xmath13 , with the one - dimensional subspace generated by a highest weight vector of an irreducible integrable highest weight @xmath12-module of level @xmath13 . such modules @xmath1 can be given the structure of a @xmath2-module and as such , satisfy the above conditions . then a special case of is a bosonic formula for the @xmath14-enumeration of level - restricted inhomogeneous paths by the energy function . in type @xmath15 this formula was conjectured in @xcite , stated there as a @xmath14-analogue of the goodman - wenzl straightening algorithm for outer tensor products of irreducible modules over the type @xmath16 hecke algebra at a root of unity @xcite . in the isotypic component of the vacuum , the bosonic formula coincides with half of the bose - fermi conjecture in ( * ? ? ? * ( 9.2 ) ) . the authors would like to thank nantel bergeron , omar foda , masaki kashiwara , atsuo kuniba , masato okado , jean - yves thibon , and ole warnaar for helpful discussions . most of the following notation is taken from ref . let @xmath17 be a dynkin diagram of affine type with vertices indexed by the set @xmath18 as in @xcite , cartan matrix @xmath19 , @xmath20 the affine kac - moody algebra , and @xmath21 the cartan subalgebra . let @xmath22 and @xmath23 be the simple coroots and roots , which are linearly independent subsets that satisfy @xmath24 for @xmath25 where @xmath26 is the natural pairing . let @xmath27 be the root lattice . let the null root @xmath28 be the unique element of the positive cone @xmath29 in @xmath30 , that generates the one - dimensional lattice @xmath31 . let @xmath32 be the canonical central element , where the integers @xmath33 are the analogues of the integers @xmath34 for the dual algebra @xmath35 defined by the transpose @xmath36 of the cartan matrix @xmath16 . let @xmath37 ( the degree derivation ) be defined by the conditions @xmath38 where @xmath39 is the kronecker delta ; @xmath40 is well - defined up to a summand proportional to @xmath41 . then @xmath42 is a basis of @xmath21 . let @xmath43 be the dual basis of @xmath44 ; the elements @xmath45 are called the fundamental weights . the weight lattice is defined by @xmath46 ; in the usual definition the scalar @xmath47 is absent . the weight lattice contains the root lattice since @xmath48 for @xmath49 . define @xmath50 . say that a weight @xmath51 has level @xmath13 if @xmath52 . consider the standard symmetric bilinear form @xmath53 on @xmath44 . since @xmath54 is a basis of @xmath44 , this form is uniquely defined by @xmath55 for @xmath25 , @xmath56 for @xmath57 and @xmath58 . this form induces an isomorphism @xmath59 defined by @xmath60 for @xmath57 and @xmath61 . also @xmath62 . the weyl group @xmath63 is the subgroup of @xmath64 generated by the simple reflections @xmath65 ( @xmath57 ) defined by @xmath66 the form @xmath53 is @xmath63-invariant . suppose @xmath67 is a real root , that is , the @xmath68-weight space of @xmath0 is nonzero and there is a simple root @xmath69 and a weyl group element @xmath70 such that @xmath71 . define @xmath72 by @xmath73 . this is independent of the expression @xmath74 . define @xmath75 by @xmath76 let @xmath11 be the derived algebra of @xmath0 , obtained by omitting " the degree derivation @xmath40 . its weight lattice is @xmath77 . denote the canonical projection @xmath78 by @xmath79 . write @xmath80 and @xmath81 for @xmath57 . the elements @xmath82 are linearly dependent . write @xmath83 for the section of @xmath79 given by @xmath84 for all @xmath57 . write @xmath85 . define the level of @xmath86 to be @xmath87 . consider the dynkin diagram @xmath88 obtained by removing the vertex @xmath89 from the diagram @xmath17 , with corresponding cartan matrix @xmath90 indexed by the set @xmath91 , and let @xmath92 be the simple lie algebra . one has the inclusions @xmath93 . let @xmath94 be the simple roots , @xmath95 the fundamental weights , and @xmath96 the root lattice for @xmath97 . the weight lattice of @xmath97 is @xmath98 and @xmath99 . the image of @xmath100 into @xmath101 is denoted by @xmath102 . we shall use the section of the natural projection @xmath103 given by the map @xmath104 that sends @xmath105 for @xmath106 . by abuse of notation , for @xmath100 , @xmath102 shall also denote the image of the element @xmath102 under the lifting map @xmath107 specified above . . for @xmath109 , denote by @xmath110 the irreducible integrable highest weight @xmath111-module of highest weight @xmath112 . let @xmath113 . one has the formulas @xmath114 , @xmath115 , and @xmath116 . observe that @xmath117 for @xmath51 let @xmath118 be the irreducible integral highest weight module of highest weight @xmath10 over the quantized universal enveloping algebra @xmath4 , @xmath119 the crystal base of @xmath118 , and @xmath120 the highest weight vector . by restriction from @xmath4 to @xmath12 , the module @xmath118 is an irreducible integral highest weight module for @xmath12 of highest weight @xmath121 , with crystal @xmath119 that is @xmath122-weighted by composing the weight function @xmath123 with the projection @xmath79 . conversely , any integrable irreducible highest weight @xmath12-module can be obtained this way . a @xmath124-weighted @xmath125-crystal @xmath126 is a colored graph with vertices indexed by @xmath127 , directed edges colored by @xmath57 , and a weight function @xmath128 , satisfying the axioms below . first some notation is required . denote an edge from @xmath129 to @xmath130 colored @xmath131 , by @xmath132 or equivalently @xmath133 . write @xmath134 ( resp . @xmath135 ) for the maximum index @xmath136 such that @xmath137 ( resp . @xmath138 ) is defined . 1 . if @xmath139 then @xmath140 . 2 . @xmath141 . an element @xmath142 is a highest weight vector if @xmath143 is undefined for all @xmath57 . the @xmath131-string of @xmath127 consists of all elements @xmath138 ( @xmath144 ) and @xmath137 ( @xmath145 ) . the nondominant part of the @xmath131-string is comprised of all elements which admit @xmath146 . we also define the crystal reflection operator @xmath147 by @xmath148 it is obvious that @xmath149 is an involution . observe that @xmath150 define the notation @xmath151 and @xmath152 . if a @xmath4-module ( resp . @xmath12-module , resp . @xmath111-module ) has a crystal base then the latter is naturally a @xmath124-weighted ( resp . @xmath122-weighted , resp . @xmath101-weighted ) @xmath125-crystal ( resp . @xmath125-crystal , resp . @xmath153-crystal ) . given crystals @xmath154 and @xmath155 , contrary to the literature ( but consistent with the robinson - schensted - knuth correspondence in type a ) , define the following crystal structure on the tensor product @xmath156 . the elements are denoted @xmath157 for @xmath158 ( @xmath159 ) and one defines @xmath160 when @xmath161 ( resp . @xmath162 ) one defines @xmath163 an element of a tensor product of crystals is called a path . the definitions here follow @xcite . suppose that @xmath154 and @xmath155 are crystals of finite - dimensional @xmath12-modules such that @xmath156 is connected . then there is an isomorphism of @xmath122-weighted @xmath125-crystals @xmath164 . this is called the local isomorphism . let the image of @xmath165 under this isomorphism be denoted @xmath166 . then there is a unique ( up to a global additive constant ) map @xmath167 such that @xmath168 this map is called the local energy function . consider @xmath169 with @xmath170 the crystal of a finite - dimensional @xmath12-module for @xmath171 . assume that for all @xmath172 , @xmath173 is a connected @xmath122-weighted @xmath125-crystal . given @xmath174 , denote by @xmath175 the @xmath176-th tensor factor in the image of @xmath129 under the composition of local isomorphisms that switch @xmath170 with @xmath177 as @xmath178 goes from @xmath179 down to @xmath180 . then define the energy function @xmath181 where @xmath182 is the local energy function . it satisfies the following property . [ energy e ] ( * ? ? ? 1.1 ) suppose @xmath57 , @xmath127 and @xmath183 is defined . if @xmath184 then @xmath185 . if @xmath186 and @xmath129 has the property that for any of its images @xmath187 under a composition of local isomorphisms , @xmath188 with @xmath189 , then @xmath190 . say that @xmath191 is classically restricted if @xmath129 is a @xmath97-highest weight vector , that is , @xmath183 is undefined for all @xmath106 . for @xmath192 denote by @xmath193 the set of classically restricted @xmath127 of weight @xmath112 . define the polynomial @xmath194 where @xmath195 is the energy function on @xmath126 . for @xmath0 of type @xmath15 @xmath196 is the generalized kostka polynomial @xcite . let @xmath126 be the crystal of a finite - dimensional @xmath12-module . say that @xmath126 is almost perfect of level @xmath13 @xcite if it satisfies the following weakening of the definition of a perfect crystal ( * ? ? ? 4.6.1 ) : 1 . @xmath197 is connected . 2 . there is a @xmath198 such that there is a unique @xmath199 such that @xmath200 and for every @xmath127 , @xmath201 . 3 . for every @xmath127 , @xmath202 . 4 . for every @xmath203 of level @xmath13 , there is a @xmath204 such that @xmath205 . @xmath126 is said to be perfect if the elements @xmath129 and @xmath130 in item 4 are unique . from now on , fix a positive integer @xmath13 ( the level ) . for @xmath171 let @xmath170 be the crystal of a finite - dimensional @xmath12-module , that is almost perfect of level at most @xmath13 . let @xmath206 , @xmath207 weights of level @xmath13 , and @xmath208 the set of paths @xmath209 such that @xmath210 is a highest weight vector of weight @xmath211 . in the special case that @xmath212 , the elements of @xmath208 are called the level-@xmath13 restricted paths of weight @xmath211 . [ level tensor ] @xcite ( * ? ? ? * appendix a ) . let @xmath0 be an affine kac - moody algebra in one of the infinite families . let @xmath126 be the tensor product of crystals of finite - dimensional @xmath12-modules , that are almost perfect of level at most @xmath13 , and @xmath203 a weight of level @xmath13 . then there is an isomorphism of @xmath122-weighted @xmath125-crystals @xmath213 where @xmath211 is of level @xmath13 . this isomorphism of @xmath122-weighted crystals can be lifted to one of @xmath124-weighted crystals by specifying an integer multiple of @xmath214 for each highest weight vector in @xmath215 . however for our purposes this should be done in a way that extends the definition of the energy function for @xmath126 . to this end , choose a perfect crystal @xmath216 of level @xmath13 , and assume that for all @xmath217 , @xmath218 is connected . let @xmath219 be the unique element such that @xmath220 . define the energy function @xmath221 by @xmath222 where @xmath223 is the energy function defined in . for @xmath224 , define an affine weight function @xmath225 . this defines the @xmath124-weight of every highest weight vector in @xmath215 and hence a @xmath124-weight function for all of @xmath215 . then one has the following @xmath124-weighted analogue of : @xmath226 where @xmath211 is of level @xmath13 . this decomposition can be described by the polynomial @xmath227 our goal is to prove a formula for the polynomial @xmath228 . let @xmath153 be the antisymmetrizer @xmath229 write @xmath230 where @xmath231 is the set of positive roots of @xmath0 and @xmath232 is the dimension of the @xmath68-weight space in @xmath0 . let @xmath233 be the unique weight defined by @xmath234 for all @xmath57 and @xmath235 . it satisfies @xmath236 where @xmath237 is the dual coxeter number . define the operator @xmath238 where @xmath239 makes sense by expanding the reciprocals of the factors of @xmath240 in geometric series . the computation is defined in a suitable completion of @xmath241 $ ] . one has @xmath242 for all @xmath51 , which is the weyl - kac character formula ( * ? ? ? * theorem 10.4 ) . [ gen boson ] let @xmath0 be an affine kac - moody algebra in one of the infinite families , @xmath243 the crystal of a finite direct sum of irreducible integrable highest weight @xmath4-modules and @xmath244 a subset such that : 1 . @xmath126 is closed under @xmath146 for all @xmath57 . @xmath243 is generated by @xmath126 . 3 . for all @xmath127 and @xmath57 , if @xmath245 then the @xmath131-string of @xmath129 in @xmath243 is contained in @xmath126 . then @xmath246 without loss of generality it may be assumed that @xmath247 for some @xmath51 . multiplying both sides of by @xmath248 , one obtains @xmath249 observe that both sides are @xmath63-alternating . the @xmath63-alternants have a basis given by @xmath250 where @xmath51 . taking the coefficient of @xmath251 on both sides , @xmath252 where @xmath253 is the set of pairs @xmath254 such that @xmath255 observe that if @xmath256 is such that @xmath129 is a highest weight vector , then @xmath257 and @xmath258 , for both of the regular dominant weights @xmath259 and @xmath260 are in the same @xmath63-orbit and hence must be equal . conditions 1 and 2 ensure that @xmath261 . let @xmath262 . it is enough to show that there is an involution @xmath263 with no fixed points , such that if @xmath264 then @xmath6 and @xmath265 have opposite sign . in this case @xmath266 is said to be sign - reversing . let @xmath267 be the set of pairs @xmath268 such that @xmath245 . define the map @xmath269 by @xmath270 . note that @xmath271 by condition 3 . the condition for @xmath272 is @xmath273 since @xmath274 , @xmath275 , so that @xmath276 . this shows that @xmath277 is well - defined . it follows directly from the definitions that @xmath277 is a sign - reversing involution . since @xmath278 it suffices to define a global involutive choice of the canceling root direction for each pair @xmath279 , that is , a function @xmath280 such that if @xmath281 then 1 . 2 . @xmath283 . let @xmath284 be an expression of @xmath10 as a sum of fundamental weights . by ( * ? * lemma 8.3.1 ) , @xmath119 is isomorphic to the full subcrystal of @xmath285 generated by @xmath286 . given @xmath279 , let @xmath287 be the image of @xmath129 in the above tensor product of crystals of modules of fundamental highest weight . let @xmath288 be minimal such that @xmath289 is not a highest weight vector . then @xmath290 is a highest weight vector , say of weight @xmath211 . let @xmath291 be a perfect crystal of the same level as @xmath292 . given any @xmath293 , the theory of perfect crystals ( * ? ? ? * section 4.5 ) gives an isomorphism of @xmath124-weighted crystals @xmath294 where @xmath295 is determined by @xmath296 and @xmath297 and @xmath298 is @xmath124-weighted using the energy function . let @xmath299 have image @xmath300 where @xmath301 . assume that @xmath297 is large enough so that @xmath302 . if one takes the image of @xmath303 in such a tensor product for @xmath304 the tensor factors @xmath305 through @xmath306 do not change . let @xmath178 be minimal such that @xmath307 is not a highest weight vector . observe that @xmath178 is independent of @xmath297 as long as @xmath297 is big enough . then @xmath308 is a highest weight vector , say of weight @xmath309 . so @xmath310 is such that @xmath311 for some @xmath57 ; let @xmath312 be the set of such @xmath57 . fix an @xmath313 . consider the same constructions for @xmath314 . let @xmath315 be the image of @xmath130 in the above tensor product of irreducible crystals of fundamental highest weights . then @xmath316 and @xmath317 is not a highest weight vector ; in particular it admits @xmath146 . take l large enough so that the image of @xmath318 in @xmath319 also has the form @xmath320 . then @xmath321 and @xmath322 admits @xmath146 . the level of the fundamental weight @xmath323 is @xmath33 . for the affine algebras @xmath324 and @xmath325 , @xmath326 for all @xmath57 . for all others @xmath327 . the theorem now follows from lemma [ lev one ] below , applied with the dominant integral weight @xmath309 . [ lev one ] for the affine kac - moody algebra @xmath0 in one of the infinite families , there exist perfect crystals @xmath126 of level one and two ( the latter case only for @xmath328 ) having the following property . let @xmath10 be a dominant integral weight of positive level , @xmath329 the set of elements @xmath330 such that @xmath331 is not a highest weight vector . then there is a map @xmath332 such that if @xmath333 then 1 . @xmath334 . 2 . for any @xmath335 , writing @xmath336 , one has @xmath337 and @xmath338 . for the involutive property 2 , it is sufficient that @xmath339 is constant on the nondominant part of every string . hence one only needs to consider @xmath340 perfect crystals of level one for @xmath324 ( @xmath341 ) , @xmath342 ( @xmath343 ) , @xmath344 ( @xmath345 ) , @xmath346 ( @xmath341 ) , @xmath347 ( @xmath343 ) and @xmath348 ( @xmath349 ) are listed in table [ table crystals ] ( see ( * ? ? ? * section 6 ) ) . note that there are no elements satisfying . this guarantees the existence of the map @xmath339 with the desired properties . .[table crystals]level one perfect crystals [ cols="^,<",options="header " , ] there are the following strings of length greater than one @xmath350{\footnotesize$\mbox{k}$ } \text(6,0)[]{\footnotesize$\mbox{k}$ } \longarrow(30,0)(60,0 ) \bboxc(96.5,0)(39,13 ) \line(90,-6.5)(90,6.5 ) \text(83,0)[]{\footnotesize$\mbox{k}$ } \text(103,0)[]{\footnotesize$\mbox{k+1}$ } \longarrow(120,0)(150,0 ) \bboxc(180,0)(52,13 ) \line(180,-6.5)(180,6.5 ) \text(167,0)[]{\footnotesize$\mbox{k+1}$ } \text(193,0)[]{\footnotesize$\mbox{k+1}$ } \ptext(45,2)(0)[b]{k } \ptext(135,2)(0)[b]{k } \end{picture } } } & & \text{for $ 1\leq k < n$ } \notag \\ & \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(240,22)(-30,-10 ) \bboxc(6.5,0)(39,13 ) \line(0,-6.5)(0,6.5 ) \text(-7,0)[]{\footnotesize$\mbox{k}$ } \text(13,0)[]{\footnotesize$\overline{\mbox{k+1}}$ } \longarrow(30,0)(60,0 ) \bboxc(90,0)(26,13 ) \line(90,-6.5)(90,6.5 ) \text(83,0)[]{\footnotesize$\mbox{k}$ } \text(96,0)[]{\footnotesize$\overline{\mbox{k}}$ } \longarrow(120,0)(150,0 ) \bboxc(173.5,0)(39,13 ) \line(180,-6.5)(180,6.5 ) \text(167,0)[]{\footnotesize$\mbox{k+1}$ } \text(186,0)[]{\footnotesize$\overline{\mbox{k}}$ } \ptext(45,2)(0)[b]{k } \ptext(135,2)(0)[b]{k } \end{picture } } } & & \text{for $ 1\leq k < n$ } \label{string gen}\\ & \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(240,22)(-30,-10 ) \bboxc(0,0)(52,13 ) \line(0,-6.5)(0,6.5 ) \text(-13,0)[]{\footnotesize$\overline{\mbox{k+1}}$ } \text(13,0)[]{\footnotesize$\overline{\mbox{k+1}}$ } \longarrow(30,0)(60,0 ) \bboxc(83.5,0)(39,13 ) \line(90,-6.5)(90,6.5 ) \text(77,0)[]{\footnotesize$\overline{\mbox{k+1}}$ } \text(96,0)[]{\footnotesize$\overline{\mbox{k}}$ } \longarrow(120,0)(150,0 ) \bboxc(180,0)(26,13 ) \line(180,-6.5)(180,6.5 ) \text(173,0)[]{\footnotesize$\overline{\mbox{k}}$ } \text(186,0)[]{\footnotesize$\overline{\mbox{k}}$ } \ptext(45,2)(0)[b]{k } \ptext(135,2)(0)[b]{k } \end{picture } } } & & \text{for $ 1\leq k < n$ } \notag \\[2 mm ] & \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(240,22)(-30,-10 ) \bboxc(0,0)(26,13 ) \line(0,-6.5)(0,6.5 ) \text(-7,0)[]{\footnotesize$\mbox{n}$ } \text(6,0)[]{\footnotesize$\mbox{n}$ } \longarrow(30,0)(60,0 ) \bboxc(90,0)(26,13 ) \line(90,-6.5)(90,6.5 ) \text(83,0)[]{\footnotesize$\mbox{n}$ } \text(96,0)[]{\footnotesize$\overline{\mbox{n}}$ } \longarrow(120,0)(150,0 ) \bboxc(180,0)(26,13 ) \line(180,-6.5)(180,6.5 ) \text(173,0)[]{\footnotesize$\overline{\mbox{n}}$ } \text(186,0)[]{\footnotesize$\overline{\mbox{n}}$ } \ptext(45,2)(0)[b]{n } \ptext(135,2)(0)[b]{n } \end{picture } } } & & \label{string n}\\[2 mm ] & \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(240,22)(-30,-10 ) \bboxc(0,0)(26,13 ) \line(0,-6.5)(0,6.5 ) \text(-7,0)[]{\footnotesize$\overline{\mbox{1}}$ } \text(6,0)[]{\footnotesize$\overline{\mbox{1}}$ } \longarrow(30,0)(60,0 ) \text(90,0)[]{\footnotesize$\emptyset$ } \longarrow(120,0)(150,0 ) \bboxc(180,0)(26,13 ) \line(180,-6.5)(180,6.5 ) \text(173,0)[]{\footnotesize$\mbox{1}$ } \text(186,0)[]{\footnotesize$\mbox{1}$ } \ptext(45,2)(0)[b]{0 } \ptext(135,2)(0)[b]{0 } \end{picture}}}.\notag & & \end{aligned}\ ] ] note that none of the elements satisfies . for type @xmath347 the crystal @xmath351 is perfect of level 2 ( * ? ? ? 1.6 and 6.7 ) . the elements are given by @xmath352{\footnotesize$\mbox{x}$ } \text(19,0)[]{\footnotesize$\mbox{y}$ } \end{picture}}}$ ] with @xmath353 and @xmath354 . the action of @xmath355 for @xmath356 is the same as for the above @xmath325 crystal of level one , and @xmath357 where @xmath358 is the involution that exchanges 1 and @xmath359 ( with appropriate reorderings ) . the strings of length greater than one are the same as in and . in addition there are the following @xmath89-strings of length 2 @xmath350{\footnotesize$\overline{\mbox{1}}$ } \text(6,0)[]{\footnotesize$\overline{\mbox{1}}$ } \longarrow(30,0)(60,0 ) \bboxc(90,0)(26,13 ) \line(90,-6.5)(90,6.5 ) \text(83,0)[]{\footnotesize$\mbox{2}$ } \text(96,0)[]{\footnotesize$\overline{\mbox{1}}$ } \longarrow(120,0)(150,0 ) \bboxc(180,0)(26,13 ) \line(180,-6.5)(180,6.5 ) \text(173,0)[]{\footnotesize$\mbox{2}$ } \text(186,0)[]{\footnotesize$\mbox{2}$ } \ptext(45,2)(0)[b]{0 } \ptext(135,2)(0)[b]{0 } \end{picture } } } & & \notag \\ & \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(240,22)(-30,-10 ) \bboxc(0,0)(26,13 ) \line(0,-6.5)(0,6.5 ) \text(-7,0)[]{\footnotesize$\overline{\mbox{2}}$ } \text(6,0)[]{\footnotesize$\overline{\mbox{1}}$ } \longarrow(30,0)(60,0 ) \bboxc(90,0)(26,13 ) \line(90,-6.5)(90,6.5 ) \text(83,0)[]{\footnotesize$\mbox{1}$ } \text(96,0)[]{\footnotesize$\overline{\mbox{1}}$ } \longarrow(120,0)(150,0 ) \bboxc(180,0)(26,13 ) \line(180,-6.5)(180,6.5 ) \text(173,0)[]{\footnotesize$\mbox{1}$ } \text(186,0)[]{\footnotesize$\mbox{2}$ } \ptext(45,2)(0)[b]{0 } \ptext(135,2)(0)[b]{0 } \end{picture } } } & & \label{string 0}\\ & \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(240,22)(-30,-10 ) \bboxc(0,0)(26,13 ) \line(0,-6.5)(0,6.5 ) \text(-7,0)[]{\footnotesize$\overline{\mbox{2}}$ } \text(6,0)[]{\footnotesize$\overline{\mbox{2}}$ } \longarrow(30,0)(60,0 ) \bboxc(90,0)(26,13 ) \line(90,-6.5)(90,6.5 ) \text(83,0)[]{\footnotesize$\mbox{1}$ } \text(96,0)[]{\footnotesize$\overline{\mbox{2}}$ } \longarrow(120,0)(150,0 ) \bboxc(180,0)(26,13 ) \line(180,-6.5)(180,6.5 ) \text(173,0)[]{\footnotesize$\mbox{1}$ } \text(186,0)[]{\footnotesize$\mbox{1}$ } \ptext(45,2)(0)[b]{0 } \ptext(135,2)(0)[b]{0 } \end{picture}}}. & & \notag\end{aligned}\ ] ] the only elements fulfilling are @xmath352{\footnotesize$\mbox{1}$ } \text(19,0)[]{\footnotesize$\overline{\mbox{1}}$ } \end{picture}}}$ ] , @xmath352{\footnotesize$\mbox{1}$ } \text(19,0)[]{\footnotesize$\mbox{2}$ } \end{picture}}}$ ] , @xmath352{\footnotesize$\mbox{2}$ } \text(19,0)[]{\footnotesize$\overline{\mbox{1}}$ } \end{picture}}}$ ] , and @xmath352{\footnotesize$\mbox{2}$ } \text(19,0)[]{\footnotesize$\mbox{2}$ } \end{picture}}}$ ] which belong to a @xmath89-string and a @xmath360-string of length two . it can be checked that setting @xmath361 for @xmath129 one of these four elements guarantees the involutive condition of @xmath339 . for type @xmath342 the crystal @xmath351 is perfect of level 2 ( * ? ? ? 1.7 and 6.8 ) . it consists of the elements @xmath352{\footnotesize$\mbox{x}$ } \text(19,0)[]{\footnotesize$\mbox{y}$ } \end{picture}}}$ ] with @xmath353 and @xmath362 ; @xmath363 is excluded . the action of @xmath355 for @xmath356 is given by the tensor product rule using the action on the level 1 crystal of @xmath342 as given in table [ table crystals ] , and @xmath357 where @xmath358 is the involution that exchanges 1 and @xmath359 ( with appropriate reorderings ) . the strings of length greater than one are those of equations and and in addition the following @xmath364-string of length four @xmath365{\footnotesize$\mbox{n}$ } \text(6,0)[]{\footnotesize$\mbox{n}$ } \longarrow(20,0)(50,0 ) \bboxc(70,0)(26,13 ) \line(70,-6.5)(70,6.5 ) \text(63,0)[]{\footnotesize$\mbox{n}$ } \text(76,0)[]{\footnotesize$\mbox{0}$ } \longarrow(90,0)(120,0 ) \bboxc(140,0)(26,13 ) \line(140,-6.5)(140,6.5 ) \text(133,0)[]{\footnotesize$\mbox{n}$ } \text(146,0)[]{\footnotesize$\overline{\mbox{n}}$ } \longarrow(160,0)(190,0 ) \bboxc(210,0)(26,13 ) \line(210,-6.5)(210,6.5 ) \text(203,0)[]{\footnotesize$\mbox{0}$ } \text(216,0)[]{\footnotesize$\overline{\mbox{n}}$ } \longarrow(230,0)(260,0 ) \bboxc(280,0)(26,13 ) \line(280,-6.5)(280,6.5 ) \text(273,0)[]{\footnotesize$\overline{\mbox{n}}$ } \text(286,0)[]{\footnotesize$\overline{\mbox{n}}$ } \ptext(35,2)(0)[b]{n } \ptext(105,2)(0)[b]{n } \ptext(175,2)(0)[b]{n } \ptext(245,2)(0)[b]{n } \end{picture}}}.\ ] ] the same four elements as for @xmath347 satisfy and again setting @xmath361 for these ensures the involutive property of @xmath339 . for type @xmath344 the crystal @xmath351 is perfect of level 2 ( * ? ? ? 1.8 and 6.9 ) . it consists of the elements @xmath352{\footnotesize$\mbox{x}$ } \text(19,0)[]{\footnotesize$\mbox{y}$ } \end{picture}}}$ ] with @xmath353 and @xmath366 , the cases @xmath367 and @xmath368 being excluded . the action of @xmath355 for @xmath356 is given by the tensor product rule using the action on the level 1 crystal of @xmath344 as given in table [ table crystals ] , and @xmath357 where @xmath358 is the involution that exchanges 1 and @xmath359 ( with appropriate reorderings ) . again the strings of length greater than one are the same as in equations and plus the following @xmath364-strings @xmath350{\footnotesize$\mbox{n}$ } \text(6,0)[]{\footnotesize$\mbox{n}$ } \longarrow(30,0)(60,0 ) \bboxc(96.5,0)(39,13 ) \line(90,-6.5)(90,6.5 ) \text(83,0)[]{\footnotesize$\mbox{n}$ } \text(103,0)[]{\footnotesize$\overline{\mbox{n-1}}$ } \longarrow(120,0)(150,0 ) \bboxc(180,0)(52,13 ) \line(180,-6.5)(180,6.5 ) \text(167,0)[]{\footnotesize$\overline{\mbox{n-1}}$ } \text(193,0)[]{\footnotesize$\overline{\mbox{n-1}}$ } \ptext(45,2)(0)[b]{n } \ptext(135,2)(0)[b]{n } \end{picture}}}\\ & \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(240,22)(-30,-10 ) \bboxc(0,0)(52,13 ) \line(0,-6.5)(0,6.5 ) \text(-13,0)[]{\footnotesize$\mbox{n-1}$ } \text(13,0)[]{\footnotesize$\mbox{n-1}$ } \longarrow(30,0)(60,0 ) \bboxc(83.5,0)(39,13 ) \line(90,-6.5)(90,6.5 ) \text(77,0)[]{\footnotesize$\mbox{n-1}$ } \text(96,0)[]{\footnotesize$\overline{\mbox{n}}$ } \longarrow(120,0)(150,0 ) \bboxc(180,0)(26,13 ) \line(180,-6.5)(180,6.5 ) \text(173,0)[]{\footnotesize$\overline{\mbox{n}}$ } \text(186,0)[]{\footnotesize$\overline{\mbox{n}}$ } \ptext(45,2)(0)[b]{n } \ptext(135,2)(0)[b]{n } \end{picture}}}\\ & \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(240,22)(-30,-10 ) \bboxc(-6.5,0)(39,13 ) \line(0,-6.5)(0,6.5 ) \text(-13,0)[]{\footnotesize$\mbox{n-1}$ } \text(6.5,0)[]{\footnotesize$\mbox{n}$ } \longarrow(30,0)(60,0 ) \bboxc(90,0)(52,13 ) \line(90,-6.5)(90,6.5 ) \text(77,0)[]{\footnotesize$\mbox{n-1}$ } \text(103,0)[]{\footnotesize$\overline{\mbox{n-1}}$ } \longarrow(120,0)(150,0 ) \bboxc(186.5,0)(39,13 ) \line(180,-6.5)(180,6.5 ) \text(173,0)[]{\footnotesize$\overline{\mbox{n}}$ } \text(193,0)[]{\footnotesize$\overline{\mbox{n-1}}$ } \ptext(45,2)(0)[b]{n } \ptext(135,2)(0)[b]{n } \end{picture}}}\ ; .\end{aligned}\ ] ] in addition to the four elements @xmath352{\footnotesize$\mbox{1}$ } \text(19,0)[]{\footnotesize$\overline{\mbox{1}}$ } \end{picture}}}$ ] , @xmath352{\footnotesize$\mbox{1}$ } \text(19,0)[]{\footnotesize$\mbox{2}$ } \end{picture}}}$ ] , @xmath352{\footnotesize$\mbox{2}$ } \text(19,0)[]{\footnotesize$\overline{\mbox{1}}$ } \end{picture}}}$ ] , and @xmath352{\footnotesize$\mbox{2}$ } \text(19,0)[]{\footnotesize$\mbox{2}$ } \end{picture}}}$ ] also the elements @xmath369{\footnotesize$\overline{\mbox{n-1}}$ } \text(38,0)[]{\footnotesize$\overline{\mbox{n-1}}$ } \end{picture}}}$ ] , @xmath369{\footnotesize$\mbox{n-1}$ } \text(38,0)[]{\footnotesize$\overline{\mbox{n-1}}$ } \end{picture}}}$ ] , @xmath370{\footnotesize$\mbox{n}$ } \text(25,0)[]{\footnotesize$\overline{\mbox{n-1}}$ } \end{picture}}}$ ] , and @xmath370{\footnotesize$\overline{\mbox{n}}$ } \text(25,0)[]{\footnotesize$\overline{\mbox{n-1}}$ } \end{picture}}}$ ] satisfy . the latter ones are contained in an @xmath371-string and an @xmath364-string . setting @xmath361 for the first four elements and @xmath372 for the last four elements ensures the involutive property of @xmath339 . the crystal @xmath373 is a level 2 perfect crystal for @xmath348 ( * * sections 1.9 and 6.10 ) . the elements of this crystal are @xmath374 , @xmath375{\footnotesize$\mbox{x}$ } \end{picture}}}}$ ] , and @xmath352{\footnotesize$\mbox{x}$ } \text(19,0)[]{\footnotesize$\mbox{y}$ } \end{picture}}}$ ] with @xmath376 and @xmath353 ; @xmath363 is excluded . the action of @xmath355 for @xmath356 is given by the tensor product rule using the action on the level 1 crystal of @xmath348 as given in table [ table crystals ] , and the action of @xmath377 is given by @xmath378{\footnotesize$\emptyset$ } \longarrow(30,0)(60,0 ) \bboxc(70,0)(13,13 ) \text(70,0)[]{\footnotesize$\mbox{1}$ } \ptext(45,2)(0)[b]{0 } \end{picture } } } & \\ \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(94,22)(-2,-10 ) \bboxc(19.5,0)(13,13 ) \text(19,0)[]{\footnotesize$\mbox{x}$ } \longarrow(30,0)(60,0 ) \bboxc(77,0)(26,13 ) \line(77,-6.5)(77,6.5 ) \text(70,0)[]{\footnotesize$\mbox{1}$ } \text(83,0)[]{\footnotesize$\mbox{x}$ } \ptext(45,2)(0)[b]{0 } \end{picture } } } & \qquad \text{for $ x\neq \overline{1}$}\\ \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(94,22)(-2,-10 ) \bboxc(19.5,0)(13,13 ) \text(19,0)[]{\footnotesize$\overline{\mbox{1}}$ } \longarrow(30,0)(60,0 ) \text(70,0)[]{\footnotesize$\emptyset$ } \ptext(45,2)(0)[b]{0 } \end{picture } } } & \\ \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(94,22)(-2,-10 ) \bboxc(13,0)(26,13 ) \line(13,-6.5)(13,6.5 ) \text(6,0)[]{\footnotesize$\mbox{x}$ } \text(19,0)[]{\footnotesize$\overline{\mbox{1}}$ } \longarrow(30,0)(60,0 ) \bboxc(70,0)(13,13 ) \text(70,0)[]{\footnotesize$\mbox{x}$ } \ptext(45,2)(0)[b]{0 } \end{picture } } } & \qquad \text{for $ x\neq 1 $ } \end{split}\ ] ] and undefined otherwise . the strings of length greater than one are given by , and @xmath379{\footnotesize$\overline{\mbox{1}}$ } \text(6,0)[]{\footnotesize$\overline{\mbox{1}}$ } \longarrow(20,0)(50,0 ) \bboxc(65,0)(13,13 ) \text(66,)[]{\footnotesize$\overline{\mbox{1}}$ } \longarrow(80,0)(110,0 ) \text(125,0)[]{\footnotesize$\emptyset$ } \longarrow(140,0)(170,0 ) \bboxc(185,0)(13,13 ) \text(186,0)[]{\footnotesize$\mbox{1}$ } \longarrow(200,0)(230,0 ) \bboxc(250,0)(26,13 ) \line(250,-6.5)(250,6.5 ) \text(243,0)[]{\footnotesize$\mbox{1}$ } \text(256,0)[]{\footnotesize$\mbox{1}$ } \ptext(35,2)(0)[b]{0 } \ptext(95,2)(0)[b]{0 } \ptext(155,2)(0)[b]{0 } \ptext(215,2)(0)[b]{0 } \end{picture } } } \\[2 mm ] & \raisebox{-0.15cm}{\scalebox{0.8 } { \begin{picture}(280,22)(-30,-10 ) \bboxc(65,0)(13,13 ) \text(66,0)[]{\footnotesize$\mbox{n}$ } \longarrow(80,0)(110,0 ) \bboxc(125,0)(13,13 ) \text(126,0)[]{\footnotesize$\mbox{0}$ } \longarrow(140,0)(170,0 ) \bboxc(185,0)(13,13 ) \text(186,0)[]{\footnotesize$\overline{\mbox{n}}$ } \ptext(95,2)(0)[b]{n } \ptext(155,2)(0)[b]{n } \end{picture}}}\end{aligned}\ ] ] there are no elements with property . the crystal @xmath373 is a level 2 perfect crystal for @xmath346 ( * ? ? ? * sec . 1.10 and 6.11 ) . the elements of this crystal are @xmath374 , @xmath375{\footnotesize$\mbox{x}$ } \end{picture}}}}$ ] , and @xmath352{\footnotesize$\mbox{x}$ } \text(19,0)[]{\footnotesize$\mbox{y}$ } \end{picture}}}$ ] with @xmath380 and @xmath353 . the action of @xmath355 for @xmath356 is given by the tensor product rule using the action on the level 1 crystal of @xmath346 as given in table [ table crystals ] , and the action of @xmath377 is the same as in . the strings of length greater than one are as in for @xmath349 , and . again there are no elements with property . [ northeast ] suppose @xmath0 is of type @xmath15 in lemma [ lev one ] . the function @xmath339 amounts to a canonical choice of a simple root @xmath131 among those such that the given element admits @xmath146 . consider @xmath381 such that @xmath382 . in addition to the realization of the crystal @xmath383 by the space of homogeneous paths using the crystal given in the proof of lemma [ lev one ] , one may also consider the realization in @xcite by @xmath364-regular partitions . suppose @xmath112 is the partition corresponding to @xmath129 . then up to the dynkin diagram automorphism that sends @xmath384 to @xmath385 modulo @xmath364 , the choice of violation @xmath339 corresponds to the corner cell of @xmath112 that is in the rightmost column of @xmath112 . this choice of corner cell is used in @xcite to define the smallest demazure crystal of @xmath383 containing @xmath129 . [ pi ] let @xmath0 be as in theorem [ gen boson ] , and @xmath126 , @xmath10 , and @xmath216 be as in . suppose in addition that for all @xmath171 and @xmath386 , if @xmath387 under the local isomorphism @xmath388 and @xmath389 then @xmath390 . then @xmath391 it is enough to verify the hypotheses of theorem [ gen boson ] , applied to @xmath392 . @xmath393 is isomorphic to a direct sum of irreducible integrable highest weight modules by theorem [ level tensor ] . @xmath394 is obviously closed under the @xmath146 . it follows from ( * ? ? ? * lemma 1 ) that @xmath395 generates @xmath393 . to check the third condition of theorem [ gen boson ] , let @xmath127 and @xmath57 be such that @xmath396 . then @xmath397 . this implies that the @xmath131-string of @xmath398 inside @xmath393 , consists of vectors of the form @xmath399 where @xmath400 . finally , lemma [ energy e ] with @xmath126 replaced by @xmath401 guarantees that the affine weight function on @xmath393 determined by its value on highest weight vectors , agrees on the subset @xmath394 with the function @xmath402 where @xmath403 is the original weight function . observe that even without the extra hypothesis on the action of @xmath404 in theorem [ pi ] , one obtains a bosonic formula . the extra condition is only needed to show that the energy function @xmath405 gives rise to the correct affine weight for all elements of the form @xmath406 and not just on the highest weight vectors . perhaps this extra condition is always a consequence of the other hypotheses . now the formula is written more explicitly . let @xmath407 and @xmath408 be of level @xmath13 . a formula equivalent to is obtained by taking the coefficient of @xmath409 on both sides : @xmath410 { k}(b,{\lambda},{\lambda}',b_0)(q ) = \sum_{(w , b)\in { \mathcal{s } } } ( -1)^w\ ] ] where @xmath253 is the set of pairs @xmath254 such that @xmath411 let @xmath412 be the sublattice of @xmath101 given by the image under @xmath413 of the @xmath414-span of the orbit @xmath415 . let @xmath416 be the group of translations by the elements of @xmath412 , where @xmath417 is translation by @xmath418 . then @xmath419 and @xmath420 . for @xmath418 and @xmath100 of level @xmath13 , one has ( * ? ? ? * ( 6.5.2 ) ) @xmath421 the action of @xmath422 on the level @xmath13 weight @xmath10 is given by @xmath423 now @xmath424 where @xmath425 is the dual coxeter number and @xmath426 is the half - sum of the positive roots in @xmath97 . recall that @xmath427 leaves @xmath428 invariant . in write @xmath429 where @xmath422 and @xmath418 , obtaining @xmath430 since both sides are weights of level @xmath13 , by equating coefficients of @xmath428 and projections into @xmath101 , one obtains the equivalent conditions @xmath431 and @xmath432 therefore one has the equality @xmath433 where @xmath127 satisfies @xmath434 let @xmath100 be a weight of level @xmath13 but not necessarily dominant . consider the weight @xmath260 . if it is regular ( not fixed by any @xmath70 ) then there is a unique @xmath70 such that @xmath437 . it follows from the definition of @xmath5 that @xmath438 then for all @xmath57 , @xmath439 suppose @xmath184 . then @xmath440 for @xmath186 , recall that @xmath441 then @xmath442 and @xmath443 these equations express the @xmath14-equivalence in @xcite . let @xmath444 have standard basis @xmath453 and @xmath101 be the subspace of @xmath444 orthogonal to the vector @xmath454 . then @xmath455 for @xmath456 , @xmath457 , @xmath53 is the ordinary dot product in @xmath444 , and @xmath427 is the symmetric group on @xmath364 letters acting on the coordinates of @xmath444 . since @xmath458 and @xmath446 is @xmath459-linear , one may replace every term @xmath460 by @xmath461 in . define the map @xmath447^{{\overline{w}}}[q]\rightarrow{\mathbb{z}}[{\overline{p}}]^{{\overline{w}}}[q]$ ] given by @xmath462 . define @xmath463 in @xmath447^{{\overline{w}}}[q]$ ] by the condition that the above linear map sends @xmath464 and @xmath465 to the same element . with this definition , we have @xmath466 it is not hard to see that this recovers the @xmath14-equivalence of schur functions given in @xcite . in this section it is assumed that @xmath0 is of type @xmath15 , @xmath467 , and the tensor factors @xmath170 are perfect crystals of the form @xmath468 in the notation of @xcite with @xmath469 for all @xmath295 . by restriction to @xmath111 , @xmath170 is the crystal of the irreducible integrable @xmath111-module of highest weight @xmath470 . in this case @xmath216 is not needed . to see this , recall that @xmath170 can be realized as the set of column - strict young tableaux of the rectangular shape having @xmath471 rows and @xmath472 columns with entries in the set @xmath473 . in @xcite the @xmath122-weighted @xmath125-crystal structure on the perfect crystals @xmath474 is computed explicitly . in particular , if @xmath386 is a tableau then @xmath475 is at most the number of ones in the tableau @xmath129 , which is at most @xmath472 by column - strictness . therefore @xmath476 never admits @xmath404 . thus the energy function @xmath195 of has the property that for any @xmath477 such that @xmath478 , one has @xmath190 . thus one obtains the bosonic formula in this case . since @xmath0 is of type @xmath15 , @xmath435 and @xmath479 . take @xmath480 in . the lattice @xmath412 is given by the root lattice @xmath481 of @xmath97 , which may be realized by @xmath482 . let @xmath483 be the set of paths @xmath127 of weight @xmath484 . then @xmath485 notice that @xmath486 is ( up to an overall factor ) the @xmath487 form of the supernomial @xmath329 of ref . @xcite so that @xmath488 equals the left - hand side of ( * ? ? ? * ( 9.2 ) ) up to an overall power of @xmath14 . this shows that the left - hand side of ( * ? ? ? * ( 9.2 ) ) is indeed the generating function of level-@xmath13 restricted paths . to establish the equality ( * ? ? ? * ( 9.2 ) ) it remains to prove that also the right - hand side equals the generating function of level - restricted paths . as in the previous section let @xmath0 be of type @xmath489 and assume that @xmath490 . fix @xmath491 and @xmath492 weights of level 1 . it is easy to verify that @xmath208 consists of at most one element @xmath493 . choose @xmath494 such that @xmath495 exists . then by and we find that @xmath496 where @xmath497 is the set of paths @xmath127 of weight @xmath498 . a similar formula exists for @xmath499 : @xmath500 where @xmath483 is the set of paths @xmath127 of weight @xmath501 . the right - hand side is the generating function of paths in @xmath126 of level zero since there are no level zero restricted paths unless @xmath126 is empty . however , the arguments of sections [ sec general bose ] and [ sec inhomo paths ] do not imply that also the left - hand side is the generating function of level zero paths since it was assumed in the proof of theorem [ gen boson ] that the level of the crystals @xmath170 does not exceed @xmath13 . we have assumed that @xmath502 which are crystals of level one . however , it is possible to define a sign - reversing involution directly on @xmath490 without using the crystal isomorphisms that are used in the proof of theorem [ gen boson ] . let @xmath127 . there exists at least one @xmath503 such that @xmath504 is defined . define @xmath505 which has the property that @xmath506 where as before @xmath507 . hence define the involution @xmath508 . it is again sign - reversing and has no fixed points when @xmath509 . this proves that the left - hand side of is the generating function of level 0 restricted paths . equation was conjectured in @xcite . for @xmath510 identity follows from the @xmath14-binomial theorem , for @xmath511 it was proven in ( * ? ? ? * proposition 5.1 ) and for general @xmath364 it was conjectured in @xcite . e. date , m. jimbo , a. kuniba , t. miwa , and m. okado , _ paths , maya diagrams and representations of @xmath513 _ , in `` integrable systems in quantum field theory and statistical mechanics '' , 149191 , adv . stud . pure math . , * 19 * , academic press , boston , ma , 1989 . v. lakshmibai , _ standard monomial theory for @xmath515 _ , in `` operator algebras , unitary representations , enveloping algebras , and invariant theory '' ( paris , 1989 ) , 197217 , progr . , 92 , birkhuser boston , boston , ma , 1990 .

we present chemodynamical simulations of a milky way - type galaxy using a self - consistent hydrodynamical code that includes supernova feedback and chemical enrichment , and predict the spatial distribution of elements from oxygen to zinc . in the simulated galaxy , the kinematical and chemical properties of the bulge , disk , and halo are consistent with the observations . the bulge formed from the assembly of subgalaxies at , and has higher [/fe ] ratios because of the small contribution from type ia supernovae . the disk formed with a constant star formation over gyr , and shows a decreasing trend of [/fe ] and increasing trends of [ ( na , al , cu , mn)/fe ] against [ fe / h ] . however , the thick disk stars tend to have higher [/fe ] and lower [ mn / fe ] than thin disk stars . we also predict the frequency distribution of elemental abundance ratios as functions of time and location , which can be directly compared with galactic archeology projects such as hermes .

1012.5144

while the evolution of the dark matter is reasonably well understood , the evolution of the baryonic component is much less certain because of the complexity of the relevant physical processes , such as star formation and feedback . with the commonly employed , schematic star formation criteria alone , the predicted star formation rates ( sfrs ) are higher than what is compatible with the observed luminosity density . thus feedback mechanisms are in general invoked to reheat gas and suppress star formation . we include the feedback from stellar winds , core - collapse supernovae ( normal type ii supernovae ( sne ii ) and hypernovae ( hne ) ) , and type ia supernovae ( sne ia ) in our hydrodynamical simulations . supernovae inject not only thermal energy but also heavy elements into the interstellar medium ( ism ) , which can enhance star formation . chemical enrichment must be solved as well as energy feedback . supernova feedback is also important for solving the angular momentum problem @xcite and the missing satellite problem @xcite , and for explaining the existence of heavy elements in intracluster medium @xcite and intergalactic medium , and possibly the mass - metallicity relation of galaxies ( * ? ? ? * hereafter k07 ) . since different heavy elements are produced from different supernovae with different timescales , elemental abundance ratios can provide independent information on `` age '' ( e.g. , * ? ? ? * hereafter kn09 ) . therefore , stars in a galaxy can be used as fossils where the chemical enrichment history of the galaxy is imprinted , and this approach is called as galactic archeology . in the milky way galaxy , the detailed structures of kinematics and metallicities have been obtained with sdss ( the sloan digital sky survey ) and rave ( the radial velocity experiment ) , and will be extended with pan - starrs and skymapper . in the next decade , high - resolution multi - object spectroscopy ( e.g. , apogee with sdss and hermes on the aat ) and space astrometry mission ( e.g. , gaia ) will provide 6d - kinematics and elemental abundances for a million stars in the local group . in order to untangle the formation and evolution history of the galaxy from observational data , a `` realistic '' model that includes star formation and chemical enrichment is required . although there are several simulations of the milky way - type galaxy with very high resolutions ( e.g. , * ? ? ? * ) , there are few studies with hydrodynamics and chemical enrichment ( e.g. , * ? ? ? * ) . in most of the previous work , however , the adopted assumptions are not appropriate for comparison with observations ; the effect of mass - loss from stars is ignored , the timescales of sne ia is given by only one parameter , the nucleosynthesis yields of supernovae are outdated , and so on . recently , several new findings have been obtained from the collaboration between observations , stellar astrophysics , and galactic chemical evolution modeling . from the observed light curves and spectra of nearby core - collapse supernovae , it has been shown that there are energetic supernovae , hne , ejecting more than ten times larger explosion energy ( @xmath3 ) and a certain amount of iron ( e.g. , * ? ? ? * ) as well as @xmath1 elements ( o , mg , si , s , ca , and ti ) . such hne play a major role in reproducing the observed [ zn / fe ] trend in the solar neighborhood , where [ zn / fe ] is @xmath4 over @xmath5 [ fe / h ] @xmath6 and possibly increases toward lower [ fe / h ] ( * ? ? ? * hereafter k06 ) . from the observations of supernova rates in various types of galaxies @xcite , it has been shown that there is a young population of sne ia in addition to the old population that is usually found in present - day elliptical galaxies . the metallicity effect on the occurrence of sne ia , which has been proposed by ( * ? ? ? * hereafter k98 ) , plays a major role in reproducing the observed [ ( @xmath1,mn)/fe]-[fe / h ] relations ( kn09 ) . including the up - to - date knowledge of chemical evolution of galaxies , we provide chemodynamical simulations of a milky way - type galaxy from cdm initial conditions . in 2 , we summarize our simulation code and chemical enrichment sources . in 3 , we show the differences between the chemical properties of the disk and bulge , namely the metallicity distribution function , the age - metallicity relation , and the [ @xmath1/fe]-[fe / h ] relation , and predict the [ x / fe]-[fe / h ] relation . we also propose a new diagram of [ @xmath1/fe]-[mn / fe ] as a sequence of the sn ia contribution . 4 gives our conclusions . the details of our grape - sph code are described in @xcite and can be summarized as follows . 1 . the gravity is computed with the tree method using the special purpose computer grape ( gravity pipe ) system at the national astronomical observatory of japan . for hydrodynamics , the smoothed particle hydrodynamical ( sph ) method is adopted , and the sph formulation is the almost same as @xcite . the grape - sph code is highly adaptive in space and time by means of individual smoothing lengths and individual timesteps , and it has very high performance ( a week for one simulation with @xmath7 ) . * radiative cooling * is computed with the metal - dependent cooling functions , which are generated with the mappings iii software @xcite as a function of [ fe / h ] . [ o / fe ] is fixed with the observed [ o / fe]-[fe / h ] relation in the solar neighborhood . our * star formation * criteria are the same as @xcite ; 1 ) convergent , 2 ) cooling , and 3 ) jeans unstable . the sfr is determined from the schmidt law ; the star formation timescale is proportional to the dynamical timescale ( @xmath8 ) , where @xmath9 is chosen from the size - luminosity relation of elliptical galaxies @xcite . in addition , we find that @xmath9 gives a better fit to the elemental abundance ratios in the milky way galaxy than @xmath10 ( see fig . [ fig : afenum05 ] ) . we also adopt the probability criterion @xcite . in our simulations , the sfr depends both on the local density and metallicity , which is different from other simplified models such as one - zone models . 4 . if a gas particle satisfies the star formation criteria , a part of the mass of the gas particle turns into a star particle . since one star particle has the mass of @xmath11 , the star particle is not a single star but an association of many stars . the masses of individual stars span according to the * initial mass function * ( imf ) . we adopt the salpeter imf that is invariant to time and metallicity with a slope @xmath12 for @xmath13 . 5 . for the * feedback * of energy and heavy elements , we do not adopt the instantaneous recycling approximation . via stellar winds , sne ii / hne , and sne ia , thermal energy and heavy elements are ejected from an evolved star particle as a function of time , and distributed to the surrounding gas particles within a sphere of feedback radius @xmath14 . the mass transfer is weighted by the sph kernel from the gas particle to the dying star particle . this feedback scheme is better than that implementing a fixed number of feedback neighbor particles ( @xmath15 ) in galaxy - scale simulations ; the metallicity of gas particles does not depend very much on the local density if @xmath15 is fixed . we find that @xmath16 kpc gives a better fit to the elemental abundance ratios than @xmath17 and @xmath18 kpc , or @xmath19 and @xmath20 ( see fig . [ fig : afenum05 ] ) . note that the diffusion between gas particles is supposed to enhance the mixing of heavy elements but is not included in our model . the * photometric evolution * of one star particle is identical to the evolution of a simple stellar population ( ssp ) . ssp spectra are taken from @xcite as a function of age and metallicity . the photometric evolution of the galaxy is calculated from the summation of the ssp spectra of star particles with various ages and metallicities . the scheme to include the detailed chemical enrichment into sph simulations has been proposed by @xcite and updated as follows . * hypernovae ( hne ) * the explosion mechanism of core - collapse supernovae is still uncertain , although a few groups have presented feasible calculations in exploding core - collapse supernovae @xcite . however , the ejected explosion energy and @xmath21ni mass ( which decays to @xmath21fe ) can be directly estimated from the observations , i.e. , the light curve and spectra fitting of individual supernova . as a result , it is found that many core - collapse supernovae ( @xmath22 ) have more than ten times larger explosion energy ( @xmath3 ) and produce a significant amount of iron @xcite . we adopt our nucleosynthesis yields of core - collapse supernovae from k06 , depending on the progenitor mass , metallicity ( @xmath23 ) , and the explosion energy ( normal sne ii and hne ) . the progenitors mass range is @xmath24 but the metal ejection is only from @xmath25 . assuming that a large fraction of supernovae with @xmath22 are hne , the evolution of the elemental abundance ratios from oxygen to zinc is in excellent agreement with observations in the solar neighborhood , bulge , halo , and thick disk . in particular , to solve the [ zn / fe ] problem , the hn fraction @xmath26 has to be as large as @xmath27 ( k06 ) . note that ti is underabundant at @xmath28[fe / h]@xmath6 , which will be solved with the 2d calculation of nucleosynthesis @xcite . the observed increasing trend toward the lower metallicity may suggest that higher energy under the assumption of inhomogeneous enrichment . at high metallicity , since neutron - rich isotopes @xmath29zn are produced , @xmath26 can be as small as @xmath30 . in this work , we assume the metal - dependent efficiency of hne ( @xmath22 ) , @xmath31 , and @xmath32 for @xmath33 , and @xmath34 is adopted independent of metallicity in k06 , k07 , and kn09 . ] , which gives better agreement with the observed present hn rate @xcite . pair - instability supernovae , which produce much more fe , more [ s / fe ] , and less [ zn / fe ] , should not contribute in the galactic chemical evolution @xcite . * type ia supernovae ( sne ia ) * the progenitors of the majority of sne ia are most likely the chandrasekhar ( ch ) mass white dwarfs ( wds ) . for the evolution of accreting c+o wds toward the ch mass , two scenarios have been proposed ; one is the double - degenerate scenario , i.e. , the merging of double c+o wds with a combined mass surpassing the ch mass limit . however , it has been theoretically suggested that it leads to accretion - induced collapse rather than sne ia @xcite , and the lifetimes are too short to reproduce the chemical evolution in the solar neighborhood ( kn09 ) . the other is the single - degenerate ( sd ) scenario , i.e. , the wd mass grows by accretion of hydrogen - rich matter via mass transfer from a binary companion . the mass accretion rate is very limited to trigger carbon deflagration @xcite , but the allowed parameter space of binary systems can be significantly increased by the wd wind effect if the metallicity is higher than [ fe / h ] @xmath35 ( k98 ) . two progenitor systems are found for red - giant ( rg ) companions with initial masses of @xmath36 and near main - sequence ( ms ) companions with initial masses of @xmath37 @xcite . in our simulations , based on the sd scenario , the lifetime distribution function of sne ia is calculated with eq.[2 ] in kn09 , taking into account the metallicity dependence of the wd winds ( k98 ) and the mass - stripping effect on the binary companion stars ( kn09 ) . the lifetimes of sne ia are determined from the lifetimes of companion stars , and thus the lifetime distribution function shows double peaks ; one is for the ms+wd systems with timescales of @xmath38 gyr , which are dominant in star - forming galaxies . the other is for the rg+wd systems with @xmath39 gyr timescales , which are dominant in early - type galaxies . although the metallicity effect has not been confirmed from the supernova survey yet , it is more strongly required in the presence of the young population of sne ia , to be consistent with the chemical evolution of the milky way galaxy . the absolute rates of each component depend on two binary parameters , which are determined by the requirement to meet the observed metallicity distribution function and [ @xmath1/fe]-[fe / h ] relations in the solar neighborhood , and the observed supernova rates in various types of galaxies . this sn ia model gives much better agreement with the observed elemental abundance ratios ( from o to zn ) than models with the double - degenerate scenario or metal - independent exponential lifetime functions ( kn09 ) . for sne ia , we take the nucleosynthesis yields from @xcite , where the metallicity dependence is not included . ni is overproduced at [ fe / h ] @xmath40 , which will be solved by tuning the propagation speed of the burning front and the central density of the white dwarf @xcite . * stellar winds * needless to say , the envelope mass and pre - existing heavy elements are returned by stellar winds from all dying stars . in addition , asymptotic giant branch ( agb ) stars with initial masses between about @xmath41 ( depending on metallicity ) produce light elements such as c and n , while the contribution of heavier elements are negligible in the galactic chemical evolution @xcite . rotating massive stars ( @xmath42 ) could also produce c and n ( e.g. , * ? ? ? * ) , but are not included . lcccc|ccc standard & @xmath43 mpc & 55,534 & @xmath44 & @xmath45 & 180,293 & 0.14 & 0.23 + wider region & @xmath46 mpc & 117,525 & @xmath47 & @xmath45 & 316,991 & 0.15 & 0.25 + higher resolution & @xmath43 mpc & 209,133 & @xmath44 & @xmath48 & 445,259 & 0.12 & 0.29 + the initial condition is generated in the same way as in @xcite and @xcite . the cdm initial fluctuation is generated by the cosmics package @xcite . the cosmological parameters are set to be @xmath49 km s@xmath50 mpc@xmath50 , @xmath51 , @xmath52 , @xmath53 , @xmath54 , and @xmath55 . we generate a periodic boundary condition with the lattice size @xmath56 mpc having a top - hat perturbation of amplitude @xmath57 in radius @xmath58 mpc at the starting redshift of @xmath59 . then , particles in a spherical region with a comoving radius that contains @xmath60 are picked up . finally , because the simulated field is not large enough to generate tidal torque , the initial angular momentum is given to the system in rigid rotation with a constant spin parameter as large as @xmath61 @xcite . such a large value is required to reproduce disk galaxies , although rare in cosmological simulations ( e.g. , * ? ? ? it seems not easy to form the galactic disk in the cdm universe as suggested by previous works ( e.g. , @xcite , but see @xcite ) . to form a disk galaxy , it is necessary that the system does not undergo the major mergers with mass ratios larger than 0.2 and at higher redshifts than @xmath62 , which destroy the disk structure and the radial metallicity gradient @xcite . even in some cases without major mergers , disk structures can not form ( e.g. , * ? ? ? * ) , although this argument depends on the modeling of feedback and numerical resolution . we have run simulations with 150 initial conditions until @xmath63 including gas dynamics and basic chemical enrichment ( fig . [ fig : ic ] ) . the disk structure with a scale length of @xmath64 kpc is seen in 48 runs , and milky way type galaxies form only in 5 runs , from which we select one realization ( 143rd in fig . [ fig : ic ] ) for detailed simulations . in order to check the resolution and boundary effects , we consider three cases ; the total number of particles , the total mass ( the half for gas and the rest for dark matter ) , and the mass resolution for each case are summarized in table 1 . the initial conditions with the wider region and higher resolution are generated with the same initial perturbation as in the initial condition of the standard model . because of star formation , the particle number increases by a factor of @xmath65 at @xmath66 . the final baryon and gas fractions of the galaxy are also summarized in table 1 and do not vary very much among these cases . for the higher resolution run , the calculation is stopped at @xmath67 . the gravitational softening length is set to @xmath68 and @xmath69 kpc for the low and high resolutions , respectively . we obtain qualitatively similar results for all three cases , and consequently , we mainly show the results with the wider initial condition in figs . [ fig : map]-[fig : afenum2 ] and [ fig : mdf2 ] , and with the standard initial condition in figs . [ fig : afenum05 ] , [ fig : afenum19 ] , [ fig : mdf ] , and [ fig : xfe]-[fig : mn ] . figure [ fig : map ] shows the time evolution of the projected density of dark matter , gas , v - band luminosity , and luminosity - weighted stellar metallicity . after the start of the simulation , the system expands according to the hubble flow . the cdm initial fluctuations grow into the structures of nodes and filaments , and small collapsed halos are realized both in dark matter and gas . in the halos , the gas is allowed to cool radiatively , and star formation takes place since @xmath70 . according to the hierarchical clustering of dark halos , subgalaxies merge to form large galaxies , which induces the initial starburst . under the cdm picture , any galaxy forms through the successive merging of subgalaxies with various masses . in this simulated galaxy , the bulge is formed by the initial starburst that is induced by the assembly of gas - rich sub - galaxies with stellar masses of @xmath71 and gas fractions of @xmath72 at @xmath73 . because of the angular momentum , the gas accretes onto the plane forming a rotationally supported disk that grows from inside out . in the disk , star formation takes place with a longer timescale , which is maintained not by the slow gas accretion , but by the self - regulation due to supernova feedback . many satellite galaxies successively come in and disrupt , but there is no major merger event after @xmath74 , which is necessary to retain the disk structure . metallicity gradients , increasing toward higher density regions , are generated both in the gas phase and stars from @xmath75 onward . figure [ fig : map2 ] shows the b - band ( upper panels ) and k - band ( lower panels ) luminosity maps at the present - day for the face - on ( left panels ) and edge - on views ( right panels ) . in the b - band , the disk component is visible because of the on - going star formation , while the bulge component is dominant in the k - band . the galaxy center and the disk plane are determined from the center of gravity and the aliment of the angular momentum vector , respectively . in the following , @xmath76 and @xmath77 denote the radius on the disk plane and the height from the disk plane , respectively . the bulge has a de vaucouleurs surface brightness profile with an effective radius of @xmath78 kpc , and the disk has an exponential profile with a scale length of @xmath79 kpc . in the surface brightness profile , an excess is seen at @xmath80 kpc , which corresponds to the remnant of satellite galaxies . the bulge in the simulated galaxy seems to be larger than the galactic bulge , and these properties may be more consistent with m31 ( e.g. , * ? ? ? note that it is not feasible to resolve the central blackhole , nuclear bulge , and bar in our simulations . the total masses inside 1 , 10 , and 100 kpc are @xmath81 , @xmath82 , and @xmath83 , respectively . the baryon fraction is @xmath84 at @xmath85 kpc , and then decreases to @xmath86 at @xmath87 kpc . the rotation velocity of the disk rapidly increases to reach @xmath88 km / s at @xmath89 kpc , stays constant until @xmath90 kpc , and then gradually decreases until @xmath91 kpc . this is roughly consistent with the observed rotation curve of the milky way galaxy , which shows a plateau at the local standard of rest ( @xmath92 km / s ) for @xmath93 kpc @xcite . in the following , we define the three major components simply from the location of the stars at the present - day : the radius of @xmath94 kpc and the height of @xmath95 kpc for the solar neighborhood , @xmath96 kpc for the bulge , and @xmath97 kpc for the halo . thick disk stars are defined from the kinematics using the toomre diagram in the cylindrical space motion @xmath98 . the number of particles significantly drops around the velocity dispersion @xmath99 km / s and @xmath100 km / s . thus , we define thick disk particles in the range of @xmath101 km / s . this range is different from the observations @xcite , but gives a consistent mass fraction of the thick disk in the simulated galaxy . with this kinematical definition , the thick disk is smaller and thicker than the thin disk . the surface density profiles ( top panel ) and density profiles ( bottom panel ) at @xmath102 kpc for the thin disk ( solid lines ) and thick disk ( dot - dashed lines ) are shown in figure [ fig : sb ] . the scale radius is @xmath103 and @xmath104 kpc , the scale height is @xmath105 and @xmath106 for the thin and thick disks , respectively . these are larger than the 2mass ( two micron all sky survey ) observations @xcite , which may be due to the lack of resolution . the mass , mass - to - light ratio , age , metallicity , and abundance ratio of each component are summarized in table 2 . figure [ fig : zg ] shows the radial and vertical metallicity gradients of stars at present ( top panel ) and at @xmath107 gyr ( bottom panel ) . the radial gradient of stars becomes flatter at higher distances from the plane , which seems to be consistent with the sdss observations @xcite . at @xmath108 kpc ( solid line ) , the stellar radial gradient is @xmath109}_*/\delta r = -0.025 $ ] dex / kpc . this is flatter than the radial gradient of the ism ( dotted line ) , which is @xmath109}_{\rm g}/\delta r = -0.049 $ ] dex / kpc at @xmath108 kpc . the metallicity radial gradients are steeper at higher redshifts . at @xmath107 gyr , @xmath109}_*/\delta r = -0.062 $ ] dex / kpc and @xmath109}_{\rm g}/\delta r = -0.094 $ ] dex / kpc for stars and the ism , respectively . the time evolution of the ism gradient is consistent with the observations with planetary nebulae @xcite . the resultant star formation histories ( fig . [ fig : sfr ] ) are different for different components . the initial starburst is induced by the assembly , and most of stars in the bulge ( dashed line ) and halo ( dotted line ) are formed at this stage . in the disk ( solid line ) , the sfr is regulated mainly by supernova feedback . figure [ fig : agecumnum ] shows the cumulative functions of ages of stars in each component in the present - day galaxy . most of stars in the present bulge ( dashed line ) have formed in the first @xmath46 gyr , 80% of bulge stars are older than @xmath110 gyr , and 90% of bulge stars are older than @xmath56 gyr . in the disk ( solid line ) , star formation takes place continuously , and 50% of solar - neighborhood stars are younger than @xmath56 gyr . in the halo ( dotted line ) , most of stars are as old as bulge , although the numerical resolution of this simulation is not enough to discuss the halo in detail . however , there does not seem to be very much difference in the sfr between the inner and outer halos . these star formation histories can be seen for other initial conditions that form disk galaxies at present in our simulations , and also for different initial conditions e.g. , those in @xcite and @xcite . in chemodynamical simulations , the ism is not homogeneous at any time . first of all , there is a local variation in star formation and metal flow by the inflow and outflow of the ism . second , the mixing of heavy elements is caused by supernova feedback . with our feedback scheme ( [ sec : model ] ) , star particles obtain heavy elements from the gas particles from which the stars form . gas particles obtain heavy elements only when they pass through within a feedback radius of dying star particles . the amount of heavy elements that the gas particles obtain is physically uncertain , and depend on the modeling of feedback . third , the elemental abundance ratios of the metals produced by supernovae depend on the metallicity and mass of progenitors ( [ sec : chem ] ) , i.e. , the metallicity and age of star particles . in our simulations , the metallicity of the first enriched stars reaches [ fe / h ] @xmath111 . at later times , the star forming region becomes denser , and both metal richer and poorer stars than [ fe / h ] @xmath111 appear . different from one - zone chemical evolution models , the following phenomena occur in the case of inhomogeneous enrichment : i ) the age - metallicity relation is weak . ii ) sne ia can affect the elemental abundance ratios at [ fe / h ] @xmath112 even with the metallicity inhibition of sne ia . iii ) the scatter of elemental abundance ratios becomes large if the supernova yield depends on progenitor metallicity . the details are described in the following sections . as discussed in [ sec : model ] , we run simulations with several feedback parameters and schemes , and show the best results in this paper . lc|ccc|cc|ccc|ccc|cc solar neighborhood & @xmath113 & 1.90 & 2.22 & 0.95 & 7.07 & 3.27 & -0.19 & -0.20 & -0.23 & -0.07 & -0.16 & -0.21 & 0.42 & 0.63 + bulge & @xmath114 & 6.92 & 6.15 & 1.51 & 11.17 & 9.03 & -0.01 & -0.03 & 0.14 & 0.01 & -0.10 & -0.32 & 0.22 & 0.62 + halo & @xmath115 & 7.92 & 6.28 & 1.69 & 12.02 & 12.07 & -0.27 & -0.38 & -1.63 & 0.24 & 0.25 & -0.85 & 0.02 & 0.06 + thick disk & @xmath116 & 5.57 & 4.53 & 1.27 & 8.61 & 7.49 & -0.23 & -0.30 & - & -0.04 & -0.08 & - & - & 0.70 + the chemical enrichment timescale is also different for the different components . figure [ fig : amr ] shows the age - metallicity relations in the present - day galaxy . in the solar neighborhood ( upper panel ) , the average metallicity reaches [ fe / h ] @xmath117 at @xmath118 gyr and does not show strong evolution for @xmath119 gyr . the scatter in metallicity at given time is caused by the inhomogeneity of chemical enrichment in our chemodynamical model ( [ sec : inhomo ] ) . as a result , both the average and scatter are in good agreement with the observations ( dots ) in spite of the uncertainties in the observational estimates of the ages . at @xmath120 gyr , a week trend is seen in our simulation , which seems to be consistent with the rave observations @xcite . in the bulge ( lower panel ) , star formation takes place more quickly , and thus the chemical enrichment timescale is much shorter than in the disk . the age - metallicity relation shows a more rapid increase than in the disk . the maximum metallicity reaches super solar ( [ fe / h ] @xmath121 ) at @xmath118 gyr . although the sfr becomes small after @xmath79 gyr , a few stars form at @xmath122 gyr . these have super - solar metallicity in general and the average metallicity does not show time evolution . the difference in the chemical enrichment timescales results in a difference in the elemental abundance ratios , since different elements are produced by different supernovae with different timescales . the best known clock is the @xmath1-elements to iron ratio [ @xmath1/fe ] since sne ia produce more iron than @xmath1 elements with longer timescales than sne ii . nevertheless it should be noted that low - mass sne ii also provide relatively low [ @xmath1/fe ] because of their smaller envelope mass compared to more massive sne ii . this mass dependence is important in a system with a low sfr such as dwarf spheroidal galaxies . figure [ fig : afecumnum ] shows the cumulative functions of [ o / fe ] of stars in each component in the present - day galaxy . in the solar neighborhood ( solid line ) , 80% of stars have [ o / fe ] @xmath123 , which indicates a significant contribution from sne ia . in the bulge ( dashed line ) and halo ( dotted line ) , the chemical enrichment timescale is so short that 60% and 80% of stars have [ o / fe ] @xmath124 , respectively , although there is a significant number of low [ @xmath1/fe ] stars . the majority of low [ @xmath1/fe ] stars also have high [ mn / fe ] because of the sn ia contribution . they are formed at @xmath125 gyr and have non - rotating ( @xmath126 ) kinematics . a small fraction of low [ @xmath1/fe ] stars , however , have low [ mn / fe ] and are formed at @xmath127 gyr . they represent the effect of low - mass sne ii in the inhomogeneous enrichment ( the third reason in [ sec : inhomo ] ) . such low [ @xmath1/fe ] stars seem to be observed in segue @xcite . the fraction of these stars is important for discussions of the formation history of the galactic halo . in order to distinguish these two enrichment sources , elemental abundance ratios of iron - peak elements such as mn are critically important . figure [ fig : afe ] shows the [ o / fe]-[fe / h ] relations , and the other @xmath1-elements show the same trends . in the solar neighborhood ( upper panel ) , at the beginning , only sne ii contribute , and [ @xmath1/fe ] shows a plateau ( [ @xmath1/fe]@xmath128 ) . around [ fe / h ] @xmath35 , sne ia start to occur , which produce more iron than @xmath1 elements . this delayed enrichment of sne ia causes the decrease in [ @xmath1/fe ] with increasing [ fe / h ] . this trend is in great agreement with the observations ( dots ) . theoretically , the decreasing point of [ @xmath1/fe ] has been used to put a constraint on the lifetimes of sne ia . @xcite and @xcite showed that [ @xmath1/fe ] decreases too early in the double - degenerate scenario and in the single degenerate without metallicity effect since the average lifetime is too short . the slope of @xmath129[@xmath1/fe]@xmath130[fe / h ] depends on the binary parameters in our sn ia model , but we should note that it is consistent with the observed supernova rates in various types of galaxies ( [ sec : chem ] ) . a significant scatter is seen in figure [ fig : afe ] , and more clearly , the distribution functions of [ o / fe ] at given [ fe / h ] in the solar neighborhood are shown in figure [ fig : afenum2 ] . at [ fe / h ] @xmath112 , [ o / fe ] is almost constant , and slightly larger for higher metallicity ( the peak [ o / fe ] @xmath131 , and @xmath132 at [ fe / h ] @xmath133 , and @xmath134 , respectively ) because of the mass dependence of the yields of sne ii . the scatter of [ o / fe ] is caused by the inhomogeneity of chemical enrichment in our chemodynamical model . at [ fe / h ] @xmath40 , the majority of [ o / fe ] is lower than @xmath135 and the peak [ o / fe ] is @xmath136 . around [ fe / h ] @xmath35 , the scatter looks a bit larger than observed ( fig . [ fig : afe ] ) . this may be because the mixing of heavy elements among gas particles is not included in our model ( [ sec : model ] ) . at [ fe / h ] @xmath112 , the scatter is caused from the following two reasons . first , although sne ia can not occur at [ fe / h ] @xmath137 in our sn ia progenitor scenario , sne ia can contribute in the abundances of stars at [ fe / h ] @xmath112 . as discussed in [ sec : inhomo ] , in the inhomogeneous enrichment , it is natural that the enrichment sources have [ fe / h ] @xmath138 but the stars forming from the ejecta have [ fe / h ] @xmath112 because of the large amount of hydrogen in protostellar clouds . in this case , [ mn / fe ] is as high as @xmath117 ( see [ sec : xfe ] for more discussion ) . secondly , there is an intrinsic variation in [ @xmath1/fe ] of supernova yields depending on the progenitor mass and metallicity , of which effect is included in our chemodynamical model . low - mass sne ii ( @xmath139 ) provide smaller [ @xmath1/fe ] than massive sne ii . in this case , [ mn / fe ] is lower than @xmath140 . the [ @xmath1/fe ] scatter in the simulation could be larger , since there is also a variation depending on the explosion energy and the remnant mass ( neutron star and blackhole ) for sne ii and hne . from the comparison of the [ @xmath1/fe ] scatter between simulations and observations , it is possible to put constraints on the unsolved physics of supernova explosion and mixing of the ism . figure [ fig : afenum05 ] shows the distribution functions of [ o / fe ] for stars in the solar neighborhood at @xmath141 [ fe / h ] @xmath142 for the standard model ( solid line ) and the models with a longer star formation timescale ( @xmath10 , dashed line ) , a smaller feedback radius ( @xmath143 kpc , dotted line ) , and a fixed feedback number ( @xmath144 , dot - dashed line ) . the smaller initial condition ( tab . 1 ) is used here . the standard model gives much better agreement with the peak value of the observations ( arrow ) than the other models . using stochastic chemical evolution model , @xcite claimed that the scatter of [ @xmath1/fe ] caused by supernova yields is too large compared to observations and an efficient mixing process is required . this is because the nucleosynthesis yields adopted in that paper @xcite were different from currently accepted values . the low [ @xmath1/fe ] values resulted from the larger amount of iron assumed for low - mass sne ii ( @xmath145 for @xmath146 ) . now the iron mass is estimated from the light curve and spectral fitting , and is @xmath147 for sne ii with @xmath148 . the high [ @xmath1/fe ] values are reduced by including hypernovae . figure [ fig : afenum19 ] shows the distribution functions at @xmath149 [ fe / h ] @xmath150 for the models with ( solid line ) and without ( dashed line ) hypernova feedback using the smaller initial condition . with hypernovae , the scatter around [ o / fe ] @xmath128 is as small as the observations @xcite . in the bulge ( lower panel of fig . [ fig : afe ] ) , the [ @xmath1/fe]-[fe / h ] relation is very much different . the chemical enrichment timescale is so short that the metallicity reaches super solar before sne ia contribute . thus , the [ @xmath1/fe ] plateau continues to [ fe / h ] @xmath151 . this is roughly consistent with the observations @xcite , except for the o observation in ( * ? ? ? * see [ sec : xfe ] for more discussion ) . in this simulated galaxy , some new stars are still forming in the bulge , which in general have super solar metallicity and low [ @xmath1/fe ] because of the large contribution from sne ia . although a small fraction of stars with the age of @xmath152 gyr ( @xmath153 gyr ) do have high [ @xmath1/fe ] ( [ o / fe ] @xmath154 ) , which is caused by the inhomogeneous enrichment , stars younger than @xmath155 gyr have @xmath141 [ o / fe ] @xmath156 and @xmath157 [ fe / h ] @xmath158 . @xcite , however , showed several young metal - rich stars with high [ @xmath1/fe ] in the galactic center , which suggests additional physical processes such as ( i ) local enrichment by the stellar winds of dying stars or sne ii , ( ii ) intrinsic variation of supernovae depending on energy and mixing , or ( iii ) the selective mass - loss of the gas enriched by sne ia . the bulge wind might be driven by sne ia , and most of fe enriched gas might be selectively blown away before these young stars are formed . if the observed scatter in [ o / h ] , [ o / fe ] , and [ ca / fe ] are real , then the scenarios ( i ) and ( ii ) may be preferable . the metallicity distribution function ( mdf ) of g - dwarf stars is a stringent constraint on chemodynamical models . figure [ fig : mdf ] shows the present mdf for three different models with the smaller initial condition . the model with the feedback from the uv background radiation ( uvb ) and hypernova ( solid line ) gives much better agreement with the observations in the solar neighborhood . however , the number of metal - poor stars is still larger than observed , in other words , there is the g - dwarf problem . in one - zone chemical evolution models , this problem is easily solved by assuming a long timescale of gas infall ( e.g. , * ? ? ? * ) , but we can not control the infall by hand in chemodynamical simulations . the mass accretion timescale is mainly determined from the hierarchical clustering of dark halos . although hypernova feedback could help in generating outflow , the mass accretion history is not affected very much . in our models , 77% of dark matter ( 45% of total mass ) is accreted inside 10 kpc in the first 1 gyr , then the rest is gradually accreted . there is a room to change the star formation and chemical enrichment timescales . with the uvb , the initial starburst is suppressed by a factor of @xmath43 , which results in the longer duration of star formation . with hne , the sfr is a little suppressed at @xmath159 gyr , and the metal ejection is significantly changed . figure [ fig : mdf2 ] shows the present mdf in the different components of the simulated galaxy for the best model with the wider initial condition . in the solar neighborhood ( solid line ) , the mdf is roughly consistent with the previous observations @xcite . however , the geneva - copenhagen survey showed a narrower mdf ( @xcite , updated from @xcite ) . if this is valid , the g - dwarf problem becomes more serious . in the bulge ( dashed line ) , @xcite and @xcite showed a broad mdf , from which @xcite discussed the metal - rich population in the bulge . also in our simulations , the metal - rich peak at [ fe / h ] @xmath160 is caused by the delayed enrichment of sne ia . however , @xcite showed a narrower mdf with a large spectroscopic sample . there seems to be the g - dwarf problem also for the galactic bulge . the lack of metal - rich tail in the mdf may suggest that the star formation is truncated in the bulge , possibly by an sne ia driven bulge wind . the other solutions of the g - dwarf problem are , as summarized in @xcite , pre - enrichment and/or variable imf . if the gas is enriched by the first stars before it accretes onto the galaxy , the number of metal - poor stars can be small . a bulge wind driven by sne ia could also work as pre - enrichment in the disk . if the imf at low - metallicity is top - heavy as suggested by the simulations of primordial star formation @xcite , larger amounts of metals and heating sources are provided . if no low - mass stars can be formed at low - metallicity , the number of metal - poor g - dwarf stars obviously becomes zero . we should note , however , that the observed abundance patterns of the extremely metal - poor ( emp ) stars rule out large contributions from pair instability supernovae that result from @xmath161 stars . stars with @xmath162 and @xmath163 are usually supposed to collapse to blackholes , but may be able to explode as core - collapse supernovae , which do not conflict with the observations @xcite . or , simply , simulations with higher resolution may solve the g - dwarf problem . finally , using our chemodynamical simulation , we predict the frequency distributions of the elements from o to zn as a function of time and location . figures [ fig : xfe]-[fig : xfe - thick ] show the mass density of stars in the [ x / fe]-[fe / h ] diagrams at present . these results can be statistically compared with the next generation of observations from high resolution and multi - object spectrographs such as hermes . these instruments enable us not only to study the formation and evolution history of the galaxy but also to update stellar evolution and supernova physics . figure [ fig : xfe ] is for the solar neighborhood . the dots are for the observations of individual stars , and they almost perfectly overlap with the contours of our simulation . in order to minimize the systematic error among various methods of observational data analysis , we consider only data from ( * ? ? * hereafter r03 ) , @xcite , @xcite at [ fe / h ] @xmath112 , @xcite at @xmath164 [ fe / h ] @xmath112 , and ( * ? ? ? * hereafter c04 ) and ( * ? ? ? * hereafter h04 ) at [ fe / h ] @xmath165 . as in k06 , the 3d corrections of @xmath166 for o in c04 , and the nlte corrections of @xmath167 and @xmath168 , respectively , for mg and al in h04 are applied . for mg , na , al , and k , c04 s data are updated by ( * ? ? ? * , hereafter a10 ) with the nlte corrections . for o , s , cu , and zn , we employ other data sources as noted below . * @xmath1 elements o , mg , si , s , and ca show the same relation ; a plateau at [ fe / h ] @xmath112 , and then a decrease due to the delayed enrichment of sne ia ( see kn09 for the sn ia progenitor models ) . for o and s , an increasing trend toward lower [ fe / h ] reported by @xcite and @xcite , respectively , is affected by the nlte and 3d effects , and is not seen in our simulation . for o , there is an offset between the lines used in the abundance analysis . to compare with the observations at low metallicity , we plot @xcite s data with the forbidden line [ oi ] 6300 since an empirical relation using [ oi ] and oi triplet at 7775 is used in r03 . for s , we plot data from @xcite and @xcite at low metallicity , and @xcite at high metallicity , where more lines are used than in r03 . in r03 , the scatter of [ s / fe ] is large and the decreasing trend toward low metallicity is not clearly seen . at [ fe / h ] @xmath40 , [ si / fe ] may be slightly larger , and [ ca / fe ] may be slightly smaller than the observations , which are due to the sn ia yields . * odd - z elements na , al , and cu show a decreasing trend toward lower metallicity since the nucleosynthesis yields of these elements strongly depend on the metallicity of progenitor stars ( see k06 for the details ) . na and al also show a decreasing trend toward higher metallicity due to sne ia , which is shallower than that of @xmath1 elements . [ na / fe ] at [ fe / h ] @xmath40 may be slightly larger than observed . the large scatter is caused by the metallicity dependence of yields . as discussed in [ sec : inhomo ] , in our chemodynamical simulations , there is only a week correlation between the metallicities of supernova progenitors and those of the surrounding ism . if the ejected metals are diluted , the ism metallicity and the metallicity of stars that form from the ism become lower than the progenitor metallicity . the scatter starts increasing at [ fe / h ] @xmath169 in this simulation , which looks consistent with the observations . observations of an unbiased sample with the nlte analysis will put a constraint on the mixing process in the chemodynamical simulations . * iron - peak elements the scatter of cr and co is very small since there is not very much variety in the sn ii yields and [ ( cr , co)/fe ] is almost @xmath117 for sne ia . for cr , since a dependence on the temperature of the observed stars has been reported , we plot the cr ii abundance from h04 for low metallicity , which is consistent with our simulation . for co , an increasing trend toward lower metallicity is seen in the observations , which is not realized in the simulation . this is due to the input of supernova yields , and can not be solved with different star formation histories or initial conditions . this could be solved if we adopt higher energy and/or higher fraction of hne where more co and zn are synthesized . * zinc zn is one of the most important elements for supernova physics . [ zn / fe ] is about @xmath117 for a wide range of metallicities , which can only be generated by such large contribution of hne . in detail , there is a small oscillating trend ; [ zn / fe ] is @xmath140 at [ fe / h ] @xmath117 , increases to be @xmath135 at [ fe / h ] @xmath170 , decreases to be @xmath140 at [ fe / h ] @xmath169 , then increases toward lower metallicity . this is characteristic in our sn ia model ( kn09 ) , and is consistent with the observations as found by @xcite . theoretically , zn production depends on many parameters ; @xmath171zn is synthesized in the deepest region of hne , while neutron - rich isotopes of zinc @xmath29zn are produced by neutron - capture processes , which are larger for higher metallicity massive sne ii . the scatter at @xmath172 [ fe / h ] @xmath173 is larger than observed , which is caused by the dependence of the zn yield on the metallicity and mass . the increasing trend toward lower metallicity in the observations @xcite could be generated with the same effects as for co. * manganese mn is a characteristic element and is more produced by sne ia relative to iron . from [ fe / h ] @xmath35 , [ mn / fe ] shows an increasing trend toward higher metallicity , which is caused by the delayed enrichment of sne ia . mn is also an odd - z element , and the mn yield depends on the metallicity both for sne ii and sne ia . @xcite showed a bit steeper slope at [ fe / h ] @xmath174 , which would be generated by the metallicity dependence of sn ia yields . at [ fe / h ] @xmath165 , the scatter in the simulation is a bit smaller than observed . since sne ia do not contribute at such low metallicity , the [ mn / fe ] scatter , if it is real , should stem from the sn ii yields depending on the remnant mass and the electron fraction @xmath175 that is related to the mechanism of supernova explosion . in the supernova ejecta , mn and cr are synthesized in the relatively outer regions compared to cr , zn and the majority of fe . therefore , relatively high [ ( mn , cr)/fe ] and low [ ( co , zn)/fe ] are obtained with faint sne with relatively large remnant masses ( blackholes ; @xcite ) . in our figures , we take the solar abundance from @xcite , which is different for some elements from the solar abundance updated by @xcite . this means the normalization should be changed both for observations and simulations . when the observational data are analyzed with 3d and nlte effects , the agreement in figures [ fig : xfe]-[fig : xfe - thick ] may be or may not be modified . for example , one of the nlte models shows [ mn / fe ] @xmath117 , [ cr / fe ] @xmath117 , and steeper increase in [ co / fe ] toward lower metallicity ( @xcite , 2010 , @xcite ) . figure [ fig : xfe - bulge ] is for the bulge , where the observational data are taken from @xcite and @xcite . as shown in fig . [ fig : afe ] , the @xmath1-element plateau continues to [ fe / h ] @xmath176 , then the second component with a mild decreasing trend are generated at @xmath177 [ @xmath1/fe ] @xmath178 from the sn ia contribution . very similar trends are seen for o , mg , si , s , and ca . this is roughly consistent with the observations , although the double peak distributions are not seen in the observations . for o , @xcite showed a steeper decrease than mg at [ fe / h ] @xmath179 . theoretically , it is not easy to produce such a different trend of o from mg since these two elements are synthesized in the similar region and ejected to the ism in a similar way . the contribution from agb stars is negligible . such a decrease in o / mg might require some additional effects that are not included in our stellar evolution models such as strong stellar winds or a process that causes the change in the c / o ratio . for other @xmath1 elements , explosion energy and remnant mass may cause a variation in ( o , mg)/(si , s ) and ( o , mg)/ca , respectively . these should be discussed with future observations with a larger sample and with nucleosynthesis yields that cover the parameter spaces . the separation between the two populations also depends on the imf , and the salpeter imf is adopted in this simulation . the imf will be constrained from the elemental abundance ratios . the odd - z elements and also zn show the increasing component from [ fe / h ] @xmath111 to @xmath117 , and the second flat component without any trend at [ ( na , al , cu , zn)/fe ] @xmath117 at @xmath180 [ fe / h ] @xmath181 . different from fig . [ fig : xfe ] , the metallicity dependence of na , al , and cu yields are not seen very much in the bulge since the chemical enrichment timescale is short . [ mn / fe ] shows the increasing component from [ fe / h ] @xmath169 to @xmath117 , which continues to the second flat component at [ mn / fe ] @xmath117 at @xmath180 [ fe / h ] @xmath181 . this is also due to the smaller contribution from sne ia in the bulge . the observations also show the high [ na / fe ] and [ al / fe ] at [ fe / h ] @xmath40 , and the increasing trend of [ mn / fe ] , although the double peak distributions are not visible . figure [ fig : xfe - thick ] is for the thick disk , where the observational data are taken from @xcite , @xcite , @xcite , and @xcite . when we select the thick disk stars from kinematics in the solar neighborhood , the [ @xmath1/fe ] plateau is dominant , which is similar to the bulge . this is also consistent with the rave observations @xcite . we should note , however , that the split trends observed in the solar neighborhood for the [ @xmath1/fe ] trends associated with the thick and thin disks are not clearly seen in the predicted distributions . the formation timescale of the thick disk is @xmath64 gyr in our simulation , which is shorter than the thin disk but longer than the bulge . in figures [ fig : agecumnum ] , [ fig : afecumnum ] , and [ fig : mdf2 ] , the functions of the thick disk are also shown ( dot - dashed lines ) . for thick disks stars , the age is older and [ o / fe ] is higher than thin disk stars , and the peak metallicity is 0.2 dex lower than the thin disk . figure [ fig : xfenum ] shows the distribution functions of [ x / fe ] in the solar neighborhood ( solid line ) , bulge ( dashed line ) , and thick disk ( dotted line ) . @xmath1 elements show the double peak distributions , and the [ @xmath1/fe]-rich peak is at the same [ @xmath1/fe ] for the solar neighborhood , bulge , and thick disk . on the other hand , na , al , cu , and zn show the single peak distributions in the solar neighborhood , but the double peak distributions in the bulge . on the average , na , al , cu , and zn are much more enhanced in the bulge , and ( na , al , cu , zn)/@xmath1 is larger than in the solar neighborhood . in the thick disk , the [ ( na , al , cu , zn)/fe ] distributions are rather similar to the solar neighborhood because the metallicity is not as high as in the bulge . the stellar population of the thick disk is neither thin - disk like nor bulge like ( see tab . [ tab : summary ] ) . for the thick disk stars , the star formation timescale is as short as in the bulge , but the chemical enrichment efficiency is not as high as in the bulge . this is because half of the thick disk stars have already formed in satellite galaxies before they accrete onto the disk , and the metals have been ejected from the satellite galaxies by the galactic winds . in figure [ fig : mn ] , for the solar neighborhood of the present - day galaxy , [ mn / fe ] is plotted against [ @xmath1/fe]=([o / fe]+[mg / fe])/2 , which clearly shows the sequence of the sn ia contribution . with sne ii and hne only , [ @xmath1/fe ] is as high as @xmath128 , and [ mn / fe ] is as low as @xmath170 . with more sne ia , [ @xmath1/fe ] decreases , while [ mn / fe ] increases . the three populations of the observed stars are along this trend ; i ) the emp stars ( o in c04 , large open circles ; mg in h04 , filled pentagons ) are found in the left - bottom region with high [ @xmath1/fe ] and low [ mn / fe ] . ii ) the thick disk stars ( small open circles ) populate the following region , [ @xmath1/fe ] @xmath182 and [ mn / fe ] @xmath183 to @xmath184 . iii ) the thin disks stars ( small closed circles ) occur at [ @xmath1/fe ] @xmath185 and [ mn / fe ] @xmath186 , which are formed from the ism largely enriched by sne ia . in other words , it is possible to select thick disk stars only from the elemental abundance ratios . the scatter should come from the variety in the sn ii yields . there is a larger scatter in h04 than c04 , which may be partially caused by the observational error ; the right - bottom star cs22952 - 015 is ( -0.1 , -0.34 ) in h06 , but ( 0.36 , -0.33 ) in c04 and a10 . on the other side , the left - top star bs16085 - 050 is at ( 0.7 , -0.01 ) , which has to be checked . if it is real , the enrichment source may be faint sne . lccccc solar neighborhood & young & low & @xmath117 & @xmath117 & @xmath117 + bulge & old & high & high & @xmath86 & low + thick disk & old & high & @xmath117 & @xmath117 & low + we present the chemodynamical simulations of a milky way - type galaxy using a self - consistent hydrodynamical code with supernova feedback and chemical enrichment . in our nucleosynthesis yields of core - collapse supernovae , the light curve and spectra fitting of individual supernova are used to estimate the mass of the progenitor , explosion energy , and ejected iron mass . a large contribution from hypernovae is required from the observed abundance of zn ( [ zn / fe ] @xmath4 ) especially at [ fe / h ] @xmath112 . in our progenitor model of sne ia , based on the single degenerate scenario , the sn ia lifetime distribution spans a range of @xmath187 gyr with the double peaks at @xmath185 and @xmath155 gyr . because of the metallicity effect of white dwarf winds , the sn ia rate is very small at [ fe / h ] @xmath112 , which plays an important role in chemical evolution of galaxies . in the simulated galaxy , the kinematical and chemical properties of the bulge , disk , and halo are consistent with the observations . the bulge formed from the assembly of subgalaxies at @xmath73 ; 80% of bulge stars are older than @xmath188 gyr , and 60% have [ o / fe ] @xmath124 . the disk formed with constant star formation over @xmath2 gyr ; 50% of solar - neighborhood stars are younger than @xmath189 gyr , 80% have [ o / fe ] @xmath123 . when we define the thick disk from kinematics , the thick disk stars tend to be older and have higher [ @xmath1/fe ] than the thin disk stars . the formation timescale of the thick disk is @xmath64 gyr . because the star formation history is different for different components , the age - metallicity relation and the metallicity distribution function are also different . the age - metallicity relation shows a more rapid increase in the bulge than in the disk . in both cases , the average metallicity does not show strong evolution at @xmath119 gyr , as in the observations . the scatter is originated from the inhomogeneity of chemical enrichment in our chemodynamical model . the observed metallicity distribution function is better reproduced with uv background radiation and hypernovae , but is still problematic . the bulge wind induced by sne ia seems to be a good solution to reduce the numbers of metal - rich stars in the bulge and of metal - poor stars in the disk . the difference in the chemical enrichment timescales results in the difference in the elemental abundance ratios , since different elements are produced by different supernovae with different timescales . we also predict the frequency distribution of elemental abundance ratios as functions of time and location , which will be statistically compared with a large homogeneous sample from galactic archeology surveys such as hermes , when they become available . * because of the delayed enrichment of sne ia , @xmath1 elements ( o , mg , si , s , and ca ) show a plateau at [ fe / h ] @xmath35 , and then the decreasing trend against [ fe / h ] , where [ mn / fe ] also shows the increasing trend . odd - z elements ( na , al , and cu ) show the increasing trend at [ fe / h ] @xmath112 because of the metallicity dependence of nucleosynthesis yields . these are in excellent agreement with the available observations . * in the bulge , the star formation timescale is so short that the [ @xmath1/fe ] plateau continues to [ fe / h ] @xmath151 . because of the smaller contribution from sne ia , the majority of stars shows high [ @xmath1/fe ] and low [ mn / fe ] . [ ( na , al , cu , zn)/fe ] are also high because of the high metallicity in the bulge . * the stellar population of the thick disk is neither disk - like nor bulge - like as summarized in table [ tab : summary ] . for thick disk stars , [ @xmath1/fe ] is higher , and [ mn / fe ] is lower than thin disk stars because of the short formation timescale . however , [ ( na , al , cu , zn)/fe ] are lower than bulge stars because of the lower chemical enrichment efficiency . this is because half of the thick disk stars have already formed in satellite galaxies before they accrete onto the disk , and the metals have been ejected from the satellite galaxies by the galactic winds . we thank the national astronomical observatory of japan for the grape system , where most of the simulations in this paper are performed . we also thank k. nomoto , m. mori , k. c. freeman , g. da costa , and d. yong for fruitful discussions . andrievsky , s. m. , spite , m. , korotin , s. a. , spite , f. , bonifacio , p. , cayrel , r. , franois , p. , & hill , v. 2010 , , 509 , 88 andrievsky , s. m. , spite , m. , korotin , s. a. , spite , f. , bonifacio , p. , cayrel , r. , hill , v. , & franois , p. 2007 , , 464 , 1081 andrievsky , s. m. , spite , m. , korotin , s. a. , spite , f. , bonifacio , p. , cayrel , r. , hill , v. , & franois , p. 2008 , , 481 , 481 argast , d. , samland , m. , thielemann , f .- k . , & gerhard , o. e. , 2002 , , 388 , 842 [ x / fe]-[fe / h ] relations in the solar neighborhood at @xmath66 . the contours show the frequency distribution of stars in the simulated galaxies , where red is for the highest frequency . the observational data ( white dots ) are taken from @xcite , large open circles ; @xcite , filled pentagons ; @xcite , crosses ; @xcite and @xcite for thin ( small filled circles ) and thick ( small open circles ) disk stars . for o , @xcite for thin ( filled triangles ) and thick ( open triangles ) disk stars . for si , s , and zn , @xcite , filled squares ; @xcite , filled pentagon ; @xcite , open squares . for cu and zn , @xcite , asterisks . , width=434 ] distribution functions of the elemental abundance ratios relative to iron [ x / fe ] in the solar neighborhood ( solid line ) , bulge ( dashed line ) , and thick disk ( dot - dashed line ) at @xmath66 . , width=434 ]

the mass - loss rate of donor stars in cataclysmic variables ( cvs ) is of paramount importance in the evolution of short - period cvs . observed donors are oversized in comparison with those of isolated single stars of the same mass , which is thought to be a consequence of the mass loss . using the empirical mass - radius relation of cvs and the homologous approximation for changes in effective temperature , orbital period , and luminosity of the donor with the stellar radius , we find the semi - empirical mass - loss rate of cvs as a function of . the derived is at and depends weakly on when min , while it declines very rapidly towards the minimum period when min , emulating the relation . due to strong deviation from thermal equilibrium caused by the mass loss , the semi - empirical is significantly different from , and has a less - pronounced turnaround behavior with than suggested by previous numerical models . the semi - empirical relation is consistent with the angular momentum loss due to gravitational wave emission , and strongly suggests that cv secondaries with are less than 2 gyrs old . when applied to selected eclipsing cvs , our semi - empirical mass - loss rates are in good agreement with the accretion rates derived from the effective temperatures of white dwarfs , suggesting that can be used to reliably infer from . based on the semi - empirical , sdss 1501 and 1433 systems that were previously identified as post - bounce cvs have yet to reach the minimal period .

1008.0955

cataclysmic variables ( cvs ) are semi - detached binaries with a white dwarf ( wd ) primary and a low - mass companion which transfers mass to a wd through roche - lobe overflow . the observed orbital period distribution of cvs has three important features : a sharp short period cut - off at the orbital period of @xmath10 min , referred to as the `` minimum period '' , the period gap , a dearth of systems with periods between @xmath11 min and @xmath12 min @xcite , and the `` period minimum spike '' , a significant accumulation of systems at @xmath13@xmath14 min @xcite . for cvs with period above @xmath15 , the dominant angular momentum loss mechanism is thought to be the magnetic braking , while the gravitational radiation may be responsible for the angular momentum loss for cvs below @xmath16 . in this work , we focus on cvs only with @xmath17 . a characteristic feature of cvs is the mass transfer from a secondary to a wd . since the gas stream carries substantial angular momentum , it forms an accretion disk around the wd . the reaction of a cv on the mass transfer can be qualitatively understood with the help of the dimensionless parameter @xmath18 where @xmath19 and @xmath20 denote the mass - loss and kelvin - helmholtz timescales of the donor , respectively @xcite . at the initial phase of mass transfer , @xmath21 , so that the donor remains close to thermal equilibrium and thus follows the mass - radius relation ( mrr ) of main - sequence stars @xmath22 with the effective mass - radius exponent @xmath23 close to @xmath24 @xcite . here , the subscript `` 2 '' denotes the quantities of cv donors . both @xmath25 and @xmath1 of the cv decrease as the donor keeps losing mass during this early phase of cv evolution . as the mass transfer continues , the binary shrinks due to the angular momentum loss , reducing @xmath26 . when @xmath19 and @xmath20 become comparable to each other ( @xmath27 ) , the donor becomes out of thermal equilibrium . since the surface luminosity is no longer in balance with the nuclear luminosity , the donor expands , causing @xmath23 to decrease with time @xcite . using the hayashi theory and homologous approximation , @xcite studied the reaction of the donor to mass loss and found that @xmath25 increases with decreasing @xmath26 . observations confirm this theoretical prediction , showing that the radii of donors in short - period cvs are @xmath28@xmath29 larger than those of isolated main - sequence stars @xcite . the effective mass - radius exponent @xmath23 becomes smaller as @xmath30 decreases towards the brawn dwarfs masses . the change in the mrr of donors slows down the temporal change in the orbital period as the donor continues losing mass . the orbital period reaches a minimum value when @xmath31 at which @xmath32 @xcite . as @xmath26 decreases further , @xmath23 declines to @xmath33 @xcite . this in turn leads to an increase in @xmath1 . such a turnaround of the orbital period at @xmath34 is referred to as the `` period bounce '' . the evolutionary scenario described above suggests that the mass - loss rates @xmath2 ( and the related thermal relaxation ) of donors control the evolution of cvs . it also implies that the observed distribution of cvs versus the orbital period may be a consequence of a certain relation between @xmath2 and @xmath1 . since @xmath35 not only provides a way to estimate the ages of cvs but also can be used to check the dominant mechanism for angular momentum loss , finding a reliable @xmath1@xmath2 relationship is of great importance in understanding cv evolution . there have been numerous studies to derive the @xmath1@xmath2 relation ; these can be categorized into two groups : purely theoretical work ( e.g. , @xcite ) and semi - empirical work ( e.g. , @xcite ) . the first , theoretical studies made use of stellar evolution codes combined with the analytical mass - transfer models of @xcite or others . the common results of these numerical experiments are that ( 1 ) the orbital period exhibits the turnaround behavior at @xmath34 that is about @xmath36 smaller than the observed @xmath37 @xcite and ( 2 ) compared to isolated stars , the secondaries in short - period cvs are oversized due to the mass loss but only by @xmath28@xmath38 , which is about a factor @xmath39@xmath40 smaller than the observed cv sizes . since the bloating of the donor depends on the efficiency of the mass loss , these results suggest either the numerical models underestimate @xmath2 above @xmath34 , or the efficiency of the angular momentum loss is higher than expected @xcite . despite the attempts to include the effects of tidal and rotational perturbations @xcite , additional angular momentum loss via magnetic braking @xcite , accretion disk winds and magnetic propeller ( see @xcite for review ) , etc . , clear explanations for the discrepancies between @xmath34 and @xmath37 and between the empirical and numerical mrrs have yet to be found . the second , semi - empirical approach uses , for instance , the effective temperatures of wds @xcite or accretion disk luminosities @xcite to infer the accretion rates @xmath41 . the physical basis for the former is that wds in cvs are observed to be hotter than expected from their ages @xcite , and such overheating " of wds may be a result of the compressional heating due to the gas piled up on their surfaces @xcite . for the latter method , the accretion luminosities can give direct information on @xmath41 if the disk properties are well constrained . when applied to dwarf novae ( dne ) that accrete sporadically and have thermally unstable disks , @xmath41 calculated from the two methods have a different scatter for given @xmath1 . the mass - transfer rates based on the accretion luminosities probably measure an accretion rate averaged over short time intervals , e.g. , a few decades , which makes them quite sensitive to short - time disk variability @xcite . this inevitably leads to a high dispersion in @xmath41 estimated from the disk luminosities . on the other hand , the effective temperatures of wds are likely to give a relatively long - term averaged @xmath41 , although this method may suffer from uncertainties related to unknown properties such as masses and radii of wds as well as some auxiliary assumptions on their structures and boundary layers @xcite . in this paper we construct another semi - empirical relation between the mass - loss rate and the orbital period , and compare the result with those of the purely theoretical and the semi - empirical approaches mentioned above . we start from the empirical mrr of @xcite for superhumping cvs , and apply the homologous approximation in order to calculate the dependencies of the effective temperature and gravothermal luminosity upon the bloating of the donor . assuming that the observed mrr of cv donors is a consequence of the mass loss and thermal relaxation processes , we calculate the mass - loss rates . we will show that the resulting mass - loss rates are broadly in line with those on the basis of the effective temperatures of wds , while different from those of the purely theoretical models of @xcite . we will also show that our results are consistent with the angular momentum loss of cvs due primarily to the emission of gravitational waves . since radii of cv donors likely trace the long - term evolution of @xmath2 @xcite and depend only weakly on details of the mass - transfer processes , the resulting mass - loss rates from the empirical mrr are less model - dependent and less sensitive to the short - term effects than the other semi - empirical estimates that rely on accretion activities . the organization of this paper is as follows . in [ sec_p_te ] , we introduce the theoretical mrr for isolated single stars and the empirical mrr for cvs donors , and define the bloating factor . we then describe a method to find the effective temperature and gravothermal luminosity of donors by using the bloating factor . in [ sec_dm ] , we calculate the semi - empirical mass - loss rate , construct the mass - loss history to estimate the cv ages , and show that the derived mass - loss rate is consistent with cv evolution by angular momentum loss via the emission of gravitational waves . in [ sec_dis ] , we apply our results to selected eclipsing short - period cvs and show that the semi - empirical mass - loss rate is consistent with the accretion rates inferred from the effective temperatures of wds . we also suggest a way to estimate the effective temperatures of donors assuming that mass is conserved . finally , we summarize our results in [ sec_con ] . we consider a cv consisting of a wd ( or accretor ) with mass @xmath42 and a secondary ( or donor ) with mass @xmath30 , radius @xmath25 , and effective temperature @xmath0 . we assume that the donor fills its critical equipotential surface . in accordance with the results of @xcite and @xcite , we assume @xmath43 throughout this paper . we use an additional subscript `` 0 '' to denote the quantities of isolated , single stars in hydrostatic and thermal equilibrium to which cv secondaries will be compared . in this section , we use the empirical relationship between masses and radii of cv secondaries to calculate the bloating factor in comparison with the single stars of the same mass . we then treat the bloating factor as perturbations to the properties of single stars in order to find the orbital period , effective temperature , and gravothermal luminosity of cvs . as the mrrs of single stars that we use as the references to those of cv secondaries , we take _ composite _ relations constructed in a manner similar to @xcite : the bcah98 isochrone of @xcite for stars with @xmath44 , the dusty isochrone of @xcite for @xmath45 ( @xmath46 k ) , and the cond isochrone of @xcite for @xmath47 . there are quite large uncertainties in the estimation of cv ages . period bouncers are thought of as a final stage of cv evolution , suggesting that they are older than their `` parent '' cvs ( see , e.g. , @xcite ) . although numerical experiments predict that cvs cross the period gap at @xmath48 gyr and reach @xmath34 in @xmath49 gyr after the onset of the mass loss below the gap ( e.g. , @xcite ) , temporal evolution of cvs depends strongly on the mass - loss rate that is not well constrained observationally . to cover a wide range of cv ages , we adopt two composite isochrones with 1-gyr and 5-gyr ages ( hereafter b1 and b5 sequences , respectively ) and 1-gyr isochrone for @xmath50 . the choice of the 1-gy isochrone in @xcite was to represent the numerical results of @xcite . ] . figure [ pic_m_r ] plots the composite @xmath51 of single stars as functions of @xmath30 for the b1 and b5 sequences , as dashed and solid lines , respectively . both sequences give almost the same radii above @xmath52 , while below this point the stellar radii in the b1 sequence are about @xmath36 larger than in the b5 sequence . as the mrr of cv secondaries , we take the relation derived by @xcite from the observed properties of superhumping cvs : @xmath53 this empirical fit assumes @xmath43 and incorporates itself the additional empirical constraints that reproduce the observed locations of the period gap and the period minimum , such that @xmath16 occurs at @xmath54 and @xmath37 at @xmath55 . the intrinsic errors in @xmath25 for this empirical mrr are estimated to be about @xmath40@xmath56 @xcite ; we take allowance for @xmath57 uncertainties in @xmath25 throughout this work . we note a caveat that our choice of the errors may underestimate real values below @xmath52 where only a few systems are observed and thus there is practically no calibrator . in addition , the masses of wds may differ from the assumed mass of @xmath58 , which also can change @xmath25 . equation ( [ eq_kn_mr_fit ] ) is plotted in figure [ pic_m_r ] as dotted line with the shaded region denoting the associated uncertainties in @xmath25 . columns ( 1 ) and ( 2 ) in table [ ta_seq ] give the @xmath30-@xmath25 relation . for comparison , figure [ pic_m_r ] also plots as a dot - dashed line the mrr of cvs with @xmath59 from the numerical models of @xcite , which clearly has smaller radii than the empirical relation , typically by @xmath28@xmath60 . this indicates that the deviation of real donors from thermal equilibrium is larger than that expected from the numerical models . we define the bloating factor @xmath61 as the relative size of a cv secondary to a single star with the same mass . figure [ pic_m_al ] plots @xmath62 as dashed and solid lines for the b1 and b5 sequences , respectively , with the shaded and hatched regions representing the corresponding uncertainties . the discontinuity of @xmath62 at @xmath63 corresponds to the period gap . since @xmath64 ( see below ) , the ratio of the bloating factors above ( @xmath65 ) and below ( @xmath66 ) the gap satisfies @xmath67 , in good agreement with the width of the gap . at the upper edge of the gap , the donors detach from their critical lobes and the mass loss stops . since the donors are thought to be oversized due to the mass loss , they shrink within the gap . when the donors come out the lower edge of the gap , they resume the mass loss and become oversized again . figure [ pic_m_al ] shows that the bloating factor increases with the decreasing donor mass for @xmath68 and then decreases with decreasing @xmath30 below this point . this can be explained qualitatively as follows . since stars with @xmath69 are fully convective and isentropic , they are well described by polytropes with index @xmath70 @xcite . the mrr for these polytropes @xmath71 with @xmath72 being the pressure constant @xcite , then gives @xmath73 the mass loss is capable of driving the donor out of thermal equilibrium . this raises the gravothermal luminosity , @xmath74 , where @xmath75 is the nuclear luminosity of the donor and @xmath76 is the surface luminosity . for fully convective stars , the change of @xmath72 depends on @xmath77 and @xmath20 as @xmath78 @xcite . combining equations ( [ eq_r_m_polytrope ] ) , ( [ eq_alpha_def_k ] ) , and ( [ eq_dk_k ] ) , one obtains @xmath79 where @xmath80 is the mass - radius exponent of unperturbed , singe stars . at the lower edge of the period gap , a donor remains close to thermal equilibrium so that @xmath81 and @xmath82 . in this case , the second term in the right - hand side of equation ( [ eq_da_a ] ) dominates and @xmath83 , indicating that @xmath62 increases as the mass loss continues . when the donor reaches the minimum hydrogen burning mass ( mhbm ) @xmath84 , the nuclear burning stops and donors are born as brown dwarfs , so that @xmath85 . in the brown dwarf regime , on the other hand , the internal pressure is dominated by the degeneracy pressure with @xmath72 depending mainly on the chemical composition @xcite . in this case , @xmath80 tends to @xmath86 and the first term in the right - hand side of equation ( [ eq_da_a ] ) dominates , resulting in @xmath87 . for brown - dwarf donors , the change in the equation of state partially compensates for the increasing tendency of the bloating factor due to the mass loss . in addition to the mass loss , tidal and rotational perturbations can also change the donor size , depending on the polytropic index and the mass ratio . we assume that the applied perturbations do not change the specific entropy , which is reasonable since the stellar distortions occur in dynamical timescales . in appendix [ ap_isentropic ] , we calculate the enlargement factor @xmath88 of a roche - lobe filling donor as a function of @xmath89 for @xmath90 . the resulting @xmath88 for @xmath70 polytropes against @xmath30 is plotted in figure [ pic_m_al ] as a dotted line . note that the expansion of low - mass donors due to the tidal and rotational distortions is @xmath49@xmath91 , insensitive to the mass ratio for @xmath92@xmath93 ( see also , e.g. , @xcite ) . it can be seen that while @xmath88 is , in general , considerably smaller than the observed bloating factor for stars with @xmath94 , it is comparable to @xmath62 when @xmath95 corresponding to @xmath16 or when @xmath96 as donors approach @xmath37 . this indicates that the effect of the isentropic expansion of donors on the bloating factor is non - negligible for cvs that just resume mass loss or are below the minimum period . for a given mrr , the orbital period can be determined by the roche lobe geometry and kepler s law as @xmath97 where @xmath98 is the mean density of the donor , @xmath99 is the relative size of the donor to the orbital separation , and @xmath100 is the mass ratio ( e.g. , @xcite ) . equation ( [ eq_mean_den_law ] ) implies that the orbital period depends on the bloating of the donor as @xmath101 clearly , the orbital period increases with the radius of a secondary . since donors below the period gap are fully convective , their luminosities are determined by the conditions at the very outermost layers @xcite , so that one can neglect the influence of the distortions on the central parts of a donor . to obtain the properties of the photospheric layers , we use the standard hayashi theory . for the opacity law of the form @xmath102 with the gas pressure @xmath103 , the ratio of the effective temperatures satisfies @xmath104 where @xmath105 ( e.g , @xcite , see also @xcite ) . for the realistic opacity laws applicable for low - mass main - sequence stars , @xmath106 has a very small value . if the opacity is governed by h@xmath107 , for example , @xmath108 and @xmath109 , leading to @xmath110 . if the molecular absorption instead is a dominant opacity source ( @xmath111 and @xmath112 ) , @xmath113 . therefore , donors in short - period cvs have approximately the same effective temperatures as the corresponding isolated single stars of identical masses ( e.g. , @xcite ) . to construct the semi - empirical @xmath1@xmath0 relations , we first use the theoretical @xmath30@xmath114 relation for isolated stars and calculate @xmath115 from equation ( [ eq_mean_den_law ] ) by fixing @xmath43 . we then use the empirical mrr and the bloating factor of donors to calculate @xmath1 and @xmath0 from equations ( [ eq_p_cor ] ) and ( [ eq_te_cor_kw ] ) . the @xmath115@xmath114 relations are plotted in figure [ pic_te_p ] as thin lines , while the thick lines are for the semi - empirical @xmath1@xmath0 relations . the dashed and solid lines correspond to the b1 and b5 sequences , respectively . again , the shaded and hatched regions give the corresponding uncertainties . columns ( 3 ) , ( 6 ) , and ( 11 ) in table [ ta_seq ] list the values of @xmath1 and @xmath0 against @xmath30 for both sequences . for a given mass , the expansion of a donor increases the orbital period while the effective temperature keeps almost unchanged . the largest difference between @xmath1 and @xmath115 occurs at @xmath116 k , corresponding to @xmath117 where the bloating factor is maximal . the dot - dashed line plots the semi - empirical result of @xcite . because of the difference in the adopted ages of the isochrones , the result of @xcite for @xmath68 ( with @xmath118 k ) and @xmath47 ( with @xmath119 k ) is in agreement with our results for the b5 and b1 sequences , respectively . the donors at @xmath37 have mass @xmath120 , corresponding to @xmath121@xmath122 k and @xmath123@xmath124 k for the b5 and b1 sequences , respectively . the right choice of the stellar age is quite important in evaluating the mass - loss rate near @xmath37 since it depends rather sensitively on @xmath0 as @xmath125 @xcite . combining equation ( [ eq_te_cor_kw ] ) with the stefan - boltzmann law @xmath126 where @xmath127 is stefan - boltzmann constant , one can show that the change in the stellar luminosity is given by @xmath128 although the effective temperature does not change much with the bloating factor , the additional @xmath129 term in the stefan - boltzmann law increases @xmath130 significantly . columns ( 5 ) and ( 10 ) in table [ ta_seq ] give the values of @xmath76 for the b1 and b5 sequences , respectively . the nuclear luminosity @xmath75 is determined by the energy produced in the central parts of a donor , so that @xmath131 , where @xmath132 is the thermonuclear reaction rate given by @xmath133 with @xmath134 and @xmath135 denoting the central temperature and density , respectively . since @xmath136 for a monatomic ideal gas , the change in the nuclear luminosity due to the bloating factor is @xmath137 theoretical models for stellar structure and evolution give @xmath138 and @xmath139 @xcite for @xmath140 k and @xmath141 g @xmath142 , typical for the centers of low - mass stars . since @xmath143 for isolated stars in thermal equilibrium , the gravothermal luminosity satisfies @xmath144 thermonuclear burning stops in cv secondaries once they reach the mhbm @xmath84 . at this point , donors are effectively born as brown dwarfs and the gravitational collapse becomes the dominant source of the stellar radiation . that is , @xmath145 for @xmath146 . figure [ pic_m_lg ] plots @xmath147 versus @xmath30 for the b1 and b5 sequences as dashed and solid lines , respectively , with the uncertainties denoted by the shaded and hatched regions . figures [ pic_m_al ] and [ pic_m_lg ] together imply that at any given mass the contribution of the gravothermal luminosity increases with the bloating factor . if @xmath25 increases by @xmath38 , more than @xmath148 of the donor luminosity is generated by the release of thermal and gravitational energies . finding the theoretical mass - loss rates of cv donors is a daunting task since it involves many complicated processes including thermal and adiabatic relaxations , nuclear evolution , effects of tidal and rotational distortions , orbital angular momentum loss , etc . , all of which may affect the internal structure significantly . although the stationary model of @xcite has been used in various theoretical models of cv evolution , it relies on the parameters describing the donor atmosphere in the vicinity of the inner lagrangian point , which are poorly known . by relating the mass - loss rate with angular momentum loss , on the other hand , the ritter model allows to check whether a proposed angular momentum loss mechanism is indeed responsible for cv evolution . as mentioned in [ sec : intro ] , the numerical calculations that incorporate the stellar evolution and the mass - loss model of @xcite produced mrrs that differ from the empirical mrr . instead of employing the ritter model to calculate the mass - loss rate , we seek the @xmath2 that is required to reproduce the empirical mrr . we make use of the well - known relation between the effective mass - radius exponent @xmath23 of the donor and the mass - loss timescale : @xmath149 @xcite , where @xmath150 is the adiabatic mass - radius exponent for fully convective stars . equation ( [ eq_ze_def ] ) dictates how the mrr should behave in response to the mass - loss and thermal relaxation processes . when the mass transfer just commences , @xmath151 and the donor response is adiabatic , leading to @xmath152 . since the ensuing deviation from the adiabatic evolution depends on how the donor reacts thermally on the mass loss , @xmath23 increases linearly with @xmath26 @xcite . combining equations ( [ eq_mean_den_law ] ) , ( [ eq_st_bl ] ) and ( [ eq_ze_def ] ) with the definitions of the kelvin - helmholtz timescale ( @xmath153 ) ) and mass - loss timescale ( @xmath154 ) , one obtains @xmath155 when @xmath156 corresponding to the period bounce , equation ( [ eq_dm_def ] ) recovers equation ( 34 ) of @xcite for the mass - loss rate at the minimum period . in figure [ pic_dm_p ] we plot the semi - empirical @xmath1@xmath2 relations calculated from ( [ eq_dm_def ] ) for the 1-gyr and 5-gyr isochrones as dashed and solid lines in red color , respectively . the corresponding uncertainties are given by the shaded and hatched regions . columns ( 7 ) and ( 12 ) of table [ ta_seq ] list the values of @xmath2 . the semi - empirical mass - loss rate is essentially non - stationary . for @xmath157 min @xmath158 min , @xmath2 is in the range of @xmath159 @xmath160 and decreases relatively slowly as @xmath161 . below 90 minutes , @xmath2 sharply declines towards @xmath37 as a result of a strong dependence of @xmath2 upon @xmath0 , emulating the @xmath1@xmath0 relation shown in figure [ pic_te_p ] . since the accretion rate decreases with @xmath2 , this rapid decline of @xmath2 is consistent with the observed low accretion activities of cvs within the period spike identified in the sdss samples @xcite . note that the mass - loss rates for the b1 and b5 sequence are almost similar to each other for masses above the mhbm , suggesting that these are perhaps permanent values . in the brown dwarf regime , on the other hand , the ambiguity in the adopted stellar age leads to a large difference in the mass - loss rate , suggesting that it is necessary to use the appropriate isochrones for an accurate estimation of @xmath2 for donors with @xmath162 . for comparison , figure [ pic_dm_p ] also plots as a dot - dashed line the @xmath1@xmath2 relation from the numerical models of @xcite , which shows a pronounced `` boomerang '' shape such that @xmath2 is nearly stationary above @xmath34 with @xmath163 . it also has @xmath2 significantly smaller for @xmath68 than our semi - empirical results . to see what causes the difference between our semi - empirical @xmath1@xmath2 relation and the numerical result of @xcite , we calculate @xmath2 from equation ( [ eq_dm_def ] ) by using the mrr of @xcite together with the 5-gyr isochrone . the resulting @xmath1@xmath2 relation is plotted in figure [ pic_dm_p ] as a dotted line , which is overall in good agreement with the numerical results of @xcite . this suggests that smaller radii of donors in their work are a primary reason for the discrepancies between the results of the current work and @xcite . this also proves to some extent that our approach provides a reasonable way to estimate the mass - loss rate . from the semi - empirical @xmath2 , we calculate @xmath26 for the b1 and b5 sequences , which are listed in columns ( 8) and ( 13 ) of table [ ta_seq ] , respectively . note that donors in cvs below @xmath16 are out of thermal equilibrium with @xmath164 , which is the main cause of the donor expansion . equations ( [ eq_lg_def ] ) and ( [ eq_dm_def ] ) imply that for given @xmath30 and @xmath23 , the mass - loss rate increases more than @xmath165 due mainly to changes in the orbital period and gravothermal luminosity . the resulting semi - empirical @xmath1@xmath2 relation still has a boomerang shape , which is much less dramatic than that of @xcite due to the donor bloating and the associated thermal relaxation . nevertheless , it is very difficult to detect the boomerang shape observationally because of the sharp decline of @xmath0 and @xmath2 below 90 min . once the @xmath1@xmath30 and @xmath1@xmath2 relations are found , it is straightforward to calculate the temporal change of the donor mass by solving @xmath166 . we assume that donors are 1 gyr old and have a fixed initial mass of @xmath167 when they emerge from the period gap . since cvs with @xmath168 h are predominantly dne @xcite , we further assume that the masses of wds are constant at @xmath43 throughout cv evolution . figure [ pic_t_m ] plots as dashed and solid lines the evolution of donor mass and orbital period calculated from the semi - empirical @xmath2 for the b1 and b5 sequences , respectively , as functions of time @xmath169 elapsed since @xmath16 ( i.e. , @xmath170 gyr ) . the dot - dashed lines show @xmath171 and @xmath172 from the mass - loss rate of @xcite for which @xmath173 and @xmath174 initially at @xmath16 . the changes in the donor mass and orbital period in the numerical model of @xcite are in general slower because of a lower mass - loss rate . our semi - empirical results indicate that @xmath30 decreases from @xmath167 to @xmath52 in @xmath48 gyr and that the minimum period is reached in @xmath48@xmath40 gyr after @xmath16 , without no essential dependence on the adopted stellar age . on the other hand , the numerical model of @xcite predicts that the same decrement of @xmath30 occurs in @xmath175 gyr , while it takes the donors @xmath49 gyr to reach @xmath37 . although uncertainties in the stellar ages make it difficult to follow @xmath171 and @xmath172 accurately below the mhbm , the close agreement between the results based on the b1 and b5 sequences suggests that the ages of cv secondaries with @xmath176 are less than 2 gyr . mass transfer in cvs below the period gap is considered to be driven by gravitational wave radiation @xcite . while equation ( [ eq_dm_def ] ) is useful to obtain @xmath2 from the empirical mrr independent of specifics of mass transfer , it does not provide any clue regarding what drives mass transfer in cvs . in this subsection , we make use of the stationary mass - transfer model of @xcite and check whether the semi - empirical @xmath2 calculated in [ ssec_dm ] is consistent with the angular momentum loss via emission of gravitational waves . @xcite presented a simple analytical model in which the mass transfer from a donor is treated as a stationary isothermal , subsonic flow of gas through the inner lagrangian point . taking allowance for the finite scale height of the donor photosphere and considering the temporal changes in the volume radii of the donor and the roche lobe , @xcite showed that the stationary mass - transfer rate is given by @xmath177,\ ] ] where @xmath178 is the angular momentum loss timescale , @xmath179 is the mass - radius exponent of the critical roche lobe . note that we neglect the nuclear - timescale term in equation ( [ eq_dm_ritter ] ) since it is much longer than the other timescales involved . in the case of dne for which the mass transfer is non - conservative , @xmath179 is given by @xmath180 @xcite . if the gravitational wave emission is the dominant mechanism for the shrinkage of the orbital radius , @xmath178 can be set equal to @xmath181 defined as @xmath182 @xcite . we use the empirical mrr of @xcite and solve equation ( [ eq_dm_ritter ] ) with @xmath183 for @xmath184 . the resulting @xmath1@xmath184 relations are plotted in figure [ pic_dm_p ] as dashed and solid lines in green color for the b1 and b5 sequences . respectively . the adopted stellar ages make a small difference in @xmath184 , since the first term in the square brackets of equation ( [ eq_dm_ritter ] ) is insensitive to the isochrones when @xmath185 mhbm , while it becomes much smaller than the second term when @xmath186 mhbm . figure [ pic_dm_p ] shows that @xmath184 agrees , within errorbars , with @xmath2 for both b1 and b5 sequences before reaching the minimum period . this verifies that our semi - empirical mass - loss rate is consistent with the angular momentum loss due to emission of gravitational waves , at least for pre - bouncers . beyond @xmath37 , @xmath2 for the b1 sequence appears to match @xmath184 better than the b5 sequence , suggesting that the cv ages are presumably less than 5 gyr . since the mass - loss rate can not be observed directly , we are only able to compare our semi - empirical results with the accretion rates estimated by means of other indicators . for this purpose , we select 10 short - period , eclipsing cvs from @xcite , @xcite , and @xcite . table [ ta_ecl ] represents a compilation of our sample cvs , with the parameters in columns ( 2 ) to ( 7 ) taken from the referred papers . to our knowledge , these are the only eclipsers with @xmath1 below 130 min , for which the reliable data for the binary parameters such as @xmath1 , @xmath42 , @xmath30 , and @xmath9 are all available . to infer the mass - accretion rates for the sample cvs , we use @xmath187 where @xmath9 and @xmath42 are the effective temperature and mass of a wd and @xmath188 is the time - averaged accretion rate , taken from @xcite . since the gravitational energy released by the accreted material onto the surface of a wd is deposited at shallow depths and soon radiated away , the quiescent surface luminosity should be proportional to the mass - accretion rate , leading to @xmath189 ( e.g. , @xcite ) . filled circles with errorbars in figure [ pic_dm_te_ecl]_a _ plot @xmath188 for the sample cvs calculated from equation ( [ eq_town_2009 ] ) ; these are also listed in column ( 8) of table [ ta_ecl ] . it can be seen that @xmath188 is broadly in line with our semi - empirical @xmath2 for both b1 and b5 sequence . this suggests that our semi - empirical @xmath1@xmath2 relation is realistic , provided that @xmath190 in these cvs during accretion disk quiescence . @xcite noted that @xmath188 significantly declines towards @xmath37 for @xmath191 min . since the accretion rate decreases with decreasing @xmath2 , this downward tendency of @xmath188 with @xmath1 is in agreement with the behavior of the semi - empirical @xmath2 . donors in short - period cvs are very faint objects , making it difficult to measure their effective temperatures @xmath0 directly from observations , especially in the brown dwarfs regime : in our sample , only two systems ( ht cas and dv uma ) have measured @xmath0 available in the @xcite catalogue . one way to estimate @xmath0 is to use the strong dependence of @xmath2 on @xmath0 . assuming that @xmath190 during the quiescence phase of dne , we combine equations ( [ eq_dm_def ] ) and ( [ eq_town_2009 ] ) to derive @xmath192 since donors with mass below the mhbm have @xmath85 , @xmath193 depends only on the orbital period and the effective mass - radius exponent . for masses above the mhbm , on the other hand , an accurate estimation of @xmath0 requires to consider the effect of @xmath194 term , which can not be determined observationally . using the method described in [ ssec_lg ] , we calculate @xmath195 for the sample cvs listed in table [ ta_ecl ] . for this , we take @xmath196 for systems with @xmath197 and @xmath198 for @xmath199 in accordance with the adopted empirical mrr . the bloating factor of each system is calculated relative to the 1-gyr isochrone . we then calculate @xmath0 from equation ( [ eq_t1_t2 ] ) and plot the results in figure [ pic_dm_te_ecl]_b _ as filled circles with errorbars ; these are also tabulated in column ( 9 ) in table [ ta_ecl ] . it is remarkable that the estimated effective temperatures , except for sdss 1507 and dv uma , are very close to the semi - empirical @xmath1@xmath0 relation , suggesting that equation ( [ eq_t1_t2 ] ) is in fact a good way to estimate @xmath0 when it can not be measured observationally . a small orbital period of sdss 1507 implies that this system is most likely formed directly from a wd / brown dwarf pair @xcite or a member of the old halo @xcite , rather than a conventional dn . in the case of dv uma , @xmath200 k calculated from equation ( [ eq_t1_t2 ] ) is @xmath201 smaller than the observed effective temperature @xmath202 k , corresponding to a spectral type of m4.5@xmath203 , from the @xcite catalogue . such a big difference in @xmath0 can not be due to ambiguity in @xmath204 or @xmath194 terms since both are close to unity for this object . @xcite noted that the best - constrained quantity from a measured @xmath9 is the accretion rate per unit area , @xmath205 , while the subsequent estimation of @xmath188 from @xmath206 requires a certain mrr of wds , which is poorly known for cvs . the wd mass in dv uma is highest among the sample , and it is highly probable that the mrr for massive wds differs from @xmath207 adopted by @xcite . based on the donor masses , @xcite claimed that sdss 1035 , sdss 1501 , sdss 1507 , and sdss 1433 possess post - period minimum cvs . although period bouncers should have low @xmath30 , the precise mass of a particular secondary does not signify its evolutionary history or status @xcite . the evolutionary status of such objects can be clarified more clearly by their positions , for instance , on the @xmath1@xmath0 or @xmath1@xmath2 diagram where the boomerang effect can be seen . figure [ pic_dm_te_ecl ] suggests that sdss 1501 and sdss 1433 among the sample of @xcite have not reached @xmath37 yet . this result does not depend on the stellar age , since these have @xmath208 for which the semi - empirical mass - loss rate is insensitive to the adopted isochrone . while the evolutionary status of sdss 1035 is unclear , the fact that the @xmath1@xmath2 dependence for the b1 sequence reconciles well with that due to gravitational wave emission indicates that it is likely a post - bounce cv . @xcite argued that the tidal and the rotational distortions do not change the value of @xmath34 appreciably because the internal structure of cv secondaries is only weakly affected by such perturbations . to show this , they compared the effective temperatures of the distorted and undistorted stars and found almost no change in @xmath0 . this result is totally expected since short - period cvs are fully convective stars . as we showed in [ ssec_te ] , the hayashi theory gives that the reaction of @xmath0 on the bloating of a donor is small for isolated , fully convective stars . we also show in appendix [ ap_hayashi_mod ] that the influence of an isentropic expansion on @xmath0 caused by tidal and rotational distortions can be neglected in most practical cases . all of these suggest that the effective temperature does not reliably measure the amount of the stellar bloating , and consequently can not be used as an indicator of the degree of tidal and rotational distortions . the mass - loss rate of donors is one of the most important parameters that govern the evolution of short - period cvs . since the mass - loss rate is not directly observable , most previous work employed either purely numerical methods or semi - empirical approaches that use the observed properties of cv accretors rather than donors . yet , the confirmation of the resulting mass - loss rates has to be made . in this work , we use the empirical mrr of short - period cvs constructed by @xcite as an input parameter and derive the semi - empirical relationship between the orbital period and mass - loss rate . we assume that the bloating of cv donors is caused mainly by the mass loss and associated thermal relaxation processes . we also calculate the responses of the effective temperature and luminosities on the bloating of the cv secondaries . we discuss our semi - empirical mass - loss rate in comparison with those from the previous numerical and semi - empirical studies , and in terms of estimating the effective temperature of the donors . our main results are summarized as follows . the semi - empirical @xmath1@xmath2 dependence shows that the mass - loss rate is essentially non - stationary . for @xmath6 min , @xmath2 is in the range of @xmath209@xmath160 and decreases with decreasing @xmath1 relatively slowly as @xmath161 . below 90 minutes , on the other hand , @xmath2 sharply declines towards @xmath37 as a result of a strong dependence of @xmath2 upon @xmath0 , emulating the @xmath1@xmath0 relation . the @xmath1-dependence and amplitudes of the semi - empirical @xmath2 are significantly different from those predicted by the purely numerical models of @xcite . the main reason for the discrepancies is small donor radii in the latter . donors in short - period cvs are out of thermal equilibrium with @xmath164 . thermal relaxation processes change the slope of the @xmath1@xmath2 and @xmath1@xmath0 diagrams significantly above @xmath37 , making the boomerang shape less pronounced . due to the sharp decline of @xmath0 and @xmath2 below 90 min , the observational detection of the boomerang shape on the @xmath1@xmath2 and @xmath1@xmath0 is highly unlikely . the semi - empirical @xmath1@xmath2 relation on the basis of the 1-gyr isochrone is consistent with that estimated from the mass - transfer model of @xcite with the angular momentum carried dominantly by gravitational waves . our semi - empirical @xmath1@xmath2 relation predicts that after emerging from the period gap , donors lose mass from @xmath167 to @xmath52 in @xmath48 gyr , and reach the minimum period in @xmath482 gyrs , suggesting that the ages of cv secondaries with @xmath210 are less than 2 gyrs . \3 . for 10 selected eclipsing cvs ( listed in table [ ta_ecl ] ) for which the reliable observational data are available , the accretion rates inferred from the observed effective temperatures of wds are in good agreement with our semi - analytic mass - loss rates , indicating that our semi - empirical @xmath1@xmath2 relation is realistic . this also suggests that the effective temperature of a cv secondary can be estimated from the effective temperature of its wd primary through equation ( [ eq_t1_t2 ] ) . when analyzed on the @xmath1@xmath2 or @xmath1@xmath0 plane , two cvs ( sdss 1501 and sdss 1433 ) that were previously considered as post - bouncers by @xcite are most likely to be systems before reaching the minimum period . the authors are grateful to the referee , dr . j. patterson , for a thoughtful report . this work was supported by the national research foundation of korea ( nrf ) grant funded by the korean government ( mest ) , no . 2009 - 0063616 . even without considering the effect of mass loss , @xcite showed that the tidal and rotational distortions alone make donors in cvs larger by about @xmath211@xmath212 for @xmath213 polytropes and @xmath214@xmath56 for @xmath215 polytropes , depending on the mass ratio . in this appendix , we evaluate the enlargement factor for polytropes with arbitrary @xmath216 . consider a polytropic star with index @xmath89 , mass @xmath30 , and pressure constant @xmath72 . its radius is @xmath51 in isolation . the mrr of the unperturbed polytropes is given by @xmath217 where @xmath218 is the dimensionless pressure constant that varies weakly with @xmath89 @xcite . when it is situated in a cv , the presence of tidal and rotational perturbations causes it to expand or contract , depending on @xmath89 , by adjusting the internal structure to restore hydrostatic equilibrium . let the new equilibrium be achieved at the volume radius @xmath25 with the corresponding dimensionless pressure constant @xmath219 . since the global internal adjustment occurs in the dynamical timescale , shorter than the thermal timescale and the mass - loss time scale , one may assume that the specific entropy of the gas inside the donor does not change much under the influence of the distortional perturbations . for isentropic responses ( i.e. , @xmath72 is unchanged ) , the enlargement factor is given simply by @xmath220 we use a self - consistent field method outlined in @xcite to construct detailed internal structure of polytropes in hydrostatic equilibrium . we then calculate @xmath218 and @xmath219 for polytropes in critical ( i.e. , roche - lobe filling ) configurations by varying @xmath89 and the mass ratio @xmath100 . figure [ pic_npl ] plots the resulting relative changes , @xmath221 , in the stellar size as a function of @xmath89 for @xmath90 . the behavior of @xmath88 with @xmath30 ( or with the mass ratio @xmath222 for @xmath43 ) for @xmath70 polytropes is plotted in figure [ pic_m_al ] as a dotted line . figure [ pic_npl ] shows that @xmath223 for @xmath224 or @xmath225 , indicating that the perturbations make the donors bigger ; the increase in the equatorial radius is larger than the decrease in the polar radius . on the other hand , critical polytropes with @xmath226 have @xmath227 , since the perturbations decrease the polar radius much more than increasing the equatorial radius . notice a discontinuity at @xmath228 for which one can not determine the stellar enlargement factor . for these polytropes , the central density is proportional to @xmath229 and the mass becomes independent of @xmath25 , making it impossible to constrain the mass from its radius ( e.g. , @xcite ) . @xmath230 diverges to either positive or negative infinity as @xmath231 , indicating that a small change of @xmath89 near 3 leads to a big difference in @xmath230 . this suggests that one should be cautious when treating @xmath228 polytropes numerically . using sph simulations , @xcite found that the critical configurations have an enlargement factor @xmath232@xmath233 for @xmath228 polytropes when @xmath234@xmath235 . this is presumably due to some ( unavoidable ) numerical errors in the sph simulations that somehow change their @xmath228 polytropes to @xmath236@xmath237 ones `` effectively '' , resulting in @xmath238@xmath239 . in [ ssec_te ] , we describe how the effective temperature @xmath0 of a fully - convective donor varies due to a homologous expansion using the hayashi theory . here we study how @xmath0 responds to an isentropic expansion . we assume that the pressure @xmath240 in the whole stellar interiors below photosphere depends on temperature as @xmath241 where @xmath242 is the mean molecular weight and @xmath243 is the gas constant . the photosphere is determined at the radius where the optical depth is @xmath244 . if the radiative pressure is unimportant , the photospheric pressure , @xmath245 , is given by @xmath246 where @xmath247 is the opacity and @xmath248 is the mean surface gravity . using a simple opacity law of the form @xmath249 , we obtain @xmath250 since we assume that the donor is in hydrostatic equilibrium , the photospheric pressure should equal the interior pressure at the interface between the convective interiors and the radiative photospheric layer . assuming @xmath251 , equations ( [ eq_p_int_def ] ) and ( [ eq_p_ph_def ] ) are combined to yield @xmath252 in contrast to the hayashi theory , the effective temperature _ decreases _ due to the donor expansion . nevertheless , @xmath253 for @xmath254 , @xmath255 , and @xmath256 , typical values for cv secondaries , so that the change in @xmath0 due to the tidal and rotational distortions are still very small and can thus be neglected in practice . andronov , n. , pinsonneault , m. , sills , a. 2003 , , 582 , 358a baraffe , i. , chabrier , g. , allard , f. , & hauschildt , p. h. 1998 , , 337 , 403 baraffe , i. , chabrier , g. , allard , f. , & hauschildt , p. h. 2002 , , 382 , 563 baraffe , i. , chabrier , g. , barman , t. s. , allard , f. , & hauschildt , p. h. 2003 , , 402 , 701 barker , j. , & kolb , u. 2003 , , 340 , 623 burrows , a. , & liebert , j. 1993 , revmodphys , 65 , 301 chandrasekhar , s. 1933 , , 93 , 390 chandrasekhar , s. 1939 , an introduction to the study of stellar structure ( chicago univ . press ) dantona , f. , mazzitelli , i. , ritter , h. 1989 , , 225 , 391 eggleton , p. p. 1983 , , 268 , 368 hansen , c. j. , & kawaler , s. , d. 1995 , stellar interiors physical principles , structure , and evolution , ed . appenzeller , i. , harvit , m. , kippenhahn , r. , strittmatter , p. , a. , ( springel - verlag , new - york , inc . ) howell , s. b. , nelson , l. a. , & rappaport , s. 2001 , , 550 , 897 king , a. r. 1988 , , 29 , 1 kippenhahn , r. , trimble , v. , & weigert , a. 1990 , stellar structure and evolution , ed . harvit , m. , kippenhahn , r. , trimble , v. , zahn , j .- p . , ( springel - verlag , berlin heidelberg ) knigge , c. 2006 , , 373 , 484 kolb , u. , & baraffe , i. 1999 , , 309 , 1034 kopal , z. 1972 , adv . astron . . 9 , 1 landau , l. d. , & lifshitz , e. m. 1975 , the classical theory of fields , 4th edn . ( pergamon , new york ) littlefair , s. p. , dhillon , v. s. , marsh , t. r. , gnsicke , b. t. , baraffe , i. , & watson , c. a. 2008 , , 388 , 1582 littlefair , s. p. , dhillon , v. s. , marsh , t. r. , gnsicke , b. t. , baraffe , i. , & watson , c. a. 2007 , , 381 , 827 paczynski , b. 1981 , acta astron . , 31 , 1 patterson , j. , et al . 1984 , , 54 , 443 patterson , j. , et al . 2005 , , 117 , 1204 patterson , j. 2009 , , submitted ( arxiv:0903.1006 ) patterson , j. , thorstensen , j. r. , & knigge , c. 2008 , , 120 , 510 rappaport , s. , joss , p. c. , & webbink , r. f. 1982 , , 254 , 616 renvoiz , v. , baraffe , i. , kolb , u. , & ritter , h. 2002 , , 389 , 485 rezzolla , l. , uryu , k. , & yoshida , s. 2001 , , 888 ritter , h. 1988 , , 202,93 ritter , h. 1996 , in evolutionary processes in binary stars , proceedings of the nato advanced study institute , cambridge , ed . wijers , r. , & davies , m. , 477 , 223 ritter , h. , & kolb , u. 2003 , , 404 , 301 sirtokin , f. v. , & kim , w .- t . 2009 , , 698 , 715(sk09 ) sion , e. , m. , 1995 , , 438 , 876 soberman , g. e. , phinney , e.s . , & heuvel e.p.j . 1997 , , 327 , 620 stehle , r. , ritter , h. , & kolb , u. 1996 , , 279 , 581 townsley d. m. , & bildsten l. 2003 , , 596 , l227 townsley d. m. , & bildsten l. 2004 , , 600 , 390 townsley , d. m. , & gnsicke , b. t. 2009 , , 693 , 1007 gnsicke , b. t. et al . 2009 , , 397 , 2170 uryu , k. , & eriguchi , y. 1999 , , 303 , 329 urban , j. a , & sion e. m. 2006 , , 642 , 1029 cccccccccccccc 0.040&0.099 & 82.769 & 3.50&-4.625&1273&-11.10&0.24&&13.08&-5.513 & 763&-11.99&0.24 + 0.044&0.101 & 81.252 & 3.10&-4.435&1407&-10.94&0.23 & & 9.74&-5.317 & 846&-11.83&0.23 + 0.048&0.103 & 79.898 & 2.80&-4.293&1513&-10.83&0.23 & & 7.31&-5.167 & 914&-11.71&0.23 + 0.053&0.105 & 78.676 & 2.55&-4.194&1588&-10.76&0.23 & & 5.47&-5.011 & 992&-11.58&0.23 + 0.057&0.107 & 77.565 & 2.33&-4.121&1645&-10.73&0.24 & & 4.08&-4.880&1062&-11.49&0.24 + 0.063&0.108 & 76.021 & 2.05&-3.945&1804&-10.67&0.29 & & 2.64&-4.507&1305&-11.23&0.29 + 0.065&0.111 & 77.472 & 1.95&-3.828&1905&-10.61&0.32 & & 2.30&-4.316&1439&-11.09&0.32 + 0.069&0.116 & 79.730 & 1.80&-3.679&2035&-10.56&0.40 & & 1.87&-4.108&1590&-10.99&0.40 + 0.075&0.121 & 82.611 & 1.70&-3.481&2226&-10.39&0.41 & & 1.72&-3.649&2021&-10.55&0.41 + 0.078&0.124 & 84.036 & 1.65&-3.384&2327&-10.33&0.44 & & 1.67&-3.461&2226&-10.39&0.43 + 0.082&0.128 & 86.094 & 1.59&-3.255&2466&-10.21&0.44 & & 1.60&-3.282&2427&-10.23&0.44 + 0.086&0.133 & 88.096 & 1.56&-3.146&2584&-10.11&0.45 & & 1.57&-3.156&2570&-10.12&0.44 + 0.090&0.137 & 90.046 & 1.53&-3.063&2671&-10.04&0.46 & & 1.54&-3.064&2668&-10.04&0.45 + 0.095&0.141 & 91.948 & 1.50&-2.987&2751 & -9.98&0.46 & & 1.52&-2.986&2751 & -9.98&0.46 + 0.099&0.144 & 93.805 & 1.47&-2.925&2812 & -9.94&0.47 & & 1.49&-2.924&2813 & -9.93&0.47 + 0.103&0.148 & 95.619 & 1.44&-2.870&2865 & -9.90&0.48 & & 1.46&-2.869&2866 & -9.89&0.48 + 0.107&0.152 & 97.394 & 1.41&-2.821&2908 & -9.86&0.49 & & 1.43&-2.820&2910 & -9.86&0.49 + 0.111&0.156 & 99.132 & 1.38&-2.778&2946 & -9.84&0.51 & & 1.40&-2.777&2948 & -9.83&0.50 + 0.116&0.160&100.835 & 1.36&-2.736&2983 & -9.81&0.52 & & 1.38&-2.735&2984 & -9.81&0.52 + 0.120&0.163&102.504 & 1.34&-2.699&3012 & -9.79&0.53 & & 1.36&-2.698&3014 & -9.79&0.53 + 0.124&0.167&104.142 & 1.32&-2.666&3037 & -9.77&0.54 & & 1.34&-2.665&3038 & -9.77&0.54 + 0.128&0.170&105.751 & 1.30&-2.636&3057 & -9.76&0.56 & & 1.32&-2.636&3058 & -9.76&0.56 + 0.132&0.174&107.331 & 1.28&-2.605&3080 & -9.74&0.57 & & 1.30&-2.605&3081 & -9.74&0.57 + 0.137&0.177&108.885 & 1.26&-2.575&3104 & -9.73&0.58 & & 1.28&-2.574&3105 & -9.72&0.58 + 0.141&0.181&110.413 & 1.24&-2.547&3124 & -9.71&0.59 & & 1.26&-2.546&3125 & -9.71&0.59 + 0.145&0.184&111.916 & 1.22&-2.521&3141 & -9.70&0.61 & & 1.24&-2.520&3143 & -9.70&0.60 + 0.149&0.188&113.396 & 1.20&-2.497&3155 & -9.69&0.62 & & 1.22&-2.496&3157 & -9.69&0.62 + 0.154&0.191&114.854 & 1.18&-2.472&3174 & -9.68&0.64 & & 1.20&-2.471&3176 & -9.68&0.63 + 0.158&0.194&116.291 & 1.16&-2.447&3191 & -9.67&0.65 & & 1.18&-2.446&3193 & -9.67&0.65 + 0.162&0.198&117.706 & 1.15&-2.425&3206 & -9.66&0.67 & & 1.16&-2.424&3208 & -9.66&0.66 + 0.166&0.201&119.103 & 1.13&-2.403&3219 & -9.66&0.68 & & 1.15&-2.403&3220 & -9.65&0.68 + 0.170&0.204&120.480 & 1.11&-2.383&3230 & -9.65&0.70 & & 1.13&-2.383&3231 & -9.65&0.70 + 0.175&0.207&121.839 & 1.10&-2.365&3240 & -9.65&0.72 & & 1.11&-2.364&3241 & -9.65&0.72 + 0.179&0.211&123.180 & 1.08&-2.345&3252 & -9.64&0.73 & & 1.09&-2.344&3253 & -9.64&0.74 + 0.183&0.214&124.505 & 1.06&-2.326&3263 & -9.63&0.74 & & 1.07&-2.325&3264 & -9.64&0.76 + 0.187&0.217&125.813 & 1.05&-2.308&3272 & -9.62&0.75 & & 1.05&-2.307&3274 & -9.64&0.78 + 0.191&0.220&127.105 & 1.03&-2.291&3281 & -9.61&0.75 & & 1.03&-2.290&3283 & -9.64&0.81 + 0.196&0.223&128.382 & 1.01&-2.275&3289 & -9.60&0.76 & & 1.02&-2.274&3290 & -9.64&0.84 + 0.200&0.226&129.563 & 1.00&-2.260&3295 & -9.60&0.78 & & 1.00&-2.260&3296 & -9.65&0.88 + [ ta_seq ] lrcccccccc sdss 1507 & 66.6119&@xmath257&@xmath258&@xmath259&@xmath260&@xmath261&@xmath262&@xmath263&3 + + sdss 1433 & 78.1066&@xmath264&@xmath265&@xmath266&@xmath267&@xmath268&@xmath269&@xmath270&3 + + sdss 1501 & 81.8513&@xmath271&@xmath272&@xmath273&@xmath274&@xmath275&@xmath276&@xmath277&3 + + sdss 1035 & 82.0896&@xmath278&@xmath279&@xmath280&@xmath281&@xmath282&@xmath283&@xmath284&3 + + sdss 1502 & 84.8298&@xmath285&@xmath286&@xmath287&@xmath288&@xmath289&@xmath290&@xmath291&3 + + sdss 0903 & 85.0659&@xmath292&@xmath293&@xmath294&@xmath295&@xmath296&@xmath297&@xmath298&3 + + sdss 1227 & 90.6610&@xmath299&@xmath300&@xmath301&@xmath302&@xmath303&@xmath304&@xmath305&3 + + oy car & 90.8928&@xmath306&@xmath307&@xmath308&@xmath309&@xmath310&@xmath311&@xmath312&1,2,3 + + ht cas & 106.0560&@xmath313&@xmath314&@xmath315&@xmath316&@xmath317&@xmath318&@xmath319&1,2 + + dv uma & 128.2800&@xmath320&@xmath321&@xmath322&@xmath323&@xmath324&@xmath325&@xmath326&1,2 + + [ ta_ecl ]

we systematically study the null - test for the growth rate data first presented in and we reconstruct it using various combinations of data sets , such as the 8 and or type ia supernovae ( snia ) data . we perform the reconstruction in two different ways , either by directly binning the data or by fitting various dark energy models . we also examine how well the null - test can be reconstructed by future data by creating mock catalogs based on the cosmological constant model , a model with strong dark energy perturbations , the and models , and the large void ltb model that exhibit different evolution of the matter perturbations . we find that with future data similar to an lsst - like survey , the null - test will be able to successfully discriminate between these different cases at the level .

1410.0338

the late - time accelerated expansion of the universe has forced cosmologists to revise our understanding of the universe either by introducing a new component in the universe called dark energy @xcite or by modifying directly the laws of gravity @xcite . within the framework of friedmann - lematre - robertson - walker ( flrw ) cosmology , we can account for a phase of accelerated expansion by simply introducing a cosmological constant ( @xmath4 ) , even though this model gives rise to severe coincidence and fine - tuning problems , observations are still consistent with a dark energy component that has the same characteristics of the cosmological constant @xcite . unfortunately , these experiments are not able to give us information either on the variation on time of a dark energy component or on the _ clustering properties _ of such dark energy component . moreover , recent observations do not have a sufficient sensitivity to be able to distinguish between a dark energy component or a modified gravity model ; even though the two classes of models can be arbitrarily alike , see @xcite , it is still important to be able to discard some of the model that manifests a different behaviour . furthermore , future experiments have been planned to collect a large amount of data with high accuracy and it would be interesting to find tests that are able to confirm our assumptions . consistency checks are usually model independent tests able to determine if the assumptions made are violated . in this paper we make use of the consistency check found in @xcite . in the latter , we introduced a new null - test specifically for the growth of matter perturbations and , as far as we know , this is the first null - test that accounts for perturbations on the matter fluids . the evolution of the matter density contrast is governed by the evolution of the hubble parameter and by the evolution of all the other clustering components @xcite . hence it is a complementary probe for the dark energy because , while many different dark energy models give the same expansion history they usually differ at perturbation level ( depending on the intrinsic characteristics of the fluid itself ) and they will affect the evolution of the matter density contrast . moreover , as it is well known , modified gravity models can also be reinterpreted as effective dark energy models with their own _ effective perturbed quantities _ and consequently the growth of matter density will be influenced by these _ effective _ perturbations ; see @xcite . hence , it was necessary to find a null - test that accounted for the growth of matter density fields . finally , the paper is organized as follows : in sec . [ settings ] we report the main equations for the growth of matter density contrast ; in sec . [ sec : lagrangianform ] we review the derivation of the null - test and generalize it to include modified gravity models , in sec . [ sec : nulltest ] we construct the null - test and discuss its implications . finally , in sec . [ sec : datanalysis ] we reconstruct the null - test with a variety of data and in sec . [ sec : conclusions ] we summarize our results . the growth of matter in the universe under the assumption of homogeneity and isotropy is governed by the second order differential equation : ( a)+(3a + ) (a)-32 = 0 , + [ ode ] where @xmath5 is the hubble parameter , @xmath6 is the matter density contrast today and @xmath7 is the hubble constant and where we introduced the effective newtonian constant @xmath8 that accounts for dark energy perturbations or for a variety of modified gravity models ; see refs . @xcite . under the assumption that the universe is currently dominated by a dark energy component with a constant equation of state and negligible dark energy perturbations , i.e. @xmath9 , then eq . ( [ ode ] ) can be easily solved analytically . the differential equation ( [ ode ] ) has in general two solutions that correspond to two different physical modes , a decaying and a growing one , that in a matter dominated universe in gr behave as @xmath10 and as @xmath11 respectively . since we are only interested in the latter , we demand that at early times @xmath12 , usually during matter domination , the initial conditions have to be chosen as @xmath13 and @xmath14 . when @xmath9 we get gr as a subcase , while in general for modified gravity theories , the term @xmath15 can be scale and time dependent . for a flat gr model with a constant dark energy equation of state @xmath16 , the exact solution of eq . ( [ ode ] ) for the growing mode is given by @xcite ( a)&= & a _ 2f_1[da1 ] + & & h(a)^2/h_0 ^ 2= a^-3+(1-)a^-3(1+w),where @xmath17 is a hypergeometric function , see ref . @xcite for more details . in more general cases , for instance admitting that the dark energy equation of state parameter is a function of time , it is impossible to find a closed form analytical solution for eq . ( [ ode ] ) , but in ref . @xcite it was shown that the growth rate @xmath18 can be approximated as f(a)&=&(a)^(a ) [ fg ] + ( a ) & & + ( a ) & = & + a more general expression for the growth index can be found in @xcite . we should note that the approximation for @xmath19 is valid at first order for a dark energy model with a constant @xmath16 , while for ( @xmath20 ) we have @xmath21 . furthermore , it is easy to convert eq . ( [ ode ] ) into an equation for the growth rate @xmath18 , which can be found to be f(a)+(+)f(a)+f(a)^2-=0[growthode ] with initial conditions @xmath22 for @xmath23 ( usually @xmath24 ) . in later sections we will use the previous equations to construct a null - test for the growth rate of matter density perturbations . the null - test is a function of redshift @xmath25 that , however , has to be constant for all @xmath25 under some assumptions , eg that gr is valid or homogeneity holds ( since eq . ( [ ode ] ) was evaluated under the assumption of homogeneity and isotropy ) . any deviation from the expected result then indicates the failure of one or more of the assumptions . typical examples in cosmology include the @xmath26 test of clarkson et al @xcite or the @xmath27 statistic of shafieloo et al . , see @xcite , @xcite . also , early examples of null - tests for the growth data were shown in refs . @xcite and @xcite . however , the former suffers from the problem that we need to know @xmath28 at @xmath29 , while the latter , known as the test requires some mild assumptions for @xmath15 . on the contrary , we will show that our new null - test does not suffer from any of these problems . in this paper we expand our work from ref . @xcite , creating a null - test that can be used also for more sophisticated cosmological models ; we also reconstruct our new null - test with a variety of both real and mock data . the last are created using different cosmologies in order to test the validity and the accuracy of our test . in this section we will review the derivation of the null - test using the lagrangian formulation and we will expand it for modified gravity theories . this can be done by again constructing a lagrangian for eq . ( [ ode ] ) and with the help of noether s theorem we can find an associated conserved quantity . if we assume that the lagrangian can be written as = ( a,(a),(a ) ) then the euler - lagrange equations become : -=0 . [ eleqs ] we can assume a lagrangian of the form & = & t - v + t&=&12 f_1(a , h(a))(a)^2 + v&=&12 f_2(a , h(a))(a)^2 where the second and third terms are the kinetic " and potential " terms respectively , and @xmath30 and @xmath31 are two functions that need to be found . substituting the last equations in the euler - lagrange eq . ( [ eleqs ] ) , we get ( a)&+ & ( + ) (a ) + & + & ( a)=0 . [ eleqs1 ] comparing eq . ( [ eleqs1 ] ) with eq . ( [ ode ] ) we immediately find that f_1(a , h(a))&= & a^3 h(a)/h_0 + f_2(a , h(a))&= & - . then the lagrangian @xmath32 and the ` hamiltonian ' @xmath33 of the system become : = t - v&=&12 a^3 h(a)/h_0(a)^2 + + & + & ( a)^2 + [ lag ] + = t+v&=&12 a^3 h(a)/h_0(a)^2- + & -&(a)^2 . [ ham ] + unfortunately , since the hamiltonian @xmath33 explicitly depends on ` time ' , i.e. the scale factor , the energy of the system is not conserved . now that we have obtained the lagrangian for the system , we can use noether s theorem to find a conserved quantity that will be later translated into the null - test . so , if we have an infinitesimal transformation @xmath34 with a generator * x*&= & ( ) + + & & (a)=(a ) , such that l_x = 0 [ lieder ] then = ( a)[const1 ] is a constant of ` motion ' for the lagrangian of eq . ( [ lag ] ) , see ref . @xcite for an application in scalar - tensor cosmology and more details . from eq . ( [ const1 ] ) we get that = a^3 h(a)/h_0 ( ) (a ) , while from eq . ( [ lieder ] ) we get & & (a ) a^3 h(a)/h_0 (a)+ + & + & = 0 . the latter can be solved in favor of @xmath35 to give ( a)=ce^-_a_0^adx where @xmath36 is an integration constant and @xmath37 can be chosen to be either 0 or 1 . then the constraint becomes = a^3 h(a)/h_0 (a)e^-_a_0^adx[const2 ] where we have redefined @xmath38 to absorb @xmath36 . choosing appropriately @xmath37 can lead to convenient values for @xmath38 , for example for @xmath39 then it is easy to see that @xmath40 and for @xmath23 then @xmath41 , while in general we have @xmath42 . we have checked numerically the validity of eq . ( [ const2 ] ) for several different cosmologies and values of the parameters . ( [ const2 ] ) can also be written in terms of the growth rate @xmath18 . as a consequence , the growth factor can be found to be @xmath43 and eq . ( [ const2 ] ) can be rewritten as /(a_0)= h(a ) f(a ) e^_a_0^a ( -)dx.[const1a ] taking into account that @xmath42 , we get that the lhs of the previous equation can be re - expressed as /(a_0)=a_0 ^ 3 h(a_0)(a_0)/(a_0)=a_0 ^ 2 h(a_0 ) f(a_0 ) , so that eq . ( [ const1a ] ) becomes : e^_a_0^a ( -)dx=1.[const2a ] taking the derivative of eq . ( [ const2a ] ) with respect to the scale factor @xmath44 , we obtain eq . ( [ growthode ] ) . this means that eq . ( [ const2a ] ) is a first integral of motion " of eq . ( [ growthode ] ) . however observations can measure directly only @xmath45 , where @xmath46 and they are not able to give directly @xmath47 , hence we need to transform eq . ( [ const2 ] ) to be able to test it directly with observations . taking into account that f_8(a)f(a ) _ 8(a)=a (a ) , where @xmath48 , we have that ( a)=(a_0)+_a_0^adx . then , eq . ( [ const2 ] ) can be written as & & + & & e^-32_m_a_0^adx=1 . [ nullf ] it is clear that the expressions of eqs . ( [ const2a])-([nullf ] ) have to be constant for all redshifts @xmath25 , so in the next section we will use them to construct a null - test . any deviation from unity will imply the presence of new physics or systematics in the data . in this section we will use eqs . ( [ const2a])-([nullf ] ) and assume @xmath49 to construct a new null - test for the growth data . since this equation only holds for gr with the flrw metric , deviations point to either new physics or systematics in the data . we have explicitly tested in the case of @xmath50 , where the analytical solution is known , that eqs . ( [ const2a ] ) - ( [ nullf ] ) are valid at all redshifts . in order to create our null - test , we implement eqs . ( [ const2a ] ) - ( [ nullf ] ) . we now have two equivalent forms of the null - test : ( z)&= & e^_a_0^a ( -)dx [ nulltest1 ] + ( z)&= & + & & e^-32_m_a_0^adx . [ nulltest2 ] both forms are totally equivalent : eq . ( [ nulltest1 ] ) is expressed in terms of the growth rate @xmath51 , which is not a direct measurable quantity but it makes the expression for the null - test much simpler and it will be useful ( as it will be clear later on ) for testing directly specific models ; eq . ( [ nulltest2 ] ) is written in terms of direct measurable quantities and it will be extremely useful to test the data . it is clear that in both cases we should have @xmath52 at all redshifts , and any deviation from unity could be due to several reasons : * detection of modified gravity and nonconstant @xmath15 . * new physics or a presence of shear or strong dark energy perturbations . * deviation from the flrw metric and homogeneity . * tension between @xmath0 ( obtained directly or derived ) and @xmath53 data . in the next section we will test the above expression with the help of mocks based on different models . in this section we schematically report the different cosmologies used to create mocks catalogs . if the universe is filled by a dark energy component , with constant equation of state parameter @xmath16 , then the hubble equation can be written as : h(a)^2/h_0 ^ 2= a^-3+(1-)a^-3(1+w ) . if the dark energy component is not a cosmological constant , i.e. if @xmath54 , then dark energy is able to cluster . the scale at which this dark energy component can cluster depends on the intrinsic characteristic of the fluid itself , namely : pressure perturbations @xmath55 which is related to the sound speed and anisotropic stress @xmath56 that are usually related to the viscosity of the fluid ( see @xcite ) . if dark energy is able to cluster at sufficiently small scales then eq . ( [ ode ] ) needs to be modified to account for the dark energy perturbations which will be an extra source term to the gravitational potential . since eq . ( [ ode ] ) has been evaluated on the limit of small scales , dark energy has to have a small value of the sound speed and zero viscosity term . in order to create mock catalogs for the @xmath57 we modify eq . ( [ ode ] ) by using the function q(a ) = 1 + a^-3w [ eq : q ] which will act as a modified newton s constant , i.e. @xmath58 . ( [ eq : q ] ) describes the amount of the dark energy perturbations and it has been evaluated under the assumption of zero dark energy sound speed and zero anisotropic stress , @xcite . if we set @xmath20 then we recover the @xmath4cdm model , i.e. with zero perturbations . in this case our default parameters for the mocks are : @xmath59 . since we are interested in examining the effect of the 8 data on the null - test , we choose an @xmath1 model that is exactly at the background level , but is significantly different at the perturbations level . this way , we can disentangle the effects of the modified gravity , @xmath60 from the background acceleration . one such degenerate model was studied in ref . @xcite , where the @xmath1 action was found to be : s= d^4 x ( f(r)/2+s_m ) , [ actionfr ] where @xmath61 is the action term for the matter fields and f(r)&=&r-2+h_0 ^ 2()^b + & & _ 2f_1(b,+b,+2b,),[frmodact1 ] where @xmath62 and the parameter @xmath63 is dimensionless and determines how strong the effects of the modified gravity are . we should note that with this lagrangian we can recover gr at early times ( @xmath64 ) , i.e. @xmath65 or @xmath66 and that it passes all criteria for viability of @xmath1 models , as shown in ref . @xcite . for this model we have by construction h(a)^2=h_0 ^ 2(a^-3 + 1- ) , [ hagr ] while newton s constant is @xcite : / g_n&= & , [ gefffr ] + m & & , + f&&f_,r=. in this case our default parameters for the mocks are : @xmath67 and we also considered the two different cases @xmath68 . another interesting case are the @xmath69 models , where @xmath70 is the gauss - bonnet term @xmath71 . again , we are primarily interested in the effects of the modification of gravity , so we will use the @xmath2 degenerate model of ref . @xcite , that is exactly at the background level . then the action is given by @xcite : s= d^4 x ( r/2+f())+s_m , [ action ] where f()=-3h_0 ^ 2 ( 1-)+ h_0 ^ 2 d. [ myaction1 ] in the last equation the first term corresponds to the cosmological constant , we have neglected a term that was just proportional to @xmath72 as it does not contribute in the field equations . the cosmological perturbations of the @xmath69 models were studied in ref . @xcite , where it was shown that the growth factor for the matter perturbations @xmath73 satisfies the evolution equation ( using the subhorizon approximation @xmath74 ) : [ per1 ] _ m+c_1(k , a ) _ m+c_2(k , a ) _ m0 , where the functions @xmath75 and @xmath76 where first derived in @xcite and are given in appendix b of ref . @xcite for completeness . in the gr limit eq . ( [ per1 ] ) reduces to _ m+2 h _ m- a^-3 _ m= 0 so comparing these two expressions we can define an effective newton s constant : ( k , a)/g_n= , [ geff ] which is valid only under the subhorizon approximation @xmath74 . even though these models suffer from instabilities in the matter density perturbations during the matter era as shown in @xcite , we still use them to make mocks since they exhibit rich phenomenology due to the presence of the second term containing @xmath75 in eq . ( [ per1 ] ) . this makes them ideal candidates for our null - test , as @xmath75 can not be described by a single @xmath15 term , thus will produce deviations from unity . in this case our default parameters for the mocks are : @xmath77 . alternatives to @xmath4 for explaining the current acceleration are inhomogeneous universe models in which the effective acceleration is caused by our special position as observers inside a huge underdense region of space . one of the simplest models to study the effect of such large inhomogeneities is the spherically symmetric lematre - tolman - bondi model @xcite ( ltb ) . in this large void model , the metric is given by ds^2 = - dt^2 + x^2(r , t)dr^2 + a^2(r , t)d^2 , where @xmath78 and the equivalent of the scale factor now depends on both time and the radial coordinate @xmath79 . we can find a relationship between @xmath80 and @xmath81 using the the @xmath82 component of the einstein equations : @xmath83 where a prime denotes a derivative with respect to coordinate @xmath79 and @xmath84 is an arbitrary function that plays the role of the spatial curvature parameter . to find the growth index in ltb cosmologies we must study linear perturbation theory in inhomogeneous universes . due to the loss of a degree of symmetry , the decomposition theorem does no longer hold . this means that , in general , our perturbations will no longer decouple into scalar , vector and tensor modes . a study of the perturbation equations in this scenario using a 1 + 1 + 2 decomposition of spacetime can be found in @xcite . however , if the normalized shear @xmath85 is small , and @xmath86 , corresponding to the transverse and along the line of sight expansion rates , respectively . ] as observations seem to confirm @xcite , we can use the adm formalism and express our perturbed ltb metric as ds^2 = -(1 + 2)dt^2 + ( 1 - 2)_ijdx^idx^j , where @xmath87 . within this formalism , the growing mode of the density contrast is given by @xcite ( r , t)= _ 2f_1 , where @xmath88 and @xmath89 specifies the local matter density today . we can now calculate the growth rate of density perturbations , noting that here the matter density parameter @xmath90 is a function of redshift via both time @xmath91 and the radial coordinate @xmath79 . in ltb models , this is in principle an arbitrary function which must be chosen appropriately in each case . in the case of the constrained gbh model @xcite the parameters are given by @xmath92\over1+\tanh[r_0/\delta r]}\\ h_0(r ) & \!=\ ! & h_0\left[{1\over1-\omega_m(r ) } - { \omega_m(r)\over(1-\omega_m(r))^{3/2}}\times \right . \nonumber\\ & & \hspace{9 mm } \left.\times{\rm arcsinh}\sqrt{1-\omega_m(r)\over\omega_m(r)}\right]\,,\end{aligned}\ ] ] with @xmath93 where these values have been chosen to best fit the supernovae and bao data @xcite . within this model , the growth function , i.e. the logarithmic derivative of the density contrast , is given by @xcite [ fzltb ] f(z)=_m^1/2(z ) , where @xmath94 are the associated legendre polynomials and @xmath95 is the fraction of matter density to critical density , as a function of redshift .. note that this is different from @xmath96 , which gives the mass radial function today , see ref . @xcite . ] this function ( [ fzltb ] ) is identical to the _ instantaneous _ growth function of matter density in an open universe , where the local matter density @xmath97 is given by @xmath95 at that redshift . this is a good approximation only in ltb models with small cosmic shear , see ref . @xcite . .@xmath98 measurements from different surveys . [ tab : fs8-data ] [ cols="^,^,^,^",options="header " , ] the binning technique to measure @xmath99 consists of evaluating it in several redshift bins by directly computing the @xmath0 values and by using the @xmath100 values measured by different experiments . the hubble parameter catalog contains @xmath101 data spanned from redshift @xmath102 up to @xmath103 , whereas the growth rate catalog contains @xmath104 data points from @xmath105 up to @xmath106 . since the growth measurements reach only up to @xmath106 we are forced to discard the last 9 data points for @xmath0 ( as we want to avoid having too wide bins ) . because the number of data for both catalogs is quite small the choice of the bins is quite restricted . we decided to opt for two different binning : first we chose @xmath107 and then @xmath108 bins , both equally spaced and we evaluate the observables at the mean redshift of the bins . it is important to notice that , in order for the consistency check @xmath109 to hold , we need to evaluate quantities at the same redshift . we show the results in fig . [ fig : binrealvsmock - null ] . as can be seen from the figure , the number of bins affect the results ; in the @xmath107-bins case the null - test @xmath109 is far from unity implying that the actual data do not give a scenario as at redshift @xmath110 the reference cosmology is at almost @xmath111 s away . however , in the @xmath108-bins case the data predict a scenario already at @xmath112 . the reason why we have such different results is due to the number of data points we are considering . at the moment , we have few data especially for the growth factor and also not uniformly distributed , leaving some bins with only two points and making the binning technique not fully reliable . as mentioned before , we also use mock catalogs based on different cosmologies to test @xmath113 for two main reasons : first , to evaluate how much the errors on the null - test will be with future experiments ; second , to examine the validity and the generality of the null - test @xmath113 . we used different cosmologies to evaluate the mock catalogs : 1 ) @xmath16cdm with @xmath20 to recover the @xmath4cdm limit and another set of data using @xmath114 which allows perturbations in the dark energy sector ; 2 ) @xmath1 model with @xmath67 and we also considered the two different cases @xmath68 ; 3 ) @xmath2 model with @xmath77 ; and 4 ) @xmath115 model with @xmath116 . the details of the models can be found in section [ theory ] . we created two different catalogs for each cosmology : the hubble parameter and the @xmath98 . since we are more interested in testing the consistency check @xmath113 rather than worrying about systematics in the data , we evaluated the hubble and growth parameters uniformly distributed in the range @xmath117 $ ] divided into @xmath118 equally spaced bins of step d@xmath119 ; the @xmath0 and the @xmath98 were estimated as its theoretical value plus a gaussian error ( that can be negative or positive ) and constant errors of 0.2 and 0.006 , respectively ; the values of the errors were obtained using the fisher matrix approach and having in mind a setup similar to euclid - like and lsst - like surveys @xcite , i.e. evaluating the sensitivity that future survey will have to measure the hubble parameter and the growth of matter . & + @xmath120 & @xmath121 & @xmath56 s & @xmath25 & @xmath121 & @xmath56 s + @xmath122 & @xmath123 & @xmath124 & @xmath125 & @xmath126 & @xmath124 + @xmath127 & @xmath128 & @xmath129 & @xmath130 & @xmath131 & @xmath132 + @xmath133 & @xmath134 & @xmath135 & @xmath136 & @xmath137 & @xmath138 + @xmath139 & @xmath140 & @xmath141 & ... & ... & ... + & & & & & + z & @xmath121 & @xmath56 s & @xmath121 & @xmath56 s & @xmath121 & @xmath56 s & @xmath121 & @xmath56 s & @xmath121 & @xmath56 s + @xmath102 & @xmath142 & @xmath143 & @xmath142 & @xmath143 & @xmath142 & @xmath143 & @xmath144 & @xmath143 & @xmath145 & @xmath143 + @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath151 & @xmath152 & @xmath153 & @xmath154 & @xmath155 & @xmath156 + @xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 & @xmath162 & @xmath163 & @xmath164 & @xmath165 & @xmath166 & @xmath167 + @xmath168 & @xmath169 & @xmath170 & @xmath171 & @xmath172 & @xmath173 & @xmath174 & @xmath175 & @xmath176 & @xmath177 & @xmath178 + + in what follows we report the results obtained by binning the data in the mock catalogs that we created using different cosmologies . we use the null - test valid for the model given by eq . ( [ nulltest2 ] ) and we analyze the mocks . in practice we ask ourselves the following : if the universe is different from @xmath4cdm , how accurately we can test it ? as we are analyzing mocks created using a cosmology different from the we expect the null - test to fail , i.e. to be different from unity at all the redshifts . to analyze the mock data we decided to use two different binning : first , we used @xmath107 bins from redshift @xmath143 until redshift @xmath179 to compare them with the results from the actual data ; second , we used @xmath180 bins using all the data , i.e. we extended our analysis up to @xmath181 . as both catalogs contain the same number of points and they are uniformly distributed , the mean redshift in each bin will be the same for each cosmology . in tab . [ tab : nulltest - results - binning - mocks ] we report the values of the @xmath113 for the different cosmologies in the @xmath107-bins case . in the same table , next to each value of the null - test , we present the confidence level , i.e. how many sigmas the values of the null - test are from unity , if the value is smaller than @xmath182 then the value of the null - test @xmath113 is within @xmath112 close to unity , if the value is large than one , then the value corresponds to the sigmas away the null - test is . in fig . [ fig : mocks1 ] we show the result for four cosmologies and in tab . [ tab : nulltest - results - binning - mocks ] we report the values found for the null - test , the corresponding errors and the confidence level . if we test the mock catalogs , we get a result that it consistent with @xmath182 already at @xmath112 , see tab . [ tab : nulltest - results - binning - mocks ] , when we use a different mock catalog for instance the @xmath16cdm one then @xmath113 is less than @xmath182 at more than @xmath183 s at almost any redshift , which is the result that we would expect as the growth of the matter density contrast increases because of the dark energy perturbations . using the @xmath1 mocks the null - test gives values closer to unity indicating that it will be more difficult to differentiate the and the @xmath1 model ; this is due to the fact that the @xmath1 model used in this paper has a hubble parameter which is exactly and an @xmath63 of @xmath184 , hence the modification to the growth @xmath98 is small . when we use the mocks from @xmath2 and ltb cosmologies , which both models give substantially different behavior of the hubble parameter and the growth of matter density contrast , the deviation from unity of the null - test @xmath113 becomes more evident , in fact we found that the @xmath2 can be ruled out at more than @xmath185 s and the ltb at more than @xmath186 s . the results up to @xmath187 can be found in tab . [ tab : nulltest - results - binning - mocks-10bins ] . an interesting alternative to binning is to fit the data , either real or mock , to the @xmath4cdm and @xmath16cdm models and then reconstruct the null - test @xmath113 . in fig . [ fig : modelfittingreal ] we show the results of reconstructing the null - test with the real data fitted by the ( left ) and @xmath16cdm ( right ) models respectively . clearly , the null - test as reconstructed with the real data seems to be compatible with unity at the @xmath188 level . next , we will also test how well the null - test will be reconstructed with future data . for this , we also consider the different cosmologies mentioned in a previous section and fit the mocks with both the and @xmath16cdm models . the reason for this is that we want to make a direct test of the standard cosmological model with as few extra assumptions as possible . we should stress that in this case any deviation from unity implies a breakdown of either the fundamental assumptions of the standard cosmological model , i.e. homogeneity , the validity of gr etc , or that the de models used are not a good description of the data . in fig . [ fig : modelsmocks1 ] we show the results for the null - test for the ( left ) and @xmath16cdm ( right ) using mocks ( first row ) and the de perturbations ( second row ) for the @xmath0 and 8 data . in fig . [ fig : modelsmocks2 ] we show the results for the null - test for the model for the mock @xmath0 and 8 data based on the @xmath1 model for @xmath189 ( first row left ) and @xmath190 ( first row right ) . on the second row we show the results for the @xmath2 @xmath0 and 8 data for the model ( left ) and @xmath16cdm ( right ) models respectively . finally , in fig . [ fig : ltbmocks ] we show the results for the null - test for the ltb @xmath0 and 8 mocks fitted with the ( left ) and @xmath16cdm ( right ) . we find that the @xmath113 null - test will be particularly successful at detecting deviations from gr at high significance @xmath191 , especially of the @xmath1 and @xmath2 types ( fig . [ fig : modelsmocks2 ] ) , but also deviations from the frw metric ( fig . [ fig : ltbmocks ] ) . this is due to the fact that these models have significantly different evolution for the matter density perturbations , which is encoded in the @xmath15 and can be detected by the null - test . + + + + as an extra check we also use alternative data instead of just the @xmath0 , namely the supernovae type ia ( snia ) to reconstruct the hubble parameter . in particular we used the latest _ union 2.1 _ set of 580 snia data of suzuki et al . @xcite that spans from redshift @xmath192 up to @xmath193 . the results for this reconstruction are shown in fig . [ fig : lcdmsniarealvsmock ] for the and in fig . [ fig : wcdmsniarealvsmock ] for the @xmath16cdm . we find that they are in excellent agreement with that of the @xmath0 data shown earlier , thus eliminating any possibility of bias due to the use of the particular data used to reconstruct the hubble expansion history . finally , we also consider @xmath1 models in the reconstruction of the null - test . specifically , in fig . [ fig : frrealsniavshz ] we show the results for the null - test for the @xmath1 model for the real snia and 8 data ( left ) and the @xmath0 and 8 data data ( right ) . again the results are in good agreement , thus demonstrating that the null - test is not particularly sensitive on the model used . + + it should be noted that unfortunately , none of the @xmath98 data used in this analysis has been evaluated in a completely model independent manner . usually an underlying cosmology has to be considered in order to extract information about the growth rate parameter , and most of the time this cosmology is the @xmath4cdm . another limitation of the @xmath53 data is that part of the measurement comes from a range of wave numbers that falls into the nonlinear regime , with the typical scales ranging from @xmath194 up to @xmath195 mpc / h , which expressed as a wavenumber is @xmath196 h / mpc . the maximum wavenumber that we can reach before entering into the non linear regime depends on the redshift of the measurements ; starting from @xmath197 h / mpc at @xmath198 up to @xmath199 h / mpc at high redshifts ( above @xmath182 ) . most surveys @xcite already do take into account non - linearities via the non linear bao diffusion . the latter is usually treated as a `` nuisance '' parameter , which will be marginalized over as it does not depend on the cosmology , and by doing so the cosmological information comes only from the linear part of the power spectrum . moreover , the growth rate @xmath200 is a linear quantity because is directly proportional to the velocity perturbations @xmath201 with no scale dependence . clearly there is still a lot room for improvement in order to have completely model independent and better measurements of the growth index . it should be stressed however , that this problem with the data could potentially be resolved in the near future if the data can be created with more model independent techniques , thus alleviating this constraint from our null - test . in this subsection we will briefly discuss another issue , that of the potential problem that the @xmath0 data might originate from a different cosmology but the growth @xmath202 data has been evaluated under the assumption of . for instance , the later time scale independent growth could be generated by @xmath1 gravity theory , but some of analysis assumed the coherent growth and then what it is measured is the mean @xmath203 in the valid @xmath204 bins . above all , uncertainties due to the non - linear or non - perturbed terms prevail through those measurements and as a result , we are not able to test anything other than simple gr models . therefore , if one wants to constrain a specific theoretical model , such as @xmath1 or @xmath2 , he / she should analyze the data based upon the model - dependent method . however , our test is very sensitive to variations of cosmologies or , in other terms , in tensions in the data , as in this particular case . an interesting game - test is the following : let us imagine we have a data set of @xmath0 measured in a model independent way and they favor a @xmath16cdm model ; then we have a different data set of @xmath205 measured assuming a model ( in this case ) so these data are clearly biased as they will favor a model rather than the true underlying cosmology ( which is @xmath16cdm with perturbations also in the dark energy sector ) . so we can think to use these two different data sets with the null - test : in practice , in order to quantify the sensitivity of our null - test , we use a mock catalog for h(z ) using the @xmath16cdm model and another mock catalog using for @xmath205 using the model . we notice that the null - test deviates from unity at the level of about @xmath206 ( @xmath207 ) and in average of about @xmath208 . we can also ask how much @xmath205 itself deviates using the wrong cosmology and test which is more sensitive . we find that for @xmath205 for the two different cosmologies , i.e. for and @xmath16cdm , the difference between the @xmath205 evaluated using two different cosmologies is of about @xmath209 and in average of about @xmath210 . we can also have a closer look at the errors on the null - test mixing - up two different cosmologies . implementing the full error propagation analysis we realize that the errors remain basically unchanged when we mix up the two different cosmologies ; the reason is that the error propagation formula does not change and the important quantities are the errors on the measured quantity , i.e. @xmath0 , @xmath98 , @xmath211 and @xmath212 . in fig . [ fig : tension - in - data ] we plot the null - test using @xmath4cdm and @xmath16cdm mock catalogs and the null - test mixing the cosmologies . in this paper we have reconstructed the null - test , that can be used to probe potential deviations from and was first presented in ref . @xcite , by using the @xmath0 , snia and 8 data . we performed the reconstruction in two different ways : by directly binning the data and by fitting the data to various dark energy models like the and @xmath16cdm and then calculating the null - test . we find that both methods have different advantages ; the former uses as few assumptions as possible while the latter directly tests the standard cosmological model . we have also generalized the null - test and extended it for modified gravity models and models with strong de perturbations , by taking into account the @xmath15 term in eq . ( [ ode ] ) . we have explicitly checked that when this term is taken into account , then the null - test is constant as expected for modified gravity models . this allows us to verify that deviations from unity in the original version of the null - test presented in ref . @xcite can indeed also be attributed to modifications of gravity . we have found that deviations from unity could be due to several reasons , either new physics including modifications of gravity and strong dark energy perturbations , or breakdowns of one of the basic assumptions of the standard cosmological model , i.e. deviation from the flrw metric and homogeneity or finally , a possible tension between @xmath0 ( obtained directly or derived ) and the @xmath53 data . in all cases due to the nature of the null - tests and that they have to be constant at all redshifts , it is enough to a have a statistically significant deviation at one redshift to detect one of the above reasons . a possible limitation at the moment is that the null - test can not tell us which of the above reasons would be responsible for that deviation though . however , our growth null - test will be extremely useful if joined with other null - tests , like the @xmath26 presented in @xcite which is able to test the assumptions of homogeneity and isotropy of the universe . we also examined how well the null - test can be reconstructed by future data by creating mock catalogs based on a lsst - like survey and on the model , a model with strong de perturbations , the @xmath1 and @xmath2 models , and the large void ltb model that exhibit different evolution of the matter perturbations . this was done so as to examine how well our null - test can be reconstructed using the data from upcoming surveys . our results were presented in figs . [ fig : binrealvsmock - null]-[fig : frrealsniavshz ] . we found that when reconstructed with real data the null - test is consistent with unity at the @xmath183 level , with both the binning and the model testing methods . however , when we reconstruct it with the mock data based on the specifications of a lsst - like survey and various models that go beyond the , ie the @xmath1 , @xmath2 models and the ltb , we find that the null - test can detect deviations from unity at the @xmath3 and also @xmath186 level . overall , the novelty of our null - test is that it can directly test the fundamental assumptions of the standard cosmological model with as few assumptions as possible . therefore , it will definitely prove to be an invaluable tool in the near future given the plethora of upcoming surveys that will produce high quality data . & & & & & + z & @xmath121 & @xmath56 s & @xmath121 & @xmath56 s & @xmath121 & @xmath56 s & @xmath121 & @xmath56 s & @xmath121 & @xmath56 s + @xmath102 & @xmath142 & @xmath143 & @xmath142 & @xmath143 & @xmath142 & @xmath143 & @xmath144 & @xmath143 & @xmath145 & @xmath143 + @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath151 & @xmath152 & @xmath153 & @xmath154 & @xmath155 & @xmath156 + @xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 & @xmath162 & @xmath163 & @xmath164 & @xmath165 & @xmath166 & @xmath167 + @xmath168 & @xmath169 & @xmath170 & @xmath171 & @xmath172 & @xmath173 & @xmath174 & @xmath175 & @xmath176 & @xmath177 & @xmath178 + @xmath213 & @xmath214 & @xmath215 & @xmath216 & @xmath217 & @xmath218 & @xmath219 & @xmath220 & @xmath221 & @xmath222 & @xmath223 + @xmath224 & @xmath225 & @xmath226 & @xmath227 & @xmath228 & @xmath229 & @xmath230 & @xmath231 & @xmath232 & @xmath233 & @xmath234 + @xmath235 & @xmath236 & @xmath237 & @xmath238 & @xmath239 & @xmath240 & @xmath241 & @xmath242 & @xmath243 & @xmath244 & @xmath245 + @xmath246 & @xmath247 & @xmath248 & @xmath249 & @xmath250 & @xmath251 & @xmath252 & @xmath253 & @xmath254 & @xmath255 & @xmath256 + @xmath257 & @xmath258 & @xmath259 & @xmath260 & @xmath261 & @xmath262 & @xmath263 & @xmath264 & @xmath265 & @xmath266 & @xmath267 + @xmath268 & @xmath269 & @xmath270 & @xmath271 & @xmath272 & @xmath273 & @xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 + the authors thank an anonymous referee for useful suggestions that improved the paper . the authors acknowledge financial support from the madrid regional government ( cam ) under the program hephacos s2009/esp-1473 - 02 , from micinn under grant fpa2012 - 39684-c03 - 02 and consolider - ingenio 2010 pau ( csd2007 - 00060 ) , as well as from the european union marie curie initial training network unilhc granto no . pitn - ga-2009 - 237920 . we also acknowledge the support of the spanish mineco s centro de excelencia severo ochoa " programme under grant no . sev-2012 - 0249 . in this section we report the null - test in terms of redshift @xmath25 that we used for binning the data and its derivatives with respect to four observables to evaluate the propagated error . the null - test @xmath109 becomes : we chose @xmath279 to be equal the first redshift available , hence all the quantities like @xmath280 and @xmath281 are the first binned values of the data . ( [ eq : nulltest - z - app ] ) depends also on @xmath7 which is in general a complicated parameter to measure , for this reason we use instead @xmath282 where @xmath211 is a parameter given by cosmic microwave background experiments and easy to measure with great accuracy . it is also important to notice that when @xmath279 approaches to 0 then we have that @xmath283 , however , this term should never be thought as the real hubble constant ( like the one appearing in the exponent ) but it has to be considered as the value of the hubble parameter at the lowest redshift because the only true hubble constant , i.e. that comes directly from the theory is the one appearing in the exponent ) . for the sake of completeness we also write the derivatives of eq . ( [ eq : nulltest - z - app ] ) with respect to the four observables that will be used to evaluate the propagated error on the quantity @xmath113 and these are : s. nesseris and d. sapone , arxiv:1409.3697 [ astro-ph.co ] . d. sapone , int . j. mod . phys . a * 25 * , 5253 ( 2010 ) . [ arxiv:1006.5694 [ astro-ph.co ] ] . s. tsujikawa , lect . notes phys . * 800 * , 99 ( 2010 ) . [ arxiv:1101.0191 [ gr - qc ] ] . p. a. r. ade _ et al . _ [ planck collaboration ] , astron . astrophys . [ arxiv:1303.5076 [ astro-ph.co ] ] . m. betoule _ et al . _ [ sdss collaboration ] , [ arxiv:1401.4064 [ astro-ph.co ] ] . @xcite m. kunz and d. sapone , phys . * 98 * ( 2007 ) 121301 [ astro - ph/0612452 ] . d. sapone and m. kunz , phys . d * 80 * , 083519 ( 2009 ) . [ arxiv:0909.0007 [ astro-ph.co ] ] . d. sapone , m. kunz and l. amendola , phys . d * 82 * , 103535 ( 2010 ) . [ arxiv:1007.2188 [ astro-ph.co ] ] . d. sapone and e. majerotto , phys . d * 85 * ( 2012 ) 123529 [ arxiv:1203.2157 [ astro-ph.co ] ] . d. sapone , e. majerotto , m. kunz and b. garilli , phys . d * 88 * ( 2013 ) 043503 [ arxiv:1305.1942 [ astro-ph.co ] ] . l. amendola , m. kunz and d. sapone , jcap * 0804 * ( 2008 ) 013 [ arxiv:0704.2421 [ astro - ph ] ] . s. tsujikawa , phys . d * 76 * ( 2007 ) 023514 [ arxiv:0705.1032 [ astro - ph ] ] . s. nesseris , phys . d * 79 * , 044015 ( 2009 ) [ arxiv:0811.4292 [ astro - ph ] ] . s. nesseris and a. mazumdar , phys . d * 79 * , 104006 ( 2009 ) [ arxiv:0902.1185 [ astro-ph.co ] ] . a. b. belloso , j. garcia - bellido and d. sapone , jcap * 1110 * ( 2011 ) 010 [ arxiv:1105.4825 [ astro-ph.co ] ] . v. silveira and i. waga , phys . d * 50 * , 4890 ( 1994 ) . w. j. percival , astron . astrophys . * 443 * , 819 ( 2005 ) [ arxiv : astro - ph/0508156 ] . abramowitz , milton ; stegun , irene a. , eds . ( 1972 ) , _ handbook of mathematical functions with formulas , graphs , and mathematical tables _ , new york : dover publications , isbn 978 - 0 - 486 - 61272 - 0 c. clarkson , b. bassett and t. h. c. lu , phys . * 101 * , 011301 ( 2008 ) [ arxiv:0712.3457 [ astro - ph ] ] . v. sahni , a. shafieloo and a. a. starobinsky , phys . d * 78 * , 103502 ( 2008 ) [ arxiv:0807.3548 [ astro - ph ] ] . l. m. wang and p. j. steinhardt , astrophys . j. * 508 * , 483 ( 1998 ) [ arxiv : astro - ph/9804015 ] . s. capozziello , s. nesseris and l. perivolaropoulos , jcap * 0712 * , 009 ( 2007 ) [ arxiv:0705.3586 [ astro - ph ] ] . s. nesseris and l. perivolaropoulos , phys . d * 77 * , 023504 ( 2008 ) [ arxiv:0710.1092 [ astro - ph ] ] . s. nesseris , phys . d * 88 * , 123003 ( 2013 ) [ arxiv:1309.1055 [ astro-ph.co ] ] . a. de felice , d. f. mota and s. tsujikawa , phys . d * 81 * , 023532 ( 2010 ) [ arxiv:0911.1811 . g. lemaitre , gen . * 29 * , 641 ( 1997 ) [ annales soc . phys . a * 53 * , 51 ( 1933 ) ] . r. c. tolman , proc . sci . * 20 * , 169 ( 1934 ) . h. bondi , mon . not . soc . * 107 * , 410 ( 1947 ) . c. clarkson , t. clifton , s. february , jcap * 0906 * ( 2009 ) 025 . [ arxiv:0903.5040 [ astro-ph.co ] ] . j. garcia - bellido , t. haugblle , jcap * 0909 * ( 2009 ) 028 . [ arxiv:0810.4939 [ astro - ph ] ] . d. alonso , j. garcia - bellido , t. haugboelle and a. knebe , phys . dark univ . * 1 * ( 2012 ) 24 [ arxiv:1204.3532 [ astro-ph.co ] ] . j. garcia - bellido , t. haugblle , jcap * 0804 * ( 2008 ) 003 . [ arxiv:0802.1523 [ astro - ph ] ] . j. garcia - bellido , t. haugblle , jcap * 0809 * ( 2008 ) 016 . [ arxiv:0807.1326 [ astro - ph ] ] . m. zumalacarregui , j. garcia - bellido and p. ruiz - lapuente , jcap * 1210 * ( 2012 ) 009 [ arxiv:1201.2790 [ astro-ph.co ] ] . d. alonso , j. garcia - bellido , t. haugblle and j. vicente , phys . d * 82 * ( 2010 ) 123530 [ arxiv:1010.3453 [ astro-ph.co ] ] . s. basilakos , s. nesseris and l. perivolaropoulos , phys . d * 87 * , no . 12 , 123529 ( 2013 ) [ arxiv:1302.6051 [ astro-ph.co ] ] . l. taddei and l. amendola , [ arxiv:1408.3520 [ astro-ph.co ] ] . m. j. hudson and s. j. turnbull , astrophys . j. * 751 * , l30 ( 2013 ) . [ arxiv:1203.4814 [ astro-ph.co ] ] . f. beutler , c. blake , m. colless , d. h. jones , l. staveley - smith , g. b. poole , l. campbell and q. parker _ et al . _ , mon . not . 423 * , 3430 ( 2012 ) . [ arxiv:1204.4725 [ astro-ph.co ] ] . l. samushia , w. j. percival and a. raccanelli , mon . not . soc . * 420 * , 2102 ( 2012 ) . [ arxiv:1102.1014 [ astro-ph.co ] ] . r. tojeiro , w. j. percival , j. brinkmann , j. r. brownstein , d. eisenstein , m. manera , c. maraston and c. k. mcbride _ et al . _ , [ arxiv:1203.6565 [ astro-ph.co ] ] . y. s. song and w. j. percival , jcap * 0910 * , 004 ( 2009 ) . [ arxiv:0807.0810 [ astro - ph ] ] . l. guzzo , m. pierleoni , b. meneux , e. branchini , o. l. fevre , c. marinoni , b. garilli and j. blaizot _ et al . _ , nature * 451 * , 541 ( 2008 ) . [ arxiv:0802.1944 [ astro - ph ] ] . c. blake , s. brough , m. colless , c. contreras , w. couch , s. croom , d. croton and t. davis _ et al . _ , mon . not . 425 * , 405 ( 2012 ) . [ arxiv:1204.3674 [ astro-ph.co ] ] . s. de la torre , l. guzzo , j. a. peacock , e. branchini , a. iovino , b. r. granett , u. abbas and c. adami _ et al . _ , [ arxiv:1303.2622 [ astro-ph.co ] ] . c. h. chuang and y. wang , mon . not . soc . * 435 * , 255 ( 2013 ) . [ arxiv:1209.0210 [ astro-ph.co ] ] . c. h. chuang , f. prada , f. beutler , d. j. eisenstein , s. escoffier , s. ho , j. p. kneib and m. manera _ et al . _ , [ arxiv:1312.4889 [ astro-ph.co ] ] . l. samushia , b. a. reid , m. white , w. j. percival , a. j. cuesta , g. b. zhao , a. j. ross and m. manera _ et al . _ , mon . not . soc . * 439 * , 3504 ( 2014 ) . [ astro-ph.co ] ] . m. moresco _ et al . _ , jcap * 1208 * , 006 ( 2012 ) [ arxiv:1201.3609 [ astro-ph.co ] ] . e. gaztanaga , a. cabre and l. hui , mon . not . * 399 * ( 2009 ) 1663 [ arxiv:0807.3551 [ astro - ph ] ] . c. blake , s. brough , m. colless , c. contreras , w. couch , s. croom , d. croton and t. davis _ et al . _ , mon . not . * 425 * ( 2012 ) 405 [ arxiv:1204.3674 [ astro-ph.co ] ] . l. anderson _ [ boss collaboration ] , arxiv:1312.4877 [ astro-ph.co ] . a. font - ribera , d. kirkby , n. busca , j. miralda - escude , n. p. ross , a. slosar , j. rich and e. aubourg _ et al . _ , jcap * 1405 * ( 2014 ) 027 [ arxiv:1311.1767 ] . l. amendola _ et al . _ [ euclid theory working group collaboration ] , living rev . * 16 * ( 2013 ) 6 [ arxiv:1206.1225 [ astro-ph.co ] ] . a. abate _ et al . _ [ lsst dark energy science collaboration ] , arxiv:1211.0310 [ astro-ph.co ] . s. nesseris , c. blake , t. davis and d. parkinson , jcap * 1107 * , 037 ( 2011 ) [ arxiv:1107.3659 [ astro-ph.co ] ] . s. nesseris and a. shafieloo , mon . not . 408 * , 1879 ( 2010 ) [ arxiv:1004.0960 [ astro-ph.co ] ] .

in the gaseous envelope of protogalaxies , thermal instability leads to the formation of a population of cool fragments which are confined by the pressure of a residual hot background medium . in order to remain in a quasi - hydrostatic equilibrium , the residual gas evolves at approximately the virial temperature of the dark matter halo . its density is determined by the requirements of thermal equilibrium . the hot gas is heated by compression and shock dissipation . the heating is balanced by direct energy loss due to bremsstrahlung emission , and by conductive losses into the cool clouds , which are efficient radiators . the cool fragments are photoionized and heated by the extragalactic uv background and nearby massive stars . several processes interact to determine the size distribution of the cool fragments . the smallest are evaporated due to conductive heat transfer from the hot gas . all fragments are subject to disruption due to hydrodynamic instabilities . the fragments also gain mass due to collisions and mergers , and condensation from the hot gas due to conduction . the size distribution of the fragments in term determines the rate and efficiency of star formation during the early phase of galactic evolution . we have performed one - dimensional hydrodynamic simulations of the evolution of the hot and cool gas . the cool clouds are assumed to follow a power - law size distribution , and fall into the galactic potential , subject to drag from the hot gas . the relative amounts of the hot and cool gas is determined by the processes discussed above , and star formation occurs at a rate sufficient to maintain the cool clouds at 10 k. we present density distributions for the two phases and also for the stars for several cases , parametrized by the circular speeds of the potentials . under some conditions , primarily low densities of the hot gas , conduction is more efficient than radiative processes at cooling the hot gas , limiting the x - ray radiation from the halo gas . lick observatory bulletin no .

astro-ph0004055

according to cdm models of galaxy formation ( blumenthal et al . 1984 ; navarro , frenk , & white 1997 ; klypin , nolthenius , & primack 1997 ) , large galaxies emerge through the coalescence of smaller systems , normally identified as dwarf galaxies , which contain mostly dark matter with a fraction of gas . the dark matter components merge with each other in a dissipationless manner to form the halos of larger galaxies . the crossing of gas streamlines leads to shocks , heating the gas to the virial temperature of the galactic halo ( binney 1977 ; rees & ostriker 1977 ; white & rees 1978 ) . in order for the heated gas in this extended halo ( hereafter , the protogalactic cloud or pgc ) to collapse , its cooling timescale @xmath2 must be shorter than its dynamical time scale , @xmath3 during each growth stage of the galaxy . when large galaxies acquire a mass comparable to the galaxy , this condition is satisfied once the characteristic length scale of the pgc is @xmath4 kpc ( blumenthal et al . 1984 ) . for typical values ( @xmath5 k ) of the virial temperature , the cooling timescale increases with temperature , and the pgc s are thermally unstable ( field 1965 ) . thermal instability leads to the rapid growth of perturbations and fragmentation of pgcs ( murray & lin 1990 ) . the result is that a two - phase medium develops during the initial cooling of the pgc , in which a population of warm fragmentary clouds ( wfc s ) are confined by the pressure of hot , residual halo gas ( rhg ) ( burkert & lin 2000 ) . the rhg is primarily heated by the release of the gravitational energy of the collapsing pgc and cooled by radiative emission and conductive transport . the wfc s settle into the central region of the halo potential . they are unable to cool below 10@xmath0 k until their density reaches a sufficiently high value that the wfc s become self - shielded from external photodissociating uv radiation ( couchman & rees 1986 ; haiman , rees , & loeb 1997 ; haiman , abel , & rees in press ) . thereafter , molecular hydrogen can form within them as a consequence of non - equilibrium cooling . efficient cooling reduces the temperature to @xmath6 k even in the absence of metals . consequently , wfc s evolve into cold molecular clouds ( cmcs ) . the formation of massive stars within the cmc s provides ultraviolet heating , limiting the formation rate of cmc s from wfc s , and leading to the formation of an equilibrium three - phase structure in the galaxy . in this paper , we discuss the evolution of the hot phase that is left after the initial thermal instability . detailed investigations of the warm and cold phases will be made in upcoming papers . for the current work , we assume a power - law size distribution for the wfc s . the cmc s are not included explicitly , but star formation within them is explicitly assumed to occur at a rate which would provide an adequate flux of uv photons to heat and ionize the wfc s . in 2 , we briefly recapitulate the physical processes associated with this scenario . in 3 , we determine the energy balance between the warm and hot phases , while in 4 , we discuss the resulting star formation rate in the wfc s . in 5 , we show model results on the evolution of the warm and hot phases . finally , we summarize our results and discuss their implications in 6 . the temperature of a large , virialized pgc , @xmath7 k , where @xmath8 is the circular velocity of the halo potential , @xmath9 is the mean atomic weight in grams , and @xmath10 is the boltzmann constant . at such temperatures , the dominant radiative cooling mechanisms are bremsstrahlung and recombination processes ( dalgarno & mccray 1972 ) . the radiative cooling rates for metal poor clouds ( with [ fe / h]@xmath11 ) are @xmath12 for @xmath13 k @xmath14 k , and @xmath15 for @xmath16 k. equation ( 1 ) is a highly simplified average representation of the behavior of the cooling rate in the given temperature range . in actuality , there is a peak due to h emission at @xmath13 k , another due to emission by he@xmath17 at 10@xmath18 k , and a minimum at @xmath19 k , before the cooling efficiency rises again , due to bremsstrahlung emission ( gould & thakur 1970 ) . the cooling timescale is @xmath20 where @xmath21 and @xmath22 are the mass and number density of the gas . the necessary collapse condition , @xmath23 where the dynamical timescale @xmath24 ( white & rees 1978 ) , for a virialized pgc is satisfied when @xmath22 exceeds a critical value @xmath25 where @xmath26 kpc , and r is the galactocentric radius . above @xmath27 k , @xmath2 is an increasing function of @xmath28 , such that the contrast between slightly cooler regions and the background grows during the cooling process . the relatively rapid loss of entropy also leads to a deficit of pressure , @xmath29 , in the clouds compared to their background . the higher pressure of the background compresses the clouds in an attempt to maintain pressure balance . the rise in @xmath22 leads to a reduction in @xmath2 for the clouds , further enhancing the dichotomy in cooling timescale between the clouds and the background . the interface separating cool clouds from the background retreats at an accelerating pace . although @xmath30 at the onset of thermal runaway , it decreases rapidly to @xmath31 as the perturbed region cools significantly below the virial temperature . clouds , with radius @xmath32 and sound speed @xmath33 , establish hydrostatic equilibrium with the rhg and undergo isobaric evolution on a timescale @xmath34 . if this timescale is @xmath35 , @xmath22 can not adjust such that cooling is isochoric . relatively large clouds undergo a transition from isobaric to isochoric cooling before their density becomes significantly larger than that of the background . this transition in small clouds occurs at a later cooling stage . their density growth becomes nonlinear before the cooling becomes isochoric . consequently , they emerge as dense , cool fragments ( burkert & lin 2000 ) . when @xmath28 in the clouds decreases to @xmath36k , recombination becomes the dominant cooling process . however , these clouds are also exposed to uv radiation due to emission from background active galactic nuclei . at redshift @xmath37 , the flux of this emission at the lyman limit is estimated to be @xmath38 ergs hz@xmath39 ster @xmath39 @xmath40 s@xmath39 , with a dropoff towards higher energies as steep as @xmath41 ( haardt & madau 1996 ) . the above value varies approximately as @xmath42 . in the outer regions of pgc where the pressure of the rhg is low , the internal density of these clouds is also relatively small , such that they may be mostly photoionized , with @xmath43k . but , in the inner regions of the pgc where the background pressure is higher , wfc s have larger internal density , such that they become self shielded against the extragalactic uv flux and their temperature cools below @xmath44 k due to emission by h@xmath45 and metal ions ( dalgarno & mccray 1972 ; hollenbach & mckee 1979 ) . when these clouds become gravitationally unstable , stars rapidly form within them . the uv flux provided by these massive stars is able to heat gas within their strmgren radii to @xmath46 k , quenching star formation in those regions . at all galactocentric radii , then , the pgc fragments into a population of wfc s which are maintained at @xmath46 k by the uv flux . in most regions , the ionizing photons are emitted by a population of massive stars whose formation rate is self - regulated . the wfc s are embedded within rhg , whose density is sufficiently low that @xmath47 , ie . it is thermally stable ( field 1965 ) , and it can remain near the virial temperature of the galaxy . the pressure of the rhg confines all but the largest wfc s , for which self gravity is important . consequently , the pgc becomes a two - phase medium , with the density contrast between the wfc s and the rhg inversely proportional to their temperature ratio ( @xmath48 ) . following its collapse into the potential of the galactic halo , the rhg is shock - heated to the virial temperature of the potential , and rapidly attains a quasistatic equilibrium . it then adjusts to have density and temperature appropriate for the maintenance of thermal equilibrium constrained by the processes discussed below . the inverse buoyancy of the pressure - confined wfc s causes them to settle towards the center of the potential provided by the dark matter halo . during their descent , the wfc s experience a drag by the rhg , reaching a terminal speed relative to the rhg given by @xmath49 where @xmath50 is the ratio of the density of the wfc s to that of the rhgs . ( in all relations below , the subscripts `` h '' and `` w '' refer to quantities of the rhg and of the wfc , respectively . ) for wfc s with sizes @xmath51 , @xmath52 . here , we neglect the effect of the clouds size evolution on their kinematics . the value of @xmath53 derived in equation ( 5 ) assumes a constant density background . in reality , the pressure gradient of the background causes the density of the rhg to rise towards the center of the galaxy . the radii of the wfc s @xmath54 , and so the pressure gradient across an individual cloud is negligible . if the rhg is approximately isothermal , and the temperature of the wfc s are determined by atomic cooling to be @xmath55 k , then @xmath56 remains constant during the motion of the clouds . as a result , the primary effect of the density gradient of the rhg upon @xmath53 is in altering the ratio @xmath57 . if @xmath58 , then , under the above assumptions , @xmath59 , @xmath60 , and @xmath61 . the motion of the wfc s through the rhg also leads to mass loss from the wfc s due to kelvin - helmholtz instability ( murray et al . 1993 ) . in the limit @xmath62 , the growth timescale of kh instability is given by @xmath63 short wavelength perturbations rapidly saturate . perturbations with wavelengths @xmath64 provide the dominant contribution to mass loss , which occurs at a rate @xmath65 for a wfc with a mass @xmath66 , the stripping timescale @xmath67 , _ i.e. _ wfc s are disintegrated before they settle to the center of the galactic potential . thus , both thermal instability associated with the cooling , and kelvin - helmholtz instability arising during the infall of the wfc s leads to the formation of wfc s with arbitrarily small sizes . the break down process increases the collective area filling factor of the wfc s , leading to an increase in their collision frequency . collisions and the resulting mergers increase the masses of the wfc s . when the rates of coagulation and disruption ( including other processes discussed below ) balance with each other , an equilibrium size distribution is established ( dong et al . in preparation ) . the existence of wfc s in the rhg also leads to heat transfer from the rhg to the wfc s through conduction ( mckee & cowie 1977 ) . this process is particularly important for small wfc s . the conductive heat flux into a wfc is given by @xmath68 where @xmath69 ( spitzer 1956 ) . to order of magnitude , @xmath70 where @xmath71 is estimated as the depth into the wfc s over which radiative emission balances conduction into the wfc s , and @xmath72 is the cooling efficiency within the wfc . in terms of the density contrast , @xmath73 where @xmath74 and @xmath75 are the mean molecular weights in the rhg and wfc s , respectively , and @xmath76 . in small clouds where the density scale height is small , the magnitude of @xmath77 is limited by the saturated flux , @xmath78 where @xmath79 is the isothermal sound speed in the rhg ( @xmath80 220 km s @xmath39 for the galaxy ) . the presence of magnetic field reduces the mean free path of the electrons and @xmath77 ( rosner & tucker 1989 ) . in our calculations , we neglect the effect of magnetic field and adopt @xmath81 based on our model parameters . large wfc s , which attain large @xmath53 , may also advect thermal energy from the rhg to the wfcs at a rate @xmath82 as discussed above , the sizes of the wfc s are modified by kelvin - helmholtz instability and coagulation before they are accelerated to @xmath53 . the wfc s may therefore spend most of their time with smaller velocities relative to the rhg ( murray et al . the advective heating is therefore expected to be weaker than that due to conduction . energy equilibrium is established within a cloud when @xmath83 the photoionization heating @xmath84 where @xmath85 is the ionization fraction in the wfc s , @xmath86 is planck s constant , @xmath87 is the mean intensity in ergs ster@xmath39 hz@xmath39 s@xmath39 , @xmath88 is the photoionization cross section as a function of frequency , @xmath89 ev , and @xmath90 depends upon the shape of the spectrum ( @xmath91 for realistic background spectra ) . this equilibrium determines the temperature of the wfc s . the above assumes a top - hat density distribution within the clouds . this is appropriate for clouds which are at most weakly self - gravitating , given the sharp temperature transition between the wfc s and the rhg resulting from thermal instability . in a mostly ionized wfc , the heat input due to conduction is larger than that due to photoionization when @xmath92 in this limit , the rate of conductive heat input exceeds that of the radiative loss due to recombination when @xmath93 in the limit of saturated conduction and @xmath94 in the limit of unsaturated conduction . the above results also apply to the limit in which the background uv flux is negligibly small , so that the gas in wfc s is mostly neutral . for the intermediate case in which photoionization contributes significantly to the heating , the additional energy input would tend to increase the value of @xmath95 and @xmath96 . wfc s with @xmath97 ( or @xmath95 ) are unstable to conductive heat transport . their inability to radiate away all the conductive heat input leads to an increase in their temperatures . in order to maintain pressure equilibrium with the rhg , these wfc s expand , decreasing their internal density . the increase in @xmath32 enhances conductive heat transport into the clouds , whereas their cooling efficiency decreases with the decrease in @xmath98 . unless @xmath72 increases rapidly with @xmath99 , conduction then leads to the total evaporation of these small wfc s . the stability criterion can be obtained by analyzing the dependence of @xmath96 , @xmath95 , and @xmath32 upon @xmath99 . if we approximate @xmath100 , then from equations 15 and 16 we find that both @xmath96 and @xmath95 vary as @xmath101 . as the pressure - confined wfc s are heated and expand , however , their radii vary as @xmath102 . the clouds are therefore unstable if @xmath103 . in that case , we can picture a cloud with @xmath104 initially . as it is heated and expands , the steep dependence of @xmath96 upon @xmath99 leads to @xmath97 . radiative cooling can then no longer offset conduction into the cloud , leading to further temperature increase and expansion . the same stability criterion of @xmath105 holds for clouds where @xmath95 is the appropriate limiting radius . this stability condition is met over the temperature range dominated by hydrogen emission , @xmath106 k , such that clouds with @xmath97 or @xmath107 remain stable until their temperatures reach 20,000 k. the minimum possible stable cloud size is therefore set by the value of @xmath96 or @xmath95 at 20,000 k , where @xmath108 is a maximum . in the above analysis , the wfc s are assumed to be confined by the ambient pressure of the rhg . if the wfc s are able to cool to @xmath109 k , then self - gravity contributes to the confinement of even relatively small clouds . in this limit , thermal equilibrium between photoionization heating and radiative losses may still be established , but it is unstable because of the negative specific heat of self - gravitating clouds ( murray & lin 1992 ) . the result is similar to that for non self - gravitating clouds , except that , for a self - gravitating cloud , the radius varies as @xmath110 , and stability requires @xmath111 , which is , again , only satisfied over a very narrow range of temperatures . while the lower limit of cloud radii is set by their ability to radiate away the energy flux due to conduction in equations ( 15 ) and ( 16 ) , the upper limit of cloud radii is set by gravitational instability . low mass clouds are confined by the pressure of the rhg . at higher masses , self - gravity becomes important . clouds become unstable above a critical mass , which depends upon the external pressure @xmath112^{1\over2},\ ] ] where @xmath113 is the pressure of the rhg , and @xmath99 is the internal temperature of the clouds ( ebert 1955 ; bonner 1956 ) . this condition corresponds to an approximate maximum cloud radius of @xmath114 we have adopted the bonner - ebert criterion for stability , rather than the jeans criterion , due to the fact that the wfc s are pressure - confined . while there might be an apparent inconsistency , in that we assume the wfc s to have nearly constant density , whereas the bonner - ebert criterion assumes a density gradient . had we used the jeans criterion , however , our value for @xmath115 would be very close to that used above , and so the choice of stability criterion does not affect our results below . the above results assume that the wfc s are subject only to heating by photoionization and by conduction from the hot gas . in addition , the drag of the clouds through the rhg , and the growth of kelvin - helmholtz instability will both lead to additional mechanical energy input into the wfc s which could , in principle , provide additional support for the clouds . these mechanisms are least important , however , in the largest clouds , which experience the least drag , and whose self - gravity stabilizes the growth of deeply - penetrating , long wavelength perturbations by kelvin - helmholtz instability ( murray et al . 1993 ) . stability of the most massive clouds should therefore still be governed by the bonner - ebert criterion . detailed models of clouds moving through background gas will be used in future work to confirm these conclusions . the physical processes involved in determining the mass spectrum of the wfc s will be examined in a future paper . in the results below , we assume that the fragmentation and coagulation of the wfc s leads to a power - law size distribution of the wfc s , in which the number density of clouds within the pgc , @xmath116 , per unit cloud radius @xmath32 is given by @xmath117 the normalization constant , @xmath118 , is related to the spatially averaged density of the warm phase by @xmath119 where @xmath120 and @xmath121 are the maximum and minimum cloud size , respectively . as discussed above , @xmath121 is determined by the smallest size of clouds which can radiate away the energy conducted into them from the rhg , while @xmath120 is determined by the bonner - ebert mass . the value of @xmath118 is a function of galactocentric radius , whereas we assume @xmath122 to be the same everywhere within the pgc . if the clouds follow a distribution close to that observed within galactic cloud complexes , for which @xmath123 ( scalo 1985 ; zinnecker , mccaughrean , & wilking 1993 ) , then the results above and below are dominated by the massive end of the distribution . alternatively , if the efficiency of cloud disruption is high relative to the rate of coagulation , then the cloud distribution will be dominated by the low - mass end . the determination of which limit is appropriate shall be made in an upcoming paper . in the current work , we treat the size distribution as an uncertainty , and consider two extreme limits . in the first , we assume @xmath124 , and all terms derived both above and below are dominated by the high mass end of the cloud distribution . in the second , we assume @xmath125 , and all terms are dominated by the low mass end of the distribution . the total rate of conductive energy transport per unit volume from the rhg into the wfc s is given by @xmath126 using the results above , this becomes @xmath127 for @xmath124 , the above result becomes @xmath128 while for @xmath129 it is @xmath130 the above results assume that @xmath131 where @xmath132 and @xmath133 . as can be seen by a comparison of equations ( 15 ) and ( 18 ) , @xmath134 and so this is a reasonable approximation . also , because @xmath135 , the greater surface - to - volume ratio of the case @xmath129 leads to a significantly greater cooling rate due to conduction into the wfc s relative to the case with @xmath124 . equations ( 23 ) and ( 24 ) give the conductive energy transfer rate from the rhg to the wfc s . the rhg also experiences radiative losses , mainly through bremsstrahlung emission , which occurs at a rate given by equation ( 2 ) . using the conductive flux given in equations ( 23 ) and ( 24 ) , we find that conductive losses into the wfc s exceed bremsstrahlung cooling in the limit that @xmath136 when conduction is saturated , and @xmath137 when conduction is unsaturated . in the above relations , we assume pressure balance between the warm and hot phases , ie . @xmath138 heat is supplied to the rhg through its own quasi - static contraction . more importantly , heat is also deposited into the rhg as a result of the frictional drag on collapsing wfc s , by which the wfc s are able to transfer a fraction of their gravitational potential energy into thermal energy of the rhg . ( in the absence of any drag , the total energy of individual wfc s is approximately conserved as they fall into the galactic halo potential . ) the heating per unit volume of a flux of wfc s moving at their terminal speeds is given by @xmath139 using @xmath53 from eq ( 6 ) , the above result becomes @xmath140 for @xmath141 , the above result is dominated by the contribution of the largest wfc s , for which @xmath53 may become comparable to @xmath8 at small @xmath142 . whether or not the clouds are able to attain @xmath53 depends upon the ratio of their acceleration timescales to the timescales upon which they are affected by conduction , collisions , or kelvin - helmholtz instability . for a cloud to be accelerated to a speed @xmath143 , it must travel a distance @xmath144 in the halo , whereas , kelvin - helmholtz instability leads to significant mass loss by the time a cloud has moved through the rhg over a distance of several times its diameter ( murray et al . 1993 ) . unless the rhg is also contracting with a velocity @xmath145 , the massive wfc s may not be able to attain their terminal velocities ( dong et al . in preparation ) . instead of @xmath53 above , we parametrize the terminal differential velocity as @xmath146 , with @xmath147 being a free parameter . using this prescription , the heating rate is given by @xmath148 in the limit @xmath125 , the heating is dominated by drag from low mass clouds , for which @xmath149 . the above result then becomes @xmath150 if the radiative and conductive cooling of the rhg exceed the thermal energy input , some of the rhg will precipitate to form additional wfc s . as the gas density of the rhg is depleted , both the conductive heat flux into , and drag on the infalling wfc s decrease . the density @xmath151 therefore adjusts so as to achieve thermal equilibrium in which @xmath152 . in the above analysis , the wfc s are generally taken to be highly ionized . to maintain this ionization state , the uv flux to which the wfc s are exposed must be sufficiently large to offset recombination . in figure 1 , we show the depths into which a cloud may be ionized by various sources of uv flux , plotted as a function of @xmath22 . these results were obtained with cloudy ( ferland 1991 ) . the solid , dashed , and dot - dashed curves represent results obtained with the standard agn spectral energy distribution , for three values of @xmath153 . the solid curve was calculated using a value of @xmath154 ergs hz@xmath39 ster@xmath39 @xmath40 s@xmath39 at 13.6 ev , representing the uv background at z=2.5 ( haardt & madau 1996 ) . the dashed and dot - dashed curves were computed using fluxes that were , respectively , smaller and larger by an order of magnitude , approximately representing redshifts @xmath155 and 5 respectively . the dotted curve was computed using the kurucz spectrum in cloudy , with the flux set to that of an o5 star at a distance of 10 pc . from figure 1 , we can infer the integrated depth into the pgc outside of which all of the warm fragments are ionized . in the early evolutionary epochs of the galaxy before most of the ordinary matter was converted into stars , only the regions outside @xmath156 kpc are photoionized by the uv background ( see below ) . at smaller galactocentric radii , local sources of uv radiation are required . in diffuse and low mass dwarf galaxies , however , the entire galaxy may be photoionized by the diffuse background radiation ( kepner , babul , & spergel 1997 ) . in addition to the extragalactic background , the rhg radiates soft x - ray photons following a bremsstrahlung spectrum . previous work ( fall & rees 1985 ) , which used analytic approximations to estimate the structure of the rhg , found the resulting x - ray flux to be insufficient to ionize large clouds beyond a few kpc from the galactic center . in the models presented below , we find the bremsstrahlung luminosities to be less than estimated by fall & rees ( 1985 ) . ionization by the rhg is therefore not a significant contributor to determining the state of the wfc s . hydrodynamic drag on wfc s also leads to energy input into them , as described in 3.4 . for relatively large wfc s @xmath157 km s@xmath39 . if these fragments are not ionized initially , then shocks will cause internal heating . from above , though , the ratio of heating by drag to that by conduction is found to be @xmath158 thus , even for velocities as high as 100 km s@xmath39 , the heating rate due to drag is negligible . for survivable clouds with @xmath159 and @xmath96 , conduction does not provide adequate heat transfer to offset radiative loses . since the dominant radiative process at @xmath36k is recombination , conductive heat flow is also insufficient to ionize the wfc s . in the absence of an adequate ionizing photon flux , the cooling efficiency of bremsstrahlung , recombination , and atomic hydrogen emission decrease rapidly below @xmath36 k. in a metal - free protogalactic cloud , however , non - equilibrium recombination leads to the formation of a small amount of @xmath160 ions which recombine with neutral @xmath161 to form @xmath162 . radiative losses due to rotational and vibrational transitions of @xmath162 reduce the gas temperature to @xmath163 k ( murray & lin 1990 ) . if [ fe / h]@xmath164 , even lower temperatures ( @xmath165 k ) are attainable due to cooling by heavy elements ( primarily by cii , co , and silicate grains ) ( hellsten & lin 2000 ) , leading to the formation of cold molecular clouds ( cmc s ) . when the temperature of the cmc s is reduced by over two orders of magnitude from that of the wfc s , their bonner - ebert masses are reduced by over four orders of magnitude ( cf . equation 17 ) . self - gravity is therefore much more important for the evolution of the cmc s than for that of the wfc s . cloudlets close to the bonner - ebert mass are centrally condensed . when their mass exceeds the bonner - ebert mass , thermal pressure can no longer support the weight of the envelope , and the cloudlets undergo inside - out collapse ( shu 1977 ) . during the collapse the jean s mass decreases with the increasing density . due to the centrally - concentrated cloud structure , however , the jeans mass is always larger than the mass contained inside any given radius . the collapse is therefore stable , and does not lead to fragmentation in the absence of further unstable cooling , such that the initial value of @xmath166 represents the minimum mass for isothermal collapsing clouds ( tsai 1992 ) . numerical simulations show that even rotating centrally - concentrated clouds have difficulties in breaking up into fragments with a small fraction of @xmath166 ( burkert & bodenheimer 1996 ) . this conjecture is in contrast to the conventional opacity - limited fragmentation scenario ( hoyle 1953 ; low & lynden - bell 1976 ) which may be more appropriate for the collapse of unstable homogeneous clouds ( burkert & bodenheimer 1993 ) . the initial conditions required by the opacity - limited fragmentation scenario may be difficult to accomplish in nature . in a metal - free environment , the temperature of the cfc s , @xmath167 k and @xmath168 . stars formed in such a metal - poor environment are massive and short - lived , consistent with their rarity today ( hellsten & lin 2000 ) . the early chemical enrichment is dominated by the output of type ii supernovae , consistent with the abundance distribution observed among stars with [ fe / h]@xmath169 ( wheeler , sneden , & truran 1989 ) . the massive stars are also copious sources of uv radiation . the effect is twofold . photons in the range 11 - 13.6 ev are able to penetrate clouds and photodissociate h@xmath45 within them , unless their column densities are so high the clouds are self - shielded . higher energy photons ionize and heat the gas to 10@xmath0 k. the loss of coolants , and increased heating means that a population of @xmath170 o5 stars is adequate to photoionize all the cmc s , re - heat them to @xmath43 k , and prevent the formation of any additional cmc s out to a few tens of kpc . photoionization heating also increases @xmath166 to @xmath171 , stablizing small cloudlets , and quenching star formation . as the massive stars evolve off the main sequence , the uv flux decreases , cooling again leads to @xmath167 k in sheltered regions , and spontaneous star formation is resumed . thus , the formation rate of massive stars may be estimated from the assumption that it is self - regulated at a level necessary to sustain marginal ionization of all wfc s . this self - regulated star formation rate naturally yields values of [ fe / h ] comparable to that observed in halo stars ( lin & murray 1992 ) . the rate at which massive stars must be formed is determined by the need to balance ionization with recombination in the wfc s . the mean free path for lyman continuum photons is large in the rhg , but not in the wfc s . for sufficiently high covering factor , we can assume that a large fraction of the uv photons produced by hot stars are absorbed locally . in equilibrium then , the total production rate of uv photons per unit volume must balance the total recombination rate per unit volume . this radiative equilibrium results in high ionization in the limit @xmath172 where @xmath173 is the number density of stars , @xmath174 is the average uv output in photons s@xmath39 per star , and @xmath175 is the case b recombination coefficient ( osterbrock 1989 ) . the production of uv photons is dominated by o stars , with lifetimes @xmath176 yr . the average mass loss rate per unit volume from the warm phase due to star formation is then given by @xmath177 where @xmath178 is the mass of stars that form to produce @xmath174 photons per unit time . if the stars follow a salpeter mass function , with a minimum mass of 0.1 m@xmath179 and maximum mass of 100 m@xmath179 , then @xmath180 the rate of mass depletion by star formation in the warm phase derived above is independent of the size distribution of the wfc s , and depends only upon their filling factor and internal density . we have performed one - dimensional lagrangian hydrodynamic models of the evolution of the rhg subject to the physical processes discussed above . the models treat the warm and hot phases as separate fluids , evolving within a fixed external gravitational potential set by the dark matter . angular momentum of the gas is not included . the models are applicable to the very early stages of star formation in galaxies or after mergers , or to cooling flows , before the gas has contracted sufficiently for its motion is dominated by angular momentum . during the later disk formation phase of a galaxy , or in the innermost regions of mergers or cooling flows , angular momentum would play an important role , whereas for the situations modelled here it would not strongly affect the physical processes under consideration . the momentum equation for the rhc @xmath181 includes contributions from the gas pressure , @xmath113 , the dark matter potential , @xmath182 , and drag from the infalling wfc s . in equation ( 35 ) , @xmath183 is the radial velocity of the rhg , and @xmath184 . for computational simplicity , the gravitational potential is taken to be isothermal out to a radius of 50 kpc , such that @xmath185 . the magnitude of @xmath182 is held constant within a core radius @xmath186 kpc and dark matter is assumed to vanish outside of 50 kpc , such that @xmath182 becomes keplerian . although this prescription of @xmath182 is highly simplified , our results do not depend sensitively upon the detailed form of @xmath182 . the form of the drag term is @xmath187 as noted above , in models where high mass clouds dominate , we set @xmath188 , whereas equation ( 5 ) is used in models where low mass clouds dominate . the equation of continuity for the rhg , @xmath189,\ ] ] includes a sink term , due to condensation of gas out of the hot phase . the amount of mass lost from the hot phase depends upon the relative magnitudes of heating and cooling . if heating either exceeds or equals cooling in the rhg , then @xmath190 . if cooling exceeds heating in the rhg , however , mass is transferred from the rhg to the wfc s at a rate given by @xmath191 where @xmath192 is the density which would be required for heating to balance cooling in the hot phase . the energy equation for the rhg , @xmath193 includes heating by drag from the warm phase , bremsstrahlung emission , and conduction losses into the warm phase . the warm phase is not pressure - supported in the galactic halo , and so the mean velocity of the wfc evolves as @xmath194 where the drag term is similar to that in equation ( 35 ) . a maximum infall velocity of @xmath146 is imposed upon the warm phase . as discussed above , the parameter @xmath147 depends upon the details of the evolution of the warm phase . in our models , we consider values of @xmath195 and 0.1 , thus examining the differences in our results for widely varying values of @xmath147 . the mass of the warm phase increases by mass loss from the hot phase , and decreases by mass loss due to star formation , following the prescription given in equation ( 33 ) , such that @xmath196.\ ] ] in principle , the smallest wfc s ( with @xmath197 or @xmath198 ) are continually being evaporated into the rhg . but , if this phase transition increases @xmath199 such that cooling exceeds heating in the rhg , then an appropriate amount of hot gas would precipitate to retain thermal equilibrium in the rhg . thus , the ratio @xmath200 depends indirectly on the size distribution of wfc s which determines the conductive flux between the two phases . finally , the wfc s are assumed to be ionized and heated by the uv flux of the nearby massive stars , such that @xmath201 k. they are also assumed to be in a pressure equilibrium with the rgh such that @xmath202 . equations ( 35)-(41 ) completely describe the @xmath199 , @xmath203 , @xmath204 , @xmath205 , and @xmath206 distribution . from these quantities , @xmath99 , @xmath207 , as well as the heat transfer luminosity due to bremstrahlung and conduction , @xmath208 and @xmath209 can be obtained . we solve these equations numerically with a 1-d lagrangian method ( richtmyer & morton 1995 ) . the warm and hot phases are treated as separate fluids , evolving and interacting as described above . the models described below use a grid of 400 cells for both fluids , with the cell spacing increasing outwards so as to give the best resolution in the region of interest , within 100 kpc . the outer boundary is set at 2 mpc , with a constant pressure . the large outer radius is chosen so that conditions at that boundary can not affect the solution . variations by factors of two in the number of cells are found to have no significant effect upon the results . the hydrodynamic equations are evolved explicitly , while the energy equation ( 38 ) is solved implicitly , allowing the use of a simple courant timestep criterion . the models are evolved until they reach a quasi - steady state , the results of which are shown in the following sections . using the approach outlined above , we now consider several simple models to describe the evolution of gaseous ordinary matter in the halo of large galaxies . in principle , we should concurrently consider the cosmological evolution which led to the formation of the halo . however , such an investigation would require not only a large dynamical range in dissipationless interaction among dark matter structure but also the nonlinear evolution of ordinary matter in evolving and asymmetrical potentials . a number of authors have performed simulations of systems which follow the evolution of both dissipational and dissipationless components , either in the context of galaxy collisions , or galaxy formation ( see , eg . barnes & hernquist 1991 ; gerber , lamb , & balsara 1996 ; lamb et al . 1997 ; abel et al . 1998 ; steinmetz & navarro 1997 , 1999 ; and references therein ) . such studies are generally either unable to resolve the full dynamical range present in a young galaxy , or do not include the full range of physical processes discussed above . fortunately , we are able to examine the importance of the processes discussed in 3 and 4 by following the standard approach in galaxy formation simulations by breaking the problem into piecemeal tasks . in contrast to the conventional simulations ( eg . yepes et al . 1997 ; elizondo et al . 1999 ) , which mainly deal with the emergence of dark matter structure and in which the star formation process is parameterized , we assume the potential is already formed and focus our attention on the microphysics of the gas . in the first model , we take @xmath210 km s@xmath39 , and assume that @xmath124 , so that the appropriate equations to use are those derived for limit in which high mass clouds dominate . for these massive clouds , the gas drag effect is relatively weak and we initially adopt @xmath195 , _ i.e. _ the wfc s move inward at speeds of @xmath8 . the initial conditions of the model are uniform density in the hot phase , with @xmath211 @xmath212 , giving a mass of @xmath213 m@xmath179 within a radius of 50 kpc . the average density of the warm phase is initially taken to be 0.01 times that of the hot phase . the temperature of the hot phase is initially set to 10@xmath214 k. we have varied the initial conditions substantially with little change in the final distribution of the hot phase . the model was evolved for 10@xmath215 yr , by which time it had reached an asymptotic quasi - stationary state . while the mass of gas continues to decrease due to star formation , the density distributions of the rhc and wfc s attain asymptotic forms . these are shown in figure 2 . figure 2a shows the final radial distributions of the density of the hot phase ( solid curve ) , the average density of the warm phase ( dashed curve ) , and the density of stars formed from the warm phase during the evolution ( dot - dashed curve ) . the stellar density is calculated using the positions of stars where they are first formed . because the stars will form on highly eccentric orbits , for which they spend most of their time near apocenter , the calculated stellar distribution should be representative of the observed distribution , after they have undergone violent relaxation ( aarseth , lin , & papaloizou 1988 ) . as can be seen from the figure , @xmath199 has a fairly shallow radial dependence , varying approximately as @xmath216 . the average density of the warm phase drops off somewhat more steeply , varying approximately as @xmath217 . the stellar mass density has the steepest radial dependence , varying approximately as @xmath218 . figure 2b shows the final radial distributions of the pressure ( solid curve ) and temperature ( dashed curve ) of the hot phase . at radii near 100 kpc , the temperature remains at @xmath219 k. at smaller radii , however , the increasing density of the warm phase , with the corresponding increase in the heating rate , leads to temperatures several times higher . the combination of the density gradient with the shallow temperature gradient leads to a relatively steep pressure gradient . the magnitude of @xmath220 at 10 kpc and it varies approximately as @xmath221 . the change in slope outside of 30 kpc is related to the form of the dark matter potential used in the model , which becomes keplerian beyond 50 kpc . figure 2c shows the radial dependences of the mass ratio of the warm to the hot phase ( solid curve ) , and of the volume filling factor , @xmath222 of the wfc s ( dashed curve ) . we define @xmath222 as @xmath223 , ie . it is the ratio of the average density of the warm gas to the internal density of the wfc s . as can be seen , most of the mass of the system is in the wfc s . it is also seen that @xmath222 shows very little variation with radius , and @xmath224 for @xmath225 kpc . the lack of variation in @xmath222 may seem somewhat counterintuitive , given that @xmath226 decreases at larger radii . it is a consequence , however , of the fact that the wfc s are in pressure balance with the rhg , and the pressure of the latter increases rapidly towards smaller galactic radii . the clouds are therefore compressed into much smaller radii at small @xmath142 . note also that the ratio @xmath227 everywhere , such that most of the ordinary matter is in the wfc s rather than residing in the rhg . this tendency arises because @xmath228 initially in this model . figure 2d shows the radial dependences of the ratio of the depletion timescale of the wfc s by star formation to the dynamical timescale ( solid curve ) , and the ratio of cooling in the rhg by bremsstrahlung emission to that by conduction into the wfc s ( dashed curve ) . the depletion timescale is defined as @xmath229 while the dynamical timescale @xmath230 the ratio @xmath231 gives a rough estimate of the ability of the system to replenish warm gas that has been lost to star formation ( it is also replaced by mass lost from the warm phase , but this can be replenished only by motions at the sound speed of the hot phase , @xmath232 ) . as can be seen from the figure , both ratios are @xmath233 between @xmath234 and 100 kpc . only in the inner regions of the halo , @xmath235 becomes a significant fraction of @xmath3 such that efficient conversion from gas into stars is possible . for the values of @xmath203 and @xmath199 in the model , most of the flux from bremsstrahlung emission is composed of x - ray photons . in the low density outer regions of the halo , bremsstrahlung is not an efficient radiative process , and a comparable amount of energy is lost to the rhg via conduction . within 100 kpc , the total luminosity of bremsstrahlung emission is @xmath236 l@xmath179 . by comparison , the total energy lost from the rhg via conduction into the wfc s ( and immediately radiated from the wfc s ) is @xmath237 l@xmath179 . in the models discussed below , a much greater fraction of the energy is lost via conduction . based on these results , we do not expect these systems to be luminous source of x - rays . we compare the results of the previous model with one having a circular velocity of 100 km s@xmath39 . such a model represents a dwarf system either at any epoch , or one which will become a larger system , but is at an early stage of evolution and has not yet undergone significant merging . in cdm models ( blumenthal et al . 1984 ; thoul , & weinberg 1996 ; navarro , frenk , & white 1997 ; klypin , nolthenius , & primack 1997 ; abel & mo 1998 ) , the first structures which form are significantly smaller than this . however , as can be seen from comparing the results below with those of model 1 , systems with significantly smaller values of v@xmath238 are unlikely to contain a significant mass of rhg , and the assumptions of the models would not be applicable to such systems . the results of model 2 are shown in figure 3 , which illustrate the same quantities as figure 2 . in figure 3a , it can be seen that the densities are reduced significantly relative to the model with higher @xmath8 . the relative sense of the steepness is retained , though the slopes are somewhat different , with @xmath199 varying approximately as @xmath239 , @xmath226 as @xmath240 and @xmath241 as @xmath242 . as shown in figure 3b , the temperature gradient in the rhg is again very shallow . the pressure of the rhg varies approximately as @xmath221 , but it is lower than in the previous model , with @xmath243 at 10 kpc . the smaller value of @xmath8 relative to the previous model reduces the heating efficiency of the wfc s . in order to balance heating and cooling , the density of the rhg is reduced relative to the previous model . this reduction in the rhg s density relative to that of the warm phase is accompanied by increases in the mass ratio of the warm to the hot phase , and of the filling factor of the warm phase . both of these quantities are larger by an order of magnitude relative to the earlier model , as can be seen in figure 3c . we also note that within 1.5 kpc , where @xmath244 , our model assumptions obviously break down . in this case , the wfc s should be considered as a coherent entity rather than an assembly of smaller clouds . the relatively small pressure of the rhg leads to low internal density of the wfc s , and a corresponding reduction in the star formation rate . this is reflected in the large values of @xmath231 shown in figure 3d . the reduced density of the rhg and the large @xmath245 ratio also increase the relative importance of conductive cooling of the rhg relative to bremsstrahlung emission . we find total luminosities out to 100 kpc of @xmath246 l@xmath179 and @xmath247 l@xmath179 . in figure 4 , we show the results of a model in which @xmath210 km s@xmath39 , and in which @xmath248 , ie . the wfc s move inwards with speeds of at most @xmath249 . this prescription would represent a situation in which disruption , coalescence , and coagulation occur on timescales short compared to the acceleration timescale of the clouds . figure 4 shows the same quantities as figure 1 . the reduction of the infall velocity has a similar affect as a relatively small @xmath8 , that is , a decrease in the heating efficiency of the rhg by the wfc s , relative to the @xmath195 case . both density and pressure of the rhg are relatively small in this model , similar to that seen in model 2 , which has a smaller value of @xmath8 . with a relatively small density , the rhg loses more energy through conduction into the wfc s than by bremsstrahlung everywhere in the system ( figure 4d ) . the total luminosities of the two processes are found to be @xmath250 l@xmath179 and @xmath251 l@xmath179 . the reduction in the infall velocity also leads to a shallower density profile in the wfc s . comparing with the earlier models , this tendency causes star formation to be enhanced at large radii and suppressed at smaller radii . consequently , the stellar density becomes relatively shallow compared with previous results . we have also considered a model in which @xmath125 so that the total mass and surface area of the cloud population is dominated by the small rather than the large clouds . in this limit , heating of the rhg , and conductive losses from the rhg into the wfc s are taken from equations ( 24 ) and ( 30 ) . the infall speed of the wfc s is limited to @xmath252 due to their relatively low infall velocities , individual small clouds are less efficient at heating the rhg than are large clouds . for a given @xmath226 , therefore , a distribution dominated by small clouds leads to a significantly smaller value of @xmath253 . in addition , the larger surface - to - volume ratio of a distribution of wfc s dominated by small clouds leads to a much greater conductive cooling rate as compared to that in a distribution dominated by large clouds . the increase in its efficiency relative to the above models means that cooling by conduction into the wfc s dominates bremsstrahlung everywhere in the rhg . the combined effects of higher cooling efficiency and lower heating efficiency lead to the depletion of the rhg and a much larger mass ratio of the wfc s to the rhg relative to the models above . the lower density in the rhg relative to earlier models leads directly to smaller pressures , causing a reduction in both @xmath98 and the star formation rate . the change was so dramatic that we found @xmath244 , ie . the rhg effectively does not exist in the system . but , in the absence of rhg , the wfc s can not fragment through kelvin - helmholz instability to offset their growth through cohesive collisions ( see 3.3 ) . thus , the large @xmath254 ratio and the dominance of the small clouds appears to be incompatible . a self - consistent treatment of both phases , including the size distribution of the wfc s will be examined and discussed in a future paper . the above models assume that @xmath255ergs @xmath256 s@xmath39 , a high value for warm clouds , near that of the peak from hydrogen emission . such would be expected , given the stability arguments presented in 3 , if the small cloud population is dominated by those clouds which are just able to radiate away the energy conducted into them from the rhg . if , however , the small cloud population is dominated by slightly larger clouds , which can radiate more efficiently , then a much small value of @xmath72 may be more appropriate . we have therefore repeated model 4 using @xmath257ergs @xmath256 s@xmath39 . the decreased emission efficiency of the small clouds led to results that were significantly different from those found above , and closely resembled those found for model 3 . the difference between these two possible cases remains an uncertainty in this work , and shall be resolved in upcoming models which examine the structure of wfc s embedded within the rhg . in this paper , we examine the microphysics of gas dynamics in the early epoch of galactic evolution . our objective is to provide a description of the dominant physical processes in ordinary matter which may be applied to a large class of galaxy formation models . for example , in the canonical hierarchical galaxy formation scenario , small dwarf - galaxy building blocks , containing non - interactive dark matter and gaseous ordinary matter , form first and subsequently merge to form larger entities such as our galaxy . in the present analysis , we neglect the dynamical evolution of the dark matter halo which is undoubtedly important in determining not only the formation process but also the present kinematic properties of galaxies . this approximation enables us to focus our attention upon the evolution of gas in the early epoch of galactic evolution , which regulates the rate and location of star formation and therefore the light distribution and chemical properties of the emerging galaxies . we discuss below three possible applications of our results . the results presented here show that thermal instability results in the formation of the residual halo gas ( rhg ) , with density and pressure appropriate for quasi - hydrostatic and energy equilibria , and warm fragmentary clouds ( wfc s ) , which are pressure confined by rhg and heated by ionizing uv photons . such systems have been proposed as being the source of observed lyman - limit systems ( mo & miralda - escude 1996 ) . the mass limits of the wfc s are set by the same criteria as in the earlier work , but in this work we include many more details of the interactions between the phases . inside @xmath258 kpc , wfc s in galaxies with masses comparable to the galaxy are self - shielded from the extragalactic uv flux ( see figure 1 ) . unless these wfc s are continually heated by the uv flux from nearby massive stars , further cooling reduces their temperature to @xmath259 k. gravitational instability in the large wfc s leads to spontaneous formation of stars among which the massive stars radiate uv photons , ionize their surroundings , and quench the formation of additional stars during their lifetime ( lin & murray 1992 ) . through such a self - regulating feedback process , the maximum rate at which gas may be converted into stars is determined by the maintenance of an adequate uv flux to photoionize all the wfc s ( 4.2 ) . in the regions far from the center of the halo where the density of the rhg is relatively low , the internal density of the wfc s and their average density @xmath260 are also low . the effect of self - regulation limits the star formation timescale @xmath261 . but , the star formation efficiency is much higher at smaller galactic distances . the results in figures 2 - 4 show that within a few kpc ( depending on the model ) , the stellar density @xmath241 already exceeds both @xmath260 and @xmath199 after 1 gyr . once the stars are formed out of gas , they can not dissipate their orbital energy such that their orbital radii can not contract further . thus , the interaction between rhg , wfc s , and the newly formed massive stars essentially determines the asymptotic surface brightness distribution in galaxies . in figure 5 , we show the computed surface density profiles computed from the stellar distributions of models 1 - 3 . as can be seen , the models are fairly well fit by devaucouleur profiles within the inner 50 kpc , in good agreement with the profiles of spheroidal systems . the onset of rapid and efficient star formation invalidates the instantaneous mixing and incremental gas - to - star conversion assumptions which are essential to the closed - box models for galactic chemical evolution and enrichment ( cf binney & tremaine 1987 ) , despite the apparent consistency between it and the observed metalicity distribution among population ii stars ( chiappini et al . if these stars are formed in a series of starburst events in which a large fraction of the remaining gas is converted into stars on a timescale shorter than the dynamical timescale of the galaxy or the lifespan of the massive stars , the metallicity distribution would reflect the metallicity inhomogeneity in wfc s . in this case , the deficiency of extremely metal poor stars ( with [ fe / h]@xmath262 - 3 ) would be consistent with an evolving initial mass function which gradually becomes less biased towards massive stars as the wfc s are chemically enriched . a necessary condition for the formation of stars with long - lived , low - mass stars is small @xmath166 ( see 4.2 ) . in a self - regulated environment , wfc s with [ fe / h]@xmath263 - 3 can spontaneously cool from @xmath46 k to @xmath165 k ( so that @xmath264 ) between successive generations of nearby massive stars ( with an interval @xmath265 a few @xmath266 yr ) provided the initial @xmath267 ( hellsten & lin 2000 ) . the corresponding external pressure needed to confine such wfc s is @xmath268 . in figures 2 - 4 , we see that these values of @xmath269 are attained outside 10 kpc , similar to the regions where the metal poor population ii stars are located . in the above scenario , we have neglected the effect of magnetic fields which stablize cold clouds against gravitational instability through field - ion coupling and ion - neutral collisions ( cf shu 1985 ; shu , adams , & lizano 1993 ) . in dense cores of molecular clouds around the solar neighborhood , low - level ionization is maintained through cosmic ray heating and the field strength declines through ambipolar diffusion ( spitzer 1978 ; shu 1985 ) . if this process is important in wfc s , the star formation rate would be much reduced from those illustrated in figures 2 - 4 . collisions between wfc s would occur at velocities in excess of 10 km s@xmath39 . shock compression and rapid cooling near the collision interface could lead to a rapid expulsion of magnetic field , rendering its support ineffective . in the models discussed above , the covering factor of large wfc s is of order unity . a large cloud would therefore collide with another large cloud only about once during a galactic crossing time . collisions between large and small clouds would , however , occur much more frequently . in the limit that the size distribution of the clouds is dominated by relatively small clouds , the area covering factor is large , and small clouds collide with each other much more frequently . these issues are beyond the scope of the present investigation and they need to be thoroughly investigated in the future . the analysis presented here can also be applied to the formation of galactic globular clusters ( lin & murray 1996 ) . prior to the conversion of ordinary matter from gas into stars , the progenitors of these clusters were protocluster clouds ( pcc s ) . the chemical homogeneities within individual clusters and the large metallicity variation among different clusters suggest that pcc s are a distinct entities which must be confined either by their own self - gravity or external pressure . but , if pcc s are entirely bound by their own self - gravity , external uv heating would not be adequate to suppress thermal instability within them ( see 3.3 ) . the magnitude of @xmath166 for the wfc s is comparable to the mass of globular clusters . we identify these warm , marginally self - gravitating and partially pressure - confined wfc s as pcc s . at galactic distances @xmath270 kpc , pcc s with mass ( @xmath271 ) @xmath272 are confined by the the pressure of the rhg , @xmath273 , depending on the galactic halo structure during the epoch of cluster formation ( figs . 2 - 4 ) . if these pcc s are completely ionized and have a @xmath274 , then @xmath275 @xmath212 . for these structural parameters , the uv flux needed is equivalent to that emitted by a few o5 stars at a distance comparable to or greater than their size ( typical a few pc ) ( see figure 1 ) . ( these stars could also reside within the pcc s ) . these clouds could persist for a significant fraction of @xmath3 if the accretion of smaller clouds or condensation from the rhg is adequate to compensate for their mass loss due to stripping by the rhg . on the observational side , in order to verify that the pcc s were pressure - confined , we first estimate their @xmath276 from the current properties of globular clusters , averaged over their half - mass radii ( @xmath277 ) ( murray & lin 1996 ) . we use these quantities because the stellar density and the velocity dispersion at @xmath277 do not change significantly during post - formation evolution . extrapolation to the stage prior to star formation is , however , highly uncertain . if , after their formation , the young stellar objects undergo collapse and virialization from rest , the clouds initial radii ( @xmath278 ) would be @xmath279 . larger @xmath278 would be expected if star formation requires dissipative collisions and coagulation of substellar fragments ( murray & lin 1996 ) . but @xmath278 is unlikely to be larger than the tidal radii of the pcc s , which are typically a few times larger than @xmath277 . thus , the initial density of the pcc s may be 1 - 3 orders of magnitude smaller than the average cluster density at @xmath277 today . based on the present velocity dispersion of the clusters , we infer the initial temperature of the pcc s to be @xmath36 k , comparable to that expected if they were photoionized . from these estimates , we infer @xmath280 ( murray & lin 1992 ) . the dependence of the pressure upon galactic radius , @xmath281 , is very poorly determined from the observational parameters . of more significance is that the magnitude of the pressure inferred from the observations is very similar to that found in the rhg in our models ( figs . 2 - 4 ) . from these results , and the cluster metallicities , we can also estimate the cooling time scale ( @xmath282 ) and dynamical time scale ( @xmath283 ) , of the pcc s . the ratio @xmath284 increases from @xmath285 near the galactic bulge to @xmath286 at @xmath287 kpc . in most pcc s , @xmath288 and thermal equilibrium is only possible in the presence of external uv photons with a flux comparable to that required by self - regulated star formation in the halo . in the hierarchical galaxy formation process , coalescence of small stellar systems ( dwarf galaxies ) occurs within a few dynamical timescales . if the ordinary matter contained initially within these building blocks is to be heated through shock dissipation to the virial temperature of the common halo , the mergers would become luminous x - ray sources ( eke , navarro , & frenk 1998 ) . but the results of our models indicate that ; 1 ) thermal instability leads to the formation of wfc s which contain most of the ordinary matter , and 2 ) even in the rhg , conduction may be a more efficient channel of heat loss than bremsstrahlung emission . as discussed in 5.2 - 5.4 , the resulting x - ray luminosities are small , ranging from 10@xmath289 l@xmath179 in model 1 to 10@xmath290 l@xmath179 in model 3 . the spectra of the models are shown in figure 6 . the spectra of the low luminosity systems may be modified somewhat by supernova emission . using the total star formation rates of models 1 - 3 , we estimate supernova luminosities of 10@xmath291 l@xmath179 for model 1 , 6@xmath29210@xmath293 l@xmath179 for model 2 , and 5@xmath29210@xmath290 l@xmath179 for model 3 . based upon these results , we expect relatively little x - ray luminosity to be released from regions with ongoing galaxy merging events , consistent with x - ray observations ( fabbiano & schweizer 1995 ; read & ponman 1998 ) . in our first attempt to investigate the complex physics of multi phase gas dynamics during the early epoch of galaxy formation , we have adopted various simplifying assumptions such as 1-d spherical symmetry and power - law size distribution for wfc s . these treatments can be improved with a self - consistent analysis of the wfc s evolution which will be presented in a follow up paper . we have also neglected the evolution of dark matter which dominates the potential . the results in figs 2 - 4 show that the distribution of pressure , density , filling factor of wfc s and the star formation rate depend , though not sensitively on the potential . a full study of this problem will require the use of multi - dimensional integrated simulations of both dark and ordinary matter . while such work has been done in a cosmological context , the extreme temperature range spanned by the gas within an individual galaxy ( at least four orders of magnitude ) places severe demands upon a code to be able to resolve structures over many orders of magnitude in size . our basic approach and the prescription provided here can be readily used in such investigations as they begin to be made . we thank drs . a. burkert , s. faber , u. hellsten , c. frenk and a. wolfe for valuable conversations . this research has been supported in part by the nsf through grant ast-9618548 and by nasa through an astrophysics theory program grant which supports a joint center for star formation studies at nasa - ames research center , uc berkeley , and uc santa cruz . this work was performed in part under the auspices of the u.s . department of energy by the university of california lawrence livermore national laboratory under contract no . w-7405-eng-48 .

we discuss the prospects of detecting the sources of ultra - high energy ( uhe ) cosmic ray ( cr ) nuclei via their emission of cosmogenic-rays in the gev to tev energy range . these-rays result from electromagnetic cascades initiated by high energy photons , electrons and positrons that are emitted by crs during their propagation in the cosmic radiation background and are independent of the simultaneous emission of-rays in the vicinity of the source . the corresponding production power by uhe cr nuclei ( with mass number and charge ) is dominated by pion photo - production ( ) and bethe - heitler pair production ( ) . we show that the cosmogenic-ray signal from a single steady uhe cr source is typically more robust with respect to variations of the source composition and injection spectrum than the accompanying signal of cosmogenic neutrinos . we study the diffuse emission from the sum of extragalactic cr sources as well as the point source emission of the closest sources .

1105.5113

the origin and chemical composition of uhe crs ( @xmath5 ev ) is a long - standing enigma in cr physics @xcite . the spectrum at these energies is expected to be dominated by extragalactic sources that have not yet been unambiguously identified . the average mass number of the composition can be inferred directly from atmospheric cr showers by measuring the elongation rate distribution and comparing this to simulations . recent results of the pierre auger collaboration @xcite indicate a transition of uhe crs within the energy range @xmath6 ev to @xmath7 ev from a light spectrum ( consistent with protons ) towards a heavier composition @xcite . in contrast , the hires collaboration @xcite finds a mass composition compatible with that of a proton - dominated spectrum @xcite . indirect evidence for the cr composition can come from various features seen in the spectrum . the `` ankle '' at about @xmath8 ev seems to be a natural candidate for the transition between galactic and extragalactic crs @xcite , but a lower energy crossover at the `` second knee '' at about @xmath9 ev has also been advocated for proton - dominated spectra @xcite . proton - dominance beyond the ankle is expected to be limited beyond the _ greisen - zatspin - kuzmin _ ( gzk ) cutoff @xcite due to resonant pion photo - production in the cosmic microwave background ( cmb ) . intriguingly , a suppression of the cr spectrum at the expected energy of about @xmath10 ev has been observed at a statistically significant level @xcite and is consistent with a proton dominance at these energies . however , a similar feature could also originate from nuclear photo - disintegration of uhe cr nuclei in the cosmic radiation background ( crb ) , or from an _ in situ _ energy cut - off of the injection spectrum . in light of these yet inconclusive and even controversial experimental findings it is important to consider alternative cosmic messengers associated with the production and propagation of uhe cr nuclei . the sources of uhe crs are also expected to emit high energy radiation in the form of neutrinos and @xmath0-rays due to their interaction with ambient matter , radiation or magnetic fields inside or in the vicinity of the acceleration region . the strength and spectral energy distribution of this emission depends , however , not only on the uncertain chemical composition and injection spectrum of uhe crs but also on the source environment . this dependence on source parameters is absent for the fluxes of cosmogenic neutrinos and @xmath0-rays that are produced by cr interactions with the crb and intergalactic magnetic fields ( igmfs ) ; these contributions can be directly related to the observed spectrum of uhe crs . the dominant interactions of uhe cr nuclei ( atomic number @xmath1 and charge @xmath2 ) with the crb are nuclear photo - disintegation @xcite , pion photo - production @xcite and bethe - heitler ( bh ) pair production @xcite . photo - disintegration in the cmb leads to a fast break - up of heavy nuclei at energies of about @xmath11 ev with an interaction length of the order of a few mpc . the direct contribution of photo - disintegration to cosmogenic neutrinos and @xmath0-rays is negligible @xcite . the dominant channel for cosmogenic neutrinos is the photo - production of charged pions in the cmb and their subsequent decay @xcite . this process can be approximated by treating the nucleons of the nuclei as free protons and neutrons with energy @xmath12 resulting in an production threshold with cmb photons of about @xmath13 ev . in the decay chain of the charged pion the three emerging neutrinos carry away about 3/4 of the total pion s energy . photo - nucleon interactions produce roughly equal number of charged and neutral pions . gamma - rays and @xmath14 produced via @xmath15 and @xmath16 decay , respectively , subsequently cascade on crb photons via repeated @xmath17 pair production and inverse compton scattering . the net result is a pile up of @xmath0-rays at gev - tev energies , just below the threshold for further pair production on the diffuse optical background . another contribution to this electromagnetic cascade comes from the bh production of @xmath18 pairs . the corresponding energy loss length of cr nuclei is minimal at energies of @xmath19 ev and decreases with charge as @xmath20 . the energy loss rate @xmath21 in electrons , positrons and @xmath0-rays determines the bolometric flux of the cosmogenic gev - tev @xmath0-ray signal and cosmogenic neutrinos . the energy loss via pion photo - production at cr energy @xmath22 can be approximated by the energy loss rate of nucleons @xmath23 as @xmath24 . since pion photo - production has a relatively high energy threshold the contribution of cosmogenic neutrinos and @xmath0-rays from this channel typically depends on the maximal energy of the uhe cr emission . this can lead to a significant model dependence of the prediction . bethe - heitler @xmath18 pairs are already produced at much lower energy . the interaction can be treated as a continuous energy loss with rate @xmath25 , where @xmath26 is the bh energy loss of protons . the @xmath27-dependence of this process and the low threshold makes this channel an important contributor to the @xmath0-ray cascade . we will study in the following the relative contributions of bh pair production and pion photo - production to the @xmath0-ray signal for cr models involving heavy nuclei . in section [ sec : propagation ] we start with a brief review of the propagation of uhe crs and the calculation of the spectra . the development of electromagnetic cascades and the corresponding diffuse @xmath0-ray signal for cr models are discussed in section [ sec : cascades ] . we will focus here on two cr models , a proton - dominated spectrum with a low crossover and an iron - dominated cr model motivated by the chemical composition inferred by auger . in section [ sec : psflux ] we discuss the prospects to detect the closest sources of uhe cr nuclei as point sources of cosmogenic @xmath0-rays . this possibility depends on the presence of an igmfs during the development of the cascade as outlined in section [ sec : magneticfields ] . we finally conclude in section [ sec : conclusion ] . the main reactions of uhe cr nuclei during their cosmic evolution are nuclear photo - disintegration @xcite , pion photo - production @xcite and bh pair production @xcite on crb photons . the angular - averaged differential rate of a transition between nuclei of type @xmath28 and @xmath29 with energy @xmath30 and @xmath31 is defined as @xmath32 where @xmath33 is the energy distribution of isotropic background photons at redshift @xmath34 and @xmath35 the photon s energy in the rest frame of the nucleus with lorentz boost @xmath36 . besides the contribution of the cmb we use the cosmic infrared / optical background ( cib ) from ref . @xcite . due to the cosmic evolution of the crb density the interaction rates scale with redshift . whereas the cmb evolution follows an adiabatic expansion , @xmath37 , we assume that the cib evolution follows the star formation rate as described in the appendices of ref . @xcite . from eq . ( [ eq : gamma ] ) we define the integrated interaction rate @xmath38 and the total interaction rate @xmath39 . photo - disintegration of nuclei with large mass number @xmath1 is dominated by the giant dipole resonance ( gdr ) with main branches @xmath40 and @xmath41 where @xmath23 indicates a proton or neutron @xcite . the gdr peak in the rest frame of the nucleus lies at about @xmath42 mev for one - nucleon emission , corresponding to @xmath43 gev in the cosmic frame with photon energies @xmath44 mev . the secondary nuclei with atomic number @xmath45 and @xmath46 inherit the boost of the initial nucleon and lie close to the next gdr at @xmath47 and @xmath48 , respectively . hence , the initial flux of nuclei emitted from cr sources rapidly cascades down to lighter nuclei . this leads to a suppression of the flux above @xmath49 . the most general evolution of primary and secondary nuclei in the crb includes all possible photo - disintegration transitions between nuclides @xmath50 competing with the decay of unstable nuclides . for simplicity , we follow the work of _ puget , stecker & bredekamp _ ( psb ) @xcite and consider only one stable isotope per mass number @xmath1 in the decay chain of @xmath51fe . at energies below 10 mev in the rest frame of the nucleus there exist typically a number of discrete excitation levels that can become significant for low mass nuclei . above 30 mev , where the photon wavelength becomes smaller than the size of the nucleus , the photon can interact via substructures of the nucleus . out of these the interaction with quasi - deuterons is typically most dominant and forms a plateau of the cross section up to the pion production threshold at @xmath52 mev . we use the reaction code talys @xcite to evaluate the cross sections @xmath53 of the exclusive processes @xmath54 , @xmath55 , @xmath56 , @xmath57 and @xmath58 ( n stands for p or n ) for nuclides of the psb - chain with @xmath59 . for the cross sections of light nuclei with mass numbers @xmath60 and @xmath61 we use the parametrization provided in ref . @xcite . resonant photo - nuclear interactions with cmb photons set in at energies of @xmath10 ev per nucleon and becomes hence more important for low mass fragments of the photo - disintegration process . we follow the approach outlined in ref . @xcite and approximate the total photo - nucleus interaction by the isospin averaged interaction rate of free nucleons as @xmath62 @xcite . we also assume that the participating nucleon is removed from the nucleus and regard this as a contribution to one - nucleon losses . the nucleon - photon interaction rates can be determined using the monte carle package sophia @xcite . we refer to the appendices of refs . @xcite for further details of the calculation . another important energy loss of uhe cr nuclei is bh pair production via scattering off the crb photons . since this is a coherent process of the nucleons the energy loss scales as @xmath27 where @xmath2 is the charge number of the nucleus @xcite . bethe - heitler pair production with differential rate @xmath63 can be treated as a continuous energy loss process with a rate @xmath64 where @xmath22 and @xmath65 are the energies of the nucleus before and after scattering , respectively . the energy loss of nuclei can be related to the loss of protons as @xmath66 . the energy loss length @xmath67 via bh pairs is typically much smaller than the interaction length of resonant photo - disintegration of heavy nuclei , which is of the order of @xmath68 mpc . hence the primary nuclei will be fully disintegrated within a few mpc . the light secondaries with small charge per nucleus , like protons and helium , do not exhibit a @xmath27-enhancement . however , photo - disintegration preserves the total number of nucleons . in particular , the full disintegration of an initial source spectrum of nuclei @xmath69 into protons is equivalent to a proton emission rate @xmath70 . we will see later that the energy loss via bh of actual models of uhe cr nuclei receives also some @xmath27-enhancement via bh loss for nearby sources . for the calculation of the diffuse spectra of uhe cr nuclei we assume that the cosmic source distribution is spatially homogeneous and isotropic . the comoving number density @xmath71 of a nuclei of type @xmath28 is then governed by a set of boltzmann equations of the form : @xmath72 together with the friedman - lematre equations describing the cosmic expansion rate @xmath73 as a function of the redshift @xmath34 . this is given by , normalized to its value today of @xmath74 kms@xmath75mpc@xmath75 , in the usual `` concordance model '' dominated by a cosmological constant with @xmath76 and a ( cold ) matter component , @xmath77 @xcite . the time - dependence of the redshift can be expressed via @xmath78 . the term @xmath79 in eq . ( [ eq : boltzmann ] ) corresponds to the emission rate of crs of type @xmath28 per comoving volume . for the cr injection spectrum of nuclei of mass number @xmath1 we use a power - law approximation with exponential cut - offs above @xmath80 and below @xmath81 , @xmath82 to account for cosmic evolution of the spectral emission rate per comoving volume we introduce an energy - independent scaling of the form @xmath83 where we use the approximation @xmath84 with @xmath85 and @xmath86 unless otherwise stated . note that our ansatz for @xmath87 does not distinguish between the cosmic evolution of the cr source density and the evolution of the intrinsic emission rate . this distinction is not important for the calculation of the diffuse cr spectra , but plays a role in the prediction of neutrino and @xmath0-ray point source fluxes associated with these cr sources . we will assume that the emission rate of cr sources is fixed and that their number density evolves with redshift . and @xmath88 . for the proton sources we use an injection spectrum with @xmath89 , @xmath90 ev , @xmath91 ev and assume strong source evolution with @xmath92 . the extragalactic iron sources assume an injection spectrum with @xmath89 , @xmath90 ev , @xmath93 ev no evolution @xmath94 . * right panel : * the corresponding spectra of cosmogenic @xmath0-rays ( dashed lines ) and neutrinos ( dotted line ) for the two models . the diffuse @xmath0-ray spectrum of the proton model is marginally consistent with the diffuse extragalactic spectrum inferred by fermi - lat @xcite and the diffuse upper limit on cosmogenic neutrinos from the 40-string configuration ( ic40 ) of icecube @xcite . the cosmogenic @xmath0-ray and neutrino spectra of the iron model are two orders of magnitude below the proton model predictions . ] in the following we are going to consider two models of extragalactic cr sources , that have been considered previously in fitting the uhe cr data @xcite . the first model consists of cr proton sources with a strong evolution ( @xmath92 ) with a relatively low crossover below the ankle . for the injection spectrum we use the power index @xmath89 and assume exponential cutoffs at @xmath90 ev and @xmath91 ev ( see eq . ( [ eq : qa ] ) ) . the spectrum of protons after propagation through the crb is shown as a red line in the left panel of fig . the second model assumes a pure injection of iron with the same spectral index @xmath89 but no evolution of the sources ( @xmath94 ) . we assume the same exponential cutoff at low energies as in the case of the proton model , @xmath90 ev , and a high energy cutoff at @xmath93 ev , motivated by the rigidity dependence of the maximal energy of cr accelerators , @xmath95 . the total spectrum of primary iron and secondary nuclei produced via photo - disintegration is shown as the blue line in the left panel of fig . [ fig1 ] . both models reproduce the uhe cr data above the ankle reasonably well . the deficit below the ankle is assumed to be supplemented by a galactic contribution . note that the crossover with the galactic component is higher for the all - iron model than for the all - proton model . the fit of the model spectra to the cr data sets the absolute normalization of the cr emission rate . this can be expressed as the required bolometric power density per cr source , which depends on the local density of source , @xmath96 . for both models we find a value of @xmath97 the evolution of cosmogenic electrons , positrons and @xmath0-rays is governed by a set of boltzmann equations analogous to eqs . ( [ eq : boltzmann ] ) . electromagnetic interactions of photons and leptons with the crb can happen on time - scales much shorter than their production rates @xcite . the driving processes of the electromagnetic cascade in the cosmic background photons are inverse compton scattering ( ics ) with cmb photons , @xmath98 , and pair production ( pp ) with cmb and cib radiation , @xmath99 @xcite . in particular , the spectral energy distribution of multi - tev @xmath0-rays depends on the cib background at low redshift . for our calculation we use the estimate of _ franceschini et al . _ @xcite . we have little direct knowledge of the cosmic radio background . a theoretical estimate has been made by _ protheroe & biermann _ @xcite of the intensity down to khz frequencies , based on the observed luminosity function and radio spectra of normal galaxies and radio galaxies although there are large uncertainties in the assumed evolution . the calculated values are about a factor of @xmath100 above the measurements and to ensure maximal energy transfer in the cascade we will adopt this estimate and assume the same redshift scaling as the cib . however , the @xmath0-ray cascade below tev does not significantly depend on the exact value of this contribution . a summary of the crb used in this calculation can be found in fig . a.6 of ref . @xcite . high energetic electrons and positrons may also lose energy via synchrotron radiation in the igmf strength @xmath101 , the strength of which is limited to be below @xmath102 g @xcite . recently , the absence of ( resolvable ) gev emission from tev @xmath0-ray blazars has been used to infer a lower limit on the igmf strength of the order of @xmath103 g @xcite . the value of the igmf has only little effect on the bolometric @xmath0-ray flux in the gev - tev energy range relevant for our discussion @xcite , but a sufficiently low value is crucial for a discussion of point - source emission . in the calculation of the cascade spectrum we will assume a weak igmf with strength of @xmath103 g unless otherwise stated . further processes contributing to the electromagnetic cascade are double pair production @xmath104 and triple pair production , @xmath105 . these contribution have only a minor effect on the @xmath0-ray flux at gev - tev and are neglected in the calculation for simplicity . the observed diffuse extragalactic @xmath0-ray flux is thought to be a superposition of various sources . besides the cosmogenic @xmath0-rays of uhe cr nuclei @xcite there are other candidates of truly diffusive processes associated with large - scale structure formation @xcite or models utilizing the decay and annihilation of dark matter @xcite . other contributors are unresolved extragalactic @xmath0-ray sources like active galactic nuclei , starburst galaxies , or @xmath0-ray bursts ( see ref . @xcite for a recent review ) . a recent analysis of the diffuse extragalactic @xmath0-ray background ( egrb ) by fermi - lat @xcite shows a @xmath0-ray spectrum that is lower and softer than previous results of egret @xcite . it has been argued that the fermi - lat flux constraints all - proton models of uhe crs extending down to energies of the `` second knee '' @xcite , though the systematics of uhe cr measurements is not sufficient to entirely exclude this model at a statistically significant level @xcite . the right panel of fig . [ fig1 ] shows the diffuse @xmath0-ray spectra ( dashed lines ) from the all - proton model ( red lines ) and the all - iron with ( blue lines ) . in both cases we normalize the cr spectra ( solid lines ) to the auger data above the ankle . the contributions from both models differ by about two orders of magnitude which is in qualitative agreement with the previous study @xcite . whereas the all - iron model has only a negligible contribution to the egrb the proton model saturates the observed background at 10 - 100 gev . ( in fact , decreasing the lower cutoff @xmath90 ev of the all - proton model would lead to an excess of the fermi - lat measurement . ) for comparison , we also show cosmogenic neutrino flux ( summed over flavors ) of these models as dotted lines . the relative contributions from the two models differ by more than two orders of magnitude similar to the case of @xmath0-rays . note , that the cosmogenic neutrino flux from the all - proton model saturates a recent upper limit on the diffuse extragalactic neutrino flux from the 40-string sub - array ( ic40 ) of icecube @xcite . the strong model dependence of the diffuse fluxes is mainly due to the evolution of the sources as we will see in the following . the cascaded diffuse @xmath0-ray flux peaks in the gev - tev region and has an almost universal shape here . its normalization can be determined by the total energy loss rate into @xmath0-rays , electrons and positrons during the propagation of uhe cr nuclei . we can define the comoving energy density at redshift @xmath34 as @xmath106\,,\ ] ] which follows the evolution equation @xmath107 the energy density ( ev @xmath108 ) of the electromagnetic background observed today is hence given by @xmath109 the relative effect of cosmic evolution on the energy density of the cascade can be estimated in the following way . the uhe cr interactions with background photons are rapid compared to cosmic time - scales . the energy threshold of these processes scale with redshift @xmath34 as @xmath110 where @xmath111 is the ( effective ) threshold today . we can hence approximate the evolution of the energy density as @xmath112 where @xmath113 denotes the energy fraction of the cr luminosity converted to the electromagnetic cascade . assuming a power - law injection @xmath114 with sufficiently large cutoff @xmath115 we see that cosmic evolution enhances the diffuse @xmath0-spectrum as @xmath116 for the proton spectrum shown in fig . [ fig1 ] this corresponds to a relative factor of @xmath117 . an additional , yet smaller relative factor depend on the chemical composition . we start with the the energy loss via photo - nucleon interactions @xmath118 . this case is particularly simple to estimate : photo - disintegration losses conserve the total number of nucleons and we have the approximate relation @xmath119 . for the injection of a primary nucleus with mass number @xmath120 and power - law index @xmath0 we expect the scaling @xmath121 for a universal high energy cutoff per nucleon . energy loss by bh pair production follows the scaling @xmath122 . in the absence of photo - disintegration and photo - nucleon interactions this would result in a simple scaling of the form @xmath123 . however , as we have already discussed in section [ sec : propagation ] , at those energies where pair - production is the dominant energy loss also nuclear photo - disintegration via the giant dipole resonance becomes important , shifting the average mass number and charge to lower values . on resonance , photo - disintegration has a typical inverse interaction rate of @xmath68 mpc and hence the primary nuclei will be fully disintegrated within a few mpc . in this case we can expect that the dominant contribution to bh loss comes from light secondary nuclei and we would obtain the scaling @xmath121 analogous to the case of photo - nuclear losses . for nearby sources the competition between photo - nuclear processes and bh loss makes it difficult to predict the exact scaling of these quantities . for distant sources and in particular for the calculation of diffuse spectra we expect that the scaling is closer to @xmath121 . for the diffuse @xmath0-spectra of the iron model shown in the right panel of fig . [ fig1 ] this corresponds to a relative factor @xmath124 compared to the proton model . together with the relative factor ( [ eq : evolfactor ] ) from cosmic evolution and the difference in the normalization of the models ( @xmath125 ) this accounts for an overall factor of @xmath126 in good agreement with the numerical value . in summary , the contribution of uhe crs to the egrb depends strongly on the underlying cr model , in particular , the evolution of the sources . the cr spectrum at the highest energy is , however , dominated by local sources . the relative contribution from these point - sources does not depend on the evolution of the full population . as we will see in the following , the predicted @xmath0-ray flux from these source is relatively robust against model variation of the emission spectrum . -ray and neutrino spectra of a cr point source of protons ( red ) , nitrogen ( green ) , silicon ( yellow ) and iron ( blue ) at a redshift @xmath127 ( @xmath128mpc ) . we assume an injection spectrum of the form @xmath129 with index @xmath0 and cutoff @xmath80 as indicated in the plots . for comparison , we chose the same normalization of the nuclei in each plot such that @xmath130 erg / s ( see eq . [ eq : l19 ] ) . the solid line shows the total flux of @xmath0-rays and the dashed line the cosmogenic neutrino flux . the dotted lines show the contribution to @xmath0-rays from bh loss alone omitting the contribution from photo - pion interactions . ] so far , we have only considered diffuse @xmath0-ray fluxes from a spatially homogenous and isotropic distribution of cr sources . however , the discreteness of cr sources can lead to local @xmath0-ray excesses , in particular for the closest sources . under optimal circumstances , _ i.e. _ sufficiently weak igmfs , these excesses may even contribute as tev @xmath0-ray point - sources ( pss ) in @xmath0-ray observatories . this possibility has been previously studied for proton sources in refs . @xcite and has been revived recently in the context of unusually bright though distant tev @xmath0-ray sources @xcite . here we extend the discussion to the case of uhe cr nuclei and study the effect of igmfs on the observability of the ps flux in detail . in the absence of an igmf the flux from a ps at redshift @xmath131 with emission rate @xmath69 [ gev@xmath132 s@xmath132 ] as in eq . ( [ eq : qa ] ) is equivalent to an integrated diffuse flux @xmath134 from a homogenous distribution on a sphere at comoving distance @xmath135 . the corresponding emission rate density is hence @xmath136 it is easy to check , that in the absence of interactions of the primary particle with the photon background the point source flux is then given as @xmath137 . in particular , this reproduces the familiar luminosity - distance relation @xmath138 with flux @xmath139 , luminosity @xmath140 and luminosity distance @xmath141 . for illustration , we show in fig . [ fig2 ] the @xmath0-ray ( solid lines ) and neutrino ( dashed lines ) ps fluxes from a source at redshift @xmath142 ( @xmath143 mpc ) emitting iron ( @xmath144 ) , silicon ( @xmath145 ) , nitrogen ( @xmath146 ) or protons . the different plots show variations of the spectral index @xmath0 of the injection spectrum and the exponential cutoff @xmath80 . in each plot the overall normalization of the proton emission rate @xmath147 is chosen such that the source luminosity is @xmath148 erg / s ( see eq . [ eq : l19 ] ) and we use the same normalization constant for the other nuclei . the top panels show the results of @xmath149 ( top left panel ) and @xmath150 ( top right panel ) for a cutoff proportional to the nucleon mass @xmath151 ev . this choice corresponds to a universal exponential cutoff for the energy per nucleon in each injection spectrum . since the energy loss @xmath152 from photo - nucleon interactions depend on @xmath12 the spectra of cosmogenic neutrinos ( dashed lines ) have an almost universal shape as expected . due to the universality of the neutrino spectra the relative normalization of their flux is in this case also given by eq . ( [ eq : omegacas ] ) and scales as @xmath153 . the contribution of bh pair production to the @xmath0-ray spectrum is shown separately in the plots as dotted lines . as discussed earlier , this contribution does in general not follow the @xmath153 behavior of the cosmogenic neutrino fluxes . if the initial nucleus is not fully photo - disintegrated , the bh contribution will be closer to @xmath154 . qualitatively , the variation of this contribution with spectral index @xmath0 and initial mass number @xmath1 is smaller than for the case of photo - pion loss and stays within a factor @xmath155 . the robustness of this contribution becomes even more apparent in the case of a fixed maximal cutoff @xmath91 ev for all nuclei which is shown in the lower panels of fig . [ fig2 ] for the same spectral indices @xmath156 ( bottom left panel ) and @xmath150 ( bottom right panel ) . since pion photo - production has a relatively high threshold of about @xmath157 ev their contribution becomes strongly suppressed as we go to heavier nuclei and hence lower maximal energy per nucleon . this is apparent from the drastic decrease of the cosmogenic neutrino flux ( dotted lines ) . ( similarly , the gzk @xmath0-ray flux @xcite at the upper end of the spectrum varies strongly with the maximal cutoff and compositon . ) the @xmath0-ray cascade , however , receives contributions from bh pair production at a lower energy threshold and this contribution is only mildly effected by the variation of the model parameters . generally , the comparatively small variation of the bh contribution in the electromagnetic cascade makes the prediction of @xmath0-rays from the sources of uhe crs more robust than cosmogenic neutrinos . the fit of uhe cr models to the data fixes the average luminosity density of cr sources . for the prediction of the average luminosity per source and hence the cosmogenic @xmath0-ray flux from the closest cr source we have to fix the local source density @xmath158 [ @xmath108 ] introduced in eq . ( [ eq : sourden2 ] ) . the local density can not be much smaller than @xmath160 mpc@xmath161 as can be estimated from the absence of `` repeaters '' in cr data @xcite . moreover , the distant to the closest source can not be much larger than , say , 100 mpc since uhe crs will unlikely survive over longer distances . .the local source density @xmath158 ( eq . [ eq : h0 ] ) and the source luminosity above @xmath162 ev ( eq . [ eq : l19 ] ) assuming a comoving distance @xmath163 to the closest cr source . [ cols="^,^,^",options="header " , ] a spatially homogenous distribution of cr sources with number density @xmath164 per comoving volume as in eq . ( [ eq : sourden2 ] ) is equivalent to a diffuse flux of @xmath165 where @xmath166 is the volume of the co - moving sphere containing the sources at redshift smaller than @xmath34 . in the the following we will assume that the closest cr source is located at a comoving distance @xmath167 . if the local source distribution with density @xmath158 is sufficiently smooth we can expect that this is the only source within a distance @xmath168 given by , @xmath169 we can then decompose the diffuse flux as @xmath170 we assume that the integral can be approximated by the diffuse flux of a spatially homogenous emission rate density @xmath171 with @xmath172 as discussed in section [ sec : propagation ] . determines the average local source density @xmath158 as in eq . [ eq : sourden2 ] . for illustration we assume local densities of @xmath173 , @xmath174 and @xmath173 mpc@xmath161 with @xmath1754 , 8 and 16 mpc , respectively . from the fit to the cr data we can determine then the average source luminosity @xmath176 * right panels : * the point - source @xmath0-ray ( blue ) and neutrino ( green ) spectra for the closest proton ( top panel ) or iron ( bottom panel ) source . we assume again three different distances @xmath1754 , 8 and 16 mpc . whereas the _ diffuse _ @xmath0-ray spectra shown in fig . [ fig1 ] of the two cr models differ by two orders of magnitude , the ps spectra are similar in magnitude for equidistant source locations . the ps fluxes of cosmogenic neutrinos depending on photo - pion interactions of protons are one order of magnitude lower for the case of an all iron source compared to an all proton source . ] the emission of the closest source at @xmath163 will only contribute to the uhe cr data at the upper end of the spectrum since @xmath177 . due to the very poor statistic at these energies the position can only be determind within large statistical uncertainties . instead as our working hypothesis we will assume in the following that the source location is fixed at @xmath178 . this corresponds to the average distance of a source uniformly distributed within a radius @xmath179 and weighted by the flux factor @xmath180 . with this choice the sum ( [ eq : psdiffuse ] ) will closely resemble the cr spectrum from a fully homogenous emission rate density with @xmath181 as we will see in the following . we will study in the following three different source locations , @xmath182 , 8 and 16 mpc . the corresponding source luminosities and volumes are tabulated in table [ tab1 ] . -ray observatories to nearby sources of uhe crs for the all - proton and all - iron model . the integrated flux @xmath183 of the point - source is almost independent of the the cr model considered . for a dilute uhe cr source density with @xmath184 mpc@xmath161 the future cta should be able to identify the cosmogenic @xmath0-ray flux of uhe cr sources within 50 hours of operation . * right panel : * the size of the @xmath0-ray halo for various igmf strengths and @xmath0-ray energies . the halo size can be well approximated by a fit @xmath185 . ] the left panels of fig . [ fig3 ] show the contribution of the closest uhe cr sources ( red lines ) to the overall diffuse flux of uhe crs for the case of the all - proton ( top panel ) and the all - iron ( bottom panel ) model . we also show as thin black lines the remaining contribution of the homogenous cr distribution beyond @xmath179 . note , that in contrast to the ps proton spectra the total ps flux form the iron sources show a strong cutoff preceded by a small bump @xcite . this limits the distance to the nearest uhe cr iron source to a few 10 mpc in this model . the right panels of fig . [ fig3 ] show the corresponding ps fluxes in cosmogenic gamma rays ( blue lines ) and neutrinos ( green lines ) . the ps @xmath0-ray flux from these close cr sources has the typical @xmath186 form extending up to several tens of tev following from inverse compton emission of a fully comptonized electron spectrum ( @xmath187 ) . the left panel of fig . [ fig4 ] shows the integrated @xmath0-ray flux of these nearby cr sources in comparison with the sensitivity of present imaging atmospheric cherenkov telescopes ( iacts ) h.e.s.s . @xcite , magic @xcite and veritas @xcite , as well as the estimated future sensitivity of the water cherenkov telescope hawc @xcite and the cherenkov telescope array ( cta ) @xcite . the solid and dashed lines show the @xmath0-ray flux from proton and iron sources , respectively , at various distances . the signal depends only weakly on the compostion of the source . more important is the increased luminosity of the source ( @xmath188 ) in the scenario of an increase local source density ( @xmath189 ) . for our three cr scenarios shown in table [ tab1 ] only a source distribution with a small local source density close to @xmath190 mpc@xmath161 ( @xmath191 mpc ) and hence a large associated cr luminosity of @xmath192 erg / s per source is expected to be visible in the future cta after 50h of observation . note , however , that our ansatz @xmath178 has been chosen for a good reproduction of the spatially homogenous emission density shown as the black lines in fig . it does not account for stochastic effects of the nearby source distribution . in general , the @xmath0-ray signal of the closest cr source at comoving distance @xmath163 and with fixed luminosity @xmath176 is proportional to @xmath193 and hence the model lines shown in fig . [ fig4 ] are expected to shift accordingly . also , the detection of these multi - tev @xmath0-ray point sources in iacts requires that the signal remains `` point - like '' , _ i.e. _ within the point - spread function ( psf ) of the telescope . in this case it is important to consider the effect of an igmfs on the development of the cascade as we will do in the next section . the cosmogenic @xmath0-ray cascade of nearby cr sources can only contribute to a gev - tev ps flux if the deflections of secondary @xmath14 in the cascade via an igmf is sufficiently small . we can estimate the extend of the cascaded @xmath0-ray emission by simple geometric arguments following @xcite . deflection of electrons and positrons will be small if the energy loss length @xmath194 of @xmath14 via inverse compton scattering ( ics ) is much smaller than the larmor radius given as @xmath195 . ( here and in the following we use the abbreviations @xmath196 , etc . ) for center of mass energies much lower than the electron mass , corresponding to energies below pev in the cmb frame , electrons and positrons interact quickly on kpc scales but with low inelasticity proportional to their energy , @xmath197 . the typical size of deflections of electrons and positrons is hence @xmath198 . deflection of @xmath14 close to the source have a smaller effect on the size of the halo then deflections close to the observer . to first order , if the cascade experiences a deflection @xmath199 at a distance @xmath200 from the observer , we can approximate the corresponding angular displacement @xmath201 in the observer s frame via @xmath202 . we can account for this scaling in the cascade equation by introducing the corresponding scaling in the larmor radius @xmath203 or , equivalently , by a scaling of the diffusion matrix of the form @xmath204 ( see appendix [ sec : appi ] ) . since the distance of the closest cr source is expected to be smaller than the energy loss length by bh pair production or by photo - pion production the electrons and positrons will be produced continuously between the source and the observer . hence , the average geometric suppression factor of the deflections is @xmath100 and hence @xmath205 . in the gev - tev energy region the size of the @xmath0-ray halo is almost independent of the source composition . this is a result of the rapid energy loss of @xmath14 via inverse compton scattering above a few tev compared to the slow production rate via bh loss of crs or via pair production of @xmath0-rays . the leptons quickly lose energy via ics with cmb photons at a rate @xmath206 ; their spectrum in quasi - equilibrium ( @xmath207 ) follows the differential equation @xmath208 . thus , the comptonized electron spectrum for @xmath209 has the form @xmath210 . the deflection of an electron of the comptonized spectrum is approximately @xmath211 following from @xmath212 and @xmath210 and @xmath205 . the typical photon energy from ics of a background photon with energy @xmath213 is given by @xmath214 and hence the halo is expected to extend up to an angle of about @xmath215 independent of cr composition and source distance . we can define the halo size more rigorously with the approach outlined in ref . we find that the halo size can be well approximated by the first moment of the angular distribution as @xmath216 the result of the diffusion - cascade equation ( see appendix [ sec : appi ] ) is shown in the right panel of fig . a fit to the data gives a numerical value of @xmath217 . this is consistent with our previous estimate for the most abundant mev photons in the cmb spectrum . the typical size of the point - spread function ( psf ) of iacts is of the order of @xmath218 . hence , an igmf with a strength less than @xmath219 g will not significantly decrease the sensitivity of future iacts to the multi - tev cosmogenic @xmath0-ray signal of nearby cr sources . on the other hand , the non - observation of cascaded @xmath0-rays as a gev - tev ps flux could imply a lower limit on the igmf strength @xcite . the observation of this effect requires that the sub - tev @xmath0-ray emission of the source is relatively quite , such that the cascaded spectrum would dominate the primary flux . recently , the absence of ( resolvable ) gev emission from tev @xmath0-ray blazars has been used to infer a lower limits on the igmf strength at the level @xmath103 g @xcite . as an example , we consider here the emission of the blazar source 1es0229 + 200 located at redshift @xmath220 , which has been detected by its tev @xmath0-ray emission by h.e.s.s . the spectrum is shown in fig . [ fig5 ] as the blue data . we consider two models for the @xmath0-ray observation . in the right panel of fig . [ fig5 ] ( from ref . @xcite ) we show a model assuming @xmath0-ray emission from the source with a rate @xmath221 ( thin gray line ) . the surviving primary @xmath0-rays are shown as a dashed green line and secondary cascaded @xmath0-rays by a solid line . the cascaded spectrum would clearly dominate the sub - tev emission and is inconsistent with upper limits from fermi lat ( from ref . however , the signal within the psf is significantly reduced by the presence of an igmf as indicated by the dotted lines . -ray spectra of the blazar source 1es0229 + 200 located at redshift @xmath220 . the green data points show the h.e.s.s . observation and the green lines the estimated upper flux limits from the non - observation by fermi - lat inferred by ref . @xcite . * left panel : * a model for the @xmath0-ray spectrum assuming @xmath0-ray emission at a rate @xmath221 . the solid green line shows the spectrum of secondary @xmath0-rays without deflections in the igmf . the dotted green lines indicate the part of the cascaded @xmath0-ray spectrum within @xmath222 around the source for an igmf with coherence length @xmath223 mpc and strength @xmath224 g , @xmath103 g and @xmath219 g , respectively . * right panel : * as in the left panel but now showing the @xmath0-ray contribution from electromagnetic cascades assuming that the blazar is a cr proton source . we assume that the primary @xmath0-ray emission is negligible and the ( beamed ) luminosity in protons is @xmath225 erg / s . in this case the @xmath0-ray flux is already below the fermi - lat upper limits and no igmf is required to explain the data . ] the right panel of fig . [ fig5 ] shows an alternative model for the @xmath0-ray emission of the blazar assuming strong emission of cr protons as in eq . ( [ eq : qa ] ) with @xmath89 and @xmath91 ev . this model has been advocated in refs . @xcite . in this case the observed spectrum is assumed to be dominated by cosmogenic @xmath0-rays emitted during propagation . however , it is apparent that this model does not necessarily require the presence of an igmf to be compatible with the fermi - lat limit as already noted by ref . we also show in this case the reduction of the ps signal via the presence of an igmf with coherence length @xmath223 mpc and a strength @xmath224 g , @xmath103 g and @xmath219 g , respectively . note , that the required luminosity of the source is high in this case , @xmath226 erg / s , which is at least three orders of magnitude larger than the average luminosity of uhe cr sources inferred from the fit to the cr spectrum assuming a local source density larger than @xmath160 mpc@xmath161 . in this model the blazar 1es0229 + 200 can hence not be a typical source of uhe crs . the cr emission along the blazar jet with opening angle @xmath227 and the increase of the effective luminosity as @xmath228 does not play a role in this consideration since the same effect will also decrease the effective local source density of the anisotropically emitting cr sources . however , this example illustrates the strong model dependence on lower limits on the igmf strength inferred by this method . we have discussed the production of cosmogenic @xmath0-rays in models of uhe cr nuclei . these @xmath0-rays are a result of electromagnetic cascades in the crb initiated by photo - pion production and bh pair production of the cr nuclei . the signal has its strongest contribution in the gev to tev range and is independent of cr interactions in the source environment prior to emission . we have discussed in detail how the @xmath0-ray flux depend on injection spectra and the chemical composition of the sources . in general , we find that the flux of cosmogenic @xmath0-rays is less model dependent than other agents of the cr interactions like cosmogenic neutrinos . as an illustration , we have studied two cr models of the uhe cr spectrum : an all - proton model with strong cosmic evolution and low transition to galactic crs and an all - iron model dominating the spectrum beyond the ankle . the diffuse @xmath0-ray flux from these models differs by two orders of magnitude . whereas the proton model saturates the diffuse extragalactic @xmath0-ray spectrum inferred by fermi - lat the iron model is practically unobservable in the background . we have shown that this large difference in the energy density of the cascade relies on the strong contribution of distant sources assumed in the all - proton model . the closest sources of uhe crs can be observed via their @xmath0-ray point source flux if the igmf is sufficiently weak ( @xmath229 g ) . we have argued that the @xmath0-ray signal is expected to show only small variations with respect to the cr emission model due to the strong contribution of bh pair production . the absolute @xmath0-ray flux depends on the cr luminosity and position of the source which can be related to the fit to the cr data and by estimates of the ( average ) local source density . we have estimated that the closest cr source should be observable via its cosmogenic @xmath0-ray emission in the future cherenkov telescope array if the local source density is small ( @xmath230 mpc@xmath161 ) and hence the average source luminosity sufficiently large ( @xmath231 erg / s ) . we have also briefly commented on the possibility that the tev emission of distant blazars can be naturally explained as a cosmogenic @xmath0-ray signal if the blazar is a strong cr proton source ( @xmath232 erg / s ) . as an example we have studied the gev - tev emission of the blazar 1es0229 + 200 . the absence of strong gev emission of this source has been used to derive lower limits on the igmf strength . in contrast , the cosmogenic @xmath0-ray emission of this source is consistent with the observation without the presence of an igmf . we have only considered in this study steady sources of crs , _ i.e. _ sources which have constant emission during the time of observation . pulsed sources of uhe crs could have a stronger @xmath0-ray emission during the time of activity that could exceed our estimates . the sources of crs are also expected to emit tev @xmath0-rays by cr interactions in the source environment . this contribution may dominate the point - source flux making the observation of cosmogenic @xmath0-rays difficult . however , in the case of pulsed cr sources , deflections of the cascade via magnetic fields can lead to a time - delay of cosmogenic @xmath0-rays with respect to the _ in situ _ @xmath0-ray emission of the source and may help to disentangle the contributions . we would like to thank l. a. anchordoqui and m. c. gonzalez - garcia for valuable comments on the manuscript . ma would like to thank the _ departament destructura i constituents de la matria and institut de ciencies del cosmos _ at the university of barcelona for its hospitality during stages of this project . js would like to thank the _ c.n . yang institute for theoretical physics _ , suny , stony brook for its hospitality . we acknowledge upported by us nsf grant no phy-0969739 and spanish micinn grants 2007 - 66665-c02 - 01 , consolider - ingenio 2010 grant csd-2008 - 0037 and cur generalitat de catalunya grant 2009sgr502 . the small magnetic deflection of electrons and positrons in weak igmf with coherence length @xmath233 and strength @xmath234 can be treated as a diffusion process . we follow ref . @xcite and define the moments of the @xmath0-ray halo as @xmath235 one can show that these moments follos the evolution equation @xmath236 with diffusion matrix @xmath237 the quantity @xmath238 denotes the inelasticity of of ics and hence @xmath239 . the first moment @xmath240 equals the ps flux @xmath241 as a solution of the boltzmann equations ( [ eq : boltzmann ] ) . we define discrete values @xmath242 , @xmath243 , etc . the combined effect of transitions and deflections within the cascade during a sufficiently small time - step @xmath244 can be described by the matrix equations @xmath245 the full cascade solution is then given by @xmath246 the @xmath247 matrizes @xmath248 and @xmath249 follow the recursive relation @xmath250 where the non - zero initial conditions are @xmath251 , @xmath252 and @xmath253 . the matrices @xmath254 and @xmath255 are the familiar transfer matrices for electromagnetic cascades in the presence of a source term . using the recursion relations ( [ eq : rec1 ] ) and ( [ eq : rec2 ] ) we can efficiently calculate the matrices @xmath256 and @xmath257 via matrix - doubling @xcite . m. nagano and a. a. watson , rev . phys . * 72 * , 689 ( 2000 ) . c. amsler _ et al . _ [ particle data group ] , phys . b * 667 * , 1 ( 2008 ) . j. abraham _ et al . _ [ pierre auger collaboration ] , phys . b * 685 * , 239 ( 2010 ) [ arxiv:1002.1975 [ astro-ph.he ] ] . j. abraham _ et al . _ [ pierre auger collaboration ] , phys . lett . * 104 * , 091101 ( 2010 ) [ arxiv:1002.0699 [ astro-ph.he ] ] . r. abbasi _ et al . _ [ hires collaboration ] , phys . lett . * 100 * , 101101 ( 2008 ) [ arxiv : astro - ph/0703099 ] . r. u. abbasi _ et al . _ [ hires collaboration ] , phys . lett . * 104 * , 161101 ( 2010 ) [ arxiv:0910.4184 [ astro-ph.he ] ] . j. linsley , _ proceedings of icrc 1963 , jaipur , india _ , pp . 77 - 99 c. t. hill and d. n. schramm , phys . rev . d * 31 * , 564 ( 1985 ) . t. wibig and a. w. wolfendale , j. phys . g * 31 * , 255 ( 2005 ) . [ arxiv : astro - ph/0410624 ] . v. berezinsky , a. z. gazizov and s. i. grigorieva , phys . d * 74 * , 043005 ( 2006 ) . [ arxiv : hep - ph/0204357 ] . z. fodor , s. d. katz , a. ringwald and h. tu , jcap * 0311 * , 015 ( 2003 ) [ arxiv : hep - ph/0309171 ] . k. greisen , phys . * 16 * , 748 ( 1966 ) . g. t. zatsepin and v. a. kuzmin , jetp lett . * 4 * , 78 ( 1966 ) [ pisma zh . eksp . fiz . * 4 * , 114 ( 1966 ) ] . j. abraham _ et al . _ [ pierre auger collaboration ] , phys . * 101 * , 061101 ( 2008 ) [ arxiv:0806.4302 [ astro - ph ] ] . f. w. stecker , phys . rev . * 180 * , 1264 ( 1969 ) . j. l. puget , f. w. stecker and j. h. bredekamp , astrophys . j. * 205 * , 638 ( 1976 ) . f. w. stecker and m. h. salamon , astrophys . j. * 512 * , 521 ( 1999 ) [ arxiv : astro - ph/9808110 ] . f. w. stecker , astrophys . j. * 228 * , 919 ( 1979 ) . g. r. blumenthal , phys . d * 1 * , 1596 ( 1970 ) . l. a. anchordoqui , j. f. beacom , h. goldberg , s. palomares - ruiz and t. j. weiler , phys . * 98 * ( 2007 ) 121101 [ arxiv : astro - ph/0611580 ] . f. aharonian and a. m. taylor , astropart . phys . * 34 * , 258 ( 2010 ) [ arxiv:1005.3230 [ astro-ph.he ] ] . a. franceschini , g. rodighiero and m. vaccari , astron . astrophys . * 487 * , 837 ( 2008 ) [ arxiv:0805.1841 [ astro - ph ] ] . m. ahlers , l. a. anchordoqui and s. sarkar , phys . d * 79 * , 083009 ( 2009 ) [ arxiv:0902.3993 [ astro-ph.he ] ] . s. goriely , s. hilaire and a. j. koning , astron . astrophys . * 487 * , 767 ( 2008 ) [ arxiv:0806.2239 [ astro - ph ] ] , http://www.talys.eu/ j. p. rachen , _ interaction processes and statistical properties of the propagation of cosmic - rays in photon backgrounds _ , phd thesis of the bonn university , 1996 . m. ahlers and a. m. taylor , phys . d * 82 * , 123005 ( 2010 ) [ arxiv:1010.3019 [ astro-ph.he ] ] . d. allard , n. g. busca , g. decerprit , a. v. olinto and e. parizot , jcap * 0810 * , 033 ( 2008 ) [ arxiv:0805.4779 [ astro - ph ] ] . s. lee , phys . d * 58 * , 043004 ( 1998 ) [ arxiv : astro - ph/9604098 ] . g. r. blumenthal and r. j. gould , rev . phys . * 42 * , 237 ( 1970 ) . r. j. protheroe and p. l. biermann , astropart . * 6 * , 45 ( 1996 ) [ erratum - ibid . * 7 * , 181 ( 1996 ) ] , [ arxiv : astro - ph/9605119 ] . m. ahlers , l. a. anchordoqui , m. c. gonzalez - garcia , f. halzen and s. sarkar , astropart . phys . * 34 * , 106 ( 2010 ) [ arxiv:1005.2620 [ astro-ph.he ] ] . kronberg , rept . phys . * 57 * , 325 ( 1994 ) . r. beck , aip conf . proc . * 1085 * , 83 ( 2009 ) [ arxiv:0810.2923 [ astro - ph ] ] . f. a. aharonian , p. s. coppi and h. j. volk , astrophys . j. * 423 * , l5 ( 1994 ) [ arxiv : astro - ph/9312045 ] . a. neronov and d. v. semikoz , jetp lett . * 85 * , 473 ( 2007 ) [ arxiv : astro - ph/0604607 ] . p. davezac , g. dubus and b. giebels , astron . astrophys . * 469 * , 857 ( 2007 ) [ arxiv:0704.3910 [ astro - ph ] ] . a. neronov and i. vovk , science * 328 * , 73 ( 2010 ) . [ arxiv:1006.3504 [ astro-ph.he ] ] . f. tavecchio , g. ghisellini , l. foschini , g. bonnoli , g. ghirlanda and p. coppi , mon . not . * 406 * , l70 ( 2010 ) [ arxiv:1004.1329 [ astro-ph.co ] ] . o. e. kalashev , d. v. semikoz and g. sigl , phys . d * 79 * , 063005 ( 2009 ) [ arxiv:0704.2463 [ astro - ph ] ] . a. loeb and e. waxman , nature * 405 * , 156 ( 2000 ) [ arxiv : astro - ph/0003447 ] . a. a. abdo _ et al . _ [ fermi - lat collaboration ] , jcap * 1004 * , 014 ( 2010 ) [ arxiv:1002.4415 [ astro-ph.co ] ] . c. d. dermer , aip conf . * 921 * , 122 ( 2007 ) [ arxiv:0704.2888 [ astro - ph ] ] . a. a. abdo _ et al . _ [ fermi - lat collaboration ] , phys . rev . * 104 * , 101101 ( 2010 ) [ arxiv:1002.3603 [ astro-ph.he ] ] . sreekumar _ et al . _ [ egret collaboration ] , astrophys . j. * 494 * , 523 ( 1998 ) [ arxiv : astro - ph/9709257 ] . v. berezinsky , a. gazizov , m. kachelriess and s. ostapchenko , phys . b * 695 * , 13 ( 2011 ) [ arxiv:1003.1496 [ astro-ph.he ] ] . r. abbasi _ et al . _ [ icecube collaboration ] , [ arxiv:1103.4250 [ astro-ph.co ] ] . e. armengaud , g. sigl and f. miniati , [ arxiv : astro - ph/0511277 ] . k. kotera , d. allard and m. lemoine , astron . astrophys . * 527 * , a54 + ( 2011 ) [ arxiv:1011.0575 [ astro-ph.he ] ] . c. ferrigno , p. blasi and d. de marco , astropart . * 23 * , 211 ( 2005 ) [ arxiv : astro - ph/0404352 ] . s. gabici and f. a. aharonian , phys . lett . * 95 * , 251102 ( 2005 ) [ arxiv : astro - ph/0505462 ] . w. essey and a. kusenko , astropart . phys . * 33 * , 81 ( 2010 ) [ arxiv:0905.1162 [ astro-ph.he ] ] . w. essey , o. e. kalashev , a. kusenko and j. f. beacom , phys . lett . * 104 * , 141102 ( 2010 ) [ arxiv:0912.3976 [ astro-ph.he ] ] . w. essey , o. kalashev , a. kusenko and j. f. beacom , astrophys . j. * 731 * , 51 ( 2011 ) [ arxiv:1011.6340 [ astro-ph.he ] ] . g. bgelmini , o. e. kalashev , d. v. semikoz , jcap * 0711 * , 002 ( 2007 ) . [ arxiv:0706.2181 [ astro - ph ] ] . e. waxman , k. b. fisher and t. piran , astrophys . j. * 483 * , 1 ( 1997 ) [ arxiv : astro - ph/9604005 ] . t. kashti and e. waxman , jcap * 0805 * , 006 ( 2008 ) [ arxiv:0801.4516 [ astro - ph ] ] . l. a. anchordoqui , m. t. dova , l. n. epele and j. d. swain , phys . d * 57 * , 7103 ( 1998 ) [ arxiv : astro - ph/9708082 ] . e. de ona wilhelmi [ hess collaboration ] , aip conf . proc . * 1112 * , 16 ( 2009 ) . m. lopez moya [ magic collaboration ] , aip conf . * 1223 * , 99 ( 2010 ) . t. c. weekes _ et al . _ [ veritas collaboration ] , int . j. mod . d * 19 * , 1003 ( 2010 ) . g. sinnis [ hawc and milagro collaborations ] , nucl . instrum . meth . a * 623 * , 410 ( 2010 ) . g. hermann [ cta collaboration ] , nucl . instrum . a * 623 * , 408 ( 2010 ) . m. ahlers , [ arxiv:1104.5172 [ astro-ph.he ] ] . f. aharonian _ et al . _ [ hess collaboration ] , astron . astrophys . * 475*,l9-l13 ( 2007 ) [ arxiv:0709.4584 [ astro - ph ] ] . r. j. protheroe and t. stanev , mon . not . * 264 * , 191 ( 1993 ) .

in a quantum theory of gravity , fluctuations about the vacuum may be considered as planck scale virtual black holes appearing and annihilating in pairs . incident fields scattering from such fluctuations would lose quantum coherence . in a recent paper hawking & ross obtained an estimate for the magnitude of this loss in the case of a scalar field . their calculation exploited the separability of the conformally invariant scalar wave equation in the electrovac c metric background , which is justified as a sufficiently good description of a virtual black hole pair in the limit considered . in anticipation of extending this result , the teukolsky equations for incident fields of spin , , , and are separated on the vacuum c metric background , and solved in the same limit . these equations are shown in addition to be valid on the electrovac c metric background when restricted to , , and . the angular solutions are found to reduce to the spin - weighted spherical harmonics , and the radial solutions are found to approach hypergeometrics close to the horizons . by defining appropriate scattering boundary conditions , these solutions are then used to estimate the transmission and reflection coefficients for an incident field of spin . the transmission coefficient is required in order to estimate the loss of quantum coherence of an incident field through scattering off virtual black holes .

hep-th9802028

the action for the gravitational field is _ not _ scale invariant , and so large fluctuations in the metric over short length scales will not have a large action . such fluctuations are therefore not precluded since they are not damped in the path integral , and indeed even variations in topology may occur since these would further increase the associated action by only an arbitrarily small amount . such considerations led wheeler @xcite to suggest that in a quantum theory of gravity , spacetime should be expected to have a ` foam - like ' structure on scales of the planck length or shorter , in which both the metric and the topology of the manifold may vary considerably . on larger scales however , spacetime would appear smooth and continuous , reverting to the manifold structure more conventionally associated with the treatment of classical general relativity . hawking @xcite and hawking , page & pope @xcite first interpreted this foam structure as made from ` quantum gravitational bubbles ' arising in three topological varieties : @xmath6 , @xmath7 , and @xmath8 . these would be the ` building blocks ' of spacetime , not themselves solutions to any field equations but rather occuring as quantum fluctuations . their non - trivial topologies do not allow complete foliation of the manifolds with a family of non - intersecting time surfaces , and consequently green functions defined on them are bound to exhibit certain acausal and frequency mixing properties more commonly associated with those defined on the euclidean black hole metrics . the interpretation of these bubbles as virtual ( or off - shell ) black holes , which materialise out of the vacuum and subsequently vanish again , would therefore appear well justified . quantum field theories defined on manifolds with non - trivial topology are necessarily non - unitary , the consequences of which include the loss of information and of quantum coherence . if spacetime is indeed composed of planck scale building blocks with non - trivial topologies , then quantum fields propagating in it ought to suffer from these effects . such a notion may navely appear to be somewhat at odds with observation ! theories relying on unitary evolution and conserved quantum coherence have been demonstrated to high degrees of accuracy , and yet the quantum bubbles picture would seem to advocate the loss of quantum coherence for fields propagating in spacetime due to their unavoidable interaction with planck scale fluctuations . how may these apparent contradictions be reconciled ? calculations in simple @xmath9 and @xmath7 models @xcite indicated that the non - trivial topologies of the quantum bubbles induced additional singularities into the green functions for particle propagators . these tended to make large contributions to the _ s_-matrix in the scalar case , but only very small contributions in higher spin cases for energies low with respect to the planck scale . in a sense then , the reconciliation is complete since all observed elementary particles have spin greater than zero . it is therefore hardly surprising that spacetime should appear smooth and nearly flat at current observational energies , since the only available probes of its structure are essentially blind to its short scale fluctuations . consideration of the pair creation of black holes led hawking @xcite to reconsider the original interpretation of spacetime foam . gibbons @xcite had suggested that the pair creation of real black holes in a background field could be described by the ernst solution @xcite and its dilaton generalisations, the euclidean section of which has topology @xmath10 . by analogy with the pair creation of ordinary particles , this topology corresponds to a black hole loop in a spacetime asymptotic to @xmath11 . however , @xmath10 may also be considered as the topological sum of the compact space @xmath9 and the non - compact space @xmath11 , leading to a reinterpretation of the @xmath9 bubbles in the spacetime foam as closed loops of virtual black holes . thus , rather than a single off - shell black hole , the quantum bubbles are to be thought of as virtual black holes appearing and annihilating in pairs . if this should occur , then it seems plausible that incident particles could be absorbed and subsequently re - emitted possibly as different particles , and with loss of quantum coherence . adopting the perspective that the compact @xmath9 bubbles occur as quantum fluctuations unaffected by incident low energy particles , a consistent scheme for estimating the scattering amplitude and loss of coherence may be constructed @xcite . this involves computing the euclidean green function in an asymptotically euclidean metric on @xmath12 , analytically continuing to the lorentzian section at infinity , and integrating with the initial and final state wave functions of the scattered field . weighting the result according to the action of the euclidean metric gives the scattering amplitude , which should then be integrated over all asymptotically euclidean metrics . however this scheme is effectively intractable since neither the calculation of the green function in a general metric or the path integration over all such metrics is possible . this problem can be at least partially circumvented by demonstrating the effects of coherence loss in any one set of appropriate metrics , since other metrics within the path integral can not then restore it . this has been carried out for incident scalar fields by hawking & ross @xcite , who exploited the separability of the scalar wave equation in the electrovac c metric for this purpose . the euclidean electrovac c metric possesses a hypersurface orthogonal killing vector , allowing analytic continuation to a well - defined real lorentzian metric in which scattering calculations may be attempted . in a sense the use of the lorentzian metric is nothing more than a mathematical trick to compute scattering effects in the euclidean section . the lorentzian section of the c metric contains a pair of black holes accelerating apart , pulled by cosmic strings extending to infinity . hawking & ross argued that , although the quantum bubble metric need not satisfy any field equations , its lorentzian section would have a structure like that of the c metric but without the axial singularities . they further argued however that the c metric was a sufficiently good model for the bubble metric in the scattering calculation since the singularities would not adversely affect the results . their calculation seemed to confirm that , at least semi - classically , quantum coherence of the incident scalar field is lost . in anticipation of extending this latter calculation , the purpose of this paper is to present , in a useful form , the classical perturbation equations for incident fields of arbitrary spin ( the teukolsky equations @xcite ) in the c metric . after briefly reviewing some properties of the c metric , in particular the ` equal temperature condition ' and the ` point - particle ' limit relevant to virtual black hole calculations , the full teukolsky equations for arbitrary spin fields are presented . the separability of these equations into one - dimensional ` radial ' and ` angular ' parts is then guaranteed ( for all fields in the vacuum case and some fields in the electrovac case ) due to the petrov type d classification ( kamran & mclenaghan @xcite ) , and this separation is subsequently demonstrated . the resulting equations are then solved in the ` point - particle ' limit to yield an approximate angular solution together with an angular momentum quantisation condition , and an approximate radial solution . transmission and reflection coefficients are then found for incident fields of a specific initial configuration , following hawking & ross . the electrovac c metric ( @xcite and further references therein ) is most commonly expressed in the form ds^2 = \ { -p(y)dt^2 + - - p(x)d^2 } . [ metric ] in this notation the ` metric polynomial ' @xmath13 is a quartic which may be written as p ( ) = \ { 1 + } \ { 1 - ^2 - ^3 } [ metpoly1 ] where @xmath14 and @xmath15 are dimensionless positive constants constrained such that @xmath13 has four real roots . this solution describes a pair of black holes of mass @xmath16 and charge @xmath17 respectively , accelerating away from each other . the parameters are related via qa = , and ma = \ { + } . the vacuum c metric solution may trivially be obtained from the above by restricting @xmath18 . the coordinates @xmath19 are suitably adapted to the timelike killing vector @xmath20 , the spacelike killing vector @xmath21 , and the nondegenerate eigenvector @xmath22 of @xmath23 . it is thus clear from the form of equation ( [ metric ] ) that the solution is static , time reversible , and axially symmetric , although these coordinates cover only the neighbourhood of one black hole . if the roots of the metric polynomial are labelled @xmath24 which , for non - extreme solutions , must satisfy @xmath25 , then equation ( [ metpoly1 ] ) may be re - expressed as p ( ) = -\{- _ 1}\{- _ 2}\{- _ 3}\{- _ 4}. [ metpoly2 ] to ensure the correct signature the coordinates must be restricted such that @xmath26 $ ] and @xmath27 $ ] . the roots then occur as shown in figure [ quartic ] where the point @xmath28 is spatial infinity , with null and timelike infinity along @xmath29 . ( 0,0)(0,0 ) ( 22,82)@xmath30 ( 137,82)@xmath31 ( 241,82)@xmath32 ( 358,82)@xmath33 ( 402,74)@xmath34 ( 316,214)@xmath13 potential conical singularities exist along the axes @xmath35 and @xmath36 . conventionally the singularity between the black holes ( along @xmath33 ) is removed by insisting periodicity in the azimuthal coordinate such that = , [ phi_period ] which unavoidably produces a conical deficit along @xmath35 . in the c metric this is generally interpreted as a cosmic string lying along the axis and terminating on the black hole . the deficit can be removed by introducing an external gauge field @xcite the ernst solution , but the lorentzian section of this metric is no longer asymptotically flat . the euclidian section of the c metric may be obtained in the conventional manner by setting @xmath37 in equation ( [ metric ] ) , where in addition the @xmath38 coordinate must be restricted to the range @xmath39 $ ] to ensure the metric is positive definite . further potential conical singularities then arise at @xmath40 and @xmath41 which must both be eliminated in order that the metric be regular . this may be achieved @xcite by requiring @xmath42 to be periodic , with = and by imposing an additional ` equal temperature condition ' on @xmath13 such that |p(_2)| = |p(_3)| . [ eqtemp1 ] the latter condition ( [ eqtemp1 ] ) , only realisable in the electrovac case , may be re - expressed as a relation between the parameters @xmath14 and @xmath15 , or alternatively as equivalent to insisting the roots @xmath24 of @xmath13 satisfy _ 2 - _ 1 = _ 4 - _ 3 . with this regularity condition the topology of the euclidean section becomes @xmath12 , where the point removed is spatial infinity @xmath28 . of particular relevance to virtual black hole calculations is the ` point - particle ' limit , in which the black holes are small on scales set by the acceleration . since an implicit relation holds between @xmath14 and @xmath15 , there exists only one dimensionless parameter in the metric , and so without loss of generality the ` point - particle ' limit may be expressed as 1 . [ pp_lim ] in this limit deviations from spherical symmetry in the @xmath43 section become small , as can be seen by making a coordinate transformation @xmath44 so that @xmath45 . this result evidently has implications for the solution of any ` angular ' equations for incident fields . equation ( [ metpoly1 ] ) may be solved for the roots @xmath24 which , under the assumption ( [ pp_lim ] ) , can be written as series in @xmath15 to give _ 1 & = & - - 2 + + ( ^2 ) , + _ 2 & = & - + + ( ^2 ) , + _ 3 & = & -1 - + ( ^2 ) , + _ 4 & = & 1 - + ( ^2 ) . clearly then , the acceleration horizon at @xmath41 is largely unaffected by variations in the parameter @xmath15 if ( [ pp_lim ] ) holds , whereas the inner and outer black hole horizons @xmath46 and @xmath31 respectively are both highly sensitive . as @xmath47 these roots @xmath48 in unison which , together with the identification of infinity at @xmath49 , supports the definition of a radial coordinate @xmath50 @xcite as r^-1 = a \{x - y}. [ r_coord ] if the axis @xmath35 is neglected all observers will intersect the acceleration horizon before reaching infinity and , in the ` point - particle ' limit , the causal structure approaches that shown in figure [ causal ] . the surface gravity @xmath51 at the outer black hole horizon and the acceleration horizon , is given by = = 1 - 2 + ( ^2 ) . [ s_grav ] ( 0,0)(0,0 ) ( 95,220)@xmath52 ( 161,220)@xmath52 ( 97,40)@xmath53 ( 155,40)@xmath53 ( 33,163)@xmath54 ( 106,163)@xmath55 ( 143,163)@xmath56 ( 215,163)@xmath57 ( 35,99)@xmath58 ( 105,99)@xmath59 ( 144,99)@xmath60 ( 214,99)@xmath61 quite apart from the additional conditions placed on the parameters to remove conical singularities and render the euclidean section regular , the c metric possesses a great deal of intrinsic symmetry which may be suitably exploited in the analysis of incident perturbing fields . in particular , the appropriately normalised null linear combinations of @xmath62 and @xmath63 are the repeated principal null vectors of both the weyl tensor @xmath64 @xcite and the maxwell tensor @xmath65 @xcite . the existence of these repeated null vectors implies the metric is of petrov type d , and perturbing fields may therefore be treated using the methods first developed by teukolsky for the kerr solution . the c metric and its associated electrovac generalisation are examples of algebraically special spacetimes , in particular they are of petrov type @xmath66 or d which amounts to saying that the weyl spinor @xmath67 may be written as _ abcd = o_(a o_b _ c _ d ) for a suitably chosen spin basis @xmath68 . this is equivalent @xcite to the existence of two real null vector fields @xmath69 and @xmath70 , everywhere satisfying l^b l^c c_abc[d l_e ] = n^b n^c c_abc[d n_e ] = 0 . @xmath69 and @xmath70 are the repeated principal null vectors of the weyl tensor @xmath64 which , for the electrovac solution , coincide with those for the maxwell tensor so that l^a f_a[b l_c ] = n^a f_a[b n_c ] = 0 . for the type d solutions the principal null congruences defined respectively by @xmath69 and @xmath70 are both geodesic and shear - free . these conditions are sufficient to guarantee , via the generalised goldberg - sachs theorem @xcite , the existence of a canonical null tetrad @xmath71 in which the newman - penrose spin coefficients @xmath72 and @xmath73 vanish identically , and in which the only non - zero tetrad components of the weyl and ricci spinors are @xmath74 and @xmath75 respectively . in the case of vacuum solutions , @xmath75 is also identically zero . assuming these results in a vacuum background , teukolsky @xcite combined the newman - penrose form of the linearised ricci and bianchi identities in pairs to generate decoupled equations satisfied by the first order gravitational perturbations @xmath76 and @xmath77 . by similarly combining the newman - penrose form of the maxwell identities he also found decoupled equations for the tetrad components @xmath78 and @xmath79 of a test maxwell field , and likewise for the tetrad components @xmath80 and @xmath81 of a test neutrino field . introducing a parameter @xmath5 as the spin - weight @xcite of the field components allows the homogeneous form of these equations to be combined and written as + & & -[-(2s-1)+-2s- ] ( + -2s ) + & & -.(2s-1)(s-1)_2}_s = 0 [ spin_plus ] for @xmath82 , and + & & -[-(2s+1)--2s+ ] ( -2s- ) + & & -.(2s+1)(s+1)_2}_s = 0 [ spin_minus ] for @xmath83 . the @xmath84 s are to be interpreted as the tetrad components of the appropriate field , which are conventionally defined as @xmath77 , @xmath79 and @xmath81 for spin - weights -2 , -1 , and -@xmath1 , and @xmath76 , @xmath78 and @xmath80 for spin - weights 2 , 1 , and @xmath1 respectively . the above results then also hold for type d electrovac solutions ( where @xmath85 ) with the exception of @xmath86 , under the assumption that the spin 1 perturbations are not in the background maxwell field but in another gauge field . in order to proceed , a choice of tetrad basis @xmath87 for the c metric must be made . the conventional reality condition on the basis vectors @xmath69 and @xmath70 requires that a different choice of tetrad basis be made for the regions in which the metric polynomial @xmath88 is greater than or less than zero . dealing first with those regions in which @xmath89 , a suitable choice is given by . ccr l_a & = & [ , 1/ , 0 , 0 ] + n_a & = & [ -,1/ , 0 , 0 ] + m_a & = & [ 0 , 0 , i/ , - ] + } . this is the canonical symmetric tetrad , in terms of which the non - zero basis components of the weyl and ricci spinors are _ 2 = a^2 \ { x - y } ^2 [ p(x ) - p(y ) ] [ psi2 ] and + & & .-2\{x - y}[p(x)-p(y ) ] ) respectively . the inherent symmetries of the chosen tetrad force the eight remaining non - zero newman - penrose spin coefficients into two real and two complex pairs , so that -= & = & a , + -= & = & a \ { \{x - y } p(y ) + p(y ) } , + = & = & a , + = & = & a \ { \{x - y } p(x ) - p(x ) } . finally , the four directional derivative operators @xmath90 , @xmath91 , @xmath92 , and @xmath93 become d & = & a\{x - y } ( - ) , + & = & a\{x - y } ( + ) , + & = & a\{x - y}(-i + ) , + & = & a\{x - y } ( i + ) . [ delta_bar ] substituting the relations ( [ psi2 ] - [ delta_bar ] ) into equation ( [ spin_plus ] ) , under the assumption that @xmath84 is functionally dependent on all coordinates , results in a second order partial differential equation for positive spin - weight fields . this may be combined with the result obtained from a similar substitution into ( [ spin_minus ] ) and written as + & & . -(2s^2 + 1 ) \{p(x)-p(y ) } ] + \{x - y } ( |s|+1 ) \ { p(y)_y + p(y ) + p(x)_x + p(x ) } + & & . - ( s^2 + 3|s|+2)\{p(x)-p(y ) } ) _ s = 0 [ teuk_eq ] which is then valid for massless fields of spin - weight @xmath94 , @xmath95 and @xmath96 . for those regions of the c metric in which @xmath97 , a suitable choice of tetrad may be obtained from the original with a boost of @xmath98 under which the newman - penrose quantities @xmath74 and @xmath75 remain unchanged , the non - zero spin coefficients transform as ccccccc & = & i , & & & = & -i , + & = & -i , & & & = & i , + & = & - , & & & = & - , + & = & - , & & & = & - , and @xmath84 acquires a phase factor _ s _ s = ( -i)^s _ s. substituting these new relations as before produces identical results to those for the @xmath99 regions , and so the teukolsky equation ( [ teuk_eq ] ) is valid throughout the c metric solution . also , by setting @xmath4 , the expression reduces to the c metric form of the conformally invariant massless scalar equation \ { - r } _ 0 = 0 , and so the domain of definition may be extended to include null scalar fields . equation ( [ teuk_eq ] ) may therefore be interpreted as describing massless fields of spin - weight @xmath0 , @xmath100 , @xmath95 and @xmath96 incident upon the vacuum c metric . in addition , it remains valid in the electrovac c metric when restricted to massless test fields of spin - weight @xmath101 , @xmath102 and @xmath95 . the form of equation ( [ teuk_eq ] ) again highlights the inherent symmetries of the chosen tetrad which effectively produce terms in matched pairs one in @xmath103 and one in @xmath38 . it is hardly surprising therefore that , with a suitably chosen ansatz for @xmath84 , the equation separates into two ordinary differential equations in the coordinates @xmath103 and @xmath38 respectively . indeed , the algebraically special character of the c metric was suitably exploited in choosing an appropriate tetrad basis for this exact purpose . this separation procedure has been shown to be possible , given a suitably chosen basis , for all type d vacuum metrics by kamran & mclenaghan @xcite , and will be explicitly demonstrated for the c metric in the next section . having derived the expressions for arbitrary spin massless perturbations in type d vacuum solutions , teukolsky demonstrated @xcite that for the kerr metric the equations separated when expressed in a kinnersley @xcite tetrad . in view of the validity of teukolsky s methods , kamran & mclenaghan conjectured that a similar separation could be performed for the entire class of type d vacuum spacetimes . this was subsequently demonstrated for @xmath4 , @xmath100 , @xmath104 and @xmath105 in @xcite . for the c metric , the nature of the conformal factor @xmath106 @xcite and the ignorable coordinates @xmath107 and @xmath108 in the metric strongly suggest an ansatz of the form _ s = [ a \{x - y } ] ^|s|+1 ^ -i(t - m ) _ s(x , y ) . [ ansatz ] since @xmath108 is a periodic coordinate ( [ phi_period ] ) , @xmath109 must be of the form m = m_0 = m_0 \ { 1 + 2 + ( ^2 ) } where @xmath110 is , without loss of generality , a positive integer . for the @xmath111 fields , acting with the first two operators in the newman - penrose form of the teukolsky equation ( [ spin_plus ] ) yields \{- 2s+ } _ s = [ a\{x - y}]^s+2 ^ -i(t - m ) _ s(x , y ) , and \ { - 2s+ } _ s = [ a\{x - y}]^s+2 ^ -i(t - m ) _ s(x , y ) , where the differential operators @xmath112 and @xmath113 may be written as _ ys_+ & = & - \ { i- s p(y ) } , + _ xs_+ & = & i + \ { m + s p(x ) } . similarly for the @xmath114 fields , acting on the same ansatz with the first two operators in equation ( [ spin_minus ] ) gives \{d - 2s- } _ s = [ a\{x - y}]^-s+2 ^ -i(t - m ) _ s(x , y ) , and \{- 2s- } _ s = [ a\{x - y}]^-s+2 ^ -i(t - m ) _ xs_- _ s(x , y ) , where now the operators @xmath115 and @xmath116 take the form _ ys_- & = & + \ { i- s p(y ) } , + _ xs_- & = & -i + \ { m + s p(x ) } . these two sets of equations may obviously be combined into a single pair valid for @xmath0 , @xmath117 , @xmath95 , and @xmath96 by defining _ ys - \ { i - s p(y ) } , and _ xs i + \ { m + s p(x ) } . the remaining operators in the newman - penrose form of the teukolsky equations ( [ spin_plus ] , [ spin_minus ] ) may be similarly expanded by allowing them to act on a coordinate - dependant function @xmath118 of the form _ s(t , y , x , ) = ^-i(t - m ) _ s(x , y ) . for the @xmath111 case this leads to + & & ^-i(t - m ) \ { a\{x - y } _ ys_+ + _ ys_+ } _ s(x , y ) , + + & & ^-i(t - m ) \ { a \{x - y } _ xs_+ + _ xs_+ } _ s(x , y ) , and for the @xmath114 case + & & ^-i(t - m ) \ { a\{x - y } _ ys_- + _ ys_- } _ s(x , y ) , + + & & ^-i(t - m ) \ { a \{x - y}_xs_- + _ xs_- } _ s(x , y ) . as for the @xmath119 operators , the @xmath120 s and @xmath121 s may be combined into pairs defined for both positive and negative @xmath5 which may be written as _ ys & & + \ { i - ( s - ) p(y ) } , + _ xs & & -i + \ { m + ( s - ) p(x ) } , and _ ys & & a \ { s + } , + _ xs & & a \ { s + } . it should be noted at this stage that each of the @xmath119 s , @xmath120 s and @xmath121 s are functions of only one coordinate , either @xmath103 or @xmath38 . if these results are combined appropriately then , given the form of the ansatz ( [ ansatz ] ) , the teukolsky equations ( [ spin_plus ] ) and ( [ spin_minus ] ) are seen to be wholly equivalent to the single expression + & & - \ { a\{x - y}_xs + _ xs } + & & . -(2|s|-1)(|s|-1)[a\{x - y } ] ^|s|+1_2 ) _ s(x , y ) = 0 . considering the action of both @xmath122 and @xmath123 on the term @xmath124^{|s|+2}$ ] and the explicit form of @xmath74 ( [ psi2 ] ) this may be substantially simplified , becoming + & & . - \ { _ xs _ xs + ( 2s^2 - 3|s| + 1)p(x ) } ) _ s(x , y ) = 0 which is clearly separable under the assumption @xmath125 . by expanding the @xmath120 and @xmath119 operators , the separated expression for the @xmath38-coordinate may be written as ( \ { p(y ) } + \ { + s p(y ) } ^2 + ( 2s^2 + 1)p(y ) + _ s ) _ s(y ) = 0 [ y_eq ] where @xmath126 is a separation constant which will in general depend on the spin - weight @xmath5 . likewise for the @xmath103-coordinate , the separated equation becomes ( \ { p(x ) } - \ { m + s p(x ) } ^2 + ( 2s^2 + 1)p(x ) + _ s ) _ s(x ) = 0 . [ x_eq ] the pair of separated equations ( [ y_eq ] ) and ( [ x_eq ] ) may be said to describe the @xmath38 and @xmath103 components respectively of fields with all spin - weights considered incident on the vacuum c metric . as before , they continue to hold in the electrovac c metric if restricted to massless test fields of spin - weight @xmath4 , @xmath117 and @xmath95 . the unification and subsequent separation of the equations describing incident fields of a variety of spin - weights evidently represents an appreciable success for the newman - penrose formalism . however the resulting expressions , when considered as a pair of second order ordinary differential equations , are seen to possess five regular singular points the four roots of @xmath13 and @xmath127 and hence can not be solved analytically . in order to furnish even an approximate solution , the ` point - particle ' limit must be exploited to suitably simplify both equations . equation ( [ x_eq ] ) for @xmath128 , together with suitable conditions of regularity at the boundaries @xmath35 and @xmath33 represents a sturm - liouville eigenvalue problem for the separation constant @xmath126 . for fixed @xmath5 and @xmath109 the eigenvalues may be labelled by @xmath129 , where it is assumed that j = j_0 + j_1 + ( ^2 ) . since the expression is self - adjoint , its eigenfunctions labelled by @xmath5 , @xmath109 and @xmath129 , will be both complete and orthogonal on the interval @xmath26 $ ] , but the existence of five regular singular points implies these solutions will not have a closed form in terms of known functions . however , considerable analytic progress may be made by moving to the ` point - particle ' limit after first changing variables x = ( x - \{_3 + _ 4 } ) where \{_4 - _ 3}. the range of @xmath103 is then mapped to @xmath130 $ ] , and the metric polynomial may be re - written as p ( ) = -^4 \ { ( + ) ^2 - 1 } \ { ^2 - 1 } with \ { _ 3 - _ 1 } . having made this change , equation ( [ x_eq ] ) becomes ( \ { p ( ) } - \ { m + s p ( ) } ^2 + ( 2s^2 + 1)p ( ) + _ s ) _ s_jm ( ) = 0 [ newx_eq ] where @xmath131 denotes differentiation with respect to @xmath103 , and @xmath132 . in the limit @xmath133 the three polynomials @xmath134 , @xmath135 and @xmath136 may be expressed as power series in @xmath15 p ( ) & = & ( 1-^2 ) \ { 1 + 2 + ( ^2 ) } , + p ( ) & = & -2 \ { - ( 1 - 3 ^ 2)+ ( ^2 ) } , + p ( ) & = & -2 \ { 1 + 6 + ( ^2 ) } . if it is further assumed that the eigenfunctions may also be written as a power series in @xmath15 = _ s^ _ jm^(0 ) + _ s^ _ jm^(1 ) + ( ^2 ) then ( [ newx_eq ] ) may be separated into a series of equations for these individual functions , the first of which is ( \ { ( 1-^2 ) } - - ( 2s^2 + 1 ) + _ s ) _ s^_jm^(0 ) ( ) = 0 . if the label @xmath129 is defined via the equation _ s = j(j+1 ) - ( s^2 - 1 ) [ gamma_s ] then , with an additional change of variables @xmath137 , this reduces to ( \ { } - \ { } ^2 - s^2 + j_0(j_0 + 1 ) ) _ s^_jm^(0 ) ( ) = 0 . since the zeroth order in @xmath15 corresponds to a limit of spherical symmetry in the @xmath138 section it is hardly surprising that the solutions are related to the spin - weighted spherical harmonics @xcite ( , ) = _ s^_jm^(0 ) ( ) ^i m . @xmath139 must therefore be half - integer , satisfying @xmath140 so in this limit @xmath139 and @xmath110 are the usual ` total angular momentum ' and ` angular momentum about the symmetry axis ' quantum numbers respectively . strictly this derivation has assumed the relation implicit between the parameters @xmath14 and @xmath15 of the metric due to the imposed ` equal temperature condition ' ( [ eqtemp1 ] ) . in this sense the solutions to ( [ x_eq ] ) , expressed in the ` point - particle ' limit as _ s(x ) = _ s_jm ( ) = ^-i m _ s y_jm ( , ) + ( ) [ x_sol ] describe the @xmath103 component of massless test fields of spin - weight @xmath141 , @xmath117 and @xmath95 only , incident on the electrovac c metric . this is the desired result for virtual black hole calculations , but for the sake of completeness , to zeroth order in @xmath15 the metric polynomial and its derivatives are identical in the vacuum ( @xmath18 ) case , and so ( [ x_sol ] ) is in addition valid for all spin - weights incident on the vacuum c metric . the evolution of fields with spin - weight @xmath5 described by equation ( [ y_eq ] ) essentially defines a one - dimensional scattering problem , upon which suitable asymptotic boundary conditions must be imposed in order that the problem be well - defined . following hawking & ross @xcite , initial data for the incident fields may be defined on an ` initial ' cauchy surface suitably constructed from the left acceleration and right black hole horizons . the scattering problem then becomes the propagation of this data forward through the spacetime to a ` future ' cauchy surface constructed from the future halves of the left and right black hole horizons and @xmath52 . the form of the initial data on the acceleration and black hole horizons is nonetheless restricted by the behaviour of the fields as the metric polynomial @xmath142 , since it is in this limit that the horizons are approached . with the exception of fields incident along the axis @xmath35 this problem may conveniently be divided into two sections propagation from @xmath60 and @xmath61 to @xmath56 and @xmath57 , followed by propagation from @xmath56 and @xmath55 to @xmath52 . the axially incident portion of the field is an exceptional case since this is the unique component which reaches @xmath52 before intersecting the acceleration horizon . however this contribution may consistently be disregarded as a set of measure zero amongst the entirety of the incident field . further insight into the asymptotic behaviour of equation ( [ y_eq ] ) may be gained by defining a conventional ` tortoise ' coordinate @xmath143 such that dy _ * = - . since @xmath144 , the range @xmath145 covers only the region between two horizons , and therefore a separate @xmath146 coordinate system is required for each ` rindler diamond ' in the spacetime ( figure [ hor_diag ] ) . attention will be focused on the first half of the scattering problem propagation through a single rindler diamond from @xmath60 and @xmath61 to @xmath56 and @xmath57 . the second half is subsequently found to be trivial . ( 0,0)(0,0 ) ( -215,-85)@xmath54 ( -158,-85)@xmath55 ( -109,-135)@xmath60 ( -78,-135)@xmath61 ( -193,-103)@xmath147 ( -169,-103)@xmath148 ( 50,-85)@xmath54 ( 103,-38)@xmath52 ( 155,-38)@xmath52 ( 200,-85)@xmath57 ( 165,-131)@xmath148 ( 188,-131)@xmath147 using the form of the metric polynomial given in equation ( [ metpoly2 ] ) the coordinate @xmath143 may be written as y _ * & = & + & = & -_1 | _ 1 - y | + _ 2 | _ 2 - y | - _ 3 | _ 3 - y | + _ 4 | _ 4 - y | where the @xmath149 are constant positive functions of the four roots @xmath24 . in the ` point - particle ' limit , they become _ i = + ( ^2 ) , [ gammas ] with a ` @xmath150 ' for @xmath151 and @xmath152 and a ` @xmath153 ' for @xmath154 and @xmath155 . clearly then , @xmath156 at the acceleration and inner black hole horizons , and @xmath157 at the outer black hole horizon . multiplying through by @xmath88 and further recognising @xmath158 allows equation ( [ y_eq ] ) to be re - written as the one - dimensional wave equation ( + \ { + s p(y ) } ^2 + v _ ) _ s(y ) = 0 where the effective potential is given by v _ = p(y ) \ { ( 2s^2 + 1 ) p(y ) + _ s } . near the outer black hole horizon @xmath142 and hence the effective potential becomes negligibly small . in addition @xmath159 where @xmath51 is the surface gravity of the horizon , and so the solution approaches . _ s(y ) |_y _ 2 ~^ i \ { - i s } y_*. similarly , near the acceleration horizon @xmath160 , and so the solution there approaches . _ s(y ) |_y _ 3 ~^ i \ { + i s } y_*. it is clear that for some values of @xmath161 the asymptotic limits of these solutions are unbounded due to terms of the form @xmath162 . such singularities are of course not physical , but instead are due to the fact that the terms @xmath163 are newman - penrose quantities , and hence projections of some physical quantity onto a tetrad which is ill - defined at the horizons . they are therefore gauge- or frame - dependent . this effect may be illustrated by re - expressing the tetrad in null coordinates and subsequently applying a position dependent boost to demonstrate that certain ` well behaved ' observers at the horizons will not witness such pathology in all of the field quantities . in ingoing coordinates @xmath164 defined via the relations dv = \ { dt + dy _ * } for the null coordinate , and dr = - \ { p(y)dy _ * + dx } for the radial coordinate ( [ r_coord ] ) , the @xmath69 and @xmath70 tetrad components may be written as l^a = [ \{x - y } , \{x - y}^-1 ^ -1 , 0 , 0 ] n^a = [ 0 , -\{x - y}^-1 ^ -1 , 0 , 0 ] where = i . both components remain singular at @xmath165 , but by applying a boost @xmath166 they become @xcite l^a = [ \{x - y } , \{x - y}^-1 ^ -2 , 0 , 0 ] n^a = [ 0 , -\{x - y}^-1 , 0 , 0 ] which are evidently ` well behaved ' at the horizons . the effect of this boost is to transform the newman - penrose quantities according to the relation _ s^(y ) = ^-s _ s(y ) , for which the @xmath38 equation becomes + & & + 1 6 ( 2s^2 - 3s+1)p(y ) + _ s^(y ) = 0 . the asymptotic form of this equation may once again be obtained with multiplication through by @xmath88 to yield ( ^2 y_*^2 + s p(y ) y _ * + \{^2 + i s p(y ) } + _ ) _ s^(y ) = 0 , the solutions to which are . _ s^(y ) |_y _ 2 ~^ -i y _ * ^\{i + 2 s } y _ * and . _ s^(y ) |_y _ 3 ~^ -i y _ * ^\{i - 2 s } y_*. in outgoing coordinates @xmath167 where @xmath147 is defined by du = \ { dt - dy _ * } , the tetrad components @xmath69 and @xmath70 are rendered ` well behaved ' at the horizons by applying a boost of @xmath168 , the effect of which is to transform the newman - penrose quantities according to _ s^(y ) = ^s _ s(y ) . in this case the asymptotic form of the @xmath38 equation is given by ( ^2 y_*^2 - s p(y ) y _ * + \{^2 + i s p(y ) } + _ ) _ s^(y ) = 0 , such that . _ s^(y ) |_y _ 2 ~^ i y _ * ^\{-i + 2 s } y _ * and . _ s^(y ) |_y _ 3 ~^ i y _ * ^\{-i - 2 s } y_*. the asymptotic behaviour of the field quantities expressed in both the symmetric tetrad and the in- and outgoing tetrads clearly exhibit the peeling behaviour expected from the peeling theorem of newman & penrose @xcite . ingoing observers ` pick out ' the ingoing component of the field to be ` non - special ' neither singular nor identically zero at the horizons @xcite , and likewise the outgoing observers ` pick out ' the outgoing components . the apparent pathology in some field components is merely a result of some gauge choice , and choosing to work in any particular gauge , with appropriately selected boundary conditions , can not therefore affect the final results . care must nonetheless be exercised when attempting to define certain quantities since @xmath38 equations such as ( [ y_eq ] ) describe the evolution of only a single spin - weight quantity @xmath163 . the failure of these equations to be self - adjoint in cases other than @xmath141 is manifested as an apparent lack of unitarity in the evolution of any one spin - weight component . however , for fields with @xmath161 there must also exist relations between the components of maximum spin - weight the teukolsky - starobinsky identities @xcite , and consequently the definitions of quantities such as conserved currents must necessarily involve both spin - weight components of a field @xcite . having laid the groundwork , the definition of asymptotic boundary conditions for the desired scattering problem may be achieved by analogy with a simple one - dimensional problem of scattering in a potential well . considering the initial hypersurfaces @xmath61 and @xmath60 , hawking & ross argued @xcite that the effects of scattering a field from @xmath61 must be the image under left - right interchange to the effects of scattering a field from @xmath60 . since the metric must be symmetric under this exchange , calculating the scattering of a field incident only from the past black hole horizon is justified since the reflection and transmission coefficients would be identical had the other choice been made instead . with this in mind , an asymptotically - plane wave outgoing from @xmath61 will propagate through the right rindler diamond and subsequently interact with some potential in the interior region . the result of this interaction must then be an ingoing reflected component through @xmath57 and an outgoing transmitted component across @xmath56 . these conditions may be written in symmetric tetrad components as _ s(y ) = ^ \ { i + s } y _ * + c_r(s ) ^ \ { -i - s } y _ * [ bh_bound ] at the black hole horizon ( @xmath169 ) , and _ s(y ) = c_t(s ) ^ \ { i - s } y _ * [ acc_bound ] at the acceleration horizon ( @xmath156 ) , where @xmath170 and @xmath171 are related to the reflection and transmission coefficients respectively , which remain to be determined . for the degenerate case of @xmath4 , the boundary conditions reduce to those chosen for the massless scalar field in hawking & ross , although the ` tortoise ' coordinate is there defined with opposite sign . having established physically consistent boundary conditions at the horizons , the remaining problem of determining the reflection and transmission coefficients reduces to that of solving ( [ y_eq ] ) subject to the constraints ( [ bh_bound ] ) and ( [ acc_bound ] ) . since this expression possesses the same regular singular points encountered in the @xmath103 equation , attempting to find an analytic solution will obviously necessitate recourse to the now familiar ` point - particle ' limit . defining a new field @xmath172 such that _ s(y ) = ^\{i - s } y _ * f_s(y ) yields a slightly more pellucid form for this problem , in which ( [ y_eq ] ) may be re - written as + & & + \ { ( i - s ) ^2 - ( i - s p(y))^2 } ) f_s(y ) = 0 . [ new_y_eq ] the boundary conditions then become f_s(y ) = ^ 2 s y _ * + c_r(s ) ^-2 i y _ * [ bound_minus ] at the black hole horizon ( @xmath157 ) , and f_s(y ) = c_t(s ) [ bound_plus ] at the acceleration horizon ( @xmath156 ) . in moving to the ` point - particle ' limit this equation is further simplified by changing coordinates y = ( y - \ { _ 3 + _ 2 } ) where \ { _ 3 - _ 2 } = \ { 1 + + ( ^2 ) } . the range of @xmath38 is then mapped to @xmath173 $ ] , and the symmetry imposed on the metric polynomial allows it to be re - written as p ( ) = -^4 \ { ^2 - ^2 } \ { ^2 - 1 } with = 1 + 4 + ( ^2 ) . with this change in place the three polynomial functions @xmath88 , @xmath174 and @xmath175 where @xmath131 denotes differentiation with respect to @xmath38 may be expressed as power series in @xmath15 which are respectively p ( ) & = & - ( 1-^2 ) \ { ( 1 - ^2 ) + 2(1 + 3 ^ 2 ) + ( ^2 ) } , + p ( ) & = & \ { ( 1-^2 ) - ( 1 - 5 ^ 2 ) + ( ^2 ) } , + p ( ) & = & \ { ( 1 - 3 ^ 2 ) + 12 ^ 2 + ( ^2 ) } . if @xmath176 then only the first terms in these series need be retained . substituting them into ( [ new_y_eq ] ) with the identification ( [ gamma_s ] ) for @xmath126 , and further assuming @xmath177 gives ( \ { ( ^2 - 1)^2 } + 2 \ { ^2 - [ 1 + 2j(j+1 ) ] } ) f_s ( ) = 0 [ mid_range ] where @xmath178 . in this approximate equation @xmath129 is used in favour of @xmath139 since regarding @xmath129 as a half - integer would introduce degeneracies that are not present in the exact equation . the solution may be expressed in the form f_s ( ) = \ { a \ { + 1 } ^2j+1 + b \ { -1 } ^2j+1 } [ centre_sol ] for arbitrary constants @xmath179 and @xmath180 . clearly this function diverges at the boundaries @xmath181 , but this is of little real concern since ( [ mid_range ] ) is not valid near those points . a separate approximation must therefore be used in the neighbourhoods of the boundaries , which may be obtained by defining a new coordinate @xmath182 where = \ { 1 - 4 q } . dealing first with the neighbourhood of @xmath183 , the leading order terms in equation ( [ new_y_eq ] ) become ( \ { q(q+1 ) } + \ { i - s } - - j(j+1 ) ) f_s(q ) = 0 . searching for a series solution of the form @xmath184 yields indicial roots @xmath185 and @xmath186 , and a general one - term recurrence relation of the form a_n+1 = a_n \ { } . this may be identified as a @xmath187 gaussian hypergeometric series , convergent throughout the unit disk @xmath188 for the non - integer parameters considered here . the solution obeying the boundary condition ( [ bound_minus ] ) is therefore + & & + c_r(s ) \{4 q}^-i f ( \{j - i + 1},\{-j - i } ; \{-s - i + 1 } ; -q ) . [ bh_sol ] moving to the neighbourhood of @xmath189 , the leading - order terms in equation ( [ new_y_eq ] ) now give ( \ { q(q+1 ) } - \ { i - s } + - j(j+1 ) ) f_s(q ) = 0 . [ plus_one ] written in this form an attempted series solution generates a two - term recurrence relation which is of little use . however , a one - term relation may be generated by instead considering the coordinate @xmath190 . in this case ( \ { ( - 1 ) } - \ { i - s } + - j(j+1 ) ) f_s(-1 ) = 0 which again gives indicial roots of @xmath185 and @xmath191 , but with a general recurrence relation of the form a_n+1 = a_n \ { } . naturally this may also be identified as a @xmath187 gaussian hypergeometric series , and the general solution for @xmath192 may be expressed as + & & + d \{q+1}^-s - i f ( \{j - i + 1 } , \{-j - i } ; \{-s - i + 1 } ; q+1 ) [ tilde_eqn ] for arbitrary constants @xmath193 and @xmath90 . if these are chosen respectively as c & = & c_t(s ) + d & = & c_t(s ) , then ( [ tilde_eqn ] ) may be transformed ( see for example @xcite ) to a solution of the form f_s(q ) = c_t(s ) \{q+1}^s f ( \{j+s+1 } , \{-j+s } ; \{s - i + 1 } ; -q ) [ acc_sol ] which clearly satisfies the boundary condition ( [ bound_plus ] ) . at this stage it should be noted that equation ( [ plus_one ] ) and its solution ( [ acc_sol ] ) will also be valid for the entire region @xmath194 $ ] . to zeroth order this solution therefore describes the form of the incident field with the desired boundary condition on @xmath56 , restricted to zero on @xmath55 , through the top rindler diamond bounded by @xmath52 . @xmath171 is therefore related to the transmission coefficient for the complete scattering problem from the previously defined initial cauchy surface to the future cauchy surface . although the definition of the gaussian hypergeometric series ensures its convergence only within the unit disk , there nonetheless exist transformation formulae for analytically continuing to the region @xmath195 @xcite . the boundary neighbourhood solutions ( [ bh_sol ] ) and ( [ acc_sol ] ) may therefore be analytically continued to large values of @xmath182 and subsequently matched to the function ( [ centre_sol ] ) which is valid in this ` central ' region . in the interest of notational convenience , the following factors are defined : u & = & , + v & = & , + w & = & , + x & = & , + y & = & , + z & = & , where it is clear from their definitions that , amongst others , the following relations hold : cccccc u(-s ) & = & z(s ) , & w(-s ) & = & x(s ) , + v^*(-s ) & = & v^-1(s ) , & y^*(-s ) & = & y^-1(s ) . [ ptm1 ] with the benefit of these definitions , the analytic continuation of @xmath192 from the black hole horizon may be written as + & & + u ( \{4 } ^-i c_r(s ) + \{4 } ^s v ) \{q}^j [ cont_bh ] for @xmath196 . likewise the analytic continuation of @xmath192 from the acceleration horizon becomes f_s(q ) = w c_t(s ) \{q}^-j-1 + z c_t(s ) \{q}^j . [ cont_acc ] by substituting for the coordinate @xmath182 defined near @xmath197 , equation ( [ centre_sol ] ) becomes + & & + ( \{-1}^j 2 ^ -j-2 b \ { q } ^-j-1 ) by the black hole horizon , and + & & + ( \{-1}^j 2^j-1 b \ { q } ^j ) by the acceleration horizon . matching these results with the corresponding analytically continued expressions ( [ cont_bh ] ) and ( [ cont_acc ] ) requires for consistency the relations u ( c_r(s ) + \{4}^s v ) = \{2 } ^2j+1 w c_t(s ) and x ( c_r(s ) + \{4}^s y ) = \ { } ^2j+1 z c_t(s ) where the phase factor @xmath198 has been absorbed into @xmath170 . these may obviously be solved for the two unknown factors . by further defining = \ { 2 } ^2 j + 1 and = they may be written as c_r(s ) = \{4 } ^s and c_t(s ) = \{4 } ^s . appropriate use of the teukolsky - starobinsky identities for the components of extreme spin - weight allows the verification of current conservation , but this relative normalisation technique may be bypassed by instead calculating @xmath199 and @xmath200 @xcite . with application of the relations ( [ ptm1 ] ) between the various factors under a replacement of @xmath201 , it is clear that , as required for consistency , _ r + _ t = 1 if _ r | c_r(s ) c_r(-s ) | and _ t | c_t(s ) c_t(-s ) | where the ` @xmath153 ' is necessary for bosonic fields , and the ` @xmath150 ' for fermionic fields . to leading order in @xmath15 this definition of the transmission coefficient reduces to _ t = ^2 | | \ { 2 - y / v - v / y } + ( ^4 ) . since @xmath5 is a discrete parameter , the term @xmath202 may be written as y / v = ( i ) where again the ` @xmath153 ' is required for bosonic fields and the ` @xmath150 ' for fermionic fields . exploiting this definition @xmath203 becomes , after a little algebra _ t = -2 ^2 \ { } ^2 | |^2 ( -1 ) + ( ^4 ) . by assumption @xmath204 where @xmath139 is a non - negative half - integer , and hence @xmath205 for integer @xmath206 . thus , in a leading order calculation , the limit may be taken in which @xmath207 tends to a non - negative integer . the first term may be re - expressed as \ { } ^2 = \ { } ^2 ^2 which in this limit becomes \ { } ^2 for all spins . the second term may be written as a finite product by recognising that | |^2 = \ { ll _ i=0^n ( i^2+n^2)^2 & + _ i = 1/2^n ( i^2+n^2)^2 & . finally , by expanding the definition of @xmath208 it is clear that @xmath209 for bosonic fields while @xmath210 for fermionic fields , and so the transmission coefficient becomes _ t = \{2 } ^4j+2 \ { } ^2 _ i=0^j_0 ( i^2+^2)^2 + for integer @xmath5 , and _ t = \{2 } ^4j+2 \ { } ^2 _ i=1/2^j_0 ( i^2+^2)^2 + for half - integer @xmath5 , where ` @xmath211 ' denotes terms of higher order in @xmath15 . clearly the largest contribution will come from the @xmath212 ( @xmath213 ) modes , for which the expressions simplify to _ t = \ { ll \{2 } ^2 ^2 & + \{2 } ^4 ( 1/4 + ^2)^2 & + \{2 } ^6 ^2 ( 1+^2)^2 & . this leading order approximation , valid for @xmath214 , may also be extended to spin @xmath3 fields incident on the vacuum c metric , although it would appear plausible to expect the overall power of @xmath15 in this approximation to remain @xmath215 when in addition these fields are considered incident on the electrovac solution . in calculating an estimate of the affect of planck size virtual black hole loops on coherence loss in incident scalar fields , hawking & ross @xcite exploited the separability of the conformally invariant scalar wave equation in one particular metric the electrovac c metric . the scattering calculation is performed in the lorentzian section , by considering the propagation of a field from an initial cauchy surface to a future cauchy surface , suitably constructed from the left and right black hole and acceleration horizons and @xmath52 . the total scattering problem is split into two distinct sections : propagation through the right rindler diamond to define reflection and transmission coefficients , followed by propagation through the top rindler diamond to @xmath52 which is found to be trivial . the transmission factor @xmath216 ( which , modulus an arbitrary phase factor , may be identified with @xmath217 ) is necessary in calculating the bogoliobov coefficients for this scattering problem and hence in estimating the number density @xmath218 of particles at @xmath52 . this procedure has herein been repeated and extended to fields of spin @xmath1 and @xmath2 by finding and separating the teukolsky equations on the electrovac c metric . ( as a byproduct the resulting equations are in addition valid for fields of spin @xmath101 , @xmath1 , @xmath2 , and @xmath3 incident on the vacuum c metric ) . it is shown that the angular parts of these solutions comprise a complete set of orthogonal polynomials @xmath219 characterised by two real numbers @xmath129 and @xmath109 and a half - integer spin - weight parameter @xmath5 . in the ` point - particle ' limit these functions reduce to the spin - weighted spherical harmonics @xmath220 with @xmath139 and @xmath110 half - integer . a suitable generalisation of the radial boundary conditions considered in hawking & ross is subsequently demonstrated to be appropriate despite the appearance of apparent pathologies in the asymptotic behaviour of non - zero spin field quantities . these effects are shown to be gauge sensitive rather than physical in nature . fixing a gauge ( or frame ) necessitates careful handling of frame independent quantities such as conserved currents , which navely appear not to be conserved since the radial equation is not self - adjoint . in the chosen gauge the radial solutions approach @xmath221 hypergeometric functions close to the horizons , and these are matched to the solution in the interior region to fix the transmission and reflection factors @xmath171 and @xmath170 . the hypergeometric approximation is in addition shown to be valid throughout the top rindler diamond and so the two factors @xmath171 and @xmath170 determine the complete scattering problem from the initial cauchy surface to the future cauchy surface . by suitably defining the transmission and reflection coefficients such that they are invariant under the transformation @xmath201 , current conservation may be demonstrated , in that @xmath222 the leading order contribution to the transmission coefficient is subsequently shown to behave as @xmath223 for @xmath214 , where @xmath15 is a small parameter . it is clear therefore that transmission is suppressed for fields of higher spin . calculations for an incident scalar field indicated that a finite non - zero number of particles should be detected at @xmath52 , implying loss of quantum coherence since each of these particles may be thought of as one of a virtual pair whose partner has fallen into the black hole . since transmission is suppressed for higher spin fields this effect will be correspondingly reduced . a full derivation of this result will be given in a subsequent paper . gibbons , how to calculate one meson exchange forces between magnetic monopoles using gravitational instantons , in _ fields and geometry 1986 _ , proceedings of the 22nd karpacz winter school of theoretical physics , karpacz , poland , ed . a. jadczyk ( world scientific , singapore , 1986 ) . n. kamran & r.g . mclenaghan , separation of variables for higher spin , zero rest - mass field equations on type d vacuum backgrounds with cosmological constant , in _ gravitation and geometry - a volume in honour of ivor robinson _ , ed . w. rindler & a. trautman ( bibliopolis , naples , 1987 ) . r. debever , n. kamran & r.g . mclenaghan , exhaustive integration and a single expression for the general solution of the type d vacuum and electrovac field equations with cosmological constant for a nonsingular aligned maxwell field , _ j math phys _ * 25 * 1955 ( 1984 ) . an excellent account of the goldberg - sachs theorem and spinor methods is given in j. stewart , _ advanced general relativity _ , cambridge monographs on mathematical physics , ( cambridge university press , cambridge , 1991 ) . the standard reference on spinor methods and the newman - penrose formalism is r. penrose & w. rindler , _ spinors and spacetime 1 : two - spinor calculus and relativistic fields _ , ( cambridge university press , cambridge , 1984 ) .

we argue that massive stars are the dominant sources of energy for the turbulent motions within giant molecular clouds , and that the primary agent of feedback is the expansion of hii regions within the cloud volume . this conclusion is suggested by the low efficiency of star formation and corroborated by dynamical models of hii regions . we evaluate the turbulent energy input rate in clouds more massive than solar masses , for which gravity does not significantly affect the expansion of hii regions . such clouds achieve a balance between the decay of turbulent energy and its regeneration in hii regions ; summed over clouds , the implied ionizing luminosity and star formation rate are roughly consistent with the galactic total . hii regions also photoevaporate their clouds : we derive cloud destruction times somewhat shorter than those estimated by williams and mckee . the upper mass limit for molecular clouds in the milky way may derive from the fact that larger clouds would destroy themselves in less than one crossing time . the conditions within starburst galaxies do not permit giant molecular clouds to be supported or destroyed by hii regions . this should lead to rapid cloud collapse and the efficient formation of massive star clusters , which may explain some aspects of the starburst phenomenon .

astro-ph0110278

giant molecular clouds ( gmcs ) are the sites of most star formation in the milky way , and the evolution of the galaxy and its stellar population are controlled in large part by the physics of gmcs . despite several decades of observations , the clouds physical nature , longevity , and modes of formation and destruction are still matters of debate . if clouds survive for more than a single dynamical time , it is plausible that their observed properties are ( like stars ) the product of internal sources and sinks of energy ( for a recent review , see * ? ? ? if instead they disappear after one crossing time @xcite then their properties will be more affected by the mechanism of their formation @xcite . however , rapid destruction also implies that the means of cloud destruction may play an equally important role . this paper considers what consequences the mechanisms of cloud destruction primarily hii regions imply for the internal motions , energy budgets , star formation rates , and lifetimes of gmcs . molecular clouds are observed to obey a set of scaling relations known collectively as larson s ( @xcite ) laws . as updated by @xcite , these can be summarized : a constant mean column density @xmath1 , corresponding to a visual extinction @xmath2 mag ; and virial balance , with a virial parameter @xmath3 @xcite of order unity @xcite . @xcite adopt the value @xmath4 , yielding the relation @xmath5 between the r.m.s . and escape velocities . ) , whose observations may be affected by beam smearing and velocity crowding ( j. p. williams 2001 , private communication ) , this assumption should be checked by future observations . ] a third but not independent property is the scaling of line width ( @xmath6 ) with cloud radius ( @xmath7 pc ) : @xmath8 km / s . as they do not represent infall , these motions are considered turbulence . the molecular gas is cold ( @xmath9 k ) and highly magnetized ( @xmath10 ) , so its motions are supersonic but roughly alfvnic . @xcite advance a model in which these properties derive from motions within the interstellar medium ( ism ) from which gmcs form . in two - dimensional simulations of turbulence within the galactic disk , they identify a population of overdensities that could be considered clouds . because of the turbulent spectrum , all of the objects formed in their simulations obey the line width - size relation noted above . however , all but the few most massive of these are transient compressions rather than self - gravitating objects ; hence their escape velocities and surface densities do not follow the virial relation . @xcite suggest that selection effects restrict observed clouds to a narrow range of inferred column densities . specifically , they note that the iras survey of @xcite may only be sensitive to an outer shell of warm dust around clouds of various columns . however , @xcite has countered that co would form and be detectable at significantly smaller columns ( @xmath11 ; * ? ? ? * ) than are typical of the observed gmcs . an alternative possibility is that the common gmc column density arises from internal cloud processes : specifically energetic feedback due to star formation ( as in the model of * ? ? ? * hereafter m89 ) . for this to be possible , two conditions must hold : 1 . stars form rapidly only in regions that exceed a critical column density ; and 2 . star formation is potentially so vigorous a source of turbulent energy that it overwhelms the natural decay of turbulence if the column is much above this critical value . under these conditions a gmc will settle into a state of energetic equilibrium much like a star s , with star formation occurring just fast enough to offset turbulent decay . the necessary column density would be roughly the critical value , hence the common value observed among gmcs . for condition ( 1 ) , @xcite proposed that star formation is inhibited in regions that are not shielded by @xmath12 visual magnitudes in each direction , because these layers are penetrated by far - ultraviolet ( fuv ) photons that elevate the level of ionization . fuv ionization thus slows ambipolar diffusion , which m89 argued to be a rate - limiting step for star formation because stellar mass regions are too highly magnetized to collapse directly . this sets the critical column density at roughly 8 visual magnitudes ( 4 on each side ) , close to the value of 7.5 observed in gmcs . corroborating this hypothesis , @xcite find a sharp distinction among substructures in the taurus clouds between those with @xmath13 mag that are actively forming stars , and those with lower extinctions that are not . in the low - metallicity environment of the smc , @xcite verified that the clouds maintain @xmath14 mag , although this required a higher column density than for milky way clouds . for condition ( 2 ) , @xcite specified protostellar winds as the agents of star formation feedback , following @xcite and @xcite , who implicated main - sequence winds , and @xcite , who realized the potential of their protostellar counterparts . subsequent numerical simulations ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) have indicated a much faster decay of turbulence than @xcite assumed , calling into question ( e.g. , * ? ? ? * ) the notion that energy injection from protostellar winds is vigorous enough to offset turbulent decay . below , we show that hii regions ( originally considered by * ? ? ? * ) represent an additional , inevitable , and more important source of turbulent energy for molecular clouds . the rapid decay of turbulent energy should logically cause an equally rapid contraction of gmcs , the end product of which is star formation . nevertheless , clouds make stars at a tiny fraction of the rate allowed by direct gravitational collapse . star formation is not only slow , but inefficient : the protostellar sources observed by @xcite compose @xmath15 of their surrounding clouds ; @xcite surveyed 54 clouds , estimating @xmath16 of their mass to be in stars , on the basis of their hii regions ; and @xcite ( hereafter , wm97 ) argued that only @xmath17 of the mass of a gmc would ever become stellar . more recently , @xcite has estimated that @xmath18 of the mass of several nearby molecular clouds is in embedded stars . the sluggishness and inefficiency of star formation can only be consistent with the rapid decay of turbulent energy if gmcs are destroyed more rapidly than they can convert themselves into stars . what process destroys molecular clouds ? @xcite note that gravitationally unbound clouds formed by turbulent compressions are easily disrupted in a single crossing time by the flows that created them . however they also show that bound objects , once formed , continue to collapse rather than re - expanding . ( in these authors simulations , such clouds are stabilized or disrupted by a local heating included to represent the action of massive star formation , a topic to be addressed in detail below . ) the @xcite survey indicates that molecular clouds are far too tightly bound to be destroyed by turbulence in the interstellar medium : their hydrostatic pressures are in excess of @xmath19 , whereas the bounding gas pressure is only @xmath20 @xcite . the ram pressure of motions in the diffuse ism ( @xmath21 , @xmath22 ) is also @xmath23 , and thus insufficient to disrupt a cloud . clouds contain a binding energy per volume that is roughly three halves their hydrostatic pressure ( somewhat less , due to magnetization ; * ? ? ? * ) ; this was radiated in the process of formation and must be resupplied to unbind the cloud . the most plausible source for this energy is young and massive stars formed within the cloud itself . several mechanisms have been considered in this regard . @xcite have shown that small clouds can be disrupted by protostellar outflows as they form low - mass stars ; however , they predict such mass loss not to be important for giant clouds . @xcite suggested that radiation pressure might unbind clouds if their stellar populations became too luminous , but neglected the reprocessing of radiation into far - infrared wavelengths to which clouds are transparent ( see * ? ? ? molecules are destroyed by fuv photons in photodissociation regions ; however , the thermal velocities of these regions are far below the escape velocities of gmcs . therefore , photodissociation regions are thought to form an atomic layer around gmcs @xcite and have only been suggested as a means of destruction for clouds that are dynamically disrupted by a different process @xcite , as @xcite may have observed around the cold cloud g216 - 2.5 . @xcite calculated the destructive effects of massive stars due to the ejection of photoionized gas , finding @xmath24 to be ejected for each blister - type hii region . supernova explosions typically add @xmath15 and at most @xmath25 to this amount , for hii regions created by single stars . a few supernovae arise from b stars in the mass range @xmath26 , which do not have appreciable hii regions @xcite ; however , these require at least @xmath27 years to evolve , and eject only @xmath28 each if they explode inside their cloud ( and much less otherwise ; ) . the main - sequence winds of massive stars emit a total energy comparable to supernovae . @xcite and @xcite have argued that stellar wind bubbles are confined within hii regions in the context of a cloudy medium . for these reasons , we concentrate on photoionization as the primary cause of destruction for gmcs ( as have @xcite , @xcite , , @xcite , and wm97 ) . the dynamical effects of stellar winds and supernovae are considered in [ winds&sne ] . the fact that massive stars are responsible for unbinding molecular clouds more rapidly than they can form stars implies that they supply more energy than is dissipated in turbulence . if a fraction of this energy is incorporated into the motions of the remaining molecular gas , it will sustain turbulence and slow the cloud s contraction . in the following sections we consider the loss and regeneration of turbulent motions ( [ sourcesandsinks ] ) , the relative importance of hii regions and protostellar winds ( [ dominance ] ) , the dynamics of an individual association ( [ obmomentum ] ) and of a population of associations ( [ pperstar ] ) , and the implications of an equilibrium between the decay of turbulence and its regeneration in hii regions for the ionizing luminosities and star formation rates ( [ sfr ] ) and lifetimes ( [ lifetimes ] ) of massive clouds . these estimates are corrected for the interaction between hii regions in [ porosity ] . finally , in [ conclusions ] we point out that the high - pressure environments of starburst galaxies prevent molecular clouds from being destroyed or even supported by photoionization . in the absence of other sources of energy , such clouds must collapse and form stars at a high efficiency ; this is a recipe for the efficient formation of massive star clusters often observed in starbursts . to consider the fates of gmcs , one must account for the gains and losses of turbulent energy @xcite . m89 assumed a range of timescales for the dissipation of turbulent energy , @xmath29 ranging from three to ten times the free - fall time of the cloud . here , @xmath30 represents both kinetic and magnetic energy associated with turbulent motions @xcite . simulations by @xcite show incomplete equipartition between these components : for the case that most resembles molecular clouds ( magnetic pressure 100 times greater than gas pressure ) , @xmath31 @xcite s assumption of a relatively long decay time scale was based on the notion that the magnetic field should cushion gas motions . testing this notion numerically , several groups @xcite have found the dissipation to be much more rapid , especially if turbulence is driven on scales smaller than the entire cloud . for the same physical situation that gives equation ( [ equipart ] ) , @xcite find @xmath32 where @xmath33 is the wavelength on which the turbulence is stirred . combining this with @xcite s virial relation @xmath34 gives @xmath35 . this dissipation time scale is shorter than the range assumed by m89 if the forcing scale for turbulence is smaller than the radius of the cloud ; the importance of the forcing scale has recently been highlighted by @xcite . to estimate how turbulence is driven one must allow for the radiative nature of the gas , which causes compressions to be very dissipative . our argument follows @xcite and m89 . an impulse ( such as a protostellar wind ) whose momentum is @xmath36 will cause a disturbance that decelerates as it sweeps into the cloud . since energy is radiated , a thin shell forms and conserves linear momentum in each direction ( e.g. , * ? ? ? * ) ; the kinetic energy associated with the motion is @xmath37 when the velocity is @xmath38 . this continues until @xmath38 has decelerated to a terminal velocity that wm97 estimate to be the effective sound speed , @xmath39 . ( note that @xmath40 far exceeds the thermal sound speed , as @xmath41 includes the total hydrostatic pressure . ) at this point the swept - up shell thickens , stalls , and loses coherence , rendering its energy to the turbulence . the increase of turbulent energy is therefore @xmath42 the efficiency coefficient @xmath43 is uncertain , and must be determined by simulation ; @xcite suggests @xmath44 to account for the energy stored in magnetic perturbations at the end of deceleration . equation ( [ regen ] ) implies that protostellar winds and hii regions generate turbulent energy in proportion to the rate at which they impart momentum to the cloud . equation ( [ regen ] ) is essentially the same formula employed previously by @xcite , m89 , @xcite , @xcite , @xcite and @xcite . it assumes that energy is injected in an explosive manner , so that the early stage of momentum - conserving thin shell expansion ( at speeds above @xmath40 ) can be separated from the later stage of turbulent dissipation ( at speeds of about @xmath45 ) . it is only valid in cases where the kinetic energy of relative gas motions is present in the center - of - mass frame of the cloud ; therefore , it does not apply to large - scale gravitational fields . for the same reason its validity is suspect if the force that gives rise to @xmath36 is applied over a large length scale compared to the cloud radius . however , we shall see in the subsequent sections that hii regions are explosive events that input momentum on scales smaller than @xmath46 , at least for clouds in our mass range of interest . what are appropriate values of @xmath36 ? in the formation of a low - mass star , a fraction of the material that accretes onto the star - disk system is redirected into winds with a characteristic velocity @xmath47 km / s . an observational analysis by @xcite implies a characteristic wind momentum of roughly @xmath48 times the mass of the star that forms ; however , this analysis is quite uncertain . models exist in which the wind removes anywhere from a tenth to half the mass flowing through the disk @xcite . we adopt one - sixth as an intermediate value ( following * ? ? ? * ) , so that the wind mass is five times smaller than the star s mass . with this choice , each star of mass @xmath49 generates a wind impulse @xmath50 where @xmath51 represents our uncertainty . in [ winds&sne ] we find that main - sequence and evolved stars contribute a comparable momentum in the six million years after their formation ; here , we restrict our attention to the more impulsive protostellar winds . if @xmath52 is the total rate at which mass is converted into stars in a cloud , equation ( [ regen ] ) gives @xmath53 as the rate of turbulence regeneration by protostellar winds . next , consider cloud destruction at a rate @xmath54 in a sequence of discrete events ( ejecting @xmath55 per event ) , no one of which completely destroys the cloud . each mass ejection delivers an equal and opposite impulse to the cloud material left behind , and each impulse increases cloud turbulence according to equation ( [ regen ] ) . since the ejection velocity must exceed @xmath56 near the surface of the cloud , @xmath57 if photoionization is the primary means of destruction , then material is ejected at about the thermal velocity of ionized gas , @xmath58 , and @xmath59 where @xmath60 is defined in analogy to @xmath51 ; by equation ( [ regen ] ) , @xmath61 an analysis of the generation of momentum in ionization fronts , presented in section [ ifmomentum ] , shows that @xmath62 . a comparison between equations ( [ pdotstar ] ) and ( [ edotionization ] ) reveals the importance of massive star formation in the support of gmcs . relevant to this comparison is the net efficiency of star formation , @xmath63 , defined as the fraction of cloud mass that will ever become stars : @xmath64 the denominator equals the initial mass of the cloud ( in the absence of ongoing accretion from the ism ) . thus @xmath65 where the brackets are time averages . @xmath63 is estimated observationally by the _ current _ fraction of mass in stars ; the two are comparable if the efficiency is low , if stars remain within their clouds @xcite . if we take @xmath66 as typical , then @xmath67 of the cloud s mass will be disrupted by blister hii regions rather than achieving stardom . equations ( [ pw ] ) , ( [ pdotionization ] ) , and ( [ sfe2 ] ) imply that the momentum contributed by protostellar winds is smaller than the contribution by hii regions so long as @xmath68 clouds that make stars as inefficiently as galactic gmcs derive more turbulent energy from blister hii regions _ alone _ than from protostellar winds . in [ pperstar ] we shall show that hii regions would be more important than protostellar winds even if star formation were efficient . the dynamical phases of expansion of an hii region have been presented in numerous prior works ( e.g. , * ? ? ? * ; * ? ? ? these authors neglected the inertia of the shell of shocked cloud gas ; we give only a cursory treatment including this inertia , for the purpose of quantifying the momentum generated . we follow @xcite ( hereafter , mw97 ) in adopting a common temperature of 7000 k for ionized gas ( and mean molecular weight 0.61 , so @xmath69 km / s ) and in approximating that @xmath70 of the ionizing photons ( emitted at a rate @xmath71 per second ) are absorbed by dust rather than gas . we will also take for the molecular gas a uniform density @xmath72 and hydrogen number density @xmath73 consistent with a mean hydrogen column density @xmath74 @xcite . after a rapid expansion to the initial strmgen radius ( using the recombination coefficient of * ? ? ? * ) @xmath75 the hii region is governed by the requirement that a very small fraction of the ionizing photons actually reach the ionization front . this causes ionized gas density @xmath76 to vary as @xmath77 where @xmath78 is the current radius of the ionization front . as the ionization front expands subsonically the density @xmath79 and pressure @xmath80 are nearly uniform within the hii region ( albeit more perfectly for embedded than for blister regions , in which a pressure gradient develops as gas accelerates away ) . so long as the hii region expands supersonically with respect to the molecular gas , it is bounded by a thin shell of dense , shocked cloud gas . the radius of this shell is nearly identical to @xmath81 , and we shall restrict our attention to the period of expansion @xmath82 , when nearly all of the mass originally within @xmath81 remains within the shell @xcite . let @xmath83 denote the shell s area and @xmath84 its mass . the shell s momentum equation is @xmath85.\ ] ] on the left of this equation is @xmath86 , the rate of increase of the shell s momentum ; on the right , the forces due to pressure ( first term in brackets ) and due to thrust caused by the exhaust of ionized gas at a velocity @xmath87 relative to the cloud . blister and embedded hii regions differ in the coefficient relating @xmath88 to @xmath81 : @xmath89 for ( blister , embedded ) hii regions whose ionization fronts are idealized as hemispheres and spheres , respectively . another difference is the relative importance of thrust and pressure in generating momentum . in an embedded region , the ionized gas is trapped and only expands homologously ; equation ( [ rhoii ] ) then implies @xmath90 , which is much less than @xmath91 when @xmath92 @xcite so that only pressure need be considered . in blister regions , on the other hand , ionized gas flows away freely allowing the ionization front to tend toward the d - critical state @xcite for which @xmath93 if the ionized gas is effectively isothermal : recoil is just as important as pressure in generating momentum . the term in brackets on the right hand side of ( [ shellmom ] ) is therefore , to a good approximation , @xmath94 for ( blister , embedded ) regions respectively , assuming a d - critical ionization front for the former . the rate at which mass is ionized is @xmath95 , or approximately @xmath96 for a blister region when @xmath97 . the ratio between the rate of momentum generation and the rate of ionization is @xmath98 , so equation ( [ shellmom ] ) implies @xmath99 where equality holds for @xmath100 , i.e. , a d - critical front . the expansion of the hii region is simplest to determine when @xmath92 . in this phase , @xmath101 assuming radial expansion . we seek a self - similar expansion of the form @xmath102 ; equation ( [ shellmom ] ) admits the solution @xmath103^{2/7 } r_{\rm st , 0}^{3/7 } ( { c_{\rm ii } } t)^{4/7 } \nonumber\\ & = & ( 23,19 ) \times \left(\frac{t}{3.7~{{\rm myr}}}\right)^{4/7 } \left(\frac{n_{h,22}}{1.5}\right)^{-3/7 } \nonumber \\ & ~ & \times { m_{\rm cl,6}}^{1/7 } s_{49}^{1/7 } ~{{\rm pc}}\end{aligned}\ ] ] for ( blister , embedded ) regions , respectively . in this equation the first line indicates @xmath104 , which fixes the coefficient in the second line . we have normalized to the ionization - averaged lifetime of rich ob associations ( m97 ) of 3.7 myr . the momentum of radial motion of the expanding shell is @xmath105 for ( blister , embedded ) regions . note that the extra thrust generated at the ionization front in a blister region compensates for the smaller working surface @xmath88 . indeed , the two results are so similar that we may estimate @xmath36 using the intermediate coefficient @xmath106 , without discriminating between the two types of regions . this will simplify the analysis in [ pperstar ] . for a blister region , the mass evaporated is @xmath107 : for @xmath108 , @xmath109 the axisymmetric calculations of give a result that is only @xmath110 lower , once we account for the differences between their ionized sound speed and recombination coefficient and ours ( without these corrections , their result would be @xmath111 higher ) . this favorable comparison gives us confidence in the approximations adopted in this section . note that the above results differ quantitatively from those given by @xcite , which were adopted by wm97 to study the erosion of molecular clouds ( see [ ifmomentum ] ) . our equation ( [ meveval ] ) for the mass evaporated agrees with these authors results within @xmath112 . however , equation ( [ riieval ] ) gives a radius for a blister or an embedded region that is smaller by a factor of 1.6 or 1.9 , respectively , than the characteristic size quoted by these authors for blister regions implying this characteristic size is best interpreted as the diameter of the hii region ( note that @xmath113 ) . the above equations must be restricted to hii regions that are still expanding and bounded within their gmc at the end of their ionizing lifetimes . if its luminosity is too low , the region will decelerate to @xmath40 and stall before its driving stars burn out ; as wm97 argue , such associations are too small to matter ( but see [ conclusions ] ) . if its luminosity is too high , the hii region will envelop its entire cloud ( @xmath114 , deforming the gmc into a cometary configuration @xcite rocketing away from the association @xcite . since this reduces the rate of photoevaporation considerably , wm97 argued that an appropriate maximum value for @xmath55 should be roughly the value predicted for a blister region at twice the time required for its size to match the cloud radius . using @xcite s theory for the size scale , wm97 identified a maximum value @xmath115 because the blister hii region radius derived in equation ( [ riieval ] ) is smaller by a factor 1.6 than in @xcite s theory , this upper limit corresponds to evaluating equation ( [ meveval ] ) at a time when @xmath116 . since this upper limit is quite uncertain , and since there is an ambiguity between the radius and diameter of the hii region in wm97 s argument , we shall simply adopt their value as given in equation ( [ mevmax ] ) . correspondingly , we shall take an upper limit for the effective injection of momentum to be @xmath117 note that these upper limits on @xmath55 and @xmath36 are important for hii regions if @xmath118 a more serious limitation on the theory presented here arises from the fact that we have ignored gravity in the evolution of an hii region . if the gmc is too dense ( @xmath119 ) , its free - fall time will be shorter than the typical ionizing lifetime of 3.7 myr . this is typically true of clouds unless @xmath120 for clouds below this limit , the orbital motion of an association within or about the cloud is likely to alter its hii region , possibly by converting it into a cometary hii region of the type discussed by , @xcite and @xcite . lastly , we have not attempted to account for the inhomogeneities of molecular cloud material in the evolution of an hii region ( e.g. , * ? ? ? * ) , nor to instabilities that may develop during its expansion @xcite . the finite porosity of the gmc ( i.e. , the interaction between hii regions ; see wm97 ) will be accounted for in an approximate manner in [ porosity ] . we now consider the effects of stellar winds and supernovae on the evolution of an hii region . results in this section are based on the stellar evolution code starburst99 @xcite , for which we have used a scalo stellar initial mass function normalized as in mw97 . in the first 3.7 million years after stars form , they inject roughly @xmath121 of momentum : @xmath122 from protostellar winds , @xmath123 from main - sequence and evolved stars , and @xmath124 in supernova ejecta . in the formation of a sufficiently massive stellar cluster , protostellar winds are mostly stopped within the self - gravitating clump from which the cluster arose @xcite and will affect this clump s dynamics @xcite more than those of the surrounding cloud . the remaining contribution is @xmath125 if , in the first 3.7 myr , this is smaller than the momentum imparted by the hii region , the ram pressure from stellar winds is lower than @xmath80 . this means that the winds will be confined within the hii region , while the ionizing stars shine , unless @xmath126 for ( blister , embedded ) hii regions , respectively . sufficiently luminous associations couple primarily through stellar ejecta rather than through their hii regions ; this is significant for the massive clusters forming in starburst environments @xcite and for the most luminous of milky way clusters . equation ( [ windrelevance ] ) accounts only for the momentum in material flung away from a star . more radial momentum can potentially be generated in a pressurized bubble or blastwave that entrains ambient mass , as the radial momentum varies with energy @xmath127 and total mass @xmath128 as @xmath129 . whereas in a thin shell @xmath36 is conserved while @xmath127 is not , in a fully adiabatic bubble or blastwave @xmath127 is conserved while @xmath36 increases as @xmath128 does . intermediate cases , such as pressure - driven snowplows @xcite and partially radiative bubbles @xcite , involve a loss of energy but a gain of radial momentum . @xcite have argued that , in the context of an inhomogeneous medium , stellar wind bubbles are either confined within their hii regions or made radiative by mass input from photoevaporating clumps . in a blister hii region , furthermore , hot gas can escape from the cloud . for these reasons we shall assume that stellar wind bubbles do not generate momentum significantly in excess of the wind momentum itself . supernovae merit special attention , as supernova remnants typically experience an adiabatic sedov - taylor phase ( @xmath130 ) followed by a pressure - driven snowplow phase ( @xmath131 ) . for an upper imf cutoff of @xmath132 , the first supernovae explode 3.6 myr after the onset of star formation and thereafter occur every @xmath133 myr for the first million years , slowing to once per @xmath134 myr thereafter . this frequency should be compared with the sound - crossing time of the hii region , @xmath135 myr , and with the @xmath136-folding decay time of the ionizing luminosity , roughly 2.5 myr . in a rich association the delay between supernovae is the shortest of these , implying that supernovae blend together with the stellar winds . the above arguments indicate that the combined effects of supernovae and winds are negligible unless equation ( [ windrelevance ] ) is satisfied . as a check , we have calculated the dynamics of the very first supernova remnant inside the hii region , using the theory of @xcite . because the progenitor mass is very high ( @xmath137 in the presupernova state ) , the remnant becomes radiative before a comparable mass has been swept up , skipping the sedov - taylor phase ( see also * ? ? ? the high progenitor mass also allows for a transition to a momentum - conserving snowplow phase while the remnant is still expanding supersonically . the remnant becomes subsonic and merges with the hii region before striking its periphery if @xmath138 for ( blister , embedded ) regions . if this happens , the remnant adds little or no momentum to the hii region . this corroborates the above conclusion that supernovae are not significant for the momentum of the hii region , except for regions smaller than the limit in equation ( [ snmerge ] ) or larger than the limit in equation ( [ windrelevance ] ) . the possibility remains that supernovae in small associations contribute non - negligibly to the total momentum input and hence reduce the ionizing flux and star formation rate we derive below by neglecting supernovae ; this merits further study . if a cluster were to deliver an impulse in excess of @xmath139 to its parent cloud in a time short compared to the cloud s dynamical time , the cloud would be dynamically unbound . this is an unattainably large value for giant clouds with masses of @xmath140 , but smaller clouds might be disrupted in this manner . the taurus - auriga and ophiucus clouds have @xmath141 , and may be susceptible to disruption . unfortunately the likelihood and frequency of this process is difficult to assess , as it depends on several uncertain elements : @xmath142 ( eq . [ [ pmax ] ] ) , the maximum size of a cluster that can form within a given cloud ( wm97 ) , and the importance of gravity for hii regions in small clouds ( [ validity ] ) . for turbulence driven by expanding shells , the relevant forcing scale is the radius of a shell once it has decelerated and become subsonic relative to @xmath40 . this is @xmath143^{1/3 } , { r_{\rm cl}}\right\}\ ] ] for ( blister , embedded ) regions , respectively . we do not allow @xmath144 to exceed the cloud radius ; ignoring stellar winds , this limit becomes important when @xmath145 our estimate @xmath146 is based on @xcite s formula for the mean pressure within gmcs . how is the effective forcing wavelength @xmath33 related to the merging radius @xmath144 in equation ( [ tdissnum ] ) ? diametrically opposed regions of a disturbance move in opposite directions , and should be considered half a wavelength apart : this suggests @xmath147 . a similar conclusion follows from the fact that a shell s expansion could be considered a quarter cycle of oscillation ( and its collapse , a second quarter cycle ) ; for these reasons we set @xmath148 and consider @xmath149 an uncertain parameter of order unity . we shall find in [ sfrgalactic ] and [ porosity ] that this is roughly consistent with the galactic ionizing luminosity and star formation rate . conversely , if future numerical simulations indicate a value of the product @xmath150 much less than unity , then feedback from hii regions would not explain the ionizing luminosity and star formation rate in the inner galaxy . this question should soon be addressed by simulations like those of @xcite and @xcite . note that this requires @xmath40 and @xmath151 to be identified in the simulation volume , so that loss of energy by decelerating shells can be discriminated from turbulent dissipation : combining equations ( [ equipart ] ) , ( [ tdissnum ] ) , ( [ regen ] ) , and ( [ lambdaandrmerge ] ) , @xmath152 where @xmath153 refers to the creation of momentum by explosive events in a simulation driven by point sources , and the coefficients 0.78 and 0.83 should be adjusted to agree with the dissipation rate and degree of equipartition , respectively , in simulations of homogeneously driven turbulence ( as in eqs . [ [ equipart ] ] and [ [ tdissnum ] ] ) that share global parameters like the ratio of magnetic to gas pressure . the numerical determination of the product @xmath150 is essential to evaluate the theory presented here , or indeed to estimate stellar feedback in general . we now wish to account for the dynamical feedback from the entire population of hii regions that will exist within a given gmc . a similar project was undertaken by mw97 , who modeled the luminosity function of hii regions in the galaxy , and wm97 , who calculated the lifetimes of gmcs after hypothesizing how the galactic population of hii regions was distributed amongst gmcs . our discussion follows these prior works to the greatest degree possible , except that we shall solve self - consistently for the rate of star formation within a given gmc . mw97 adopted a @xcite stellar initial mass function ( imf ) , with an upper cutoff at @xmath132 , normalized for a mean stellar mass @xmath154 . they used the results of @xcite for the ionizing fluxes and lifetimes of massive stars : averaged over the imf , this gives a mean ionizing flux @xmath155 and a mean ionizing lifetime @xmath156 myr . they assumed that the fraction of ob associations born with more than @xmath157 stars , which we denote @xmath158 , satisfies @xmath159 from a lower limit of 100 up to a maximum of @xmath160 stars . with these limits , mw97 were able to fit surveys of luminous hii regions ( e.g. , * ? ? ? * ) , the total star formation rate in the galaxy , the birthrate of nearby associations , and the galactic recombination radiation . a cluster s ionizing luminosity is dominated by its most massive members ; this implies a distinction between _ rich _ clusters ( @xmath161 ; * ? ? ? * ) , which are populous enough to sample the imf up to the upper cutoff , and _ poor _ clusters , which are not . the ionizing luminosity of a poor cluster reflects its most massive member , and rises only statistically with @xmath157 . the cluster s ionizing lifetime likewise reflects this star s main - sequence lifetime ( @xmath162 ; @xcite , mw97 ) . in contrast , a rich cluster s ionizing luminosity and lifetime represent averages over the imf : @xmath163 and @xmath164 , respectively . this distinction introduces a turnover in the cluster luminosity function @xmath165 at the boundary between rich and poor clusters : @xmath166 for rich clusters , but statistical fluctuations cause a flattening of @xmath165 for poor clusters ( @xcite and mw97 ) . associations near this turnover dominate the feedback from hii regions on gmcs , because @xmath36 ( and for blister regions , @xmath55 ) scale as @xmath167 . for instance , wm97 found that half of all photoevaporation in clouds with @xmath168 is accomplished by regions with @xmath169 . it will be useful to define the cluster - weighted mean of a quantity @xmath170 as @xmath171 for mw97 s galactic hii region luminosity function the mean mass per association is @xmath172 , corresponding to @xmath173 . other useful averages are listed in table [ moments ] . ccccc @xmath174 & 4/7 & 9/7 & 0.499 & ( [ vchariiestimate ] ) , ( [ mdestiiestimate ] ) + @xmath175 & 16/21 & 12/7 & 0.572 & ( [ ndotassocs ] ) , ( [ mdotstargmc ] ) + @xmath176 & 1/14 & 9/7 & 1.856 & ( [ lifetime ] ) + @xmath177 & 3/7 & 19/7 & 2.128 & ( [ q0 ] ) + the galactic population of gmcs , @xmath178 , satisfies @xmath179 where @xmath180 from an undetermined lower limit to an upper limit of @xmath181 ( mw97 ) . mw97 argue that a given cloud can not make arbitrarily large ob associations . taking the maximum cluster mass to be @xmath17 of the cloud mass , they derive the upper limit @xmath182 which we also adopt . hii regions must therefore occur within a given gmc in different proportions than they are found in the galaxy . wm97 construct a cloud s luminosity function @xmath183 by assuming : 1 . no clusters form above @xmath184 ; 2 . since gmcs give birth to hii regions , the galactic luminosity function must equal the sum of all gmcs luminosity functions ; and 3 . within the clouds that can form a given size of cluster , the birthrate of those clusters is proportional to cloud mass . these assumptions led to mw97 s equation ( 23 ) for the population of associations forming within a given cloud . below , we shall solve for the star formation rate in a given cloud under the hypothesis that it is supported by hii regions forming within it ; this requires that assumption ( 3 ) be dropped . if the star formation rate within clouds scales as @xmath185 , a derivation analogous to mw97 s equation ( 16 ) gives @xmath186 } { 1 - ( { \cal n}_\star/1.2\times 10 ^ 6)^{\beta-\alpha } } f_a(>{\cal n}_\star),\ ] ] where @xmath187 for @xmath188 is the step function and @xmath189 . mw97 assumed @xmath190 ; below we derive @xmath191 for the mass range of interest , and in [ porosity ] we find @xmath192 after accounting for the interaction between hii regions . the above birthrate distribution is used in monte carlo simulations ( as described by mw97 ) to produce the figures in this section . however , the following argument indicates that the difference between @xmath193 and @xmath194 is unimportant for clouds in the mass range where our theory is valid . for analytical estimates it is useful to note that the associations responsible for most of the mass ejection and energy injection ( @xmath195 ) are much smaller , for giant clouds , than those associations whose existence or behavior are affected by their sizes or the finite sizes of their parent clouds . specifically , we noted in [ validity ] that our neglect of cloud gravity is only valid for clouds with @xmath196 . for such clouds , @xmath197 ; the hii regions whose maximum sizes ( eq . [ [ s49mevmax ] ] ) and merging radii ( eq . [ [ rmergeisrcl ] ] ) approach @xmath46 have @xmath198 , increasing rapidly with @xmath199 ; and those for which winds and supernovae are important ( eq . [ [ windrelevance ] ] ) have @xmath200 . these facts have two implications : 1 . a cloud s luminosity function @xmath183 only differs from the galactic form @xmath165 for luminosities near @xmath184 , so we may neglect this difference in analytical estimates ; and 2 . we may assume that hii regions are entirely contained within their parent clouds . these approximations , to which we refer collectively as the _ giant - cloud approximation _ , are the same that led to mw97 s equation ( 40 ) . for clouds with @xmath201 it might be more appropriate to take a small - cloud approximation in which all hii regions are assumed to outgrow their clouds : @xmath202 , @xmath203 , and @xmath204 ( cf . . this would be very uncertain , however , both because our estimates of @xmath205 and @xmath142 are uncertain and also because we have neglected cloud gravity , which should be important when @xmath206 . the performance of the giant - cloud approximation is illustrated in figure [ fig0 ] , where we plot its prediction for @xmath207 ( _ dashed line _ , eq . [ [ momentablemb ] ] ) against the results of a monte - carlo simulation ( _ solid line _ ) . the simulation accounts for the difference between the luminosity distribution of ob associations within a given gmc ( @xmath194 , eq . [ [ f_a , m ] ] ) and that within the galaxy as a whole ( @xmath193 , eq . [ [ faform ] ] ) , whereas this distinction is neglected in the giant - cloud approximation . this is the primary cause for the difference between the two curves for @xmath208 , to the right of the _ vertical dotted line _ ; for instance , the kink in the simulation at @xmath209 is caused by the kink in @xmath194 in equation ( [ f_a , m ] ) . the simulation also accounts for finite - size effects ( @xmath142 , eq . [ [ pmax ] ] ) ; these dominate the difference between the curves to the left of the dotted line . however , our neglect of cloud gravity is not valid in that region ( eq . [ [ mcllimitfromgrav ] ] ) . a comparison between @xmath210 and @xmath211 gives the effective input of momentum per stellar mass averaged over hii regions . in the giant - cloud approximation , an integral of equation ( [ momentablemb ] ) over @xmath165 gives @xmath212 the characteristic velocity ( momentum per stellar mass ) identified in this equation is much larger than the corresponding coefficient @xmath213 that we have estimated for protostellar winds in equation ( [ pw ] ) , or the value @xmath214 estimated for main - sequence and evolved star ejecta in equation ( [ pej ] ) . since equation ( [ vchariiestimate ] ) applies to both blister - type and embedded hii regions , it implies that hii regions are more important than the combined effects of protostellar winds , stellar winds , and supernovae in driving turbulence within gmcs , regardless of the efficiency of star formation . figure [ fig0 ] makes this point graphically . our estimate of @xmath215 indicates that hii regions dominate over protostellar winds so long as @xmath216 . this , in turn , suggests that protostellar winds may support small clouds and the self - gravitating clumps within gmcs @xcite , whereas hii regions support the gmcs themselves . if we assume that all hii regions evolve into blister regions , as have @xcite , @xcite , and wm97 , we arrive at an upper limit to the amount of mass than can be removed from a gmc by photoevaporation . since @xmath217 for a blister region , equation ( [ vchariiestimate ] ) gives , in the giant - cloud approximation , @xmath218 implying a typical instantaneous star formation efficiency @xmath219 @xmath63 and @xmath220 are related by @xmath221 @xcite ; however , a sudden disruption of the cloud ( a precipitous drop of @xmath220 , as discussed in [ oblit ] ) would reduce @xmath63 below the typical value of @xmath220 . the lines marked _ a _ in figure [ fig1 ] compare equation ( [ sfetotestimate ] ) with the results of a monte - carlo simulation which accounts for the birthrates of ob associations within individual clouds as given by equation ( [ f_a , m ] ) . the lines marked _ b _ account self - consistently for the frequent interaction between hii regions , in a manner we discuss in [ porosity ] . unfortunately , there are insufficient observations of clouds in the mass range @xmath222 to test equation ( [ sfetotestimate ] ) directly . observations of nearby clouds , which are too small for an application of our theory , indicate a significantly lower star formation efficiency than an extrapolation of our theory would predict : for instance , observations of nearby small clouds @xcite indicate that only @xmath223 of the mass is stellar and only @xmath224 will ever be . besides gravity , a number of other effects may be important for clouds this small ; wm97 have emphasized their disruption by hii regions that outgrow their boundaries ( see also * ? ? ? * ) , as well as the likelihood that this disruption will render them susceptible to photodissociation . as a cloud s turbulence decays it must be replenished , if virial balance is to be maintained . energy can be derived from gravitational contraction , winds and supernovae , and hii regions . contraction is problematic as it would imply a collapse rate close to free - fall , hence rapid star formation ; this may hold for the formation of stellar associations , but can not for entire gmcs . we have shown that hii regions are the most effective of the remaining sources . accordingly we will match the loss of turbulent momentum with its regeneration in hii regions . we must account for a population of hii regions with different luminosities and therefore different values of @xmath36 , @xmath144 , and @xmath33 . equations ( [ diss ] ) and ( [ tdissnum ] ) predict that energy decays in a time proportional to @xmath33 . it is most consistent to assume that the contribution from each type of hii region decays independently on its own timescale ; the total turbulent energy is then the sum of these decaying contributions . taking the decay time from equation ( [ tdissnum ] ) , and taking the energy input @xmath225 from equation ( [ regen ] ) , @xmath226 where @xmath227 is given by equations ( [ rmerge ] ) and ( [ lambdaandrmerge ] ) and @xmath228 is the formation rate of associations . applied to a gmc , equation ( [ eturbdiss ] ) must be consistent with the observed kinetic energy of the cloud and the expected degree of equipartition between kinetic and magnetic energy ( eq . [ [ equipart ] ] ) . equating expressions ( [ equipart ] ) and ( [ eturbdiss ] ) gives the formation rate of associations : @xmath229 if one were to vary the ionizing flux of all stars in the imf by the same factor , then @xmath228 would vary as @xmath230 . this scaling holds also for the star formation rate @xmath231 . but , the mean ionizing flux produced by these associations would vary much less : @xmath232 . this is not surprising , as the ionizing photons are directly responsible for sustaining equilibrium . a cloud s ionizing luminosity is therefore quantity most tightly constrained by the assumption of equilibrium : in the giant - cloud approximation , @xmath233 for hii regions that are primarily in the ( embedded , blister ) state , respectively . this equation will be adjusted for the interaction between hii regions in [ porosity ] . to compare with observations of individual clouds one must account for the leakage of ionizing photons ; mw97 estimate that @xmath234 escape the observed hii regions . the star formation rate is @xmath235 . using the giant - cloud approximation and mw97 s imf , @xmath236 for ( embedded , blister ) regions . recall that @xmath237 is predicted less robustly than @xmath238 due to uncertainties in the mean stellar mass and ionizing luminosity . @xcite find that the ratio @xmath239 varies by about @xmath240 among the forms of the imf and sets of stellar tracks they consider ; the predicted @xmath52 is therefore uncertain by at least this amount . the mean ionizing luminosity and lifetime depend on the metallicity @xmath241 of the stellar population , and these effects cause a shift in the star formation rate relative to equation ( [ mdotstargmc ] ) even if the imf remains constant . numerical integrations of equation ( [ shellmom ] ) , performed for starburst99 populations of various @xmath241 , exhibit @xmath242 implying @xmath243 . in any situation where m89 s theory holds , @xmath244 and therefore @xmath245 . our hypothesis that the star formation rate is determined by a balance between the dissipation of turbulence and its regeneration in hii regions can only make sense if it predicts more than one hii region per dynamical time of the cloud . this is true for all the clouds within the regime of validity of our theory , as we find that @xmath246)$ ] so long as @xmath196 . in figure [ fig2 ] we plot @xmath247 , the time scale for gas to be converted into stars . equation ( [ mdotstargmc ] ) is shown as the _ dashed curve _ marked _ a_. the _ solid curve _ marked _ a _ is the results of a monte - carlo simulation which incorporates equation ( [ faform ] ) and the effects of finite cloud size ( eqs . [ [ pmax ] ] and [ [ rmerge ] ] ) . the curves marked _ b _ are adjusted to account for the interaction between hii regions , as described in [ porosity ] . also plotted ( _ dash - dot line _ ) is value of @xmath248 appropriate to the entire inner galaxy , and the range suggested by @xcite s observations of clouds around @xmath249 , 190 - 440 myr ( corresponding to the acceptable age range 3 - 7 myr for the embedded population ; see below ) . equations ( [ stotgmc ] ) and ( [ mdotstargmc ] ) refer only to a single molecular cloud . integrating over the gmc population for the inner galaxy , we arrive at the total inner - galactic ionizing luminosity @xmath250 and star formation rate @xmath251 . wm97 model the gmc population of the inner galaxy as @xmath252 ; setting @xmath253 as observed , and considering _ only _ giant clouds with @xmath254 for which we are relatively confident of the star formation rate , we find @xmath255 and @xmath256 these results would be @xmath257 higher if we were to extrapolate the giant - cloud approximation below the lower limit of @xmath258 . as before , @xmath250 is much better constrained than @xmath251 by our hypothesis that hii regions dominate feedback . we may use observations to check that clouds with @xmath259 , whose evolution we can not address , do not dominate the galactic star formation rate or ionizing flux . consider @xcite s observations of the perseus , orion a , orion b , and mon r2 clouds ( @xmath260 for each ) . @xcite estimated the completeness of the 2mass survey and the resulting stellar fraction in these clouds , a slowly increasing function of the assumed age of the observed stellar population . @xcite argues that that these clouds have created a stellar mass equal to @xmath261 of their own mass in a period between 3 and 7 myr . extrapolating this star formation rate per unit molecular mass to all clouds with @xmath259 , using mw97 s cloud mass function , gives a range of 0.7 to 1.7 @xmath262 per year . this estimate should be considered very uncertain , as it assumes @xmath263 for small clouds ; however it indicates that clouds in the mass range @xmath264 , which constitute one third of the molecular mass , most likely do not dominate the galactic star formation rate . mw97 estimate the ionizing flux of the inner galaxy at @xmath265 ( assuming all ionizing photons are caught within the disk ) and derive a star formation rate of @xmath266 . [ @xcite estimates the star formation rate in the inner galaxy to be @xmath267 . ] equations ( [ stgalaxy ] ) and ( [ mdottgalaxy ] ) are comparable to this value , with @xmath44 as suggested by @xcite , and with @xmath268 ( close to unity as suggested in [ forcing ] ) . given the approximations that led to equation ( [ mdottgalaxy ] ) , this degree of agreement is quite remarkable ; it supports our proposition that feedback from hii regions regulates the rate of star formation in giant molecular clouds . we may combine the star formation rate given in equation ( [ mdotstargmc ] ) with the upper limit for mass evaporation from gmcs given in equation ( [ mdestiiestimate ] ) , to arrive at a lower limit for the lifetime of gmcs assuming all hii regions evolve into blister regions . in the giant - cloud approximation , this gives @xmath269 for @xmath270 . a significant population of hii regions that remain embedded rather than evolving into blister regions would extend this lifetime , but we argue in [ porosity ] that such a population is not likely . another effect that extends cloud lifetimes relative to @xmath176 is the finite porosity of hii regions within their volumes ( mw97 ) , for which we account in [ porosity ] . note that the estimate of the cloud destruction time in equation ( [ lifetime ] ) is roughly @xmath271 times longer than the cloud s free - fall time ; the two time scales are equal for a mass @xmath272 , very close to the upper mass limit for galactic gmcs ( @xmath273 ; wm97 ) . this raises the possibility that the upper mass limit derives from the difficulty of assembling an object that destroys itself rapidly compared to its free - fall time scale , a hypothesis that should be tested against extragalactic observations . we refine this argument in [ porosity ] , where we adjust the destruction time for finite cloud porosity . ( but note , a possible counterargument is raised in [ conclusions ] . ) cloud destruction timescales are plotted in figure [ fig3 ] , where our monte - carlo simulation ( _ solid lines _ ) is compared with the results of the giant - cloud approximation as given by equations ( [ lifetime ] ) and ( [ lifetimewithq ] ) for cases _ a _ and _ b _ , respectively . the curves marked _ b _ are adjusted for the interaction between hii regions ( [ porosity ] ) . the free - fall time is also plotted , making visible the crisis of rapid destruction for clouds above the galactic upper mass limit . our equation ( [ lifetime ] ) is based on the same physics that led mw97 to estimate a destruction time of 20 - 25 myr for clouds with @xmath274 . our estimate is shorter because mw97 assumed @xmath263 , whereas an energetic balance requires @xmath275 in our giant - cloud approximation ; this redistribution shortens the lives of the massive clouds to which our theory applies . the frequency with which hii regions form within gmcs implies that they are likely to interact ( wm97 ) . defining the porosity @xmath276 as the time - averaged volume filling factor of hii regions , @xmath277 dt\ , df_a$ ] , we estimate from the above equations @xmath278 for ( blister , embedded ) regions respectively . this expression is not yet fully self - consistent ; we find @xmath276 in terms of @xmath177 below . the large value of @xmath177 for embedded regions implies that fully embedded hii regions are ruled out , as hii regions will percolate and allow their gas to vent . this reinforces our neglect of embedded regions in equation ( [ lifetime ] ) . wm97 argue that the effect of interactions is equivalent to regrouping stars into non - interacting regions that are larger by the factor @xmath279 , i.e. , @xmath280 and @xmath281 . for each of the regrouped regions , @xmath282 , @xmath283 , and , for the giant - cloud approximation ( @xmath284 ) , @xmath285 . whereas wm97 assumed a star formation rate within a given cloud , our theory demands that we determine the star formation rate self - consistently by matching turbulent decay with driving by hii regions . applying these transformations to equation ( [ ndotassocs ] ) and eliminating @xmath286 , we find that @xmath287 where @xmath288 is the uncorrected value in equation ( [ ndotassocs ] ) . but , the porosity is proportional the star formation rate : @xmath289 . a self - consistent value of @xmath276 therefore satisfies @xmath290 so that @xmath291 when @xmath292 , but @xmath293 when @xmath294 . we find that the useful relation @xmath295 holds within @xmath112 in the cloud mass range of interest . the analytical ( equation [ [ q0 ] ] ) and numerical values of @xmath177 and @xmath276 are plotted in figure [ fig4 ] . to account self - consistently for finite porosity , we must increase @xmath228 , @xmath296 and @xmath52 by @xmath297 relative to equations ( [ ndotassocs ] ) , ( [ stotgmc ] ) , and ( [ mdotstargmc ] ) , increase the cloud lifetimes by @xmath298 relative to equation ( [ lifetime ] ) , and increase the stellar mass per ejected mass @xmath299 by @xmath300 relative to equation ( [ sfetotestimate ] ) . the revised equations are @xmath301^{26/21 } } \left(\frac{n_{h,22}}{1.5}\right)^{1.75 } \left(\frac{{c_{\rm eff}}}{0.57{v_{\rm rms}}}\right)^{-0.83 } \nonumber \\ & ~ & \times { m_{\rm cl,6}}^{1.38 } ~{m_\odot}~{\rm yr}^{-1},\end{aligned}\ ] ] @xmath302^{26/21 } } \left(\frac{n_{h,22}}{1.5}\right)^{1.75 } \left(\frac{{c_{\rm eff}}}{0.57{v_{\rm rms}}}\right)^{-0.83 } \nonumber \\ & ~ & \times { m_{\rm cl,6}}^{1.38 } ~{m_\odot}~{\rm yr}^{-1},\end{aligned}\ ] ] @xmath303^{0.81 } \left(\frac{n_{h,22}}{1.5}\right)^{-0.85}\nonumber \\ & ~&\times \left(\frac{{c_{\rm eff}}}{0.57{v_{\rm rms}}}\right)^{0.54 } { m_{\rm cl,6}}^{-0.35 } ~{{\rm myr } } , \end{aligned}\ ] ] and @xmath304^{-3/7 } } \left(\frac{n_{h,22}}{1.5}\right)^{0.89 } \nonumber \\ & ~&\times \left(\frac{{c_{\rm eff}}}{0.57{v_{\rm rms}}}\right)^{-2/7 } { m_{\rm cl,6}}^{1/28 } , \end{aligned}\ ] ] respectively . the corrected lifetime is @xmath305 , i.e. , just over one free - fall time at the upper mass limit for galactic gmcs . the total ionizing luminosity and star formation rate in the inner galaxy due to clouds in the mass range @xmath306 become @xmath307 and @xmath308 , respectively . these are consistent with the observed rate if @xmath44 and @xmath309 . thus , our estimates of the galaxy s ionizing luminosity and star formation rate ( with @xmath310 ) are comparable to the values adopted by mw97 if cloud porosity is not accounted for ( consistency requiring @xmath311 ) and somewhat higher ( consistency requiring @xmath312 ) after the porosity correction is applied . this level of agreement suggests hii regions are indeed responsible for maintaining energetic equilibrium within gmcs . what might explain the remaining discrepancy ? first , numerical estimates of turbulent decay rates may decrease as studies improve in resolution and less diffusive codes are used , or as turbulent anisotropy is included @xcite . second , the galaxy s ionizing luminosity could be higher than mw97 found , if a significant fraction of ionizing photons escape the galactic disk ( unlikely , but controversial ; * ? ? ? * ) but conversely , additional ionization from sources like supernova remnants @xcite would increase the discrepancy . third , the discrepancy results from our method of accounting for finite porosity , a crude approximation when @xmath313 . fourth , we were not entirely able to exclude a contribution of momentum from supernovae in [ winds&sne ] , especially for small associations . lastly , we have followed mw97 in approximating clusters ionizing luminosity as a step function of duration @xmath314 myr . in reality , a numerical integration of equation ( [ shellmom ] ) using the ionization history predicted by a starburst99 synthesis indicates that @xmath36 continues to grow ( albeit more slowly ) , gaining another @xmath240 after 6.6 myr . this should effectively increase @xmath315 in the largest clouds , whose free - fall times are long enough to permit this expansion . all of these topics merit further study . the main results of this paper are as follows . 1 . hii regions are the most plentiful sources of energy for the turbulence within giant molecular clouds ; they are more significant than the combined effects of protostellar winds , main - sequence and evolved - star winds , and supernovae . this result was indicated by the low efficiency of star formation in gmcs in [ dominance ] and demonstrated on more general grounds in [ winds&sne ] and [ populationmomentum ] . 2 . the input of turbulent energy by hii regions occurs on scales comparable to , but somewhat smaller than , the cloud radius . large - scale forcing minimizes the rate of turbulent decay , as recently emphasized by @xcite . 3 . a balance between turbulent decay and the regeneration of turbulence by hii regions allows a prediction of the stellar ionizing luminosity and ( less robustly ) the star formation rate . we present these results for clouds in the mass range @xmath306 , in which the stellar ionizing lifetime is briefer than the free - fall time . the results are roughly consistent with the total ionizing luminosity and star formation rate of the inner galaxy , provided that future numerical simulations verify our estimate of the coupling between momentum input and turbulent energy ( eq . [ [ suggestnumerics ] ] ) . our estimate of the ionizing flux in the inner galaxy is @xmath234 higher than the observed value ; we list in [ porosity ] several possible resolutions of this discrepancy . because hii regions also evaporate their clouds , an energetic balance also implies a rate of photoevaporation and a destruction time scale for gmcs . we derive a destruction time of 17 to 24 times @xmath316 million years , somewhat shorter than was found by wm97 . the upper mass limit for milky way gmcs most likely derives from the difficulty of assembling an object that destroys itself in a single crossing time . however , less massive clouds ( at least down to @xmath317 ) survive for many crossing times , produce many hii regions per crossing time , and are in both energetic and dynamical equilibrium . so long as there exists a minimum optical depth or column density required for star formation , the vigorous energetic feedback by hii regions provides a mechanism for the maintenance of cloud column densities near the critical value , and hence , for the gmc line width - size relation . massive clouds therefore follow the scenario proposed by m89 , but with hii regions rather than protostellar winds as the primary agents of feedback . this conclusion is not assured for clouds less massive than @xmath317 , for which cloud gravity must be considered in the dynamics of hii regions . the theory we have presented is robust , in the sense that it applies to massive gmcs ( containing the bulk of the molecular mass ) and derives from the flattening of the luminosity function between rich and poor ob associations a product solely of the upper mass limit for stars rather than the detailed birthrate distribution of ob associations . so long as a galaxy s molecular mass is concentrated in the most massive clouds , and so long as the birthrate of stellar associations drops with ionizing luminosity more steeply than @xmath318 , energetic equilibrium within molecular clouds determines its ionizing flux . this assertion can be tested by extragalactic observations . starbursts may however be an exception to this rule if , as discussed below , their gmcs are not in equilibrium . other observational tests of the theory presented here include the variation of ionizing luminosity and star formation rate ( figure [ fig2 ] ) with cloud mass and the high porosity of hii regions ( figure [ fig4 ] ) for massive gmcs . the variations of these quantities with mean cloud column density can potentially be tested by observations of the smc . the star formation efficiency ( figure [ fig1 ] ) and cloud lifetime ( figure [ fig3 ] ) will be more difficult to verify . the inhomogeneity of giant molecular clouds is the greatest source of uncertainty in the present work . the interaction of hii regions one source of inhomogeneity was treated in an approximate manner in [ porosity ] , but future work must treat the interaction of hii regions with a realistic background cloud . we have noted that clouds more massive than @xmath319 should not form in the milky way because they would be disrupted by hii regions in the time needed to assemble them . however , we must also note that clouds more massive than about @xmath320 may be incapable of driving champagne flows because their escape velocities exceed the exhaust velocity @xmath321 of ionized gas . this fact does not affect our derivation of the star formation rate in such clouds , but it does call into question our derivation of the lifetime for the most massive clouds ( and hence our suggestion for the origin of the upper mass limit ) . further work will be needed to resolve this issue . similarly , clouds more massive than about @xmath322 can not have supersonic hii regions , as their effective sound speeds exceed @xmath323 , the sound speed of ionized gas . objects created in this state or pushed into it by an increase in external pressure can neither be disrupted nor even supported by photoionization , and must collapse as rapidly as their turbulence decays ; this is a recipe for the efficient production of massive star clusters observed to occur in starburst galaxies . for instance , @xcite find values of @xmath324 in the range @xmath325 to @xmath326 for the _ mean _ central molecular gas in a number of starbursts ; this is sufficient to crush all gmcs in the mass range @xmath327 . it is a pleasure to thank chris mckee , jonathan williams , and the referee , mordecai - mark mac low , for substantial comments that led to improvements of this paper . i am also grateful to peter goldreich , rob kennicutt , peter martin , eve ostriker , nick scoville , and jim stone for insightful discussions and suggestions . this research was supported by an nserc fellowship .

b0218 + 357 is one of the most promising systems to determine the hubble constant from time - delays in gravitational lenses . consisting of two bright images , which are well resolved in vlbi observations , plus one of the most richly structured einstein rings , it potentially provides better constraints for the mass model than most other systems . the main problem left until now was the very poorly determined position of the lensing galaxy . after presenting detailed results from classical lens modelling , we apply our improved version of the lensclean algorithm which for the first time utilizes the beautiful einstein ring for lens modelling purposes . the primary result using isothermal lens models is a now very well defined lens position ofmas relative to the a image , which allows the first reliable measurement of the hubble constant from the time - delay of this system . the result of is very high compared with other lenses . it is , however , compatible with local estimates from the _ hst _ key project and with wmap results , but less prone to systematic errors . we furthermore discuss possible changes of these results for different radial mass profiles and find that the final values can not be very different from the isothermal expectations . the power - law exponent of the potential is constrained by vlbi data of the compact images and the inner jet to be , which confirms that the mass distribution is approximately isothermal ( corresponding to ) , but slightly shallower . the effect on is reduced from the expected 4 per cent decrease by an estimate shift of the best galaxy position of ca . 4mas to at most 2 per cent . maps of the unlensed source plane produced from the best lensclean brightness model show a typical jet structure and allow us to identify the parts which are distorted by the lens to produce the radio ring . we also present a composite map which for the first time shows the rich structure of b0218 + 357 on scales ranging from milli - arcseconds to arcseconds , both in the image plane and in the reconstructed source plane . finally we use a comparison of observations at different frequencies to investigate the question of possible weakening of one of the images by propagation effects and/or source shifts with frequency . the data clearly favour the model of significant ` extinction ' without noticeable source position shifts . the technical details of our variant of the lensclean method are presented in the accompanying paper i. quasars : individual : jvas b0218 + 357 gravitational lensing distance scale techniques : interferometric

astro-ph0312263

the lensed b0218 + 357 @xcite is one of the rapidly growing but still small number of gravitational lens systems with an accurately known time - delay @xcite . it is therefore a candidate to apply refsdal s method to determine the hubble constant @xmath4 @xcite . besides the time - delay and an at least partial knowledge of the other cosmological parameters , only good mass models of the lens are needed to accomplish this task . no other complicated and potentially incompletely known astrophysics enters the calculation or contributes to the errors . since the cosmological parameters are believed to be known with sufficient accuracy now , e.g. from the wmap project @xcite , the only significant source of errors lies in the mass models themselves . the difficulties in constraining the mass distribution of lenses should not be underestimated , but the lens method is still much simpler to apply than classical distance - ladder methods , where errors difficult to estimate can enter at each of the numerous steps . the classical methods still fight with the problem of results which are incompatible within the formal error bars . the hst key project @xcite e.g. obtains a value of @xmath5 ( @xmath6 ) which is often used as a reference . it should not be forgotten , however , that other groups obtained results which are significantly different , even though the general distance ladder used is very similar . @xcite determines a value of @xmath7 ( @xmath6 ) , @xcite obtain @xmath8 ( 90% ) . the differences in the analyses are various , ranging from the correction of selection effects and the cepheid period - luminosity relation to instrumental effects . the fact that a number of world - expert groups spent such an amount of effort but still do not agree on the final result proves how difficult the problems of distance ladder methods really are . see the references above for a discussion of possible reasons for the discrepancies . in addition we refer to @xcite for an extensive discussion of this subject and an additional result of @xmath9 . after finishing a first version of this paper , the analysis of the results of the first year of wmap observations have been published . although @xmath4 is not the primary goal of cmb observations , determinations of this parameter are possible with additional assumptions . assuming an exactly spatially flat universe ( @xmath10 ) and a cosmological constant for the ` dark energy ' ( equation of state parameter @xmath11 ) , @xcite obtain a value of @xmath12 . for different @xmath13 , the result for @xmath4 can be significantly smaller . the assumptions of @xmath14 and @xmath10 can only be dropped if additional information from other astronomical fields are included . with the local information from the 2df galaxy redshift survey @xcite , a constraint on @xmath13 is obtained . wmap alone can constrain @xmath15 only weakly , leading to wide allowed ranges for the density of dark energy or matter . for a determination of all free parameters , more astronomical data has to be included as discussed in detail by @xcite . the constraints for all cosmological parameters obtained from the combination of many data sets are very impressive , but it should be kept in mind that the situation now becomes similar to the distance ladder methods where model assumptions in many astrophysical fields are necessary , making realistic error estimates of the final result extremely difficult . therefore the more parameters that go into cmb models have accurate independent determination , the better . this allows the cmb analyses to concentrate on other parameters which can not be determined in any other way . taking these difficulties into account , we believe that the lens method is superior to most other methods not only in terms of the possible accuracy of @xmath4 but also , and not less important , in terms of estimating the possible errors reliably . we emphasize that the effort needed is small when compared with large projects like the hst key project or wmap plus 2df survey etc . the only serious problem when using the lens effect to determine the hubble constant lies in the mass models for the lenses which directly affect the resulting @xmath4 . to be able to constrain a mass model accurately , two conditions should be met . first , the lens should be simple by which we mean the galaxy should be normal without nearby companions and without a surrounding cluster . only in these cases can the mass distribution be described with confidence by a small number of parameters . second , as many observational constraints as possible should be provided by the system . advantageous is a high number of highly structured images . extended lensed sources and einstein rings can provide constraints for a wide range of positions and are therefore preferred to point - like multiple images . in the case of more than two images , multiple time - delays can also be used as constraints for the mass model . in addition to these pure lensing constraints , additional information from stellar dynamics or surface photometry of the lens can be included in the modelling . radio loud lensed sources have the potential to provide a whole set of independent observational constraints additional to those provided by optical observations . flux ratios of the images are less affected by microlensing and extinction at radio wavelengths than optical ones . radio observations also have the advantage that they can reach much higher levels of resolution than in the optical . much smaller substructures in the images can therefore be detected and used for the lens models . the lens b0218 + 357 , discovered in the jvas survey @xcite , meets most of the criteria for a ` golden lens ' which allows an accurate determination of @xmath4 . the lens is an isolated single galaxy , and no field galaxies nearby contribute to the lensing potential significantly @xcite . the lens shows only two images of the core of the source , but substructure of the images can be used as further constraint . the main bonus is the presence of an einstein ring which shows a lot of substructure in detailed radio maps @xcite . this paper deals with utilizing the information from the ring to constrain the mass models and to determine @xmath4 . b0218 + 357 has only one disadvantage : it is the system with the smallest image separation known , which makes useful direct measurements of the lensing galaxy s position relative to the lensed images extremely difficult . since this parameter is of fundamental importance for the determination of @xmath4 , no reliable results have been possible so far . @xcite present a galaxy position which may be close to the correct value , but their error bars are highly underestimated as was already stated by @xcite . in this paper we will first discuss ` classical ' lens models which only use parametrized data of the compact images to constrain the models . we will see that with this approach the constraints are not sufficient to determine the galaxy position with any accuracy . to be able to determine the hubble constant , other information has to be included which is naturally provided by the highly structured radio ring . the best method available to utilize this information is lensclean@xcite . because the method in its original form has serious shortcomings which prevent its use in a system like b0218 + 357 , where the dynamic range between the compact images and the ring is very high , we had to improve the algorithm considerably . the lensclean method itself is discussed in paper i @xcite while the results are presented here ( paper ii ) . we will use the improved version of lensclean to determine the galaxy position for isothermal models with an accuracy that is sufficient to achieve a result for the hubble constant which is competitive with other methods but avoids their possible systematic errors . possible deviations from isothermal mass distributions , which are the main source of systematic error when combining the results from many lenses , will be discussed by presenting a preliminary analysis of vlbi observations of the substructure of the images . these will allow us to constrain the radial mass profile for power - law models . we will learn that deviations from isothermal mass distributions do _ not _ play a major role in b0218 + 357 , because the constraints are already quite strong and in addition the effect on @xmath4 is in this special case , where the galaxy position is determined indirectly with lensclean , much weaker than in most other systems . we nevertheless discuss future work which will improve the results even more . as a secondary result we will use our best lens models and newly developed methods to produce maps of the source plane as it would look like without the action of the lens . we will show the source as well as its appearance in the lensed image plane on scales from milli - arcseconds to arcseconds . finally we use lensclean to shed some light on the question of scattering induced effective ` extinction ' in the a component of b0218 + 357 by comparing results for data sets of different frequencies . this paper ii as well as paper i are condensed versions of major parts of @xcite . for more details the reader is referred to that work . in the absence of an accurate position of the lens relative to the position of the lensed images there is no need to use very complicated and general lens models . the effect of the lens position shows its full impact already with simple standard models . we therefore restrict most of our work to isothermal elliptical mass models . to introduce the notation , we write the lens equation as @xmath16 the true source position @xmath17 is the image position @xmath18 shifted by the apparent deflection angle @xmath19 . in static single - plane lenses , the latter can be written as gradient of a potential @xmath20 : @xmath21 we use this potential to parametrize the mass models . general power - law models have @xmath22 the radial power - law index is @xmath3 in the isothermal case . the arbitrary azimuthal function @xmath23 describes the asymmetry . mass distributions _ can be described by the formalism presented in @xcite or @xcite . we prefer the approximation of elliptical _ potentials _ @xcite , which have practical advantages over the true elliptical mass distributions , because the inversion of the lens equation is possible analytically for elliptical potentials . for sufficiently small ellipticities , both approaches are equivalent . a singular elliptical power - law potential ( ` siep ' in the isothermal case ) is given by @xmath24 for small ellipticities @xmath25 and moderate @xmath26 , this model corresponds to an elliptical mass distribution with axial ratio @xmath27 , or @xmath28 for isothermal models . we do not include external perturbations , because the external shear and convergence introduced by field galaxies and large scale structure are expected to be very small . @xcite estimate them to be of the order @xmath29 to @xmath30 , which is negligible in our context . the time - delay @xmath31 between images @xmath32 and @xmath33 is generally related to the potential and the apparent deflection angle in the following way : @xmath34 here @xmath35 is the redshift of the lens and @xmath36 , @xmath37 and @xmath38 are the normalized angular size distances to the lens , source and from lens to source , respectively . isothermal models have an interesting analytical property , first described by @xcite . for fixed positions of the images relative to the mass centre , the time - delay for given @xmath4 is uniquely determined and does not depend on any other parameters of the model , as long as the images are fitted exactly . in our case this means that , given the time - delay , @xmath4 for isothermal models can be calculated directly from the lens position @xmath39 using the equation @xmath40 the dependence of @xmath41 on @xmath39 is linear . these simple relations show the importance of measuring the lens position directly . once this position is known with sufficient accuracy , no explicit model fits are necessary , but the hubble constant can be derived directly from the positions of the images and the lens . the resulting hubble constant for isothermal models using the time - delay of @xmath42 @xcite is shown in fig . [ fig : h0 ] as a function of @xmath39 . these values are calculated for a low - density flat universe ( @xmath43 , @xmath44 ) . they would be smaller by 5.5 per cent for an einstein - de sitter universe ( eds ) . we use a lens redshift of @xmath45 ( @xcite ; and others ) and a source redshift of @xmath46 @xcite . the older value of @xmath47 @xcite would lead to a 4.2 per cent smaller hubble constant . included in the plot are estimates for the lens position from different sources . the optical positions from @xcite are clearly not accurate enough . the centre of the ring is relatively well defined , but it can deviate from the position of the mass centre by 003 or even more , depending on the ellipticity of the lens and the size and shape of the source . we will see later that the error bars of @xcite are underestimated significantly so that their result for the lens position can not be used . the alternative time - delay of @xmath48 ( @xmath49 ) from @xcite is compatible with the result from @xcite which we used for our work because of the smaller error bars . the result for @xmath4 obtained from the @xcite time - delay would be 4 per cent higher . our first modelling attempts for b0218 + 357 followed the classical route of using only parameters of the two compact components as constraints . this is the same approach as presented in @xcite . we will learn that it is not possible to determine uniquely the lens position with this information . the results shown in this section were obtained using isothermal models because the ( small , see below ) deviations from isothermality are not of importance as long as the galaxy position is not known with high accuracy . once this parameter is determined with lensclean , constraints on these deviations as well as the impact on @xmath4 will be discussed in sec . [ sec : non - iso ] . all coordinates in this paper are measured with respect to the a image . the @xmath50 coordinate is measured eastwards ( increasing right ascension ) , @xmath51 is measured northwards . table [ tab : pos ] shows a compilation of known relative positions of the a and b images as well as of the subcomponents 1 and 2 which were revealed by vlba observations @xcite . there are significant differences in the given positions , but they are too small to affect the results seriously . for the b@xmath52a separation , we used the 15ghz vlba positions for the core components a1 and b1 with their formal accuracy . the relative positions of the _ _ sub__components offer an additional possibility to constrain the lens models . for the modelling we assumed an accuracy of 0.1mas as in @xcite and made independent calculations for the data from @xcite and @xcite . the difference between the two sets of positions is of the same order as the assumed error bar . the new vlbi data presented in @xcite show , in addition to the already known two close subcomponents , the inner jet in both images with subcomponents at separations of up to ca . 10mas which could potentially provide additional constraints . these data are not used for our classical modelling because the accuracy of parametrized model fits to the additional components are influenced significantly by the underlying smooth emission of the jet . the true errors of such fits are expected to be much higher than formal estimates so that we decided to exploit this new data set only later with lensclean . rl@ cd3.2d3.2c & & & ( sub)comp + 5ghz & evn & accuracy ] & 308.5 & 130.3 & b@xmath52a + 8.4ghz & vlbi & & 309.00 & 127.30 & b1@xmath52a1 + & & & 309.32 & 126.37 & b2@xmath52a2 + 15ghz & vlba & accuracy ] & 309.2 & 127.4 & b1@xmath52a1 + & & & 309.6 & 126.6 & b2@xmath52a2 + 15ghz & vla & & 310.56 & 127.11 & b@xmath52a + optical & hst & & 307 & 126 & b@xmath52a + 8.4ghz & vlbi & & 1.18 & 0.87 & a2@xmath52a1 + & & & 1.50 & -0.06 & b2@xmath52b1 + 15ghz & vlba & & 1.072 & 0.868 & a2@xmath52a1 + & & & 1.470 & 0.000 & b2@xmath52b1 the magnification ratio of the two compact images is more difficult to obtain . several effects can influence the results . one problem is scattering in the lensing galaxy which seems to decrease the flux density of a. this effect is stronger for lower frequencies and ( with regard to the fluxes ) thought to be almost absent at 15ghz . secondly , flux densities of compact images embedded in a smooth surface brightness background can not be determined unambiguously . we always have to expect an uncertainty of about the background surface brightness integrated over the beam . this error should be smaller than 10 per cent for the lower resolution data sets at lower frequencies and much less than that in other cases . finally we observe the source at different epochs in the two images because of the time - delay . from measured light curves and the known time - delay @xcite , the typical error of this effect is estimated to be less than 5 per cent . table [ tab : flux ] shows flux density ratios for different frequencies , epochs and resolutions . there seems to be a systematic increase in the ratio with increasing frequency ( discounting optical ) , which hints at scattering . also the spread in values for 8.4ghz and above is relatively small . for both these reasons , and because there are values with the effects of delay removed , we chose to use a value of 3.75 which is compatible with most of the measurements at higher frequencies . please note that the results depend only very weakly on small changes of this ratio . for the lensclean work presented below , no explicit measurement of the flux density ratio is required . resolution effects and contamination by the ring are then taken into account automatically . d1@ll@ ccd1.3@l & & epoch & + 1.7&ghz & vlbi & & 1992 jun 19 & 2.62 + 5&ghz & merlin & & 1991 aug 26 & 2.976 + 5&ghz & merlin & & 1992 jan 13 & 3.23 + 5&ghz & merlin & & 1992 mar 27 & 3.35 + 5&ghz & evn & & 1990 nov 19 & 3.185 + 8.4&ghz & vla & & 1991 aug 1 & 3.247 + 8.4&ghz & vla & & 1996/1997 & 3.57 & @xmath53 + 8.4&ghz & vla & & 1996/1997 & 3.2 & @xmath54 + 8.4&ghz & vlbi & & 1995 may 9 & 3.18 & @xmath55 + & & & & & 3.72 & @xmath56 + 15&ghz & vla & & 1991 aug 1 & 3.690 + 15&ghz & vla & & 1992 nov 18 & 3.79 + 15&ghz & vla & & 1996/1997 & 3.73 & @xmath53 + 15&ghz & vla & & 1996/1997 & 4.3 & @xmath57 + 15&ghz & vlba & & 1994 oct 3 & 3.623 & @xmath58 + 22&ghz & vla & & 1991 aug 1 & 3.636 + & hst & & & 0.14 & the subcomponents are only marginally resolved and even the formal uncertainties of fitted gaussians are too high to serve as additional constraints . furthermore , the apparent sizes and shapes might be influenced by scatter broadening in the interstellar medium of the lensing galaxy , which is known to be gas rich @xcite . we nevertheless tried to include the shapes of the subcomponents into the modelling , without any gain in accuracy of @xmath4 . the actual observational data as given by @xcite and @xcite are shown in table [ tab : subcomp shapes ] . the fit of the best source shapes for each lens model is performed analytically with a cartesian parametrization of the ellipses . mathematical details can be found in appendix [ sec : app shapes ] . .shapes of the subcomponents [ cols="^,^,^,^,>,>",options="header " , ] the hubble constant @xmath4 can now be determined from the distances and from the parameter @xmath41 given by the lens model . for the final result we only use the naturally weighted data since the uncertainties are smaller here . for an eds world model , the result is @xmath59 . for a low - density flat standard model with @xmath43 and @xmath44 , it increases by about 6 per cent : . ] @xmath60 the error bars are formal @xmath49 limits which include the uncertainty of the lens model ( 6 per cent , contributed almost exclusively by the uncertainty of the lens position ) and the time - delay ( 4 per cent ) . we stress that they do not include possible systematic errors which might result if the true mass distribution can not be described by our isothermal models ( siep ) . as we show below , no large effects from this are expected . using the time - delay from @xcite , our result becomes @xmath61 with the largest error contribution ( 15 per cent ) coming from the time - delay . standard quintessence models ( e.g. * ? ? ? * ; * ? ? ? * ) have parameters between the eds and the low - density @xmath62 world models and would lead to hubble constants between the two extremes . for @xmath63 , the result would be indistinguishable from the eds case ( equivalent to @xmath64 ) while for @xmath65 the hubble constant would be 2.1 per cent higher than for eds . the cosmological constant corresponds to @xmath11 . it is a well known fact the the radial mass distribution is the most important source of uncertainties in the @xmath4@xmath66 relation for lens systems where the lens position is known accurately ( see e.g. references in the introduction of * ? ? ? * ) . for power - law models a simple scaling law of @xmath67 is valid in many cases @xcite while in quadruple lenses an even stronger dependence of @xmath68 was found under certain circumstances by @xcite for generalized power - law models following eq . . as a generalization of the siep models , we use elliptical power - law potentials which ( if not too far from isothermal and not too elliptical ) are a good approximation to elliptical power - law mass distributions . the two families of models differ only in the special form of the azimuthal function @xmath23 . we will see that deviations from isothermality are very small so that our lenscleanresults for isothermal models can still be used as a first approximation . the deviations can thus be treated as a small perturbation . the subcomponents of both images are oriented more or less radially with respect to the expected lens position . this has the disadvantage that their positions can not be used to constrain the galaxy position for isothermal models as shown before . for non - isothermal models , on the other hand , this insensitivity can be used to determine constraints for the radial mass profile from these data relatively independent of the true lens position . if this position is taken to be close to the lensclean result and compatible with the vlbi constraints , @xmath26 can be estimated with some accuracy using the method presented in appendix [ sec : app vec ] . if we use the 8.4ghz vlbi data from @xcite , the result is @xmath69 ( @xmath6 limits ) . the error bar does not include uncertainties of the galaxy position but the values are relatively independent of @xmath39 anyway . from the 15ghz data from @xcite , a slightly larger value of @xmath70 can be estimated . @xcite were able to detect the jet with many subcomponents over a length of 10mas in both images with global vlbi observations at 8.4ghz . a systematic relative stretching of the b jet image by about 10 per cent can be accounted for by a non - isothermal lens model with @xmath71 , in very good agreement with the other results . we conclude that the lensing galaxy seems to be close to isothermal with a deviation in @xmath26 of only about 4 per cent , but have to keep in mind that these results are only preliminary estimates which are not based on a self - consistent lens model but used the lens position from lensclean with isothermal models to estimate deviations from isothermality . the errors of this method are expected to be only small ( see also below ) but a combined fit of non - isothermal lens models to the vla and vlbi data should be aimed for in the future . as explained in paper i , lensclean relies on a very robust method for the inversion of the lens equation . using the newly developed lentil method implemented into lensclean , we were now able to perform a few tests with non - isothermal lens models . we repeated the calculations described before with the 15ghz vla data but used different fixed values of @xmath26 and compared the results afterwards . we noticed that the best lens position does indeed shift slightly eastwards when increasing @xmath26 , by about 1mas for 1 per cent change of @xmath26 . see the arrow in fig . [ fig : vla i ] for an illustration . to our relief , this effect acts on @xmath4 in the opposite direction as the scaling with @xmath72 , which we expect for fixed galaxy positions , so that the two cancel almost completely . a rather conservative estimate is that the @xmath72 scaling is reduced by at least a factor of two as a result of the shifting lens centre . this means that for a value of @xmath73 a change of the hubble constant by at most 2 per cent is expected compared to the isothermal models . this is much smaller than the statistical uncertainties of the result for isothermal models and thus not of concern yet . when _ increasing _ @xmath26 slightly , starting from the isothermal @xmath3 , the resulting shift of the best lensclean lens position has the tendency to _ decrease _ the optimal @xmath26 in classical fits . combined fits of both data sets should therefore be possible and lead to very stable results . in the near future we will use lensclean and lentil to obtain more accurate constraints on the radial mass profile from the existing very high - quality vlbi observations at 8.4ghz @xcite . in combination with the existing vla data which are more sensitive to the galaxy position , a self - consistent lens model can be found and accurate error estimates will be possible . in paper i we discussed a new method to reconstruct source plane maps and superresolved lens plane maps from emission models determined with lensclean . we present the resulting lens plane maps of b0218 + 357 in fig . [ fig : lens map ] . one version ( a ) is restored with the nominal clean beam while the other ( b ) is a superresolved version with anisotropic and varying beams calculated as explained in paper i. the superresolved map is not easy to interpret because of the varying and anisotropic beams . the resolution is generally higher close to the critical curve i.e. in parts of the ring . the elliptical shape of the b image is also a result of the local elliptical beams . the reconstructed source plane map is shown in fig . [ fig : src map ] . the left map ( a ) is reconstructed with the source plane beams calculated from eqs . and in paper i. the radial streaks emerging from the region near the caustic are artifacts of the varying and highly anisotropic beams . the elliptical shape of the core is also a result of the highly eccentric local beam and should not be misinterpreted as a resolved elliptical core . the right map ( b ) was restored with varying but circular beams ( the smallest circles which still cover the local elliptical beams completely ) to make interpretations easier . unfortunately the greatest part of the lensing magnification and improvement of resolution is lost in this map . we see that the jet , which by coincidence emerges from the inner core in a direction very similar to the major axis of the elliptical core image in fig . [ fig : src map](a ) , bends southwards and crosses the caustic of the lens at the bend . it is this part of the source which forms the ring in the lens plane and which is used as constraint by lensclean . on larger scales ( shown especially well by longer wavelength observations , e.g. the 8ghz vla data in fig . [ fig : comp ] ) , the jet proceeds in a southern direction only to bend eastwards again at a distance of about @xmath74 . the jet then seemingly ends in a relatively bright blob @xmath75 south - east of the core . [ [ sec : comp ] ] the composite colour map + + + + + + + + + + + + + + + + + + + + + + + + in fig . [ fig : comp ] shows b0218 + 357 on different scales both in the lens plane and in the source plane ( reconstructed with lensclean ) . the main maps showing the outer parts of the jet on arcsec scales are made from a 8.4ghz data set obtained at the vla simultaneously with a global vlbi observation ( * ? ? ? we show a regular lens plane map and a source plane reconstruction with circular beams . the insets on the left are made from the 15ghz vla data set that was used for our main lensclean work . the lens and source plane maps show the same data as fig . [ fig : lens map](a ) and [ fig : src map](b ) . we also show the critical curve in the vla lens plane maps and the tangential caustic ( diamond shape ) and cut ( elliptical ) in the vla source plane maps . the insets on the right ( upper : b , lower : a ) are regular lens plane cleanmaps of a global vlbi data set @xcite . the source plane map is a projection into the source plane of the a image alone . the b image is less resolved and would not provide much additional information . the map was restored from the clean components with the projected clean beam . 5 in @xcite shows projections of images a _ and _ b restored with equal beams to show the similarities . true lenscleansource plane maps of the vlbi data have not been produced yet . the magnifications relative to the main maps are 3 for the left insets and 130 for the jet insets on the right . the scales are the same for lens plane and source plane . the contour lines start at three times the rms noise and double for each new level . the surface brightness levels in the lens plane and source plane maps are directly comparable . the magnified sections are marked by black rectangles in the vla maps . it is a generally accepted fact that the apparent flux ratio a@xmath76b changes systematically with the frequency at which it is measured . at the high end the value seems to reach its final value of 3.8 for frequencies @xmath77 . on the other end the ratio goes down to 2.6 at @xmath78 ( compare table [ tab : flux ] ) . at even lower frequencies it is difficult to separate the compact emission from the ring which becomes more dominant at low frequencies . there is unambiguous evidence that the ism of the lensing galaxy is very rich and that optical and radio emission are affected to a significant degree @xcite . @xcite estimated the scattering measure from the broadening of image a relative to b ( taking into account the magnifying effect of the lens ) . they obtain a value of @xmath79 which is very high compared to typical lines of sight along our galaxy but comparable to lines of sight through the galactic centre . one possible explanation for the low flux ratios at low frequencies is therefore the action of the very rich ism of the lensing galaxy in front of image a. this can cause an effective extinction of a either by a high amount of scattering or by other physical effects . the refractive index of an interstellar plasma goes proportional to @xmath80 which would explain the stronger effects for larger @xmath81 . another possible explanation is a very strong frequency dependence of the source structure which would effectively result in a position shift along the jet for lower frequencies . because the spectrum of the jet is much steeper than the spectrum of the core , such an effect is not implausible . when the source is shifted in this direction , the situation becomes more symmetric and the magnification ratio decreases . for a sufficiently large shift ( @xmath82 ) the differential magnification gradient could , in combination with this frequency dependent source position , well explain the observed flux ratio changes . there are , however , strong arguments against this scenario . if the shift is as large as required , there should be significant differences in the appearance of the innermost components in vlbi maps at 8.4ghz ( see e.g. * ? ? ? * ; * ? ? ? * ) and 15ghz ( see e.g. * ? ? ? * ) , but these maps look remarkably similar . both show two strong inner components with similar shapes and compatible relative positions . their separation of @xmath83 is far too small to explain the demanded shift by different spectral indices . if the shift is caused by another component which is not seen at 15ghz but becomes stronger at low frequencies , such a component should be detectable on a scale of @xmath84 to @xmath85 in the existing very deep 8.4ghz vlbi observations . @xcite do indeed detect more jet components but these are far too weak to account for a shift of the required magnitude . a more exotic alternative scenario would be refraction caused by large scale systematic trends in the column density of the lens galaxy s ism . to investigate the question of extinction and/or source shifts with lensclean , we compare our best lens models fitted to the 15ghz data which are known to be unaffected by both effects with a merlin data set at 5ghz . this multi - frequency data set was used to produce the maps in @xcite and is described in detail there . we used only the merlin part because the vla resolution at 5ghz is too low to be of any help here . unfortunately , this data set has some disadvantages for lensclean . while the resolution is better than in the 15ghz vla data , and the flux in the ring is higher due to the lower frequency , it was not possible to obtain reliable direct constraints for the lens model with lensclean . one problem is the possibility of extinction itself . another very serious problem is the frequency - dependence of the emission , which is different for the ring and the compact images . @xcite approximately corrected for this effect by first mapping the three frequencies independently . the clean components responsible for the compact images were then subtracted from the data so that only the ring remains . the three data sets were rescaled in magnitude to obtain a consistent total flux density at all frequencies . finally , the compact components from the central frequency were added back to the complete data set . this process may introduce some distortions to the lowest and highest frequencies , because it combines data for the ring from these frequencies with data for the compact components from the central frequency . the combined data were then successively mapped and self - calibrated to obtain the final data set and map . we used the resulting @xmath86 data set as basis for our computations . the spectral index correction seems to work very well for making maps , but introduces errors in lensclean which are not fully understood . this shows especially in incompatible results from the individual three frequencies and the combination of all three . on the other hand , tests showed that the data can nevertheless be used to fit a subset of the lens model parameters if the other ones are fixed . at most three parameters can be determined accurately because they are already well defined by the two bright images alone and the ring is only a relatively small correction . we therefore used a variety of lens models compatible with the 15ghz vla data and fitted the same models to the 5ghz data set allowing for a relative shift of the two data sets and a possible extinction in the a image at 5ghz . these fits are much more robust than complete lens model fits from this data set alone . in particular , the results do not depend on which of the frequencies is used . the combination of all three also leads to the same result . our strategy to determine shift and extinction works like this . first we took one of the best lens models from lenscleaning the 15ghz data . then we fixed all parameters but @xmath39 and made a fit with the 5ghz data including a ( for the moment fixed ) extinction in a. the difference between the two values of @xmath39 is then a measurement of the shift in the lens plane because in reality the lens should have the same position at both frequencies but the source may shift . we then varied the extinction to find the best value by minimizing the residuals . vla data.,width=302 ] the results for one of the best vla 15ghz lens models are shown in fig . [ fig : vlarefit ] . the position of the minimum changes somewhat when other lens models still compatible with the 15ghz data are used . the most extreme values for the extinction are 15 and 35 per cent ; most probable is a value between 20 and 30 per cent . the data sets where initially registered to obtain the same position for the a image . the relative shift in the fit ( measured as a shift of the lens centre merlin @xmath52 vla , equivalent to a shift of the source by the same amount with opposite direction ) in the region of the minimum is negligible . the fact that the shifts in @xmath50 and @xmath51 have their zeros at almost the same extinction which itself is furthermore compatible with the residual minimum gives strong evidence that the real shift is indeed small and that the brightness of a at 5ghz is reduced by @xmath87 per cent by some propagation effect , possibly scattering . for zero extinction , the required shift would be about @xmath88 eastwards , but the residual curve seems to be incompatible with this scenario . it seems as if extinction in a , caused by whatever physical process , accounts for a least most of the frequency dependent flux ratio . this has an important consequence for further modelling work . although the ring becomes brighter at lower frequencies , only high frequencies can be used in a simple way to fit lens models . observations at medium frequencies may on the other hand be used to study the propagation effects in the ism of the lensing galaxy in more detail now that the lens model is known with sufficient accuracy . b0218 + 357 is one of the most promising lens systems for determining the hubble constant with the method of @xcite . this method has , compared with other approaches , the advantage that it is a very simple one - step determination and relies on the understanding of only very little astrophysics . in this way the systematic uncertainties can not only be minimized but ( equally important ) also be estimated much better than in distance ladder methods which have their problems with several astrophysical processes at each step . the only serious uncertainty left is the mass distribution of the lens . in this respect b0218 + 357 is a close to optimal ` golden lens ' . the lensing galaxy seems to be a regular isolated galaxy without contributions from a group or cluster nearby . this not only avoids the inclusions of further parameters to describe the external mass distribution but also allows the assumption that the mass distribution of the galaxy itself is that of an unperturbed smooth ellipsoid . b0218 + 357 provides a wealth of potential constraints for the mass models in the form of highly accurate image positions , a measurable flux density ratio with probably negligible influence by microlensing , substructure on scales resolved by vlbi and a well resolved and structured einstein ring on the scale of the image separation . without using the ring , ` classical ' lens modelling allows us to tightly constrain the ellipticity and mass of the galaxy as long as its position is known . the vlbi structure of several subcomponents along the jet in a radial direction relative to the lens allows us to determine the radial mass distribution with an accuracy not possible in most other lenses ( see e.g. * ? ? ? * and references therein ) . this most important general degeneracy of the lens method can therefore be broken in the case of b0218 + 357 . the only real disadvantage is the small size ( @xmath89 ) of the system . b0218 + 357 is indeed the lens with the smallest image separation known to date . although the images and the lens are detected with optical observations , it was until now not possible to use these observations to accurately measure the galaxy position directly . _ hst_/acs observations obtained recently are currently analysed to allow a first useful direct measurement . the model lens positions from @xcite are probably not too far from the truth and are indeed consistent with our results but their error bars are seriously underestimated . we have shown that , without using the ring , no useful estimate of the galaxy position and thus @xmath4 is possible . previous efforts have not utilized the most striking feature of b0218 + 357 , the beautiful einstein ring shown by radio observations ( e.g. * ? ? ? doing this is much more difficult than using parameters of compact components because the source itself has to be modelled as well , either implicitly or explicitly . the method best suited for this task is lensclean which surprisingly has been used only for very few cases before . this is partly a result of the high numerical demands and partly of serious shortcomings of the original method as it was implemented by @xcite and @xcite . in paper i we discuss the development of our new improved variant of lensclean . one of the most important improvements is the implementation of a new unbiased selection of components . the problem of determining the lens position is almost degenerate because the bright components , which contribute most of the signal , provide no information for this parameter . it is only the relatively weak and fuzzy ring which can be used as constraint . in other words the residuals have a very strong dependency on two directions ( defined by the compact components ) of the five dimensional parameter space but change only weakly in the other three directions as a result of the constraints provided by the ring . finding minima of functions like this is a numerically difficult problem and relies on the very accurate knowledge of the residual function . any numerical noise produces local minima which make the finding of the global minimum difficult . special care was therefore needed to produce the best possible results with the given limited computing resources . the modified lenscleanalgorithm was applied to a 15ghz vla data set resulting in good constraints for all parameters of an _ isothermal _ lens model , including the lens position of @xmath90mas , @xmath91mas relative to the a image . this model was then used to determine the hubble constant from the time - delay of @xcite to be @xmath92 . the error bars are @xmath49 confidence limits which for the hubble constant include the error of the lens position ( 6 per cent ) and the time delay ( 4 per cent ) . the accuracy for all other lens model parameters is much higher as long as the position is fixed . the value for @xmath4 is in agreement with the results from the _ hst _ key project @xcite and the wmap project @xcite but incompatible with the lower distance - ladder results of @xcite , @xcite or @xcite . recently , a series of papers was published by @xcite , see also @xcite , in which it is claimed that a number of gravitational lens time - delays is compatible with the hubble constant from the _ hst _ key project only if the mass concentration is as compact as the light distribution ( @xmath93 in the picture of power - law models ) . our result for b0218 + 357 does not confirm this view and does not give rise to a ` new dark matter problem ' since it is significantly higher than other results from lenses . our isothermal models lead to a value absolutely compatible with the one preferred in @xcite . one could now in a similar way compare with the lower @xmath4 results of @xcite , @xcite or @xcite . however , as long as the local determinations do not agree with each other , we consider such an exercise to be of only limited value . rather than using locally measured values for @xmath4 to constrain lens mass distribution , we prefer a more direct approach , either using the lens effect itself ( see below ) , or by including additional information . the latter approach is followed by thy ` lenses structure and dynamics ' survey lsd @xcite . the general idea is simple : the total mass within the einstein radius is well constrained by lensing , in contrast to the radial mass _ profile_. for the given mass , the stellar velocity dispersion of the lensing galaxy depends strongly on this profile . the more the mass is concentrated in the centre , the deeper the potential well and the higher the velocity dispersion has to be . with some additional assumptions , it is indeed possible to obtain valuable constraints which can then again be included in the lens models to determine @xmath4 . for the important system pg 1115 + 080 , @xcite obtain a power - law index of @xmath94 which increases the otherwise very low estimate of @xmath4 for this system to a value of @xmath95 ( @xmath6 ) . this steep mass profile is quite unusual and moves the galaxy significantly off the fundamental plane . nevertheless it is not as steep as the _ light _ profile and the result for @xmath4 correspondingly lies between the isothermal and constant @xmath96 model results of @xcite . for b1608 + 656 , @xcite are able to constrain the mass profile of the main lensing galaxy quite well and find that it is consistent with isothermal . the resulting hubble constant is @xmath97 ( @xmath98 , inconsistent with the very low mean value of @xmath99 obtained by @xcite for isothermal models for a number of lenses , but consistent with our result for b0218 + 357 . however , b1608 + 656 has the major disadvantage of a second lensing galaxy very close to the primary . the two galaxies might even be interacting which could cause complicated deviations from usual galaxy mass distributions . these examples show that the general picture of lensing constraints for @xmath4 and mass profiles is not as consistent as it appeared not long ago ( e.g. * ? ? ? * ) . in our opinion the only way to resolve the discrepancies is to try and use the best available constraints for all applicable lens systems individually . our work on b0218 + 357 shows this for one example , although there are still many open questions . our alternative way of constraining mass profiles relies on the lens effect alone and has the advantage that additional astrophysics ( galactic dynamics in lsd work ) and corresponding additional model assumptions are not needed , keeping the method simple and clean . it is easy to understand that lenses showing only two or four images of one compact source can not provide sufficient information to constrain the mass profile tightly , especially since quads usually have all images close to the einstein ring so that they probe the potential at only one radius . far better suited are systems with extended sources or at least with substructure in the compact images . b0218 + 357 is a good example which offers both . using our galaxy position as an estimate , non - isothermal power - law models could be constrained with the vlbi substructure . these are not the final results since no self - consistent fitting of all available data has been performed so far , but the weak dependency of the radial mass exponent @xmath26 on the galaxy positions shows that the estimate is nevertheless quite good . we learned that the mass distribution in the lens is close to isothermal ( which would be @xmath3 ) but is slightly shallower ( @xmath100 ) . preliminary calculations with lensclean showed that the shift in the best lens position caused by this small deviation partly compensates for the expected scaling with @xmath72 for a fixed position . we therefore do not expect very significant changes of the result for @xmath4 from the slight non - isothermality in this system . a very conservative error estimate disregarding this compensation effect would be a possible 4 per cent error for @xmath4 . it has to be kept in mind , however , that all results presented so far depend on the assumption of a power - law for the mass profile ( and potential and deflection angle ) . for more general mass models the constraints on @xmath26 are still interesting , but they only measure the _ local _ slope of the profile rather than the global mass profile . it is expected that power - laws can be used as a very good approximation for more general models for limited ranges of radii , and our model fits confirm this view in the case of b0218 + 357 . the lensclean results for the lens position presented here are therefore also valid for other mass profiles , but the value for the hubble constant may well have to be modified slightly . the mass models discussed here do not include external shear because it seems to be very small . the expected true external shear of 12 per cent @xcite would change the result of @xmath4 by about the same amount at most . more difficult to estimate is the possible influence of differences in the ellipticity of the inner and outer parts of the lensing galaxy which can effectively also act as external shear . the possible impact of the remaining higher - order aspects of the mass profile degeneracies and the effects of multi - component mass models ( bulge+disc+halo ) are currently under investigation , especially the aspect of what apparent isothermality in the central part of the galaxy , which should be dominated by the luminous matter in the bulge , means for the true mass distribution . in the future we will avoid the model fitting of vlbi components which are then used as constraints and instead use lensclean itself also on the existing vlbi data that show a wealth of structure in the images ( see fig . [ fig : comp ] ) . to be able to use non - isothermal models with lensclean , we developed the new method lentil for the inversion of the lens equation . with these two methods combined , a simultaneous fit of medium resolution vla data ( sensitive to the lens position ) and the existing high resolution vlbi data of @xcite , which are more sensitive to the radial mass profile , will not only improve the results for all parameters but will , by using self - consistent lens models , also provide realistic error estimates including all uncertainties . to be able to use the 8.4ghz data set it will be necessary to include the extinction and scatter broadening in the a image in the models . this is possible because the 15ghz data are only little affected by extinction so that the combination of the two allows estimates of this effect . since the vlbi constraints for the lens models are mainly given by the positions of subcomponents but not by their size , which is relatively independent of the lens model , the different sizes of components in a and b can be used to obtain better constraints on the scattering measure than the simple estimates from @xcite . the application of lensclean to the vlbi data will also show whether any significant amount of substructure in the mass distribution of the lens is needed to explain all features of the jet . the first results from @xcite seem to be compatible with no substructure , but a quantitative analysis can be done in the future . if significant substructure effects are present , it will also be necessary to correct the medium resolution data for them before using lensclean to determine the galaxy position . clumps of matter close to one of the images could change the magnification significantly which would then mimic a different lens model . the analysis of the vlbi data will show whether this may be the explanation for the ( compared with some other lenses ) relatively high result for @xmath4 . lensclean can also be used to study substructure from vlbi observations of other lenses . it will then be possible to do an analysis similar to e.g. @xcite but without any assumption about the true source structure . to improve the medium resolution side , we recently performed observations of b0218 + 357 at 15ghz using the fibre link between the vla and the vlba telescope pie town . this combination provides a resolution 1.52 times better than the vla alone . the long track observation of the total accessible hour angle range of 14 hours also improves the sensitivity which in combination with the improved resolution can reduce both statistical and ( even more important ) systematical errors . simulations performed before the observations showed that an accuracy of about 12@xmath101 for the lens position should be achievable . this would in comparison to the existing data be an improvement of a factor of @xmath102 . the remaining uncertainty in @xmath4 from the lens position alone will then shrink to below 2 per cent . possible uncertainties caused by calibration errors will also be reduced significantly . this new data set is currently being calibrated and will be analysed with lenscleansoon , alone and in combination with the vlbi data . as a secondary result of lensclean we presented maps of the brightness distribution of the source as well as improved lens plane maps , both produced with a method newly developed together with lensclean in paper i. this allowed the first view of b0218 + 357 s source as it would be seen without the distortion of the lens , but with improved resolution . the ring itself is caused by a bending jet which crosses the tangential caustic of the lens . in the future we will improve these results and also produce source plane polarization maps . the ring shows an interesting radial polarization pattern @xcite and the data will , when projected to the source plane , give us a very detailed and magnified view of the polarization structure in the jet of the source . finally we used lensclean to investigate the puzzling changes of the flux density ratio a@xmath76b with frequency . to test the two theories of either scattering induced extinction caused by the ism in the lensing galaxy or of an effective shift of the source with frequency , we compared our best lensclean models , which were fitted to 15ghz data where the fluxes are expected to be unaffected by propagation effects , with merlin data at 5ghz taking into account a possible relative shift and extinction in the 5ghz data . although the 5ghz data set has some calibration problems , the evidence for a significant extinction ( @xmath103 per cent ) of the a component at 5ghz is very strong . the relative shift of the source between these two frequencies seems to be negligible . this work will be extended in the future by comparing medium resolution data sets at different frequencies in order to study the propagation effects in the lensing galaxy in more detail . observations with merlin at different frequencies will allow us to measure the position dependent faraday rotation and depolarisation , providing invaluable information about the ism of the lensing galaxy . a direct measurement of the galaxy position in b0218 + 357 by _ hst_/acs observations will be available soon . the formal uncertainty of this measurement may be larger than the lensclean model constraints but it will be an absolutely independent and therefore complementary direct measurement . the comparison will either confirm the lens models or give information about deviations of the true mass distribution from the relatively simple models . we will then reach the point were the uncertainty in @xmath4 is no longer dominated by the lens models but by the time - delay uncertainties themselves . a reanalysis of existing monitoring data @xcite and new monitoring campaigns will therefore allow further improvements . with all the constraints available now or soon and the simplicity of the possible mass models , b0218 + 357 has the potential to lead to the most robust measurement of @xmath4 of all time - delay lenses . the authors would like to thank the anonymous referee for a very helpful report and the royal astronomical society for covering the cost of the colour figure . o.w . was funded by the ` deutsche forschungsgemeinschaft ' , reference no . re 439/261 and 439/264 ; european commission , marie curie training site programme , under contract no . hpmt - ct-2000 - 00069 and tmr programme , research network contract erbfmrxct96 - 0034 ` ceres ' ; and by the bmbf / dlr verbundforschung under grant 50 or 0208 . a. d. , browne i. w. a. , helbig p. , koopmans l. v. e. , wilkinson p. n. , perley r. a. , 1999 , , 304 , 349 a. d. , browne i. w. a. , muxlow t. w. b. , wilkinson p. n. , 2001 , , 322 , 821 biggs a. d. , wucknitz o. , porcas r. w. , browne i. w. a. , jackson n. , mao s. , patnaik a. r. , wilkinson p. n. , 2003 , , 338 , 599 i. w. a. , patnaik a. r. , walsh d. , wilkinson p. n. , 1993 , , 263 , l32 c. l. , rupen m. p. , yanny b. , 1993 , , 412 , l59 a. s. , hewitt j. n. , moore c. b. , haarsma d. b. , 2000 , , 545 , 578 cohen j. g. , lawrence c. r. , blandford r. d. , 2002 , , in print , also available as astro - ph/0209457 e. a. , browne i. w. a. , wilkinson p. n. , 1996 , in @xcite , p. 37 j. d. , kochanek c. s. , hewitt j. n. , 1996 , , 464 , 556 w. l. , et al . , 2001 , , 553 , 47 a. , kovner i. , 1993 , , 417 , 450 kemball a. j. , patnaik a. r. , porcas r. w. , 2001 , , 562 , 649 kochanek c. s. , 2002a , , submitted , also available as astro - ph/0204043 kochanek c. s. , 2002b , , 578 , 25 kochanek c. s. , 2003 , , 583 , 49 c. s. , blandford r. d. , lawrence c. r. , narayan r. , 1989 , , 238 , 43 kochanek c. s. , falco e. e. , impey c. , lehr j. , mcleod b. , rix h .- w . , , 2002 , castles website , http://cfa-www.harvard.edu/castles/ c. s. , hewitt j. n. , eds , 1996 , astrophysical applications of gravitational lensing iau symp . 173 c. s. , narayan r. , 1992 , , 401 , 461 kochanek c. s. , schechter p. l. , 2003 , in freedman w. l. , ed . , measuring and modeling the universe , carnegie observatories astrophysics series , vol . 2 . in print , also available as astro - ph/0306040 the hubble constant from gravitational lens time delays l. v. e. , fassnacht c. d. , 1999 , , 527 , 513 l. v. e. , treu t. , 2002 , , 568 , l5 l. v. e. , treu t. , 2003a , astro - ph/0308056 l. v. e. , treu t. , 2003b , , 583 , 606 l. v. e. , treu t. , fassnacht c. d. , blandford r. d. , surpi g. , 2003 , astro - ph/0306216 r. , schneider p. , bartelmann m. , 1994 , , 284 , 285 i. , 1987 , , 312 , 22 c. r. , 1996 , in @xcite , p. 299 j. , et al . , 2000 , , 536 , 584 e. v. , 1988a , , 206 , 175 e. v. , 1988b , , 206 , 190 metcalf r. b. , 2002 , , 580 , 696 j. r. , et al . , 2000 , , 529 , 786 b. r. , saha a. , sandage a. , tammann g. a. , 2000 , , 540 , 634 a. r. , browne i. w. a. , king l. j. , muxlow t. w. b. , walsh d. , wilkinson p. n. , 1993 , , 261 , 435 a. r. , porcas r. w. , 1999 frequency dependent radio structure of the gravitational lens system b0218 + 357 . 156 , p. 247 a. r. , porcas r. w. , browne i. w. a. , 1995 , , 274 , l5 w. j. , et al . , 2001 , , 327 , 1297 s. , 1964 , , 128 , 307 a. , 1999 , , 527 , 479 d. n. , et al . , 2003 , , 148 , 175 g. a. , reindl b. , thim f. , saha a. , sandage a. , 2002 , in metcalfe n. , shanks t. , eds , a new era in cosmology cepheids , supernovae , @xmath4 , and the age of the universe , also available as astro - ph/0112489 . 283 t. , koopmans l. v. e. , 2002a , , 575 , 87 t. , koopmans l. v. e. , 2002b , , 337 , l6 l. , et al . , 2002 , , 335 , 432 s. , kochanek c. s. , narayan r. , 1996 , , 465 , 64 j. , paczyski b. , 1994 , , 108 , 1156 t. , combes f. , 1995 , , 299 , 382 h. j. , mao s. , keeton c. r. , 2000 , , 544 , 98 h. j. , mao s. , schechter p. l. , 1995 , , 443 , 18 wucknitz o. , 2002a , , 332 , 951 wucknitz o. , 2002b , phd thesis , universitt hamburg , germany , available from the author or from http://www.astro.physik.uni-potsdam.de/~olaf/ wucknitz o. , 2003 , , also available as astro - ph/0312262 ( paper i ) o. , refsdal s. , 2001 , in brainerd t. g. , kochanek c. s. , eds , gravitational lensing : recent progress and future goals , asp conf . series , vol . 237 degeneracies of the radial mass profile in lens models . applying the lens equation for two nearby positions gives to first order : @xmath104 here @xmath105 denotes the magnification matrix @xmath106 clearly any information about the small scale structure of the images a and b ( flux density ratio , subcomponent positions and shapes ) can only provide constraints for the _ relative _ magnification matrix @xmath107 . it would be possible to use the subcomponents of a and b as images of two independent sources to constrain the models in the usual way . since the separation of the subcomponents is much smaller than the typical scale of the system , however , these data can only constrain the first derivative of the deflection angle , i.e. the magnification matrix . a different approach to constrain this matrix directly is therefore more appropriate . given a model value for the relative position vector of @xmath108 in the source plane , the corresponding vectors in the lens plane are given by eq . , leading to a contribution to @xmath109 of @xmath110 with the measured lens plane vectors of @xmath111 , @xmath112 and corresponding covariance matrices @xmath113 and @xmath114 . this equation defines an ordinary linear least squares problem leading to a weighted mean of the backprojected @xmath115 for the solution : @xmath116 with the common polar ellipticity parameters , fits can only be done numerically and are potentially very unstable . the following cartesian formalism allows a direct analytical fit for the elliptical shape of the source to minimize the deviations from the observed images . an ellipse with major and minor axis @xmath117 and @xmath118 with direction of the major axis @xmath119 ( mathematical sense ) can be described by the matrix @xmath120 . for all points @xmath121 located on the ellipse , the following equation holds : @xmath122 { { \mathbfss{e } } } = \begin{pmatrix } r_+ + r_- \cos2\phi & r_- \sin2\phi \\ r_-\sin2\phi & r_+ - r_-\cos2\phi \end{pmatrix } { } \end{gathered}\ ] ] @xmath123 the astronomical position angle ( north through east ) is related by @xmath124 . transformation between source and lens plane is linear for the ellipticity parameters @xmath125 : @xmath126 we can therefore use the same simple general linear least squares formalism as in appendix [ sec : app vec ] with transformation matrices for the three - dimensional ellipticity vector @xmath127 . in this way the best source plane ellipticity matrix / vector and the residual contributions can be calculated analytically . the cartesian approach also avoids singularities and degeneracies for certain shapes , e.g. for circular sources .

dihadron and isolated direct photon - hadron angular correlations are measured in collisions at gev . correlations of charged hadrons of gev/ with mesons of gev/ or isolated direct photons of gev/ are used to study nonperturbative effects generated by initial - state partonic transverse momentum and final - state transverse momentum from fragmentation . the nonperturbative behavior is characterized by measuring the out - of - plane transverse momentum component perpendicular to the axis of the trigger particle , which is the high- direct photon or . nonperturbative evolution effects are extracted from gaussian fits to the away - side inclusive - charged - hadron yields for different trigger - particle transverse momenta ( ) . the gaussian widths and root mean square of are reported as a function of the interaction hard scale to investigate possible transverse - momentum - dependent evolution differences between the-h and direct photon - h correlations and factorization breaking effects . the widths are found to decrease with , which indicates that the collins - soper - sterman soft factor is not driving the evolution with the hard scale in nearly back - to - back dihadron and direct photon - hadron production in collisions . this behavior is in contrast to drell - yan and semi - inclusive deep - inelastic scattering measurements .

1609.04769

in the last two decades , the study of quantum chromodynamics ( qcd ) bound states has evolved from static , one - dimensional snapshots of quarks and gluons to focus on multidimensional structure and the dynamics of partons . the theoretical framework that has been developed to describe parton dynamics in hadrons involves transverse - momentum - dependent ( tmd ) parton distribution functions ( pdfs ) and fragmentation functions ( ffs ) . in traditional collinear pdfs and ffs , any momentum of the partons transverse to the hadron boost axis is integrated over . in tmd pdfs or ffs , the transverse momentum of the partons is not integrated out and instead remains explicit in the pdf or ff , offering a means of describing the transverse momentum distribution of unpolarized partons within an unpolarized hadron , as well as a variety of spin - momentum correlations when polarized hadrons and/or partons are considered . early theoretical work in tmd pdfs took place in the 1980s by collins , soper , and sterman @xcite , with extensive further development in the 1990s ( see e.g. @xcite ) . however , some theoretical details regarding the definition of tmd pdfs within a perturbative qcd ( pqcd ) framework have only been clarified in the last five years @xcite . we note that due to confinement , the behavior of partons within hadrons is nonperturbative in that it can not be calculated theoretically within pqcd . collinear or tmd pdfs are nonperturbative functions that can be constrained by and/or used to predict high - energy scattering processes within a pqcd framework . in such a framework , the nonperturbative functions such as pdfs as well as ffs factorize from the perturbatively calculable partonic hard scattering cross section and from each other . lattice qcd offers an alternative , complementary approach to pqcd , performing numerical nonperturbative calculations directly . in the past lattice qcd could only calculate moments of pdfs , integrated over parton collinear momentum fraction @xmath12 as well as parton transverse momentum . however , recent developments have demonstrated the potential to go beyond these limitations . these efforts are still in very early stages @xcite . there is already experimental evidence from semi - inclusive deep - inelastic scattering ( sidis ) and drell - yan ( dy ) measurements that several tmd pdfs describing spin - momentum correlations are nonzero @xcite . in addition , there is empirical evidence for nonzero spin - momentum correlations in the process of hadronization from electron - positron annihilation as well as sidis @xcite . furthermore , transverse single - spin asymmetries up to @xmath1340% have been measured in inclusive hadron production in hadronic collisions , indicating large nonperturbative spin - momentum correlations in these processes ( see e.g. @xcite ) . however , these measurements can not probe tmd functions directly because there is no simultaneous observation of perturbative and nonperturbative momentum scales . the recent focus on multidimensional structure and parton dynamics has not only offered richer information on the behavior of partons confined within hadrons , but has moreover brought to light fundamental predictions regarding qcd as a nonabelian gauge - invariant quantum field theory . in particular , the role of color interactions due to soft gluon exchanges with the remnants of the hard scattering have become clearer . because tmd functions preserve more nonperturbative information compared to collinear functions , tmd functions can differ from collinear ones with regards to universality and factorization . for example , the sivers tmd pdf @xcite , a correlation between the proton spin and quark transverse momentum , was shown to possibly be nonzero due to phase interference effects from soft gluon exchanges in sidis @xcite . shortly afterward , @xcite showed that , due to the gauge invariant nature of qcd and the parity and time ( pt ) odd nature of the sivers tmd pdf , the function should be the same magnitude but opposite in sign when measured in drell - yan vs. sidis processes because of the different color flows possible in the initial state vs. final state . twist-2 tmd pdfs that involve one polarization vector are odd under pt transformations , leading to this predicted effect . the nonvanishing nature of the sivers function has already been measured in polarized sidis @xcite ; there is not yet a measurement of this function in polarized drell - yan . a first indication from the drell - yan like w boson production exists @xcite . the results favor a sign - change if tmd evolution effects are small , but at this stage the error bars are still large enough that a definitive statement can not be drawn from this single measurement . it is only for the tmd pdfs odd under pt transformations , where such sign - change behavior is expected , that gluon exchanges can not be completely eliminated via a gauge transformation . in the more complicated qcd process to hadrons , soft gluon exchanges in both the initial and final state are possible , leading to new predicted effects for observables sensitive to a small transverse momentum scale . in such processes , factorization breaking has been predicted @xcite in both polarized and unpolarized interactions . here the nonperturbative objects in the cross section become correlated with one another and can not be factorized into a convolution of tmd pdfs or tmd ffs . however , there are no theoretical claims that the perturbative partonic cross section does not factorize from the nonperturbative physics . similarly to the case of the tmd pdfs that are odd under pt transformations , gluon exchanges that lead to the predicted factorization breaking can not be eliminated via a gauge transformation . it is important to recognize that the ideas behind the predicted sign change of certain tmd pdfs and factorization breaking represent a major qualitative departure from previous purely perturbative approaches that do not account for soft gluon exchanges with remnants of the hard scattering . possibly related effects known as color coherence " have been studied and observed in multijet states in hadronic collisions @xcite , but these types of effects have not been rigorously treated in a tmd framework . in calculations of tmd processes where factorization is predicted to hold , the evolution with the hard scale of the interaction is known to be governed by the collins - soper ( cs ) evolution equation @xcite . note that the cs evolution equation comes directly out of the derivation of tmd factorization @xcite . in contrast to the dglap collinear evolution equations @xcite , which are purely perturbative , the kernel for the cs evolution equation for tmd processes involves the collins - soper - sterman ( css ) soft factor @xcite , which generally contains nonperturbative contributions . the soft factor is understood to be strongly universal , the same for unpolarized and polarized processes , pdfs and ffs , with the only difference being between quarks and gluons @xcite . because lattice calculations of the soft factor are currently not possible , the soft factor must be extracted from parameterizations of experimental measurements within a pqcd framework . for a discussion of the css soft factor and tmd evolution phenomenology , see ref . @xcite . the theoretical expectation from css evolution is that any momentum width sensitive to nonperturbative @xmath14 would grow as the hard scale increases . this can be understood intuitively as a broadening of the phase space for gluon radiation with increasing hard scale . in addition this has been studied and observed in multiple phenomenological analyses of drell - yan and z boson data ( see e.g. @xcite ) , as well as phenomenological analyses of sidis data , where factorization is also predicted to hold ( see e.g. @xcite ) . as mentioned above , because the cs evolution equation comes directly out of the derivation of tmd factorization , it then follows that a promising avenue to investigate factorization breaking effects is by looking for qualitative differences from css evolution in processes where factorization breaking is expected , such as nearly back - to - back dihadron correlations produced in collisions . to have sensitivity to possible factorization breaking and modified tmd evolution effects , a particular observable must be sensitive to a small scale on the order of @xmath15 and measured over a range of hard scales . nearly back - to - back dihadron production has long been used as a proxy for measuring initial - state partonic transverse momentum @xmath14 @xcite , which is defined in fig . [ fig : ktkinematics ] . first used in predictions by ref . @xcite as a method for understanding large differences in hard scattering cross sections between theory and data , nearly back - to - back two - particle and dijet angular correlations have since been used to measure @xmath14 over a large range of center of mass energies @xcite . direct photon - hadron correlations are of particular interest because the photon comes directly from the partonic hard scattering , and thus carries initial - state information without any final - state fragmentation effects . the direct photon approximates the away - side jet energy at leading order ( lo ) while still being directly sensitive to the partonic transverse momentum scale . direct photons also give an interesting comparison to dihadron production because they do not carry color charge , thus , assuming factorization holds , only two tmd pdfs and one tmd ff are necessary in the cross section calculation compared to two tmd pdfs and two tmd ffs in dihadron production . therefore , there should be more avenues for gluon exchange in nearly back - to - back dihadron events when compared to direct photon - hadron events . figure [ fig : ktkinematics ] shows the hard scattering kinematics of a nearly back - to - back dihadron event in the transverse plane . the effect of initial - state @xmath14 and final - state @xmath16 , the transverse momentum of the hadron with respect to the jet axis , can be probed in hadronic collisions by measuring the out - of - plane momentum component @xmath17 with respect to the near - side hadron or direct photon , collectively referred to as the trigger particle . @xmath17 thus quantifies the acoplanarity of the two - particle pair , with @xmath18 signifying exactly back - to - back particle production . using the trigger particle as a proxy for the jet , the 1-dimensional quantity @xmath17 is transverse to the of the trigger particle , @xmath19 , and has a magnitude of : @xmath20 where @xmath21 is the of the associated hadron and @xmath22 is the azimuthal angular separation between the trigger and associated particle as shown in fig . [ fig : ktkinematics ] . reference @xcite has shown that the root mean square of @xmath17 and @xmath14 are related by @xmath23 where @xmath24 and @xmath25 , and quantities with a hat indicate partonic - level quantities . all the quantities on the left side of eq . [ eq : kteqn ] are partonic , while those on the right side can be measured via the correlated away - side hadron . equation [ eq : kteqn ] gives a clear definition for how to relate the root mean square initial - state @xmath14 and final - state @xmath16 to the observable @xmath17 . a diagram showing the hard - scattering kinematics of ( a ) dihadron and ( b ) direct photon - hadron event in the transverse plane . two hard - scattered partons with transverse momenta @xmath26 and @xmath27 [ red lines ] are acoplanar due to the initial - state @xmath28 and @xmath29 from each parton . these result in a trigger and associated jet fragment @xmath19 and @xmath21 with a transverse momentum component perpendicular to the jet axis @xmath30 , which is assumed to be gaussian such that @xmath31 . for ( b ) direct photons only one jet fragment @xmath21 is produced because the direct photon is produced from the hard scattering . the quantity @xmath17 [ blue ] is the transverse momentum component of the away - side hadron perpendicular to the trigger particle axis . , title="fig : " ] + a diagram showing the hard - scattering kinematics of ( a ) dihadron and ( b ) direct photon - hadron event in the transverse plane . two hard - scattered partons with transverse momenta @xmath26 and @xmath27 [ red lines ] are acoplanar due to the initial - state @xmath28 and @xmath29 from each parton . these result in a trigger and associated jet fragment @xmath19 and @xmath21 with a transverse momentum component perpendicular to the jet axis @xmath30 , which is assumed to be gaussian such that @xmath31 . for ( b ) direct photons only one jet fragment @xmath21 is produced because the direct photon is produced from the hard scattering . the quantity @xmath17 [ blue ] is the transverse momentum component of the away - side hadron perpendicular to the trigger particle axis . , title="fig : " ] the relativistic heavy ion collider ( rhic ) is an ideal facility to study nonperturbative factorization breaking effects because they are only predicted in hadronic collisions where at least one final - state hadron is measured , and the measurement has sensitivity to a small initial- and final - state transverse momentum scale . observables of interest are final states where at least one particle has a large , defining a hard scale , at least one final - state hadron is measured , and the observable is also sensitive to initial- and final - state @xmath14 and @xmath16 . at rhic energies , the @xmath9 reach for direct photons and pions is sufficiently large to have separation from the nonperturbative momentum scale . direct photon - hadron and @xmath32-hadron correlations were chosen specifically because of experimental capabilities and because of the differing number of final - state hadrons in the event ; the @xmath32-hadron correlations probe an extra nonperturbative function , assuming factorization , and thus one more gaussian @xmath16 convolution than the direct photon - hadron correlations . in 2012 and 2013 the phenix experiment collected data from @xmath33 collisions at @xmath34 gev . after data quality and vertex cuts , integrated luminosities of approximately 30 pb@xmath35 in 2012 and 152 pb@xmath35 in 2013 were used for the analysis of dihadron and direct photon - hadron correlated pairs . the measured @xmath17 distributions presented here are at a higher center of mass energy and have significantly reduced statistical uncertainties compared to @xcite . the phenix detector can measure two - particle correlations between photons and hadrons with its electromagnetic calorimeter ( emcal ) and drift chamber ( dc ) plus pad chamber ( pc ) tracking system located in two central arms . the central arms are nearly back - to - back in azimuth , with each arm covering approximately @xmath36 radians in azimuthal angle and 0.7 units of pseudorapidity about midrapidity @xcite . a schematic showing the two central arms is shown in fig . [ fig : phenix ] . the emcal @xcite is located at a radial distance of approximately 5 meters from the beam pipe and is composed of 8 sectors , 4 in each arm . six sectors are lead - scintillator ( pbsc ) sampling calorimeters , and the other two are lead glass ( pbgl ) @xmath37erenkov calorimeters . the pbsc and pbgl calorimeters measure electromagnetic showers with intrinsic resolution @xmath38 and @xmath39 , respectively . high energy photons are identified with a cluster shower shape cut and charged particle veto . the shower shape cut also removes most high energy photons that overlap too closely with another photon , which helps eliminate @xmath32 merging effects at energies greater than @xmath1312 gev in the pbsc and @xmath13 17 gev in the pbgl . the high granularity of the emcal , @xmath40 for pbsc and @xmath41 for pbgl , along with the shower shape cut allows for @xmath32 and @xmath42 reconstruction via the diphoton channel up to @xmath43 17 gev . previous direct - photon , @xmath42 , and @xmath32 cross sections measured in the phenix central arm can be found in @xcite . the @xmath32 and @xmath42 mesons are tagged in the emcal via their two - photon decay for the purposes of removing decay photon background to identify direct photons and constructing the @xmath32-h@xmath11 correlated pairs . to reduce the combinatorial background , only photons with energy greater than @xmath44 gev are considered . the invariant mass windows were 120160 mev/@xmath45 for @xmath32 and 500600 mev/@xmath45 for @xmath42 mesons . cross section view along the beam line of the phenix detector , showing the detectors composing the central arms in 2012 and 2013 . the relevant subsystems for this analysis are described in the text . ] the phenix tracking system @xcite allows charged hadron detection via a drift chamber ( dc ) in each central arm along with two pad chambers directly behind the drift chambers . the momentum resolution was determined to be @xmath46 with @xmath0 in gev/@xmath4 . tracks are identified via the dc , covering a radial distance of @xmath47 meters from the beam pipe . secondary tracks from decays or conversions are reduced by a condition that matches tracks in the dc to hits in the outermost pc3 , located at a radial distance of 4.98 meters from the beam pipe . the charged particle veto suppresses hadronic showers in the emcal by matching tracks from the full tracking system to clusters in the emcal . the correlation functions are constructed following the methods of refs . @xcite . the number of correlated hadrons per trigger particle is referred to as the per trigger yield , and is collected for the different types of trigger particle - associated hadron pairs . to quantify the inefficiencies of the phenix detector , the hadron yields are corrected by a charged hadron efficiency determined from a monte carlo single - particle generator and a geant - based simulation of the detector . additionally , due to the limited acceptance of the phenix detector , the per - trigger yields are divided by a mixed event distribution . mixed event distributions are collected on a run - by - run basis to quantify any changing inefficiencies with time in the acceptance . the collected trigger particle is mixed with charged hadrons from different events and a mixed event correlation function is constructed to correct for the acceptance of the detector . in total the full correlation function is determined by the following equation @xmath48 where @xmath49 is the hadron efficiency described above . note that this definition is general for any observable that could be constructed in a two - particle correlation , so it applies to the determination of the @xmath17 distributions also . for a complete description of two - particle correlation analyses in the phenix central arms , see refs . @xcite . to identify direct photons , ref . @xcite used a method that is based upon identifying a total sample of inclusive per - trigger yield correlations , then subtracting the decay component . from ref . @xcite , the yield of charged hadrons per direct photon was determined with the following equation @xmath50 here @xmath51 is the per - trigger yield where the trigger particle for each per - trigger yield is indicated as direct , inclusive , or decay , and @xmath52 is the relative contribution of direct photons to decay photons such that @xmath53 . the total yield of photons , the inclusive photons , comes from adding all of the decay and direct photons , @xmath54 . in ref . @xcite direct photons are defined as any photon not from a decay process , which includes next - to - leading order ( nlo ) photons that emerge from parton - to - photon fragmentation . to eliminate the presence of nlo fragmentation photons , ref . @xcite implemented isolation and tagging cuts ; thus eq . [ eq : statsubrgamma ] was modified to include these cuts . to determine the per - trigger yield of isolated direct photons , the number of isolated decay photons was subtracted from the isolated inclusive photon sample , where @xmath55 . the subtraction procedure results in the following equation for per - trigger yields of isolated photon quantities @xcite @xmath56 where the trigger particles are noted as direct , inclusive , or decay for a given per - trigger yield @xmath51 and iso " refers to isolated " . @xmath57 is the relative contribution of isolated direct and decay photons , where @xmath58 and indicates isolated direct photon production for @xmath59 . the subtraction procedure eliminates remaining background due to isolated decay photons that appear direct , which are due most often to asymmetric @xmath60 decays where the low photon is not detected . to suppress sources of background photons , tagging and isolation cuts are implemented at the event - by - event level . to reduce the contribution from decay photons , candidate inclusive photons are tagged and removed if a partner photon of @xmath61 gev is found such that the invariant mass of the pair falls within the regions of 118162 or 500600 mev/@xmath45 . the tagging cuts use a larger @xmath32 invariant mass range than for identifying @xmath32 for dihadron correlations to err on the side of removing more decay photons . an isolation cut further suppresses decay photons as well as nlo fragmentation photons by requiring that the sum of the emcal energy deposits and of charged tracks within a radius of 0.4 radians around the candidate photon be less than 10% of the photon s total energy . to reduce the impact of detector acceptance effects , photons that pass the isolation and tagging cuts are also required to be @xmath130.1 radians from the edge of the detector in both @xmath42 and @xmath62 forcing a large portion of the isolation cone to fall inside the phenix acceptance . because the number of isolated decay photons is not _ a priori _ known , the decay photon per - trigger yield is determined with a probability density function . isolated @xmath32-h@xmath11 correlated pairs are weighted by a probability density function to map these per - trigger yields to the isolated decay photon hadron correlated per - trigger yields . this function , determined in ref . @xcite , gives the probability of an isolated @xmath32 with @xmath63 to decay to a photon with @xmath64 in the phenix acceptance where the photon was unable to be tagged as a decay photon . in the phenix central arms , the inability to tag a decay photon happens most often from asymmetric @xmath32 decays , where one photon misses the detector completely . a 4% systematic uncertainty was assigned to the decay photon statistical subtraction method as a whole , which includes not considering backgrounds due to higher mass states such as the @xmath42 , @xmath65 , and @xmath66 . to determine the per - trigger yield of isolated decay pairs , the number of isolated @xmath32s is mapped via the probability function to the number of isolated decay pairs in a given bin . the per - trigger yield of isolated decay pairs is then @xmath67 where @xmath68 is the probability density function described above and contains all of the dependence and efficiencies of the detector on @xmath63 and @xmath69 . the @xmath70 and @xmath71 are simply the number of isolated @xmath32-trigger particles measured and the number of isolated @xmath32-h@xmath11 pairs measured , respectively . the @xmath57 is determined by measuring @xmath52 and correcting @xmath52 with tagging and isolation efficiencies . because the quantity is the ratio of the inclusive photons to decay photons after tagging and isolation cuts , it can be written as @xmath72 where `` niso '' refers to `` not isolated . '' @xmath57 is now written in terms of values that can be measured . @xmath52 and the tagging efficiency @xmath73 can be determined without the probability function . the right - most fraction in eq . [ eq : rprimeeq ] is simply the number of photons that pass the isolation and tagging cuts divided by the total number of inclusive photons . the efficiency with which the isolation cut removes decay photons @xmath74 is determined by applying the probability function at the level of the isolated parent meson and mapping the effect to the daughter photon @xmath75 each of these quantities is found by counting the number of photons that pass the various cuts except for the isolation efficiency @xmath76 , which is found by measuring the number of isolated and not isolated @xmath32 that pass the cuts . @xmath52 was found by dividing the number of inclusive photons by the number of decay photons , and correcting for higher mass states and the phenix single and diphoton detection efficiencies derived from a geant - based simulation . as a cross check , systematic uncertainties on @xmath52 were evaluated using the direct photon and @xmath32 pqcd cross sections with the ct10 pdfs @xcite and dss14 ffs @xcite . the tagging efficiency is 0.360.43 and the isolation efficiency is 0.610.69 from the lowest to highest @xmath19 bins . figure [ fig : rgammaprime ] shows the values of @xmath57 as a function of @xmath64 ; for which values greater than unity indicate isolated direct - photon production . measured @xmath57 for use in the statistical subtraction , eq . [ eq : statsubeqn ] . the boxes quantify the systematic uncertainty . ] figure [ fig : dphicorrelations ] shows a few examples of the per - trigger yields of associated charged hadrons for both @xmath32 and direct photon triggers as a function of @xmath22 in bins of @xmath19 and @xmath21 . the azimuthal correlations as a function of @xmath22 show standard jet structure characteristics . the @xmath32 yields have clear peaks at both @xmath77 and @xmath78 , indicating nearly back - to - back jet production . the near - side yields of the isolated direct photons are not plotted , similarly to ref . @xcite , because the yields within the isolation cone are physically uninterpretable . in addition , the effect of @xmath14 smearing is characterized with the away - side peaks . the near - side @xmath32 peaks are larger than the away sides due to the effect of so called trigger bias , " discussed in ref . the away - side yields of the direct photons are smaller than those from the @xmath32 triggers due to the smaller jet energy sampled by direct photons ; the @xmath32 has some fractional energy @xmath79 where @xmath80 refers to the momentum fraction of the @xmath32 from the scattered parton and @xmath81 is the energy of the jet , whereas the direct photon approximates the away - side jet energy at lo . the underlying event levels for @xmath32 and direct photon triggers are similar , which would be expected if the underlying event structure is completely uncorrelated from the partonic hard scattering . a 9% normalization uncertainty on the charged hadron yields is not shown on this figure or any of the following per - trigger yields . all of the per - trigger yields as a function of @xmath22 can be found in the supplemental material @xcite . the value @xmath82 was determined by the widths of gaussian fits to the near - side of the @xmath32 correlation functions , similarly to ref . examples of the fits are shown on the near - side @xmath32 peaks in fig . [ fig : dphicorrelations ] . values of @xmath82 were calculated with the following equation @xmath83 where @xmath84 is the gaussian width . previous measurements have shown @xmath82 to be approximately constant with @xmath85 and @xmath19 in a similar @xmath19 range to that examined here @xcite . only bins @xmath86 gev/@xmath4 were used to satisfy the assumption @xmath87 which was made to determine eq . [ eq : jteqn ] . each @xmath19 bin was fit to a constant and averaged @xmath82 over @xmath21 . figure [ fig : jts ] shows the results , which were then fit with a constant to average over @xmath19 , which shown as a dotted line in fig . [ fig : jts ] . after averaging , @xmath82 was determined to be @xmath88 where the systematic uncertainty is due to the momentum resolution of the detector as well as approximations made to determine eq . [ eq : jteqn ] in ref . recent atlas results show a similar fragmentation variable over a significantly larger range of hundreds of gev/@xmath4 in , and show that the average transverse momentum with respect to the jet axis rises slowly with @xmath89 over this significantly larger range @xcite . the @xmath82 shown as a function of @xmath19 is determined with eq . [ eq : jteqn ] and has been shown to be approximately constant with @xmath85 and in the limited range examined here . the line shows a constant fit to average over @xmath19 . ] the quantity @xmath90 was extracted from the @xmath22 correlations as was done in previous measurements in the error function in the denominator , not the numerator as was done in the previous references . in order for the normalization of the function to be unity across the range @xmath91 $ ] this @xmath92 should be in the denominator of the error function . we have studied the effect of this change and find that it does not change the value of @xmath90 extracted . this is because , as fig . 31 of ref . @xcite shows , the quantity @xmath90 is determined from the exponential component of the fit function . the yield parameter extracted in ref . @xcite changes slightly , but within the quoted systematic uncertainties . ] @xcite . the value of @xmath90 quantifies the width of the away - side jet . the correlation functions are fit in bins of @xmath19 and @xmath21 with the following function in the range @xmath93 : @xmath94 with @xmath95 were the parameters @xmath96 , @xmath97 , and @xmath90 are left as free parameters , with @xmath96 quantifying the underlying event , @xmath97 a normalization constant , and @xmath90 the parameter of interest . the fit extends to @xmath98 and @xmath99 in order to accurately quantify the underlying event . example fits are drawn on the correlation functions in fig . [ fig : dphicorrelations ] . systematic uncertainties were evaluated by altering the fit region by @xmath1000.2 radians and taking the absolute value of the difference of the resulting @xmath90 . the systematic uncertainties for the direct photons are larger due to the increased fluctuations in the underlying event due to the statistical subtraction technique . @xmath90 for @xmath32 and direct photon triggers as a function of @xmath21 in several @xmath19 bins . ] figure [ fig : rmspoutvsptassoc ] shows @xmath90 as a function of @xmath21 in several @xmath19 bins for both @xmath32 and direct photon triggers . all of the @xmath90 values can be found in tables i and ii of the supplemental material @xcite . in the following figures showing measured quantities , filled points are for isolated direct photons and open points are for @xmath32 triggers . both @xmath90 distributions for @xmath32 and direct photon triggers show a clear dependence on @xmath21 , with the direct photons having a stronger dependence . the direct photon @xmath90 quantities are larger due to the smaller jet energy being sampled compared to the @xmath32 jet energies . the strong dependence of @xmath90 on @xmath21 is anticipated as could be ascertained from the definition of @xmath101 . in the same @xmath22 region , as @xmath21 gets larger , @xmath17 will also get larger . figure [ fig : rmspoutvspttrig ] shows a subset of the @xmath90 results for both direct photon and @xmath32 triggers as a function of @xmath19 in the @xmath21 range 24 gev/@xmath4 . the @xmath32-triggered @xmath90 decreases with @xmath19 , although this dependence is small . the direct photons clearly have a strong dependence on @xmath19 relative to the @xmath32 triggers . the @xmath32 data shown in fig . [ fig : rmspoutvspttrig ] contain a dependence on the fragmentation function not present in the direct photon data because the direct photons emerge directly from the hard scattering . to explore this dependence , pythia 6.4 @xcite hard - scattered qcd events were analyzed to determine the average @xmath102 of a @xmath32 where the hat quantity refers to the hard scattered parton . @xmath103 was determined in the same bins used in the data to correct the @xmath32 @xmath19 to an estimated jet in order to make a better comparison between the direct photons and @xmath32 . figure [ fig : rmspoutvsptjet ] shows the same @xmath90 plotted as a function of @xmath89 , where @xmath89 refers to @xmath19 for the direct photons and to @xmath104 for the @xmath32 . the @xmath103 correction ranges from 0.450.63 as a function of @xmath19 . after the correction , the @xmath32 and direct photon @xmath90 do not appear to form a single continuous function ; rather the @xmath32 @xmath90 continue approximately linearly to lower @xmath89 . @xmath90 for @xmath32 and direct photon triggers as a function of @xmath19 in two different @xmath21 bins . ] @xmath90 for @xmath32 and direct photon triggers as a function of @xmath89 in two different @xmath21 bins . for @xmath32 triggers , @xmath105 where @xmath103 was determined from pythia . ] figure [ fig : poutdistributions ] shows the per - trigger yields of the @xmath17 distributions for @xmath32 and direct photon triggers . only away - side hadrons were used in making the distributions , with the requirement that the correlated hadron satisfy @xmath106 . the underlying event was statistically subtracted out from the @xmath17 per - trigger yields using the parameters from the fits to the @xmath22 correlations with eq . [ eq : rmspouteq ] in order to identify only charged hadron yield associated with the hard scattering . the underlying event yield for a given bin was statistically subtracted by applying a factor @xmath107 where @xmath108 is the correction function determined by @xmath96 divided by the fits to the @xmath22 correlations . for the smaller @xmath19 bins , this is an important subtraction because in the signal region @xmath109 the underlying event contributes roughly 50% of the away - side yield , as could be ascertained from fig [ fig : dphicorrelations ] . the yield corrected by the underlying event factor @xmath110 is then subjected to the usual construction of the correlation function outlined in sec . [ statistical_subtraction ] . systematic uncertainties on the underlying event background subtraction were evaluated by performing the subtraction after changing the underlying event parameter @xmath96 to @xmath111 , where @xmath112 is the error on @xmath96 from the fit . these uncertainties were found to be on the order of tenths of a percent in the @xmath113 region . the values of the @xmath17 distributions can be found in tables iii - xiii of the supplemental material @xcite . note that a 4% systematic uncertainty is assigned to @xmath17 due to the detector resolution on @xmath21 and @xmath22 . the distributions are fit with a gaussian at small @xmath17 in the region [ -1.1,1.1 ] gev/@xmath4 as well as a kaplan function over the whole range , with the kaplan function parameterized by @xmath114 where @xmath115 , @xmath116 , and @xmath4 are free parameters . in fig . [ fig : poutdistributions ] the solid lines are fits to the @xmath32 distributions and the dashed lines are fits to the isolated direct photon distributions . the gaussian functions clearly fail past @xmath131.3 gev , showing a transition to power law behavior which the kaplan functions accurately describe . the power law behavior is generated from hard gluon radiation in the initial state or final state , whereas the gaussian behavior is generated from the soft @xmath14 and @xmath16 and is demonstrated in the nearly back - to - back hadrons that are produced around @xmath113 . the evolution of @xmath17 as a function of @xmath19 was characterized by the gaussian widths at small @xmath17 . figure [ fig : gausswidths ] shows the widths from gaussian fits to both @xmath32 and direct photon triggers as a function of @xmath19 . systematic uncertainties were evaluated by altering the gaussian fit region by @xmath100 0.15 gev/@xmath4 and taking the absolute value of the difference of the resulting widths . as the systematic uncertainties dominate the uncertainties of the widths , the error bars shown in fig . [ fig : gausswidths ] are the statistical and systematic uncertainties combined in quadrature . similarly to @xmath90 , the direct photons and @xmath32 both show decreasing widths with @xmath19 . linear fits to the two sets of widths give slopes of @xmath117 for @xmath32 mesons and @xmath118 for direct photons . systematic uncertainties on the slopes were conservatively estimated by evaluating the fit when the points were placed at the limits given by the systematic uncertainties , and then taking the difference of the slopes . similarly to @xmath90 the @xmath32 triggers were corrected by the same @xmath103 corrections from pythia . the result is shown in fig . [ fig : gausswidthscorrected ] ; again the @xmath103 correction amounts to a scale factor of approximately two for the @xmath19 of the @xmath32 triggers . when plotted against @xmath104 the magnitude of the slope for the @xmath32 triggers is @xmath119 . it should be noted that the slope of the widths changes if the minimum @xmath21 cut is increased , but that the slope always remains negative . integrating over the full range of @xmath120 gev/@xmath4 allowed by the phenix detector gives the smallest magnitude slope , thus it is the most conservative measurement for comparing to css evolution . for example , the slope of the gaussian widths of @xmath17 for @xmath121 gev/@xmath4 was determined to be @xmath122 for @xmath32-meson and @xmath123 for direct - photon triggers . the same behavior can be seen in the values of @xmath90 in fig . [ fig : rmspoutvspttrig ] and in the supplemental material @xcite . gaussian widths of @xmath17 for @xmath32 and direct photon triggers as a function of @xmath19 . ] gaussian widths of @xmath17 for @xmath32 and direct photon triggers as a function of @xmath89 , where @xmath89 is @xmath19 for the direct photons and @xmath104 for the @xmath32 triggers . @xmath103 was determined from a pythia simulation . ] ccc trigger type & @xmath124 [ gev/@xmath4 ] & gaussian width [ gev/@xmath4 ] + @xmath32 & 4.49 & 0.518 @xmath126 + & 5.46 & 0.514 @xmath127 + & 6.45 & 0.507 @xmath128 + & 7.44 & 0.502 @xmath128 + & 8.44 & 0.493 @xmath129 + & 10.1 & 0.487 @xmath130 + & 13.1 & 0.473 @xmath131 + + direct photon & 7.43 & 0.535 @xmath132 + & 8.44 & 0.476 @xmath133 + & 10.1 & 0.498 @xmath134 + & 13.2 & 0.456 @xmath135 + figures [ fig : rmspoutvspttrig ] and [ fig : gausswidths ] show that , consistent with previous rhic measurements , @xmath90 and the gaussian widths of @xmath17 sensitive to initial - state and final - state @xmath14 and @xmath16 decrease with the hard scale . interpretation of @xmath90 is slightly different than that of the gaussian widths from the @xmath17 distributions , because the gaussian widths are extracted from fits to the nearly back - to - back region , which is generated only by nonperturbative @xmath14 and @xmath16 . the @xmath90 values are extracted from fits to the entire away - side jet region in the @xmath22 correlations ; therefore , these quantities inherently include the charged hadrons in the perturbatively generated tail away from @xmath109 whereas the gaussian widths measured from the @xmath17 distributions _ only _ have contributions from @xmath109 . nonetheless the values of @xmath90 are dominated by the nearly back - to - back region as this is where most of the away - side charged hadrons are , but this subtle difference between the two observables should be noted . the widths quantified by @xmath90 have the benefit that they can be extracted from the finely binned @xmath136 @xmath22 angular correlations . throughout this discussion , we will use the term width " to refer to both the @xmath90 and gaussian widths extracted from @xmath17 . there is a difference in the mix of scattered away - side partons probed by includive-@xmath32 and direct - photon triggers . figure [ fig : pi0_partonic ] shows the fractional contribution to the total cross section calculated in pqcd for the lo diagrams for ( a ) inclusive-@xmath32 and ( b ) direct - photon production . the cteq6l1 @xcite pdfs were used for the calculations in addition to the dss14 ffs @xcite for the @xmath32 . at lo , quark - gluon compton scattering accounts for approximately 85% of direct photons produced at midrapidity , while @xmath32 triggers are instead generated by a significant contribution of @xmath137 , @xmath138 , and @xmath139 scatterings . therefore , any comparison between direct photon and @xmath32 triggers could be affected by the fact that the away - side charged hadrons are produced by quark jets @xmath1385% of the time for direct photons and a mix of gluon and quark jets for the @xmath32 . the fractional contribution of partonic scattering processes to ( a ) inclusive @xmath32 and ( b ) direct - photon production at lo in collisions at @xmath34 gev in the phenix pseudorapidity region . note that the process @xmath140 is not drawn in panel ( a ) because its contribution is less than one percent in this @xmath9 range . , title="fig : " ] the fractional contribution of partonic scattering processes to ( a ) inclusive @xmath32 and ( b ) direct - photon production at lo in collisions at @xmath34 gev in the phenix pseudorapidity region . note that the process @xmath140 is not drawn in panel ( a ) because its contribution is less than one percent in this @xmath9 range . , title="fig : " ] many correlation measurements similar to the one presented here have been made at rhic @xcite , as discussed in the introduction . although the same conclusion regarding the evolution of the widths found here can be drawn from these measurements , they were made with different physics goals ; examples include understanding partonic energy loss in a nuclear medium or characterizing fragmentation functions . earlier correlation measurements were largely motivated by the heavy ion or hard scattering high - energy - physics community , and it was not until recently that the nucleon - structure community began to understand how to look for possible factorization breaking effects in these types of measurements @xcite . this came as a result of the recent interest in understanding tmd evolution , especially understanding the nonperturbative contributions to tmd evolution . reference @xcite gives a comprehensive discussion of phenomenology including tmd evolution and how this phenomenology came to the forefront in 2011 . consistent with previous measurements , the data presented here clearly show that momentum widths sensitive to nonperturbative @xmath14 and @xmath16 decrease with the hard scale in @xmath32- and direct photon - charged hadron correlations . as was mentioned in the introduction , the expectation from css evolution is that momentum widths sensitive to nonperturbative transverse momentum scales should increase with the hard scale . to compare to what is predicted by css evolution , the slopes were compared to a slope of zero as this quantifies the boundary between narrowing and increasing widths with @xmath19 . the confidence interval excludes a slope of zero at the 2.6@xmath112 level for both the @xmath32 and direct photon triggered correlations . the likelihood ratio from a slope of zero was calculated to be 0.03 for both the @xmath32 and direct photon triggered correlations , which implies that the data is not consistent with a flat line . because @xmath14 and @xmath16 have been measured to be approximately constant in the @xmath19 region probed here @xcite , kinematically it would be expected that the acoplanarity decrease with @xmath19 . however , this same argument would apply for both dy and sidis , showing that the effect of decreasing widths seen in @xmath141 and direct @xmath142 correlations can not be a kinematic or fragmentation effect alone . it is also interesting that ref . @xcite shows that in dijet correlations at very high and , momentum widths sensitive to initial - state @xmath14 increase with the of the jet . these measurements are at large and sensitive to large @xmath14 at the higher of the large hadron collider , and thus follow the leading - log approximation which is also purely perturbative and predicts increasing widths with the hard scale ( see e.g. @xcite and references within ) . the css evolution framework was motivated by understanding perturbative qcd dynamics . at this time , qcd was still in the early stages of development , and nonperturbative dynamics were not the focus within the framework of pqcd . as qcd became well established as the theory of the strong force , measurements performed at high enough energies for perturbative techniques to be applicable began to be used routinely to constrain nonperturbative physics in the form of collinear pdfs and ffs . it is only in the last two decades that there has been increasing focus on using perturbative techniques to understand nonperturbative parton dynamics . the study of such nonperturbative dynamics provides information on parton behavior within bound states and the process of hadronization by defining and constraining tmd pdfs and ffs . importantly , it is additionally offering new insights on fundamental aspects of qcd as a nonabelian gauge - invariant quantum field theory , for example through the predicted relative sign difference of the sivers tmd pdf when probed via sidis versus dy @xcite , and through tmd factorization breaking in certain processes @xcite . factorization breaking results from basic qcd principles . namely , nonabelian phase interferences from the exchange of gluons between colored objects can not , in general , be disentangled . similarly , phase interferences from gluon exchange play a role in the sivers effect where it implies a sign change for the sivers function between dy and sidis interactions . the reason that gluon exchange in dy and sidis does not lead to factorization breaking is because dy and sidis are both quantum electrodynamic processes at lo , so there are limited paths for gluon exchange . only initial - state exchange in dy and final - state exchange in sidis are possible , whereas in hadronic collisions with a final - state hadron measured both initial- and final - state exchanges are possible . observing differences in the evolution of momentum widths as a function of the hard scale is a powerful observable due to the qualitative conclusion that can be drawn from the data when comparing to the expectation of css evolution . before the recent interest in understanding tmd evolution , measured deviations from calculations at some given scale assuming factorization holds were the only obvious way to look for factorization breaking effects . such calculations are not available . simply looking for qualitative differences in the evolution of the observable gives a clear discrepancy with the expectation from css evolution , and this is significantly more powerful than trying to compare with a calculation that requires greater knowledge of the nonperturbative functions . it is furthermore interesting to point out that the inclusive hadron transverse single - spin asymmetries in hadronic collisions measured at forward rapidities also deviate from the expectation provided by standard perturbative evolution . in charged pion production , the asymmetry changes strikingly little from @xmath143 gev to @xmath144 gev @xcite . asymmetries have been measured to be nonzero at center of mass energies up to 200 gev and appear to plateau at up to 5 gev/@xmath4 @xcite , while perturbative techniques give a clear prediction that the asymmetry should fall off as increases @xcite . without any available theoretical calculations of our observable , the results found in the data as well as the expectation from css evolution were investigated with a pythia simulation . factorization breaking is not predicted in dy production because there are no final - state hadrons produced directly from the hard scattering . the same observable @xmath17 can be constructed in dy events with two nearly back - to - back leptons . for dileptons , it would be expected that the gaussian width of @xmath17 would broaden as one increases the hard scale of the interaction , as dy is known to follow css evolution . the perugia0 tune @xcite should be ideal to study this because it was tuned to the cdf z@xmath145 cross section data at low @xcite . therefore the perugia tunes should be reasonably adequate at reproducing dy events where the total is small . gaussian widths extracted from pythia drell - yan @xmath17 distributions . in drell - yan factorization breaking is not predicted . here the widths show a positive slope with the invariant mass of the dilepton , as predicted by css evolution . the solid red line shows a log fit over the full invariant mass range and the blue dotted line shows a linear fit in the region 510 gev/@xmath45 . ] pythia 6.4 dy events were generated and @xmath17 was determined for the correlated dileptons to confirm the expectation from css evolution for this observable . @xmath17 is defined similarly to eq . [ eq : pout ] , @xmath146 where the higher - lepton is taken as the near - side trigger particle and the lower - lepton is taken as the away - side associated particle used in the determination of @xmath17 . the distributions were fit with gaussian functions in the nonperturbative nearly back - to - back region , and the widths of the dilepton @xmath17 distributions are shown in fig . [ fig : dygausswidths ] . pythia reproduces the expectation that the widths of the dy pairs increase with the @xmath147 of the interaction when @xmath17 is only sensitive to initial - state @xmath14 and there are no final - state hadrons . the widths are quantitatively much larger than the dihadron or direct photon - hadron widths because the dy dileptons emerge from the virtual photon , which means that , in the phenix pseudorapidity region , their is large . when measuring a final - state hadron , @xmath17 by definition must be smaller for the case of a measured final - state hadron vs. a dy lepton because the of the charged hadron must be smaller than or equal to the of the scattered parton due to the fragmentation process . any quantitative value of @xmath17 will naturally be dependent on the @xmath21 measured ; see fig . [ fig : rmspoutvsptassoc ] . in dy , each lepton will have approximately half the momentum of the interaction hard scale , and the larger the momentum the larger @xmath17 can be while still being in the nearly back - to - back region @xmath109 , i.e. nonperturbatively generated . what is relevant is the evolution of this width with the hard scale of the interaction , not the quantitative value , as this is just indicative of what away - side @xmath21 is observed . the dy widths were fit with a linear function shown as the blue dotted line in the region which was most linear , 510 gev/@xmath45 , and the slope of the line was determined to be 0.146@xmath1000.004 . additionally the red solid line shows a log fit over the full invariant mass range . the dy slope is the opposite sign from the direct photon - hadron and dihadron correlations and it is also approximately one order of magnitude larger making it significantly different from the dihadron and direct photon - hadron slopes . similarly , pythia direct photon and dijet events were generated at @xmath34 gev with the perugia0 tune , changing the gaussian intrinsic @xmath14 parameter parp(91 ) setting to @xmath148 gev/@xmath4 as should be expected at @xmath34 gev from ref . the perugia0 tune was used again for the dihadron and direct photon - hadron correlations so that a direct comparison could be made to the dy perugia0 tune gaussian widths . the direct photons were required to be isolated similarly to what was done in data . correlated pairs of @xmath32 or direct photon and @xmath149 , @xmath150 , and @xmath0,@xmath151 were collected in the phenix pseudorapidity , and the observables @xmath22 and @xmath17 were determined from the correlated pairs . similarly to what was done in data , the background from the underlying event was statistically subtracted out to make the @xmath17 distributions . pythia correlated pairs show the same features as the data do as can be seen in fig . [ fig : pout_pythia ] ; @xmath17 exhibits a gaussian shape at small @xmath17 which transitions to a power law shape at large @xmath17 . gaussian widths were extracted from the pythia - generated correlations in the same way that was done for the data . the widths from the pythia @xmath17 distributions are shown with the measured widths in fig . [ fig : pythia_gauss ] . remarkably , the pythia results reproduce the measured slopes in both sign and magnitude for both @xmath32 and direct photon triggers . the slope values from pythia were @xmath152 for @xmath32-meson @xmath153 for direct - photon triggers . the measured slopes are @xmath117 for @xmath32-meson and @xmath118 for direct - photon triggers . the negative sign of the slope was found in both the quark - gluon compton and quark - antiquark annihilation processes for isolated direct photon production , indicating that the effect in pythia is not due to a difference in quark vs. gluon fragmentation . one noticeable difference between pythia and the data is the quantitative values of the widths . here the results from pythia differ by about @xmath1315% for both the @xmath32 and direct photon triggers , depending on the @xmath19 bin . gaussian widths extracted from pythia @xmath17 distributions are shown compared to the results measured at phenix . pythia produces consistent evolution rates as what was determined in phenix , although there is a 10%15% difference in the quantitative values in each @xmath19 bin . ] the nonperturbative gaussian behavior of @xmath17 is generated by the soft initial - state @xmath14 and final - state @xmath16 as indicated in fig . [ fig : ktkinematics ] . in the nearly back - to - back region @xmath154 gev/@xmath4 , @xmath17 is small and thus can only be generated by soft gluon radiation because the two particles are nearly coplanar . it is unsurprising that pythia does not replicate the quantitative values of the gaussian widths well as there is little data that would offer constraints to this region . what is striking is that pythia replicates the evolution rate for both @xmath32 and direct photon gaussian widths . while pythia certainly does not explicitly consider analytical factorization breaking effects as it assumes collinear factorization , in contrast to a collinear pqcd calculation it does include initial- and final - state interactions . after a parton interacts in pythia , the remnants of the two protons are free to interact with other objects in the event , and every object in the interaction is forced to color neutralize . factorization breaking effects are predicted in dihadron and direct photon - hadron correlations due to the possibility of gluon exchange in both the initial and final states . this includes gluon exchange with remnants of the interaction , because the remnants of the interacting protons can exchange gluons with partons in both the initial and final states . sensitivity to these effects requires a small transverse momentum scale ; in the transverse - momentum - integrated case observables no longer have this sensitivity . for this reason it is plausible that pythia could be sensitive to these effects because of interactions between the proton remnants and partons involved in the hard scattering . because pythia allows initial- and final - state interactions via gluon exchanges , the necessary interactions to allow for factorization breaking effects are present within the pythia framework . it should also be noted that pythia replicates the color coherence effects in refs . @xcite as well . the underlying mechanism that leads to the prediction of the sign change in the sivers function or factorization breaking is gluon exchange between partons associated with the hard scattering and colored remnants . in cases where factorization breaking is predicted , it implies that the traditional organization of the nonperturbative objects into separate pdfs and ffs for each colliding proton and produced hadron no longer holds . however , so far we have no knowledge of how to approach a reorganization of the nonperturbative objects , which would presumably include novel correlation functions describing partons correlated across the colliding protons . the fact that pythia accurately describes both the qualitative and quantitative nature of the slopes of the widths as a function of hard scale offers a potential path forward to greater understanding and further advancing what can be calculated within the rigors of pqcd . dihadron and direct photon - hadron correlations sensitive to nonperturbative transverse momentum effects have been measured in the phenix experiment at rhic in @xmath34 gev @xmath33 collisions , motivated by the prediction of factorization breaking in such processes @xcite . correlations between @xmath32 or direct photons with charged hadrons were measured . the azimuthal angular separation @xmath22 and out - of - plane transverse momentum component @xmath17 for the correlated pairs were measured . @xmath17 has sensitivity to nonperturbative transverse momentum in the initial state , as well as in the final state when a produced hadron is measured . the @xmath90 and gaussian widths of the @xmath17 distributions were measured from the correlations , and both observables decrease with the hard scale of the interaction @xmath19 . the direct photons exhibit a larger dependence than the @xmath32 triggers on @xmath19 for both @xmath90 and the gaussian widths of @xmath17 . the narrowing of the gaussian widths as a function of @xmath19 indicates that the collins - soper - sterman soft factor can not be driving the evolution , in contrast with drell - yan dilepton production and sidis where factorization is predicted to hold and the widths are empirically known to increase with hard scale . study of the same observables via the pythia event generator , which allows for gluon exchange between partons involved in the hard scattering and the proton remnants , reveals strikingly similar characteristics . the similarity between pythia and the experimental data offers a promising path forward to understand the mechanism in qcd driving the observed evolution in more detail . we thank the staff of the collider - accelerator and physics departments at brookhaven national laboratory and the staff of the other phenix participating institutions for their vital contributions . we also thank t.c . rogers , j.c . collins , and l. lnnblad for valuable discussions regarding the interpretation of these results , as well as t. kauffman for providing several pqcd calculations . we acknowledge support from the office of nuclear physics in the office of science of the department of energy , the national science foundation , abilene christian university research council , research foundation of suny , and dean of the college of arts and sciences , vanderbilt university ( u.s.a ) , ministry of education , culture , sports , science , and technology and the japan society for the promotion of science ( japan ) , conselho nacional de desenvolvimento cientfico e tecnolgico and fundao de amparo pesquisa do estado de so paulo ( brazil ) , natural science foundation of china ( p. r. china ) , croatian science foundation and ministry of science , education , and sports ( croatia ) , ministry of education , youth and sports ( czech republic ) , centre national de la recherche scientifique , commissariat lnergie atomique , and institut national de physique nuclaire et de physique des particules ( france ) , bundesministerium fr bildung und forschung , deutscher akademischer austausch dienst , and alexander von humboldt stiftung ( germany ) , national science fund , otka , kroly rbert university college , and the ch . simonyi fund ( hungary ) , department of atomic energy and department of science and technology ( india ) , israel science foundation ( israel ) , basic science research program through nrf of the ministry of education ( korea ) , physics department , lahore university of management sciences ( pakistan ) , ministry of education and science , russian academy of sciences , federal agency of atomic energy ( russia ) , vr and wallenberg foundation ( sweden ) , the u.s . civilian research and development foundation for the independent states of the former soviet union , the hungarian american enterprise scholarship fund , and the us - israel binational science foundation . 74ifxundefined [ 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 10.1016/0550 - 3213(81)90339 - 4 [ * * , ( ) ] , link:\doibase 10.1016/0550 - 3213(82)90021 - 9 [ * * , ( ) ] link:\doibase 10.1016/0550 - 3213(85)90479 - 1 [ * * , ( ) ] link:\doibase 10.1103/physrevd.41.83 [ * * , ( ) ] link:\doibase 10.1103/physrevd.43.261 [ * * , ( ) ] \doibase doi:10.1016/0550 - 3213(95)00632-x [ * * , ( ) ] , http://www.cambridge.org/de/knowledge/isbn/item5756723[__ ] ( , ) link:\doibase 10.1103/physrevd.85.094510 [ * * , ( ) ] link:\doibase 10.1103/physrevd.83.094507 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.262002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.94.012002 [ * * , ( ) ] link:\doibase 10.1016/j.physletb.2015.03.056 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.102.182001 [ * * , ( ) ] link:\doibase 10.1007/bf01549713 [ * * , ( ) ] link:\doibase 10.1103/physrevd.39.92 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.116.132301 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.072003 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.262002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.96.232002 [ * * , ( ) ] link:\doibase 10.1103/physrevd.90.052003 [ * * , ( ) ] link:\doibase 10.1016/j.physletb.2010.08.012 [ * * , ( ) ] link:\doibase 10.1103/physrevd.90.072008 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.92.171801 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.042001 [ * * , ( ) ] link:\doibase 10.1016/s0550 - 3213(03)00121 - 4 [ * * , ( ) ] link:\doibase 10.1016/s0370 - 2693(02)01320 - 5 [ * * , ( ) ] link:\doibase 10.1016/s0370 - 2693(02)01819 - 1 [ * * , ( ) ] link:\doibase 10.1103/physrevd.81.094006 [ * * , ( ) ] link:\doibase 10.1140/epjc / s2006 - 02554 - 2 [ * * , ( ) ] link:\doibase 10.1103/physrevd.75.114014 [ * * , ( ) ] @noop `` , '' ( ) , link:\doibase 10.1103/physrevd.50.5562 [ * * , ( ) ] link:\doibase 10.1016/s0370 - 2693(97)01190 - 8 [ * * , ( ) ] link:\doibase 10.1140/epjc / s10052 - 014 - 2901 - 8 [ * * , ( ) ] @noop `` , '' ( ) , link:\doibase 10.1016/0550 - 3213(77)90384 - 4 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevd.91.074020 [ * * , ( ) ] link:\doibase 10.1103/physrevd.67.073016 [ * * , ( ) ] link:\doibase 10.1016/j.physletb.2005.12.063 [ * * , ( ) ] link:\doibase 10.1103/physrevd.81.094019 [ * * , ( ) ] link:\doibase 10.1103/physrevd.89.094002 [ * * , ( ) ] link:\doibase 10.1103/physrevd.64.059903 , 10.1103/physrevd.61.014003 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.74.072002 [ * * , ( ) ] link:\doibase 10.1103/physrevd.81.012002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.81.2642 [ * * , ( ) ] link:\doibase 10.1016/0370 - 2693(80)90572 - 9 [ * * , ( ) ] link:\doibase 10.1103/physrevd.18.3320 [ * * , ( ) ] link:\doibase 10.1103/physrevd.82.072001 [ * * , ( ) ] link:\doibase 10.1016/j.physletb.2015.05.033 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1016/s0168 - 9002(02)01954-x [ * * , ( ) ] link:\doibase 10.1103/physrevd.86.072008 [ * * , ( ) ] link:\doibase 10.1103/physrevd.93.011501 [ * * , ( ) ] link:\doibase 10.1103/physrevd.83.032001 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevd.82.074024 [ * * , ( ) ] link:\doibase 10.1103/physrevd.91.014035 [ * * , ( ) ] link:\doibase 10.1016/j.physletb.2012.11.003 [ * * , ( ) ] @noop @noop * * , ( ) link:\doibase 10.1140/epjc / s10052 - 011 - 1795-y [ * * , ( ) ] @noop `` , '' @noop `` , '' link:\doibase 10.1103/physrevlett.112.122301 [ * * , ( ) ] link:\doibase 10.1103/physrevd.83.114042 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.102.232002 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.85.655 [ * * , ( ) ] link:\doibase 10.1103/physrevd.90.012006 [ * * , ( ) ] link:\doibase 10.1103/physrevd.88.014002 [ * * , ( ) ] link:\doibase 10.1103/physrevd.82.074018 [ * * , ( ) ] link:\doibase 10.1103/physrevd.86.052010 [ * * , ( ) ]

we characterize the finite intervals of the muchnik lattice by proving that they are a certain proper subclass of the finite distributive lattices . + muchnik lattice finite distributive lattices turing degrees 03d28 , 03d30 .

math0606728

the medvedev lattice and the muchnik lattice are structures from computability theory that were originally defined for their connections with constructive logic , but that are of independent interest as well . both can be seen as generalizations of the turing degrees , and for example when muchnik presented his solution to post s problem he phrased it as a result of the medvedev lattice . in terwijn @xcite the structure of the medvedev lattice @xmath0 was investigated , and it was proven there that the finite intervals of @xmath0 are precisely the finite boolean algebras , and that the infinite intervals of @xmath0 all have cardinality @xmath1 ( cf . theorem [ extendeddichotomy ] ) . it was noted there that this strong dichotomy does not hold for the muchnik lattice @xmath2 , and that there are many more possibilities for intervals in @xmath2 , both for the finite and for the infinite ones . in this paper we characterize the finite intervals of @xmath2 by proving that they are a certain subclass of the finite distributive lattices that can be described using elementary lattice theory . in the rest of this section we will repeat the necessary definitions and some further preliminaries . the medvedev lattice , introduced by medvedev @xcite , is a particular way of specifying kolmogorov idea of a calculus of problems . let @xmath3 denote the natural numbers and let @xmath4 be the set of all functions from @xmath3 to @xmath3 ( baire space ) . mass problem _ is a subset of @xmath4 . every mass problem is associated with the `` problem '' of producing an element of it . a mass problem @xmath5 _ medvedev reduces _ to mass problem @xmath6 , denoted @xmath7 , if there is a partial computable functional @xmath8 defined on all of @xmath6 such that @xmath9 . that is , @xmath10 is a uniformly effective method for transforming solutions to @xmath6 into solutions to @xmath5 . the relation @xmath11 induces an equivalence relation on mass problems : @xmath12 if @xmath13 and @xmath14 . the equivalence class of @xmath5 is denoted by @xmath15 and is called the _ medvedev degree _ of @xmath5 . we denote medvedev degrees by boldface symbols . there is a smallest medvedev degree , denoted by @xmath16 , namely the degree of any mass problem containing a computable function , and there is a largest degree @xmath17 , the degree of the empty mass problem , of which it is absolutely impossible to produce an element . a meet operator @xmath18 and a join operator @xmath19 are defined on mass problems as follows : for functions @xmath20 and @xmath21 , as usual define the function @xmath22 by @xmath23 and @xmath24 . let @xmath25 , where @xmath26 denotes concatenation . define @xmath27 and @xmath28 the structure @xmath0 of all medvedev degrees , ordered by @xmath11 and together with @xmath19 and @xmath18 is a distributive lattice ( medvedev @xcite ) . the muchnik lattice , introduced by muchnik @xcite , is a nonuniform variant of the medvedev lattice . it is the structure @xmath2 resulting from the reduction relation on mass problems defined by @xmath29.\ ] ] ( the `` w '' stands for `` weak '' . ) that is , every solution to the mass problem @xmath6 can compute a solution to the mass problem @xmath5 , but maybe not in a uniform way . it is easy to check that @xmath2 is a distributive lattice in the same way that @xmath0 is , with the same lattice operations and @xmath30 and @xmath17 . notice that @xmath31 in @xmath2 simplifies to @xmath32 . an m - degree is a _ muchnik degree _ if it contains a mass problem that is upwards closed under turing reducibility @xmath33 . the muchnik degrees of @xmath0 form a substructure that is isomorphic to @xmath2 . for any mass problem @xmath5 , let @xmath34 denote the upward closure of @xmath5 under @xmath33 . we have the following embeddings : @xmath35 more discussion about the elementary properties of @xmath0 and @xmath2 can be found in in rogers textbook @xcite and the survey paper by sorbi @xcite . previous results about embeddings of lattices and algebras into @xmath0 and @xmath2 can be found in sorbi @xcite . binns and simpson @xcite contains results about lattice embeddings into the lattice of @xmath36-classes under @xmath11 and @xmath37 . our notation is mostly standard and follows odifreddi @xcite . @xmath38 is the @xmath39-th partial computable functional . for countable sets @xmath40 and mass problems @xmath41 , @xmath42 , we have the meet operator @xmath43 note that for finite @xmath44 this is m - equivalent to an iteration of the meet operator @xmath18 . if @xmath45 in some partial order , we use the interval notation @xmath46 = \{x : a{\leqslant}x{\leqslant}b\}$ ] . similarly @xmath47 denotes an interval without endpoints , and @xmath48 $ ] denotes the set @xmath49 . we say that @xmath50 _ covers _ @xmath51 if @xmath52 and there is no @xmath53 with @xmath54 . in the final section of @xcite some consequences of the results of that paper for the muchnik lattice @xmath2 were listed . some of these consequences were : * in contrast to @xmath0 , the lattice @xmath2 contains nonempty linear intervals . * every finite boolean algebra is isomorphic to an interval of @xmath2 . * every muchnik degree is as large as set - theoretically possible : for every mass problem @xmath55 we have @xmath56 . * whereas in @xmath0 only countable boolean algebras can be embedded , the dual of @xmath57 is embeddable into @xmath2 as a boolean algebra . a medvedev degree is a _ degree of solvability _ if it contains a singleton mass problem . a mass problem is called _ nonsolvable _ if its m - degree is not a degree of solvability . for every degree of solvability @xmath58 there is a unique minimal m - degree @xmath59 that is denoted by @xmath60 ( cf . medvedev @xcite ) . if @xmath61 then @xmath60 is the degree of the mass problem @xmath62 note that @xmath63 is m - equivalent to the set of all noncomputable functions . we will also denote this set by @xmath64 . note further that @xmath65 so that in @xmath2 we can use this simplified version of @xmath66 . dyment @xcite proved that the degrees of solvability are precisely characterized by the existence of such an @xmath60 . namely , the degrees of solvability are first - order definable ( both in @xmath0 and in @xmath2 ) by the formula @xmath67 thus the turing degrees form a first - order definable substructure of both @xmath0 and @xmath2 . this has many immediate corollaries , for example that the first - order theories of the structures @xmath68 and @xmath69 are undecidable . ( dyment @xcite , cf . * theorem 2.5 ) ) [ dyment ] for medvedev degrees @xmath70 and @xmath71 with @xmath72 it holds that @xmath73 if and only if there is a degree of solvability @xmath58 such that @xmath74 , @xmath75 , and @xmath76 . theorem [ dyment ] also holds for @xmath2 , with a much easier proof . we will include a proof here , as a warm - up for section [ finite ] . [ subst ] suppose that @xmath5 and @xmath6 satisfy @xmath77 then there exists @xmath78 such that @xmath79 and @xmath80 . if moreover @xmath81 then the interval @xmath82 is infinite . since from ( [ new ] ) it follows that @xmath83 , there is @xmath84 such that @xmath85 . again by ( [ new ] ) we have that @xmath86 , so we can take @xmath87 . if in addition @xmath81 then we have @xmath88 . since @xmath5 and @xmath89 also satisfy ( [ new ] ) we can by iteration of the first part of the lemma obtain an infinite downward chain in @xmath82 . ( dyments theorem for @xmath2 ) [ dymentmw ] for muchnik degrees @xmath70 and @xmath71 with @xmath90 it holds that @xmath91 if and only if there is a degree of solvability @xmath58 such that @xmath92 , @xmath93 , and @xmath94 . ( if ) suppose that @xmath95 is as in the theorem and suppose that @xmath96 , @xmath97 , and @xmath98 . if @xmath99 does not contain any element of turing degree @xmath100 then it follows that @xmath101 , because the elements of @xmath99 that get sent to the @xmath102-side are all strictly above @xmath20 , hence included in @xmath66 . so in this case @xmath103 by @xmath104 . otherwise @xmath99 contains an element of turing degree @xmath100 , and consequently @xmath105 . hence @xmath106 . ( only if ) suppose that @xmath107 . then by lemma [ subst ] , @xmath5 and @xmath6 do not satisfy condition ( [ new ] ) , hence there is a finite set @xmath108 such that @xmath109 . there is also an @xmath110 such that @xmath111 , for otherwise we would have @xmath112 . because the interval is empty and @xmath113 we must have @xmath114 since there is no other possibility for @xmath89 . we also have @xmath115 because both @xmath111 and @xmath116 . hence @xmath117 , again by emptiness of the interval , and in particular @xmath104 . so we can take @xmath58 to be @xmath118 . let @xmath20 and @xmath21 be t - incomparable . then the interval @xmath119 $ ] contains exactly two intermediate elements , cf . figure [ fig ] . ( 24 , 29 ) ( 12,0 ) ( -6,12 ) ( -6,16)(2,1)12 ( -6,10)(2,-1)12 ( 30,12 ) ( 30,16)(-2,1)12 ( 30,10)(-2,-1)12 ( 12,24 ) this can be generalized to obtain finite intervals of size @xmath120 for any @xmath121 as follows : [ exactlyn ] let @xmath6 be any mass problem . let @xmath122 and let @xmath123 be t - incomparable such that @xmath124 for every @xmath125 . then the interval @xmath126\ ] ] in @xmath2 is isomorphic to the finite boolean algebra @xmath127 . this was proved in @xcite for @xmath0 . it holds for @xmath2 with the same proof . platek @xcite proved that @xmath0 has the ( for a collection of sets of reals maximal possible ) cardinality @xmath1 by showing that @xmath0 has antichains of that cardinality . ( he mentions that the result was noted independently by elisabeth jockusch and john stillwell . ) in fact , in @xmath0 such large antichains occur in every infinite interval : ( terwijn @xcite ) [ extendeddichotomy ] let @xmath128 $ ] be an interval in @xmath0 with @xmath72 . then either @xmath128 $ ] is isomorphic to the boolean algebra @xmath127 for some @xmath122 , or @xmath128 $ ] contains an antichain of size @xmath1 . in the latter case , assuming ch , it also contains a chain of size @xmath1 . in particular , @xmath0 s version of theorem [ exactlyn ] is the _ only _ way to generate finite intervals of @xmath0 . as we will see in what follows , the situation for @xmath2 is rather different . although every countable linear order can be embedded into @xmath2 ( because by lachlan ( cf . * p529 ) ) this already holds for the turing degrees ) , the following result shows that not every countable linear order is _ isomorphic _ to an interval in @xmath2 . ( from theorem [ main ] it will follow that every finite linear order is isomorphic to an interval in @xmath2 . ) not every countable linear order is isomorphic to an interval in @xmath2 . consider the linear order @xmath129 ( that is , a copy of @xmath3 followed by a reverse copy of @xmath3 ) . suppose that @xmath130 and @xmath131 , @xmath132 , are mass problems such that for all @xmath121 and @xmath133 @xmath134 let @xmath135 . then for all @xmath121 , @xmath136 , so the interval @xmath137 $ ] is not isomorphic to @xmath129 . [ downwardchain ] @xmath2 contains linear intervals that are countably infinite . by lachlan , cf . * p529 ) , every countable distributive lattice with a least element is embeddable as an initial segment in the turing degrees . consider the linear order @xmath138 ( a least element plus a reverse copy of @xmath3 ) and embed this in the t - degrees : let @xmath139 be such that @xmath140 and such that @xmath141 for every @xmath142 . let @xmath143 and @xmath144 . ( note that @xmath5 is in fact w - equivalent to @xmath64 . ) now if @xmath145 $ ] then @xmath99 can be split in disjoint parts @xmath146 and @xmath147 such that @xmath146 is maximal with the property @xmath148 and @xmath149 . then @xmath150 and @xmath151 . so it suffices to analyze all subclasses of @xmath152 : for every @xmath40 consider @xmath153 . if @xmath44 is infinite then @xmath154 , hence @xmath155 . for @xmath44 and @xmath156 finite we have @xmath157 whenever @xmath158 . so we see that the interval @xmath159 contains only the countably many elements @xmath160 , @xmath161 . by proposition [ downwardchain ] there are linear nonempty intervals in @xmath2 . this contrasts the situation for @xmath0 , where by theorem [ extendeddichotomy ] all the linear intervals are empty . so here we already see that theorem [ exactlyn ] is not the only way anymore to generate finite intervals . @xmath2 contains antichains of size @xmath1 , using the same argument that platek used for @xmath0 , ( starting with an antichain of size @xmath162 in the turing degrees , form @xmath1 incomparable combinations ) but proposition [ downwardchain ] shows that they do not occur in every infinite interval , as we had for @xmath0 ( cf . theorem [ extendeddichotomy ] ) . in fact there are intervals with maximal antichains of every possible size : [ antichainsinmw ] each of the following possibilities is realized by some interval @xmath128 $ ] in @xmath2 : 1 . @xmath128 $ ] contains an antichain of size @xmath121 , but not of size @xmath163 , 2 . @xmath128 $ ] contains an antichain of size @xmath164 , but no uncountable antichain , 3 . @xmath128 $ ] contains an antichain of size @xmath162 , but not of size @xmath1 , 4 . ( platek @xcite ) @xmath128 $ ] contains an antichain of size @xmath1 . _ this follows from theorem [ exactlyn ] . _ let @xmath165 be an increasing chain of elements in some lattice and let @xmath166 be a decreasing chain of elements in the same lattice such that @xmath167 for all @xmath121 . let @xmath168 be the free distributive lattice on these sets of elements with an additional least element . then @xmath168 is a countable bottomed distributive lattice , so by lachlan ( * ? ? ? * p529 ) , @xmath168 is embeddable into the turing degrees as an initial segment . let @xmath169 and @xmath170 be representatives from the image @xmath171 of @xmath168 corresponding to the sequences @xmath172 and @xmath173 respectively , so that @xmath174[c]{$|_t$}}g_j$ ] for all @xmath125 and @xmath175 and such that for all @xmath121 , @xmath176 and @xmath177 . let @xmath178 then every @xmath179 $ ] can be split as @xmath180 , with @xmath181 maximal with the property that @xmath148 and @xmath182 . claim : the only elements of @xmath171 that are not in @xmath6 are the @xmath183 , @xmath184 , @xmath161 . to see the claim , note that the nonzero elements of @xmath171 are free combinations of the @xmath183 and @xmath184 . clearly @xmath6 is closed under joins . by freeness of @xmath185 it also easily follows that @xmath186 is closed under meets . hence @xmath187 is closed under meets and joins , and from this it easily follows by induction on the complexity of the elements that every element in @xmath188 is in @xmath6 . this proves the claim . as a consequence , we have ( by maximality of @xmath146 ) that @xmath189 . now @xmath190 is determined by @xmath191 : one easily checks that if @xmath192 $ ] are split as above as @xmath180 and @xmath193 then @xmath194 implies that @xmath195 . in its turn , @xmath191 is determined by the minimal @xmath121 ( if any ) such that @xmath196 and by whether @xmath147 contains infinitely or finitely many @xmath197 s , and in the latter case by the maximal @xmath133 ( if any ) such that @xmath198 . so we see that there are only countably many possibilities for the degree of @xmath147 , and hence for the degree of @xmath99 , and hence @xmath199 $ ] is countable . now consider the mass problems @xmath200 clearly @xmath201 if @xmath202 . so @xmath199 $ ] is countable and contains an infinite antichain . _ let @xmath168 be a countably infinite distributive lattice with a least element and an infinite antichain . by lachlan ( * ? ? ? * p529 ) , @xmath168 is embeddable into the turing degrees as an initial segment . let @xmath183 , @xmath132 , be a set of representatives of all the degrees in the image of @xmath168 . consider the interval @xmath203 $ ] , where @xmath204 and @xmath205 . then @xmath203 $ ] contains an infinite antichain of elements of the form @xmath89 because @xmath168 contains a corresponding infinite antichain . for @xmath206 let @xmath207 . then for incomparable sets @xmath44 , @xmath208 it holds that @xmath209 . so @xmath203 $ ] contains an antichain of size @xmath162 . now if @xmath210 $ ] then @xmath180 , with @xmath181 maximal with the property that @xmath148 and @xmath211 . so the muchnik degree @xmath190 of every @xmath210 $ ] is determined by a countable set @xmath147 , hence there are at most @xmath162 many elements in @xmath203 $ ] . _ _ we can apply plateks argument to any interval that contains an antichain of size @xmath162 of singletons : suppose that the interval @xmath199 $ ] contains the elements @xmath212 , @xmath213 , such that the @xmath214 form an antichain in the turing degrees . for @xmath215 let @xmath216 . clearly @xmath217 $ ] . now for incomparable sets @xmath44 , @xmath218 it holds that @xmath209 , so it suffices to note that there is an antichain of size @xmath1 in @xmath57 . ( for some general notes on chains and antichains we refer to @xcite . ) from the proof of theorem [ antichainsinmw ] we can also read off some consequences for chains in @xmath2 : 1 . by theorem [ exactlyn ] there are intervals containing chains of size @xmath121 but not of size @xmath163 , 2 . by the proof of item 2 . , and also proposition [ downwardchain ] , there are countable intervals with an infinite chain , 3 . the example of an interval given in the proof of item 3 . contains also a chain of size @xmath162 , but not of size @xmath1 . this is because @xmath219 has a chain of size @xmath162 so the same holds with @xmath3 replaced by @xmath220 . a chain in the interval of item 3 . can not be bigger since the interval itself was of size @xmath162 . ch implies that @xmath2 has a chain of size @xmath1 , cf . the conditions for the existence of chains of size @xmath1 in @xmath57 , in @xmath0 , and in @xmath2 are the same . the consistency of the existence of chains of this size also follows from the fact that the dual of @xmath57 is embeddable into @xmath2 , cf . ( sorbi @xcite ) [ embedding ] a countable distributive lattice with 0,1 is embeddable into @xmath0 ( preserving 0 and 1 ) if and only if 0 is meet - irreducible and 1 is join - irreducible . sorbi proved theorem [ embedding ] by embedding the ( unique ) countable dense boolean algebra into @xmath0 . since this algebra is embeddable into @xmath0 it also embeds into @xmath2 . ( this is because @xmath11 is stronger than @xmath37 and because for any mass problems @xmath5 and @xmath6 the sets @xmath221 and @xmath31 are the same in @xmath0 and in @xmath2 . ) in particular every finite distributive lattice is embeddable into @xmath2 . in the following we consider lattices that are _ isomorphic _ to an interval of @xmath2 . in theorem [ extendeddichotomy ] we saw that for @xmath0 these were precisely the finite boolean algebras . of course no nondistributive lattice can be isomorphic to an interval in @xmath0 or @xmath2 since both structures are distributive ( medvedev @xcite ) . in this section we characterize the finite intervals of @xmath2 as a certain subclass of the finite distributive lattices ( theorem [ main ] ) . we start with some illustrative examples . [ exdiam ] that the diamond lattice is isomorphic to an interval in @xmath2 was already shown in theorem [ exactlyn ] . for later purposes we show that this way of obtaining a diamond is essentially the only way . suppose that @xmath222 $ ] is an interval in @xmath2 containing precisely two intermediate elements @xmath6 and @xmath99 , and that @xmath6 and @xmath99 are incomparable , cf . figure [ fff ] . then @xmath5 and @xmath223 do not satisfy property ( [ new ] ) of lemma [ subst ] because the interval is finite . so there is a finite set @xmath224 such that @xmath225 , and hence @xmath226 . without loss of generality the elements of @xmath227 are pairwise t - incomparable and @xmath228 for every @xmath229 . since for every @xmath230 we have @xmath231 we see that @xmath227 can contain at most two elements . if @xmath227 would contain only one element @xmath20 then @xmath232 $ ] would be an initial segment of the interval @xmath222 $ ] , which is a contradiction . so @xmath227 contains precisely two elements , @xmath233 and @xmath234 say . then @xmath222 $ ] contains the interval @xmath235 $ ] , which by theorem [ exactlyn ] is isomorphic to the diamond lattice . so we must have that @xmath236 . let @xmath233 and @xmath234 be t - incomparable and let @xmath237[c]{$|_t$}}f_0 $ ] be minimal over @xmath234 , i.e. @xmath238 and there is no function of t - degree strictly between @xmath234 and @xmath239 . ( in lattice theoretic terminology @xmath239 is said to _ cover _ @xmath234 . ) let @xmath240[c]{$|_t$}}g_1\}$ ] . we then have @xmath241 , as is easy to check . define @xmath242 then by example [ exdiam ] the intervals @xmath222 $ ] and @xmath243 $ ] are both isomorphic to a diamond , hence the whole interval @xmath244 $ ] is isomorphic to the configuration of figure [ ff2 ] . as in example [ exdiam ] one can also show that the above method is essentially the _ only _ way of obtaining this configuration as an interval of @xmath2 . assume for a contradiction that the interval @xmath246 $ ] is isomorphic to the lattice of figure [ figccc ] . as in example [ exdiam ] we can argue that there is a finite set @xmath224 such that @xmath247 . using the same reasoning as before we can argue that @xmath227 contains precisely two t - incomparable elements @xmath233 and @xmath234 with @xmath248 . ( if @xmath227 contained at least three of such elements then by theorem [ exactlyn ] the interval @xmath246 $ ] would contain a copy of @xmath249 , but the interval contains only @xmath250 elements so this is impossible . ) since by example [ exdiam ] there is only one way of obtaining a diamond , there are t - incomparable @xmath251 and @xmath239 with @xmath252 such that @xmath253 from the two equations for @xmath223 it follows that @xmath254 . now there are two cases : * both @xmath255 s are t - above both @xmath256 s . but then we have @xmath257 a contradiction . ( the second to last inequality is strict since @xmath258 because @xmath251 and @xmath239 are incomparable . ) * the @xmath255 s are not both above @xmath233 and @xmath234 . hence either there is precisely one @xmath255 above each @xmath256 , or there are precisely two @xmath255 s above one @xmath256 . in both cases there is at least one @xmath255 t - incomparable to an @xmath256 , say that @xmath259[c]{$|_t$}}g_1 $ ] . now consider the element @xmath260 . clearly @xmath261 $ ] . but @xmath262 is w - incomparable to @xmath223 : @xmath263 because @xmath264 , and @xmath265 because @xmath266 . so again we have reached a contradiction , because @xmath246 $ ] does not contain an element incomparable to @xmath223 . let us recall some elementary lattice theory from grtzer @xcite . let @xmath185 be a distributive lattice . @xmath267 denotes the set of all _ nonzero _ join - irreducible elements of @xmath185 . @xmath267 is a poset under the ordering of @xmath185 . for @xmath268 define @xmath269 for a poset @xmath270 let @xmath271 be the collection of downwards closed subsets of @xmath270 , partially ordered by inclusion . then @xmath271 is a distributive lattice , and we have say that a lattice @xmath185 contains another lattice @xmath275 as a _ subinterval _ if there is an interval @xmath46 \subseteq l$ ] such that @xmath46 \cong l'$ ] . note that this is not the same as saying that @xmath275 is a sublattice of @xmath185 . for example , the free distributive lattice on three elements @xmath276 , depicted in figure [ fd3 ] , contains the double diamond of figure [ figccc ] as a sublattice , but not as a subinterval . we call a finite distributive lattice _ double diamond - like _ if it has at least two elements , it has no largest and smallest nonzero join - irreducible element , and @xmath277 is not the meet of two nonzero join - irreducible elements one of which is maximal . ( 110 , 70 ) ( -20,0 ) ( -30,30 ) ( -10,30 ) ( -20,20 ) ( -20,40 ) ( -20,20)(-1,1)10 ( -20,20)(1,1)10 ( -20,40)(-1,-1)10 ( -20,40)(1,-1)10 ( -10,10 ) ( -30,10 ) ( -20,0)(-1,1)10 ( -20,0)(1,1)10 ( -20,20)(-1,-1)10 ( -20,20)(1,-1)10 ( -20,0 ) ( -10,10 ) ( 0,40 ) ( -10,30 ) ( -10,50 ) ( -10,30)(1,1)10 ( -10,50)(-1,-1)10 ( -10,50)(1,-1)10 ( 0,20 ) ( -10,10)(1,1)10 ( -10,30)(1,-1)10 ( -10,10 ) ( 30,0 ) ( 30,0)(0,1)14.14 ( 20,24.14 ) ( 40,24.14 ) ( 30,14.14 ) ( 30,34.14 ) ( 30,14.14)(-1,1)10 ( 30,14.14)(1,1)10 ( 30,34.14)(-1,-1)10 ( 30,34.14)(1,-1)10 ( 50,5 ) ( 50,5)(0,1)14.14 ( 40,29.14 ) ( 60,29.14 ) ( 50,19.14 ) ( 50,39.14 ) ( 50,19.14)(-1,1)10 ( 50,19.14)(1,1)10 ( 50,39.14)(-1,-1)10 ( 50,39.14)(1,-1)10 ( 50,53.28 ) ( 50,39.14)(0,1)14.14 ( 60,34.14 ) ( 80,34.14 ) ( 70,24.14 ) ( 70,44.14 ) ( 70,24.14)(-1,1)10 ( 70,24.14)(1,1)10 ( 70,44.14)(-1,-1)10 ( 70,44.14)(1,-1)10 ( 70,58.28 ) ( 70,44.14)(0,1)14.14 ( 30,0)(4,1)20 ( 20,24.14)(4,1)40 ( 40,24.14)(4,1)40 ( 30,14.14)(4,1)40 ( 30,34.14)(4,1)40 ( 50,53.28)(4,1)20 ( 120,0 ) ( 110,30 ) ( 130,30 ) ( 120,20 ) ( 120,40 ) ( 120,20)(-1,1)10 ( 120,20)(1,1)10 ( 120,40)(-1,-1)10 ( 120,40)(1,-1)10 ( 130,10 ) ( 110,10 ) ( 120,0)(-1,1)10 ( 120,0)(1,1)10 ( 120,20)(-1,-1)10 ( 120,20)(1,-1)10 ( 120,0 ) ( 130,10 ) ( 140,40 ) ( 130,30 ) ( 130,50 ) ( 130,30)(1,1)10 ( 130,50)(-1,-1)10 ( 130,50)(1,-1)10 ( 140,20 ) ( 130,10)(1,1)10 ( 130,30)(1,-1)10 ( 130,10 ) ( 100,40 ) ( 120,40 ) ( 110,50 ) ( 110,30)(-1,1)10 ( 110,50)(-1,-1)10 ( 110,50)(1,-1)10 ( 120,20 ) ( 100,20 ) ( 110,10)(-1,1)10 ( 110,30)(-1,-1)10 ( 110,10 ) ( 120,60 ) ( 120,60)(-1,-1)10 ( 120,60)(1,-1)10 ( 120,20 ) suppose that @xmath267 is an initial segment of an upper semilattice and let @xmath46 $ ] be an interval in @xmath185 . we prove that @xmath46 $ ] is not double diamond - like . to this end , suppose that @xmath278 and that @xmath279)$ ] has no largest and no smallest element . we have to prove that then the 0 of @xmath46 $ ] , which is @xmath51 , is a meet of two elements of @xmath279)$ ] , one of which is maximal . by the assumption @xmath278 we have @xmath279)\neq\emptyset$ ] , and therefore there are at least two maximal nonzero join - irreducible elements @xmath280 and @xmath281 in @xmath46 $ ] and at least two minimal ones , @xmath282 and @xmath283 say . suppose that both @xmath284 s are above both @xmath53 s , so that @xmath279)$ ] contains the configuration of figure [ jl ] . we can not immediately conclude from this that @xmath267 contains the same configuration , for @xmath279)$ ] and @xmath267 can even be disjoint . nevertheless , suppose that @xmath280 is join - reducible in @xmath185 as @xmath285 , with @xmath286 . by lemma [ hulp ] we can choose @xmath287 and @xmath288 such that @xmath289 and @xmath290 . then in @xmath185 the set @xmath291 , is partially ordered as in figure [ jl ] . continuing in this way we can reduce the configuration until the top element @xmath280 has become join - irreducible , and of course we can reduce @xmath281 in the same way . then @xmath185 contains the configuration of figure [ jl ] with both top elements in @xmath267 . but the bottom elements always bound nonzero join - reducible elements , so we see that @xmath185 contains figure [ jl ] with all four elements in @xmath267 . but this contradicts that @xmath267 is an initial of an upper semilattice and hence that the bottom two elements should have a least upper bound in @xmath267 . we conclude that in the original configuration @xmath292 in @xmath279)$ ] it is not possible that both @xmath284 s are above both @xmath53 s , so there is at least one @xmath53 that is not below a @xmath284 , say @xmath293 . but then for the join - irreducible elements @xmath282 and @xmath281 in @xmath46 $ ] we have @xmath294 : if @xmath295 then @xmath282 would bound some join - irreducible element @xmath296 , contradicting that @xmath282 is minimal in @xmath279)$ ] . this shows that @xmath46 $ ] is not double diamond - like . conversely , if @xmath185 is a lattice such that @xmath267 is not an initial segment of any upper semilattice then it must contain two incomparable elements @xmath282 and @xmath283 without a least upper bound . if @xmath282 and @xmath283 would have no or only one upper bound in @xmath267 then @xmath267 could be consistently extended to an upper semilattice , so there must be at least _ two _ incomparable minimal upper bounds @xmath280 and @xmath281 for both @xmath282 and @xmath283 . but then the poset @xmath292 is isomorphic to the configuration in figure [ jl ] ( with possibly other elements in between the @xmath53 s and @xmath284 s ) . denote this subposet of @xmath267 by @xmath270 . now it is not hard to check that @xmath185 has a double diamond - like subinterval . namely by theorem [ isom ] we have @xmath297 . consider the interval @xmath298 $ ] in @xmath273 defined by @xmath299 then clearly @xmath298 $ ] is double diamond - like , because @xmath300 and @xmath298 = h(p)$ ] and so @xmath301)$ ] @xmath302 @xmath303 . suppose that @xmath285 in @xmath185 , with @xmath286 . note that @xmath287 and @xmath288 can not be both in @xmath46 $ ] . suppose that @xmath305 \big).\ ] ] consider @xmath287 and @xmath306 . if @xmath307 then this contradicts ( [ star ] ) ( because both @xmath308 they must be incomparable in this case ) . if @xmath309 then by @xmath310 we again contradict ( [ star ] ) . hence ( [ star ] ) is false , and if @xmath285 with @xmath286 in @xmath185 we can always choose @xmath311 $ ] and @xmath312 $ ] . in this case @xmath313 , for if @xmath314 then by @xmath315 we would have @xmath280 join - reducible in @xmath46 $ ] , contradiction . hence for every @xmath316 $ ] it holds that @xmath317 , and in particular @xmath318 now we also have @xmath319 because otherwise @xmath320 , contradiction . [ exmain ] before giving the general result of how to obtain lattices as intervals of @xmath2 we give a specific example to illustrate the method . figure [ procedure ] depicts the procedure to obtain a given lattice @xmath185 as an interval of @xmath2 . ( 100,120 ) ( -20,116)@xmath185 ( 0,80 ) ( -10,110 ) ( -10,110 ) ( 10,110 ) ( 0,100 ) ( 20,100 ) ( 20,100 ) ( 0,120 ) ( 0,100)(-1,1)10 ( 0,100)(1,1)10 ( 20,100)(-1,1)10 ( 20,100)(-1,-1)10 ( 0,120)(-1,-1)10 ( 0,120)(1,-1)10 ( 10,90 ) ( 10,90 ) ( -10,90 ) ( -10,90 ) ( 0,80)(-1,1)10 ( 0,80)(1,1)10 ( 0,100)(-1,-1)10 ( 0,100)(1,-1)10 the top left side of the picture shows an example of a finite distributive lattice , with its nonzero join - irreducible elements circled . the partial order @xmath267 is depicted on the top right . now for the lattice @xmath185 in this particular example we can map the poset @xmath267 to an isomorphic configuration @xmath326 in @xmath330 . ( the picture remains the same , so we drew it only once . ) this means that the only relations are the ones indicated in the picture , @xmath251 covers @xmath233 and @xmath234 , and @xmath239 covers @xmath234 . next we can form the distributive lattice @xmath331 , which is isomorphic to @xmath185 by theorem [ isom ] . finally we apply the mapping @xmath332 defined as follows . first define @xmath333[c]{$|_t$}}g_0 \big\ } } \cup { \big\ { h { \geqslant}_t f_1 : h { \makebox[4mm][c]{$|_t$}}g_0 \wedge h { \makebox[4mm][c]{$|_t$}}g_1 \big\}}.\ ] ] this has the effect that modulo @xmath227 we have @xmath334 and @xmath335 . for every @xmath336 define @xmath337 finally define @xmath338[c]{$|_t$}}\widehat{a } \big\}.\ ] ] here @xmath339[c]{$|_t$}}\widehat{a}$ ] denotes that @xmath339[c]{$|_t$}}g$ ] for every @xmath340 . we thus obtain the lattice @xmath329 on the bottom left of the picture , with @xmath341 using examples [ exdiam ] and [ exdouble ] one can check that @xmath342 is an isomorphism between @xmath274 and @xmath329 , so that the interval @xmath343 = \big [ f(\emptyset ) , f(i(j(l ) ) ) \big]$ ] is indeed isomorphic to @xmath185 . we follow the procedure depicted in figure [ procedure ] . let @xmath185 be as in the hypothesis of the theorem . since every finite upper semilattice with a least element is isomorphic to an initial segment of the turing degrees @xmath330 ( cf . lerman ( * ? ? ? * p156 ) ) we have a finite poset @xmath326 in @xmath330 that is isomorphic to @xmath267 . this means that if that if @xmath21 covers @xmath20 in @xmath267 then the image of @xmath21 is a minimal cover of @xmath20 in @xmath330 . furthermore , the minimal elements of @xmath326 can be chosen to be of minimal t - degree ( so that in particular they are all noncomputable ) . next we form the distributive lattice @xmath331 , which is isomorphic to @xmath185 by theorem [ isom ] . finally we define the mapping @xmath332 as follows . for a given @xmath344 let @xmath345 be all elements of @xmath326 covering @xmath20 . define @xmath346[c]{$|_t$}}g_0 \wedge \ldots \wedge h { \makebox[4mm][c]{$|_t$}}g_m \big\ } } , \\ { \mathcal{x}}&= & \displaystyle \bigcup_{f\in i(j(l ) ) } { \mathcal{x}}_f . \end{array}\ ] ] notice that if @xmath20 is maximal in @xmath326 then there are no elements of @xmath326 covering @xmath20 , hence @xmath347 . next , for every @xmath336 define @xmath348 @xmath349[c]{$|_t$}}\widehat{a } \big\}.\ ] ] here @xmath339[c]{$|_t$}}\widehat{a}$ ] denotes that @xmath339[c]{$|_t$}}g$ ] for every @xmath340 . by definition , @xmath339[c]{$|_t$}}\emptyset$ ] holds for every @xmath20 , so we have that @xmath350 we thus obtain the lattice @xmath329 . note that @xmath274 has @xmath351 as least element and @xmath326 as largest element we prove that @xmath342 is an isomorphism from @xmath274 to the interval @xmath352\subseteq { { \mathfrak m}}_w$ ] . since @xmath274 is isomorphic to @xmath185 this suffices to prove the theorem . _ @xmath342 is injective . _ suppose that @xmath353 . note that since @xmath354 and @xmath355 are downwards closed , @xmath356 if and only if @xmath357 . so it suffices to show that if @xmath358 then @xmath359 . suppose that @xmath360 . since @xmath344 we have @xmath361 . since @xmath362 we also have @xmath363 because @xmath364 is an antichain . if @xmath339[c]{$|_t$}}{\widehat{b}}$ ] then @xmath365 so in this case @xmath359 . if @xmath366[c]{$\not|_t$}}{\widehat{b}}$ ] then there is @xmath367 with either @xmath368 or @xmath369 . in the first case we have @xmath370 ( because @xmath21 is in @xmath371 and @xmath20 is not ) , hence @xmath372 , and again we can conclude that @xmath359 . in the second case , since @xmath373 we have @xmath374 , but @xmath375 because @xmath367 and @xmath371 is an antichain , so again @xmath359 . _ @xmath342 is monotone . _ we claim that @xmath376 implies that @xmath377 . suppose that @xmath376 and that @xmath378 . we prove that @xmath379 . we have the following three cases , corresponding to the three components of @xmath380 : * if @xmath381 then we are immediately done . * if @xmath382 , @xmath383[c]{$|_t$}}{\widehat{b}}$ ] then we have one of the following three options : * * @xmath384[c]{$|_t$}}{\widehat{a}}$ ] . in this case we are done immediately . * * @xmath385 . in this case we can not have @xmath386 because @xmath376 and @xmath383[c]{$|_t$}}{\widehat{b}}$ ] , so we have @xmath387 and hence @xmath379 . * * @xmath388 . this case can not occur because @xmath354 is downwards closed , hence @xmath142 would be in @xmath354 , hence in @xmath355 , contradicting @xmath383[c]{$|_t$}}{\widehat{b}}$ ] . * @xmath389 for some @xmath390 . when @xmath391 we are done . if @xmath392 then since @xmath376 and @xmath20 is maximal in @xmath355 we have @xmath393 , so either @xmath339[c]{$|_t$}}{\widehat{a}}$ ] , in which case we are done or @xmath394 , in which case @xmath395 since @xmath392 , and again we are done . @xmath396 : by monotonicity of @xmath342 we have @xmath397 , hence also @xmath398 . for the other direction @xmath399 , suppose that @xmath400 . we consider the three cases corresponding to the three components of @xmath401 . * if @xmath381 then we immediately have that @xmath402 . * suppose that @xmath400 because @xmath403 for some @xmath404 . if @xmath391 or @xmath390 then @xmath405 or @xmath406 , hence we are done . if @xmath407[c]{$|_t$}}{\widehat{a } } \}$ ] or @xmath407[c]{$|_t$}}{\widehat{b } } \}$ ] then we are also done . otherwise , in particular both @xmath408[c]{$\not|_t$}}{\widehat{a}}$ ] and @xmath408[c]{$\not|_t$}}{\widehat{b}}$ ] , say that @xmath409 and @xmath410 are such that @xmath408[c]{$\not|_t$}}g_0 $ ] and @xmath408[c]{$\not|_t$}}g_1 $ ] . it is impossible that @xmath411 because then ( because @xmath354 , @xmath355 downwards closed ) @xmath412 , contradicting @xmath404 . so at least one of @xmath413 and @xmath414 must hold . but in the first case we have @xmath379 and in the second @xmath415 . * finally suppose that @xmath384[c]{$|_t$}}{\widehat{a\cap b}}$ ] . when @xmath384[c]{$|_t$}}{\widehat{a}}$ ] or @xmath384[c]{$|_t$}}{\widehat{b}}$ ] then we are done , so suppose that neither of these hold , say @xmath408[c]{$\not|_t$}}f \in { \widehat{a}}$ ] and @xmath408[c]{$\not|_t$}}g \in { \widehat{b}}$ ] . when either @xmath20 or @xmath21 is in @xmath416 then @xmath408[c]{$\not|_t$}}{\widehat{a \cap b}}$ ] contrary to assumption , so we have that @xmath20 , @xmath417 . when @xmath418 then @xmath419 , and because @xmath417 there is then @xmath420 with @xmath421 , contradicting @xmath422 . likewise , @xmath423 is impossible , so we have @xmath339[c]{$|_t$}}g$ ] . hence either @xmath424 or @xmath425 . in the latter case @xmath426 , contradicting @xmath384[c]{$|_t$}}{\widehat{a\cap b}}$ ] , and in the former case we have @xmath402 . @xmath428 : by monotonicity of @xmath342 we have @xmath429 , hence also @xmath430 . for the other direction , suppose that @xmath431 . we prove that @xmath432 . if @xmath433 we are immediately done , so assume that @xmath434 . we have to prove that either @xmath435[r]{\makebox[10mm][l ] { or } } \exists f { \makebox[4mm][c]{$|_t$}}{\widehat{a\cup b } } \;\ ; h { \geqslant}_t f. \label{done2 } \end{aligned}\ ] ] we have the following cases , corresponding to the four remaining ways in which @xmath142 can be above the components of @xmath427 and @xmath380 that are different from @xmath227 : * @xmath436[c]{$|_t$}}{\widehat{a}}\ } , \{f \in i(j(l ) ) : f { \makebox[4mm][c]{$|_t$}}{\widehat{b}}\}$ ] . suppose that @xmath437 , @xmath259[c]{$|_t$}}{\widehat{a}}$ ] and @xmath438[c]{$|_t$}}{\widehat{b}}$ ] . note that it is not possible that @xmath439 for some @xmath440 because then @xmath441 would be below an element of either @xmath364 or @xmath371 , contradicting that @xmath259[c]{$|_t$}}{\widehat{a}}$ ] and @xmath438[c]{$|_t$}}{\widehat{b}}$ ] . so we either have @xmath442 for some @xmath440 or @xmath443[c]{$|_t$}}{\widehat{a\cup b}}$ ] . in the first case we have @xmath421 and we are done by way of ( [ done1 ] ) . in the second case we are done by way of ( [ done2 ] ) . * @xmath436[c]{$|_t$}}{\widehat{a}}\}$ ] and @xmath444 for some @xmath445 . if there is such @xmath233 with @xmath446 then we are done by way of ( [ done1 ] ) , so assume without loss of generality that @xmath447 . if @xmath384[c]{$|_t$}}{\widehat{a\cup b}}$ ] then we are done by ( [ done2 ] ) , so assume that @xmath408[c]{$\not|_t$}}{\widehat{a\cup b}}$ ] , say @xmath448 , @xmath449[c]{$\not|_t$}}f$ ] . we have one of the following two cases : * * @xmath450 . in this case @xmath390 is impossible by @xmath445 , so we must have @xmath362 . but this contradicts @xmath436[c]{$|_t$}}{\widehat{a}}\}$ ] . * * @xmath389 . in this case we are done by way of ( [ done1 ] ) . * @xmath436[c]{$|_t$}}{\widehat{b}}\}$ ] and @xmath444 for some @xmath451 . this is completely symmetric to the previous case . * @xmath452 for some @xmath451 and @xmath453 . if either @xmath233 or @xmath234 is in @xmath454 then we are done by way of ( [ done1 ] ) . otherwise @xmath455 . when @xmath384[c]{$|_t$}}{\widehat{a\cup b}}$ ] we are done by ( [ done2 ] ) , so assume that @xmath408[c]{$\not|_t$}}{\widehat{a\cup b}}$ ] , say @xmath449[c]{$\not|_t$}}f$ ] with @xmath448 . because @xmath456 it is impossible that @xmath457 , for then either @xmath233 would not be in @xmath364 or @xmath234 would not be in @xmath371 . hence @xmath458 , and we are done by way of ( [ done1 ] ) . _ @xmath342 is surjective . _ it remains to check that if @xmath459 $ ] then there is @xmath336 such that @xmath460 . to this end , let @xmath354 be a maximal subset of @xmath326 such that @xmath461 . we claim that then also @xmath462 . namely we have @xmath463 because @xmath464 . as for the other components of @xmath427 , suppose that @xmath344 , @xmath339[c]{$|_t$}}\widehat{a}$ ] and suppose that @xmath99 contains no element of degree @xmath100 . then @xmath465 , for the elements of @xmath99 that are mapped to @xmath20 in the reduction @xmath461 are all @xmath466 . but @xmath465 contradicts the maximality of @xmath354 . it follows that @xmath467[c]{$|_t$}}\widehat{a } \big\}$ ] . we also have @xmath468 . namely suppose not , that is , suppose there is @xmath469 such that @xmath470 . such @xmath20 can not be maximal in @xmath326 because @xmath464 . hence the set @xmath471 of all elements of @xmath326 covering @xmath20 is nonempty . we have @xmath472 because @xmath473 and because of the minimality of the @xmath255 over @xmath20 . but then @xmath474 , contradicting the maximality of @xmath354 . we have thus proved that @xmath475 . this concludes the proof of the surjectivity of @xmath342 and of the theorem . we start by arguing as in theorem [ noccc ] . if @xmath267 is not an initial segment of a finite upper semilattice then then it must contain two incomparable elements @xmath282 and @xmath283 without a least upper bound . if @xmath282 and @xmath283 would have no or only one upper bound in @xmath267 then @xmath267 could be consistently extended to an upper semilattice , so there must be at least _ two _ incomparable minimal upper bounds @xmath280 and @xmath281 for both @xmath282 and @xmath283 . but then the poset @xmath292 is isomorphic to the configuration in figure [ jl ] . ( note that the intervals between the @xmath53 s and the @xmath284 s need not be empty though . ) consider the subinterval @xmath298 \subseteq h(j(l))$ ] , where @xmath299 suppose that @xmath476\subseteq { { \mathfrak m}}_w$ ] is isomorphic to @xmath298 $ ] from this assumption we will derive a contradiction . the lattice @xmath298 $ ] starts and ends with a diamond , namely @xmath298 $ ] has at the top the diamond with top @xmath477 and bottom @xmath478 and at the bottom the diamond with top @xmath479 and bottom @xmath480 . hence , because by example [ exdiam ] there is only one way to implement the diamond in @xmath2 , we can argue as before that there are @xmath481 with @xmath482 , @xmath483[c]{$|_t$}}f_1 $ ] , and @xmath484 , @xmath485[c]{$|_t$}}g_1 $ ] , such that @xmath486 we now have two cases : _ case 1 : @xmath487 . _ note that the interval @xmath488\subseteq [ x , y].\ ] ] under the assumed isomorphism of @xmath298 $ ] with @xmath476 $ ] must correspond to the interval @xmath489 \subseteq [ { \mathcal{x}},{\mathcal{y}}].\ ] ] because @xmath490 ( since @xmath484 ) we have by dyments theorem [ dymentmw ] that the subinterval @xmath491 is empty . hence we see that the element @xmath492 in the interval ( [ interval2 ] ) is join - irreducible in @xmath2 , and hence also in @xmath476 $ ] . we derive a contradiction by showing that the interval ( [ interval1 ] ) contains only elements that are join - reducible in @xmath298 $ ] . the elements of this interval are all supersets @xmath493 that because of the special form of @xmath292 contain no elements both above @xmath282 and @xmath283 . hence every such @xmath354 can always be split into a nonempty part @xmath494 of elements above @xmath282 and a nonempty part @xmath495 of elements above @xmath283 . in particular every element of ( [ interval1 ] ) is join - reducible in @xmath298 $ ] . _ if case 1 . does not obtain there must be at least one @xmath20 incomparable to a @xmath21 , say that @xmath259[c]{$|_t$}}g_1 $ ] . it is clear that @xmath298 $ ] is double diamond - like . we derive a contradiction by showing that @xmath476 $ ] is not double diamond - like . let @xmath496 consist of @xmath233 plus the set of all elements @xmath142 with @xmath497 that are minimal with respect to the property @xmath498 . ( note that there can be only finitely many @xmath142 in @xmath499 $ ] since otherwise @xmath476 $ ] would be infinite , in which case we would have reached a contradiction right away . ) then we have @xmath500 we prove that the two elements on the right hand side are join - irreducible in @xmath476 $ ] , and that one is maximal with this property . since @xmath227 is the 0 of the lattice @xmath476 $ ] this proves that this interval is not double diamond - like . _ @xmath501 is join - irreducible : _ when @xmath502 $ ] and @xmath503 then @xmath504 : we have @xmath505 , and if in this reduction all elements of @xmath5 that are mapped to @xmath233 are strictly above it then @xmath506 , contradicting the assumption . hence @xmath5 contains an element of @xmath507 , and because @xmath508 we then have @xmath504 . _ @xmath509 is a maximal join - irreducible element of @xmath476 $ ] : _ join - reducibility is seen with an argument similar to the previous one : when @xmath502 $ ] and @xmath510 then @xmath511 . namely , we clearly have @xmath512 . if @xmath5 would not have an element of degree @xmath513 then contrary to assumption we would have @xmath514 : by definition of @xmath496 and because @xmath515 , any element in @xmath5 that is not below @xmath239 can be mapped to either @xmath516 or @xmath496 . any element in @xmath5 that can be mapped to @xmath239 is actually strictly above @xmath239 , so can be mapped to @xmath517 by the identity . hence we also have @xmath518 . to see the maximality : suppose @xmath502 $ ] is such that @xmath519 . then @xmath5 is of the form @xmath520 , with @xmath521 . but then it is easy to check that @xmath522 using that @xmath523 . the two components on the right hand side are w - incomparable and in @xmath476 $ ] , so @xmath5 is join - reducible in @xmath476 $ ] . this concludes case 2 . summarizing , we have seen that @xmath273 contains a subinterval @xmath298 $ ] that can not be an interval in @xmath2 . thus @xmath273 , and hence by by theorem [ isom ] also @xmath185 , can not be an interval of @xmath2 . we can extend the definition of the mapping @xmath342 in the proof of theorem [ main ] as follows . define @xmath524 as before and let @xmath525[c]{$|_t$}}f \mbox { for all $ f$ minimal in } i(j(l ) ) \big\ } } , \\ { \mathcal{x}}&= & \displaystyle { \mathcal{x}}_0 \cup \bigcup_{f\in i(j(l ) ) } { \mathcal{x}}_f . \end{array}\ ] ] then for every @xmath526 define @xmath427 as before , using this new definition of @xmath227 . this addition does not change anything in the proof of theorem [ sufficient ] , but now we have that @xmath527 , as is easily checked , using that we chose the minimal elements of @xmath326 of minimal t - degree . thus we obtain that a finite distributive lattice has no double diamond - like subinterval if and only if it is isomorphic to an interval of the form @xmath528 $ ] in @xmath2 . from this the corollary follows immediately . andrea sorbi , _ the medvedev lattice of degrees of difficulty _ , in : s. b. cooper , t. a. slaman , and s. s. wainer ( eds . ) , computability , enumerability , unsolvability : directions in recursion theory , london mathematical society lecture notes 224 , cambridge university press , 1996 , 289312 .

the paper is devoted to penalty robin robin domain decomposition methods ( ddms ) , proposed by us for the solution of unilateral multibody contact problems of elasticity . these ddms are based on the penalty method for variational inequalities and some stationary and nonstationary iterative methods for nonlinear variational equations . the main result of the paper is that we give mathematical justification of proposed ddms and prove theorems about their convergence . we also provide numerical investigations of the efficiency of these methods using finite element approximations . * key words : * elasticity , multibody contact , variational inequalities , penalty method , iterative methods , domain decomposition * msc2010 : * 65n55 , 74s05

1208.6478

contact problems of elasticity are widely used in many fields of science and engineering , especially in machine science , structural mechanics , geology and biomechanics . the brief overview of existing numerical and analytical methods for the solution of contact problems can be found in @xcite . efficient approach to the solution of multibody contact problems is the use of domain decomposition methods ( ddms ) . ddms are well developed for the solution of linear boundary value problems , particularly for linear poisson and linear elasticity problems @xcite . the construction of ddms for unilateral contact problems , which are nonlinear , are much more complicated . among domain decomposition methods for unilateral two - body contact problems obtained on continuous level , one should mention signorini neumann @xcite , signorini dirichlet @xcite and signorini signorini @xcite iterative algorithms . all of these methods in each iteration require to solve a nonlinear one - sided contact problem with a rigid surface ( signorini problem ) for one of the bodies , and a linear elasticity problem with neumann @xcite or dirichlet @xcite boundary conditions on possible contact area for other body , or require to solve nonlinear signorini problems for both of the bodies @xcite . moreover , to increase the convergence rate of signorini dirichlet and signorini signorini algorithms , it is recommended to perform an additional iteration , in which we have to solve linear elasticity problems with neumann boundary conditions for both of the bodies @xcite . domain decomposition method presented in work @xcite for two - body unilateral contact problem , is also obtained on continuous level . it is based on the use of augmented lagrangian variational formulation and uzawa block relaxation method . this domain decomposition method in each iteration require to solve linear elasticity problems for both of the bodies . on contrary , ddms can be constructed on discrete level , after a discretization of corresponding continuous boundary - value problem . among discrete ddms for unilateral contact problems , one should mark out substructuring and feti methods @xcite . in works @xcite we have proposed on continuous level a class of penalty parallel robin robin type domain decomposition methods for the solution of unilateral multibody contact problems of elasticity . these methods are based on penalty method for variational inequalities and some stationary and nonstationary iterative methods for nonlinear variational equations . in each iteration of proposed ddms we have to solve in parallel some linear variational equations in subdomains , which correspond to elasticity problems with robin boundary conditions , prescribed on some subareas of possible contact zones . these ddms do not require the solution of nonlinear one - sided contact problems in each step . the main result of this paper is that we have proved theorems about convergence of proposed penalty robin robin domain decomposition methods . the paper is organized as follows . in section 2 the classical formulation of multibody contact problem in the form of the system of second order elliptic partial differential equations with inequality and equality constrains is given . in section 3 we consider the variational formulation of this problem in the form of convex minimization problem and in the form of elliptic variational inequality at closed convex set , and formulate theorem about a unique existence of the solution of this inequality . in section 4 we use the penalty method to reduce the variational inequality to an unconstrained minimization problem , which is equivalent to a nonlinear variational equation in the whole space . later , we prove a theorem about a unique existence of the solution of the penalty variational equation and a theorem about strong convergence of this solution to the solution of initial variational inequality . in section 5 we consider stationary and nonstationary iterative methods for the solution of abstract nonlinear variational equations in reflexive banach spaces . we prove theorems about convergence of these methods , and show that the convergence rate of stationary methods in some energy norm is linear . we also prove theorem about stability of stationary iterative methods to errors which may occur in each iteration . in section 6 we present parallel stationary and nonstationary penalty robin robin domain decomposition methods for the solution of nonlinear penalty variational equations of multibody unilateral contact problems . we prove theorem about convergence of these methods , and show that the convergence rate of stationary robin robin methods in some energy norm is linear . in section 7 we provide numerical investigations of proposed domain decomposition methods using finite element approximations . the penalty parameter and mesh refinement influence on the numerical solution , as well as the dependence of the convergence rate of domain decomposition methods on iterative parameters are investigated . in conclusion section we summarize all results presented in the paper . introduce the cartesian coordinate system @xmath0 with basis vectors @xmath1 and consider the problem of frictionless unilateral contact of @xmath2 elastic bodies @xmath3 with piecewise smooth boundaries @xmath4 , @xmath5 ( fig . 1 ) . denote @xmath6 . the stress - strain state in point @xmath7 of each solid @xmath8 is descried by the displacement vector @xmath9 , the symmetric tensor of strains @xmath10 , and the tensor of stresses @xmath11 . these quantities satisfy cauchy relations , hook s law and the equilibrium equations : @xmath12 @xmath13 @xmath14 where @xmath15 are the components of the volume forces vector @xmath16 . the elastic coefficients @xmath17 are measurable , symmetric , and bounded with constants @xmath18 in the following sense @xcite : @xmath19 suppose that the boundary @xmath20 of each solid consists of three parts : @xmath21 , @xmath22 , @xmath23 , such that @xmath24 , @xmath25 , @xmath26 , @xmath27 , @xmath28 . boundary @xmath29 is the possible contact area of body @xmath30 with other bodies , @xmath31 is the possible contact area of body @xmath8 with body @xmath32 , and @xmath33 is the set of the indices of all bodies which are in contact with body @xmath34 . on each boundary @xmath20 let us introduce the local orthonormal coordinate system @xmath35 , where @xmath36 is outer unit normal to @xmath20 . then the vectors of displacements and stresses on the boundary can be written in the following way : @xmath37 @xmath38 we assume that the surfaces @xmath39 and @xmath40 are sufficiently close @xcite . therefore @xmath41 , where @xmath42 is the projection of point @xmath43 on the surface @xmath44 . we denote by @xmath45 the distance in @xmath46 between bodies @xmath8 and @xmath32 before the deformation . the sign of @xmath47 depends on a statement of the problem . on the part @xmath21 the kinematical ( dirichlet ) boundary conditions are prescribed : @xmath48 and on the part @xmath49 we consider the static ( neumann ) boundary conditions @xmath50 where @xmath51 and @xmath52 are given boundary displacements and stresses . further , for the simplicity of variational formulations and proofs , we assume that all bodies are rigidly fixed on the surface @xmath21 , i.e. @xmath53 on the possible contact areas @xmath54 , @xmath55 , @xmath56 the following unilateral contact conditions hold : absence of extension @xmath57 absence of friction @xmath58 mutual nonpenetration of the bodies @xmath59 contact alternative @xmath60 where @xmath61 . the system of second order partial differential equations ( [ eq_pii_1 ] ) ( [ eq_pii_3 ] ) with boundary conditions ( [ eq_pii_5 ] ) ( [ eq_pii_11 ] ) is the mathematical formulation of frictionless unilateral multibody contact problem of elasticity . note that the contact problem ( [ eq_pii_1 ] ) ( [ eq_pii_3 ] ) , ( [ eq_pii_5 ] ) ( [ eq_pii_11 ] ) is nonlinear , since the real contact areas are unknown . let us give the weak formulation of the contact problem ( [ eq_pii_1 ] ) ( [ eq_pii_3 ] ) , ( [ eq_pii_5 ] ) ( [ eq_pii_11 ] ) in the form of variational equality and convex minimization problem . for each body @xmath8 , @xmath5 consider sobolev space @xmath62^{3 } $ ] with scalar product @xmath63 , + @xmath64 and norm @xmath65 . introduce the following closed subspace in @xmath66 : @xmath67 where @xmath68^{3 } $ ] is surjective , linear and continuous trace operator @xcite . space @xmath69 is a hilbert space with the same scalar product and norm as in @xmath66 . consider the space @xmath70 , which is the direct product of spaces @xmath69 : @xmath71 and define in it the scalar product and norm : @xmath72 , @xmath73 , @xmath74 . note , that hilbert space @xmath75 is closed reflexive banach space . now , let us introduce the closed convex set of all displacement vectors in @xmath70 which satisfy nonpenetration contact conditions ( [ eq_pii_10 ] ) : @xmath76 where @xmath77 is the set of all possible unordered pairs of the subscripts of bodies which are in contact with each other , and @xmath78 , @xmath79 , @xmath80 , @xmath5 . the quantities @xmath81 , @xmath5 in ( [ eq_pii_13 ] ) have to be understood in the following way @xmath82 where @xmath83^{3 } $ ] is surjective , linear and continuous trace operator on surface @xmath84 @xcite , and @xmath85^{3 } $ ] . note , that all equalities and inequalities in spaces @xmath86 , @xmath87 , @xmath88 and @xmath89 hold almost everywhere . since set @xmath90 is a closed convex subset of hilbert space @xmath70 @xcite , it is weakly closed @xcite . in space @xmath91 consider bilinear form @xmath92 , which represents the total deformation energy of the system of bodies : @xmath93 @xmath94 in space @xmath70 define linear form @xmath95 , which is equal to the external forces work : @xmath96 @xmath97 where @xmath98^{3 } $ ] , @xmath99^{3 } $ ] , @xmath5 . * lemma 1 . * _ if @xmath100 _ , @xmath5 _ are piecewise smooth , @xmath101 _ , @xmath27 , @xmath102^{3 } $ ] , @xmath103^{3},$ ] _ and condition ( [ eq_pii_4 ] ) holds , then the bilinear form @xmath104 is symmetric , continuous and coercive , and the linear form @xmath105 is continuous , i.e. _ @xmath106 @xmath107 @xmath108 @xmath109 * proof . * the symmetry of bilinear form @xmath110 follows from the symmetry of elastic constants in the hook s law ( [ eq_pii_2 ] ) , and its continuity and coercivity follows from the continuity and coercivity of bilinear forms @xmath111 for each body . the continuity of @xmath112 follows from the continuity of linear forms @xmath113 , which can be proved using trace theorems @xcite . the detailed proof of this lemma can be found in @xcite . @xmath114 according to @xcite , the original contact problem ( [ eq_pii_1 ] ) ( [ eq_pii_3 ] ) , ( [ eq_pii_5 ] ) ( [ eq_pii_11 ] ) has an alternative weak formulation as the convex minimization problem of the quadratic functional on the set @xmath90 : @xmath115 using the general theory of variational inequalities @xcite we have proved in @xcite the next theorem . * theorem 1 . * _ suppose that bilinear form @xmath104 and linear form @xmath105 satisfy properties ( [ eq_pii_18 ] ) ( [ eq_pii_21 ] ) , and set @xmath90 is convex closed subset of hilbert space @xmath70 . then the minimization problem ( [ eq_pii_22 ] ) has a unique solution @xmath116 , and this problem is equivalent to the following variational inequality : _ @xmath117 to obtain a minimization problem in original space @xmath70 , we apply a penalty method @xcite to convex minimization problem ( [ eq_pii_22 ] ) . for the violation of nonpenetration conditions we use the penalty in the following form @xcite : @xmath118^{2}\ , ds,\ ] ] where @xmath119 is a penalty parameter , @xmath120 . let us consider the following minimization problem with penalty in space @xmath70 : @xmath121 note , that the introduction of penalty corresponds to the introduction of conditional intermediate winkler layer between bodies with stiffness coefficient @xmath122 . the quantity @xmath123 has a sense of the normal contact stress between bodies @xmath8 and @xmath124 , and the penalty @xmath125 represents the total normal contact stresses work . now consider the properties of penalty term ( [ eq_pii_24 ] ) in more detail . functional @xmath125 is nonnegative @xmath126 and gateaux differentiable in @xmath70 : @xmath127 furthermore , the gateaux differential @xmath128 is linear in @xmath129 . * lemma 2 . * _ if surfaces @xmath31 _ , _ @xmath130 are piecewise smooth , and @xmath131 _ , _ then @xmath128 satisfy the following properties : _ @xmath132 @xmath133 @xmath134 * proof . * it is obvious to see , that functional @xmath128 is linear in @xmath129 . at first , let us show the satisfaction of property ( [ eq_pii_29 ] ) . let us write @xmath128 in the extended form : @xmath135 , where @xmath136 and @xmath137 . taking into account the following inequality for real numbers : @xmath138 and schwarz inequality , we obtain @xmath139^{m } } ^{2},\ ] ] where @xmath140^{3 } } ^{2 } + \varepsilon _ { \alpha \beta } $ ] , @xmath141 , @xmath142 , @xmath55 . from trace theorems in @xcite , it follows , that @xmath143^{3 } } ^{2 } \le t_{\alpha \beta } ^{2 } \left\| { \bf v}_{\alpha } \right\| _ { v_{\alpha } } ^{2 } \right\}.\ ] ] substituting ( [ eq_pii_35 ] ) into ( [ eq_pii_34 ] ) , we come to inequality @xmath144 , where @xmath145 , @xmath142 , @xmath5 . similarly to this , we obtain the same inequality for the second term of relationship ( [ eq_pii_32 ] ) . hence @xmath146 as a result , we find @xmath147 where @xmath148 . inequality ( [ eq_pii_29 ] ) is proved . now , we prove that condition ( [ eq_pii_30 ] ) holds . for this we use the next inequality @xmath149z\le 0\right\},\ ] ] rewrite @xmath150 in the following way : @xmath151 where @xmath152 , @xmath153 in view of property ( [ eq_pii_36 ] ) , we obtain @xmath154 therefore , since @xmath119 , we come to inequality @xmath155 finally , let us prove the satisfaction of property ( [ eq_pii_31 ] ) . using the next inequality for real numbers @xmath156 we obtain @xmath157 taking into account , that the components of outer unit normal to @xmath158 satisfy property @xmath159 , and using inequalities ( [ eq_pii_33 ] ) and ( [ eq_pii_39 ] ) , we find @xmath160^{3 } } ^{2 } + \left\| { \rm{tr}}_{\beta } ^{0 } ( \ , { \bf w}_{\beta } ) \right\| _ { [ l_{2 } ( s_{\alpha \beta } ) ] ^{3 } } ^{2}\right).\ ] ] now , let us write @xmath161 as follows : @xmath162 consider the term @xmath163 in more detail . let us use inequalities ( [ eq_pii_33 ] ) , ( [ eq_pii_40 ] ) , and schwarz inequality : @xmath164 @xmath165^{3 } } ^{2 } + \left\| { \rm{tr}}_{\beta } ^{0 } ( \ , { \bf w}_{\beta } ) \right\| _ { [ l_{2 } ( s_{\alpha \beta } ) ] ^{3 } } ^{2 } \right)\ , \ , \left\| { \rm{tr}}_{\alpha } ^{0 } ( \ , { \bf v}_{\alpha } ) \right\| _ { [ l_{2 } ( s_{\alpha \beta } ) ] ^{3 } } ^{2 } .\ ] ] using inequality ( [ eq_pii_35 ] ) , we find further @xmath166 where @xmath167 , @xmath168 . in much the same way we come to inequality @xmath169 where @xmath170 . taking into account the last two inequalities , we establish @xmath171 @xmath172 where @xmath173 . as a result , we obtain @xmath174 @xmath175 @xmath176 where @xmath177 , @xmath178 . @xmath114 + * theorem 2 . * _ suppose that @xmath70 is closed reflexive banach space , @xmath179 is gateaux differentiable , and conditions ( [ eq_pii_18 ] ) ( [ eq_pii_21 ] ) , ( [ eq_pii_26 ] ) , ( [ eq_pii_29 ] ) ( [ eq_pii_31 ] ) hold . then there exist a unique solution of nonquadratic minimization problem ( [ eq_pii_25 ] ) in @xmath180 and this problem is equivalent to the following nonlinear variational equation : _ @xmath181 * proof . * as shown in @xcite , due to the properties ( [ eq_pii_18 ] ) ( [ eq_pii_21 ] ) , the functional @xmath182 is strictly convex , and coercive ( @xmath183 ) , and the differential @xmath184 is linear and continuous in @xmath129 . from the property ( [ eq_pii_30 ] ) , it follows that the penalty term @xmath185 is convex in @xmath70 @xcite . now consider the properties of functional @xmath186 this functional is gateaux differentiable in @xmath70 : @xmath187 and strictly convex , as the sum of convex functional @xmath125 and strictly convex functional @xmath182 . in addition , since functionals @xmath188 and @xmath128 are linear and continuous in @xmath129 , it follows that @xmath189 is also linear and continuous in @xmath190 . hence , according to @xcite , functional @xmath191 is weakly lower semicontinuous . due to the coercivity of @xmath182 and property ( 26 ) , we obtain that @xmath192 . since @xmath70 is closed reflexive banach space , @xmath191 is weakly lower semicontinuous and coercive , then according to the theorem of existence of the minimum in reflexive banach space @xcite , there exists a solution of minimization problem ( [ eq_pii_25 ] ) in the space @xmath70 . furthermore , the functional @xmath191 is strictly convex and gateaux differentiable in @xmath70 . therefore , according to the theorem about necessary and sufficient conditions of the minimum in reflexive banach space @xcite , the solution of problem ( [ eq_pii_25 ] ) is unique , and this problem is equivalent to the variational equation ( [ eq_pii_41 ] ) . @xmath114 now let us prove that the solution of penalty variational equation ( [ eq_pii_41 ] ) converges strongly to the solution of original variational inequality ( [ eq_pii_23 ] ) as @xmath193 . let us rewrite variational equation ( [ eq_pii_41 ] ) in the following equivalent form @xmath194 where @xmath195 is the action of a functional @xmath196 on an element @xmath197 , @xmath198 is the space dual to @xmath70 , @xmath199 is gateaux derivative of functional @xmath200 , and @xmath201 . we have proved the next lemma . * lemma 3 . * _ suppose that properties ( [ eq_pii_30 ] ) and ( [ eq_pii_31 ] ) hold . then operator @xmath202 in problem ( [ eq_pii_42 ] ) is a penalty operator for the kinematically allowable displacements set @xmath90 , i.e. _ 1 ) . _ @xmath203 is monotone in @xmath70 : _ @xmath204 2 ) . _ @xmath203 satisfies the lipschitz condition in @xmath70 : _ @xmath205 3 ) . _ the kernel of operator @xmath203 is equal to set @xmath90 : _ @xmath206 * proof . * the monotonity of operator @xmath203 follows from the condition ( [ eq_pii_30 ] ) , and the satisfaction of lipschitz condition follows from the property ( [ eq_pii_31 ] ) . if @xmath116 , then @xmath207 , and , on the contrary , if @xmath208 , we have @xmath116 . hence , @xmath209 . for more details see @xcite . @xmath114 now , using the results of works @xcite , let us prove the proposition about the convergence of penalty method , applied to the variational inequality ( [ eq_pii_23 ] ) . * theorem 3 . * _ suppose that @xmath210 is a convex closed subset of hilbert space @xmath70 _ , _ bilinear form @xmath110 _ , @xmath211 _ and linear form @xmath95 _ , @xmath212 _ satisfy properties ( [ eq_pii_18 ] ) ( [ eq_pii_21 ] ) , @xmath202 is penalty operator for set @xmath90 _ , @xmath213 _ is a unique solution of variational inequality ( [ eq_pii_23 ] ) , and @xmath214 is a unique solution of penalty variational equation ( [ eq_pii_42 ] ) with penalty parameter @xmath215 . then @xmath216 strongly in @xmath70 _ , _ i.e. @xmath217_. * proof . * in works @xcite it is proved that , if conditions ( [ eq_pii_19 ] ) ( [ eq_pii_21 ] ) hold , @xmath218 is a penalty operator for set @xmath90 and there exist solutions of problems ( [ eq_pii_23 ] ) and ( [ eq_pii_42 ] ) , then the sequence @xmath219 is bounded : @xmath220 and there exists such subsequence @xmath221 , which converges weakly in @xmath70 to some solution of variational inequality ( [ eq_pii_23 ] ) , i.e. @xmath222 moreover , in @xcite it is shown that any weakly convergent subsequence of sequence @xmath223 converges weakly in @xmath70 to some solution of variational problem ( [ eq_pii_23 ] ) . now let us assume that variational problems ( [ eq_pii_23 ] ) and ( [ eq_pii_42 ] ) have unique solutions . then , as follows from above , the sequence @xmath224 has a unique partial weak limit @xmath213 . since the sequence @xmath219 has a unique weak limit point , and is bounded , then according to theorem in @xcite , it is weakly convergent to this point , i.e. @xmath225 where @xmath213 is the unique solution of variational inequality ( [ eq_pii_23 ] ) . further , let us show that @xmath219 converges strongly to @xmath213 as @xmath193 . due to ( [ eq_pii_43 ] ) , we get @xmath226 where @xmath227 is the gateaux derivative of functional @xmath228 , @xmath229 . since @xmath230 is a solution of penalty variational equation ( [ eq_pii_42 ] ) , it is obvious that @xmath231 taking into account the monotonity of penalty operator @xmath232 and the property @xmath233 , we obtain @xmath234 in view of this property and the nonnegativity of bilinear form @xmath104 , we get the inequality @xmath235 . hence @xmath236 passing to the limit in expression ( [ eq_pii_45 ] ) as @xmath193 , and taking into account property ( [ eq_pii_44 ] ) , we obtain @xmath237 further , in view of coercivity of bilinear form @xmath104 , it follows that @xmath238 as a result , we establish that @xmath239 . @xmath114 thus , using the penalty method we have reduced the solution of original variational inequality ( [ eq_pii_23 ] ) to the solution of the nonlinear variational equation in the whole space @xmath70 , which depends on the penalty parameter @xmath119 . we also have proved the existence of the unique solution of penalty variational equation ( [ eq_pii_42 ] ) and its strong convergence to the solution of initial variational inequality ( [ eq_pii_23 ] ) as penalty parameter @xmath240 tends to zero . consider an abstract nonlinear in @xmath241 variational equation in form ( [ eq_pii_41 ] ) , where @xmath70 is closed reflexive banach space , @xmath110 is bilinear form , set in @xmath242 , @xmath243 is linear functional , and the term @xmath128 is linear in @xmath129 and nonlinear in @xmath241 . suppose that conditions ( [ eq_pii_18 ] ) ( [ eq_pii_21 ] ) , ( [ eq_pii_29 ] ) ( [ eq_pii_31 ] ) are satisfied . hence , there exists a unique solution of the problem ( [ eq_pii_41 ] ) . for the numerical solution of the nonlinear variational equation ( [ eq_pii_41 ] ) let us use the following iterative method @xcite : @xmath244 , \,\ , k=0,1, ... \,,\ ] ] where @xmath245 is some bilinear form assigned in @xmath91 , @xmath246 , @xmath247 is the @xmath248-th approximation to the exact solution @xmath249 of the problem ( [ eq_pii_41 ] ) , @xmath250 is an initial guess , and @xmath251 is an iterative parameter . we have proved the following proposition about the convergence of the iterative method ( [ eq_pii_46 ] ) . * theorem 4 . * _ suppose that bilinear form @xmath245 is symmetric , continuous and coercive : _ @xmath252 @xmath253 @xmath254 _ properties ( [ eq_pii_19 ] ) ( [ eq_pii_21 ] ) , ( [ eq_pii_29 ] ) ( [ eq_pii_31 ] ) are satisfied , and the iteration parameter lies in the interval @xmath255 _ , @xmath256 , @xmath257 . _ _ _ then the sequence @xmath258 _ , _ obtained by iterative method ( [ eq_pii_46 ] ) converges strongly in @xmath70 to the exact solution @xmath249 of the variational equation ( [ eq_pii_41 ] ) , + i.e. @xmath259 _ , _ and the convergence rate in the energy norm + @xmath260 is linear : _ @xmath261 _ moreover , the maximal convergence rate reaches as @xmath262_. _ _ * proof . * since bilinear form @xmath245 is symmetric , continuous and coercive , we may introduce a scalar product and norm @xmath263 from properties ( [ eq_pii_48 ] ) and ( [ eq_pii_49 ] ) , it follows that norms @xmath264 and @xmath265 are equivalent in space @xmath70 . in each step @xmath266 of method ( [ eq_pii_46 ] ) , we have to solve the linear variational problem : @xmath267 where @xmath268 $ ] is linear in @xmath269 , @xmath270 . using properties ( [ eq_pii_19 ] ) , ( [ eq_pii_21 ] ) , ( [ eq_pii_29 ] ) and ( [ eq_pii_48 ] ) , we obtain that @xmath271 is continuous : @xmath272 where @xmath273 , @xmath274 . since conditions ( [ eq_pii_47 ] ) ( [ eq_pii_49 ] ) and ( [ eq_pii_52 ] ) are satisfied , we see that the problem ( [ eq_pii_51 ] ) has a unique solution @xmath275 . now let us show , that the sequence of solutions of problems ( [ eq_pii_51 ] ) converges strongly to the solution of initial variational equation ( [ eq_pii_41 ] ) . suppose that @xmath276 is the exact solution of problem ( [ eq_pii_41 ] ) . introduce a notation @xmath277 , @xmath278 , and rewrite ( [ eq_pii_46 ] ) as follows : @xmath279.\ ] ] subtracting from this expression the identity @xmath280 $ ] , we obtain @xmath281.\ ] ] let us define a functional @xmath282 , @xmath283 , which is linear in @xmath269 . due to properties ( [ eq_pii_30 ] ) and ( [ eq_pii_31 ] ) , the following conditions hold : @xmath284 @xmath285 let us rewrite expression ( [ eq_pii_53 ] ) in the form @xmath286 , \,\ , \forall \ , { \bf v}\in v_{0 } .\ ] ] if we take @xmath287 in ( [ eq_pii_56 ] ) , we shall have @xmath288 taking into account the continuity of bilinear form ( [ eq_pii_19 ] ) , and property ( [ eq_pii_55 ] ) , we get @xmath289 further , in view of relation between norms @xmath290 we come to inequalities @xmath291 now let us take @xmath292 in expression ( [ eq_pii_56 ] ) . then @xmath293 $ ] . this relation can be written as @xmath294 @xmath295.\ ] ] in view of properties ( [ eq_pii_19 ] ) and ( [ eq_pii_55 ] ) , we come to inequality @xmath296 suppose that @xmath297 . then , taking into account the coercitivity of bilinear form ( [ eq_pii_20 ] ) , property ( [ eq_pii_54 ] ) , and the previous inequality , we obtain @xmath298.\ ] ] in view of inequalities ( [ eq_pii_57 ] ) and ( [ eq_pii_58 ] ) , we find further @xmath299 if the following inequality @xmath300 holds , then the sequence @xmath301 will be monotonically nonincreasing : @xmath302 , and , hence , @xmath303 , where @xmath304 . passing to the limit as @xmath305 in expression ( [ eq_pii_59 ] ) , we obtain @xmath306 , i.e. @xmath307 , and , therefore @xmath308 . from inequality ( [ eq_pii_60 ] ) we establish the interval of allowable values of iteration parameter @xmath309 : @xmath310 since the norms @xmath264 and @xmath311 are equivalent , we have @xmath312 , and , hence , @xmath313 . using inequality ( [ eq_pii_59 ] ) , we find estimate @xmath314 it is not hard to show , that @xmath315 for @xmath316 . we obtain this from the next relations : @xmath317 as follows from estimate ( [ eq_pii_61 ] ) , the convergence rate is maximal if the parameter @xmath318 is minimal , i.e. , if @xmath319 . @xmath114 * remark 1 . * _ suppose that term @xmath128 is gateaux differentiable in @xmath241 . then conditions ( [ eq_pii_29 ] ) , ( [ eq_pii_30 ] ) in _ * _ theorem 4 _ * _ can be replaced by the following properties _ @xmath320 @xmath321 * proof . * let us apply to @xmath322 the lagrange formula of finite increments @xcite : @xmath323 then the satisfaction of conditions ( [ eq_pii_62 ] ) and ( [ eq_pii_63 ] ) yields the satisfaction of properties ( [ eq_pii_30 ] ) and ( [ eq_pii_31 ] ) . @xmath114 now let us investigate the stability of iterative method ( [ eq_pii_46 ] ) to errors . suppose that conditions of * theorem 4 * are satisfied . then for any @xmath324 , @xmath278 there exists a unique solution @xmath325 of problem ( [ eq_pii_51 ] ) . therefore , there exists an operator @xmath326 , which maps every element @xmath327 to the solution @xmath328 of the problem ( [ eq_pii_51 ] ) , and the iterative method ( [ eq_pii_46 ] ) can be written in the following form @xmath329 now assume that in each step @xmath248 of iterative method ( [ eq_pii_64 ] ) we get some computational errors . then this iterative method will take the form : @xmath330 @xmath331 where @xmath332 , @xmath278 is an approximate solution of problem ( [ eq_pii_51 ] ) , @xmath333 , @xmath278 is a computational error , which occurs in each step @xmath248 , and @xmath334 is an initial guess error . * corollary 1 . * _ suppose that conditions of _ * _ theorem 4 _ * _ are satisfied , and errors which occur in each step @xmath248 of iterative method ( [ eq_pii_46 ] ) are uniformly bounded , i.e. _ @xmath335 _ then the following estimates hold : _ @xmath336 @xmath337 @xmath338 @xmath339 _ where @xmath276 is the exact solution of problem ( [ eq_pii_41 ] ) . _ the proof of this proposition follows from property ( [ eq_pii_50 ] ) . thus , the errors , which occur in each step of iterative method ( [ eq_pii_46 ] ) , do not accumulate . now consider the nonstationary iterative method for the solution of nonlinear variational equation ( [ eq_pii_41 ] ) , where bilinear forms @xmath340 are different in each iteration @xcite . in space @xmath91 introduce a sequence of bilinear forms @xmath341 , @xmath278 , which satisfy the property @xmath342 for the solution of nonlinear variational equation ( [ eq_pii_41 ] ) , we have proposed the following nonstationary iterative method @xcite : @xmath343,\,\ , k=0,1, ... \,,\ ] ] where @xmath270 is the @xmath248-th approximation to the exact solution of problem ( [ eq_pii_41 ] ) , and @xmath251 is an iterative parameter . we also have proved the next proposition about the convergence of this method . * theorem 5 . * _ suppose that conditions ( [ eq_pii_19 ] ) ( [ eq_pii_21 ] ) , ( [ eq_pii_29 ] ) ( [ eq_pii_31 ] ) hold , bilinear forms @xmath344 satisfy properties _ @xmath345 @xmath346 @xmath347 @xmath348 _ and iterative parameter @xmath309 lies in the interval @xmath349 _ , @xmath350 . _ _ _ then the sequence @xmath258 obtained by the nonstationary iterative method ( [ eq_pii_71 ] ) converges strongly in @xmath70 to the exact solution @xmath276 of the variational equation ( [ eq_pii_41 ] ) , i.e. @xmath351_. _ _ the proof of this theorem is similar to the proof of * theorem 4*. we omit the details . note , that in iterative methods ( [ eq_pii_46 ] ) and ( [ eq_pii_71 ] ) we can also take the parameter @xmath309 differently in each iteration . the purpose of the nonstationary choice of @xmath309 might be the improvement of the convergence rate of iterative methods ( [ eq_pii_46 ] ) and ( [ eq_pii_71 ] ) . note , that in the most general case iterative methods ( [ eq_pii_46 ] ) and ( [ eq_pii_71 ] ) applied to solve nonlinear penalty variational equations ( [ eq_pii_41 ] ) for multibody contact problems do not lead to domain decomposition . therefore , we now consider such variants of these methods , which lead to domain decomposition , namely , which reduce the solution of the original multibody contact problem in @xmath352 to the solution of a sequence of separate linear variational problems in subdomains @xmath8 , @xmath5 . let us take the bilinear form @xmath353 in the iterative method ( [ eq_pii_46 ] ) as follows @xcite : @xmath354 where @xmath355 is the next bilinear form @xcite : @xmath356 here @xmath357 are characteristic functions of some given subareas @xmath358 of possible contact zones @xmath359 , @xmath5 , @xmath142 . the iterative method ( [ eq_pii_46 ] ) with bilinear form ( [ eq_pii_77 ] ) can be written in the following equivalent way : @xmath360 @xmath361 * lemma 4 . * _ suppose that surfaces @xmath31 _ , _ @xmath362 are piecewise smooth . then bilinear form ( [ eq_pii_77 ] ) is symmetric , continuous and nonnegative , i.e. _ @xmath363 @xmath364 @xmath365 * proof . * it is obvious , that conditions ( [ eq_pii_80 ] ) and ( [ eq_pii_82 ] ) hold . thus , let us show the continuity of bilinear form ( [ eq_pii_77 ] ) . we can write @xmath366 , where @xmath367 the first term can be written in the following form : @xmath368 \left[{\bf n}_{\alpha } \cdot { \rm{tr}}_{\alpha } ^{0 } ( \ , { \bf v}_{\alpha } ) \right]\ , ds,\ ] ] taking into account , that the functions @xmath369 and the components of unit normals @xmath370 are bounded , and using inequality ( [ eq_pii_33 ] ) and schwarz inequality , we obtain @xmath371^{3 } } ^{2 } \left\| tr_{\ , \alpha } ^{0 } ( { \bf v}_{\alpha } ) \right\| _ { [ l_{2 } ( s_{\alpha \beta } ) ] ^{3 } } ^{2 } .\ ] ] in view of inequality ( [ eq_pii_35 ] ) , we find further @xmath372 similar inequality can be obtained for the second term of @xmath373 . thus @xmath374 as a result , we establish @xmath375 @xmath376 where @xmath377 , @xmath378 . @xmath114 from this * lemma * and * lemma 1 * , it follows that bilinear form ( [ eq_pii_76 ] ) is symmetric , continuous and coercive with constants @xmath379 and @xmath380 respectively . in addition , due to * lemmas * * 1 * , * 2 * , and * 4 * , functionals @xmath95 , @xmath381 , and @xmath382 are linear and continuous in @xmath269 . therefore , there exists a unique solution @xmath383 of variational problem ( [ eq_pii_78 ] ) . thus , the conditions of * theorem 4 * are satisfied , and we obtain the next proposition . * theorem 6 . * _ suppose that @xmath255 _ , @xmath384 . _ then the sequance @xmath385 _ , _ obtained by iterative method ( [ eq_pii_78 ] ) ( [ eq_pii_79 ] ) , which is equivalent to iterative method ( [ eq_pii_46 ] ) with bilinear form ( [ eq_pii_76 ] ) , converges strongly in @xmath70 to the exact solution @xmath249 of nonlinear penalty variational equation ( [ eq_pii_41 ] ) for unilateral multibody contact problem , i.e. _ @xmath386 . _ moreover , the convergence rate in norm @xmath264 is linear ( [ eq_pii_50 ] ) , where @xmath387 _ , _ and the maximal rate reaches as @xmath388 _ , @xmath257 . now let us show that iterative method ( [ eq_pii_78 ] ) ( [ eq_pii_79 ] ) leads to domain decomposition . due to relationships @xmath389 @xmath390 method ( [ eq_pii_78 ] ) ( [ eq_pii_79 ] ) rewrites as follows : @xmath391 @xmath392 @xmath393 since the common quantities of the subdomains are known from the previous iteration , the variational equation ( [ eq_pii_83 ] ) splits into @xmath2 variational equations in separate subdomains @xmath8 . therefore , method ( [ eq_pii_83 ] ) ( [ eq_pii_84 ] ) can be written in the following equivalent form : @xmath394 @xmath395 @xmath396 @xmath397 since bilinear forms @xmath398 , @xmath399 are symmetric , continuous and coercive , and functionals @xmath400 , @xmath401 , @xmath402 are linear and continuous in @xmath403 , it follows that there exists a unique solution @xmath404 of each variational equation ( [ eq_pii_85 ] ) . furthermore , it is obvious to see that the unique solution @xmath405 of variational equation ( [ eq_pii_78 ] ) , i.e. ( [ eq_pii_83 ] ) , takes the form @xmath406 . therefore , the solution of variational equation ( [ eq_pii_78 ] ) is equivalent to the solution of @xmath2 variational equations ( [ eq_pii_85 ] ) in separate subdomains , and iterative processes ( [ eq_pii_78 ] ) ( [ eq_pii_79 ] ) and ( [ eq_pii_85 ] ) ( [ eq_pii_86 ] ) are equivalent . now , consider iterative method ( [ eq_pii_85 ] ) ( [ eq_pii_86 ] ) in more detail . in each iteration of this method we have to solve in parallel @xmath2 variational equations ( [ eq_pii_85 ] ) , which correspond to some elasticity problems in subdomains with prescribed robin boundary conditions on possible contact areas : @xmath407 here @xmath408 are unknown normal stresses on possible contact areas @xmath31 . therefore , iterative method ( [ eq_pii_85 ] ) ( [ eq_pii_86 ] ) refers to parallel robin robin type domain decomposition schemes . since domain decomposition method ( [ eq_pii_85 ] ) ( [ eq_pii_86 ] ) and iterative method ( [ eq_pii_78 ] ) ( [ eq_pii_79 ] ) are equivalent , the convergence * theorem 6 * also holds for the method ( [ eq_pii_85 ] ) ( [ eq_pii_86 ] ) . taking different characteristic functions @xmath409 in ( [ eq_pii_85 ] ) , i.e. different subareas @xmath410 of possible contact zones @xmath31 , we can obtain different particular cases of domain decomposition method ( [ eq_pii_85 ] ) ( [ eq_pii_86 ] ) . thus , taking @xmath411 , @xmath412 , i.e. @xmath413 , we get domain decomposition scheme with robin boundary conditions on whole possible contact areas : @xmath414 therefore , we have named this domain decomposition method as full parallel robin robin domain decomposition scheme @xcite . taking @xmath415 , @xmath416 , i.e. @xmath417 , we get parallel neumann neumann domain decomposition scheme @xcite : @xmath418 @xmath419 in each step @xmath248 of this scheme we have to solve in parallel @xmath2 variational equations ( [ eq_pii_88 ] ) , which correspond to elasticity problems in subdomains with neumann boundary conditions on possible contact areas : @xmath420 note , that in the most general case , we can choose functions @xmath409 , i.e. surfaces @xmath410 , differently for each @xmath421 . moreover , we can choose functions @xmath409 differently at each iteration @xmath248 , i.e. @xmath422 where @xmath423 , @xmath278 are some given subareas of possible contact zones @xmath31 , @xmath5 , @xmath424 . as a result we obtain following nonstationary robin robin type domain decomposition scheme @xmath425 @xmath426 @xmath427 this domain decomposition scheme is equivalent to nonstationary iterative method ( [ eq_pii_71 ] ) with bilinear forms @xmath428 where @xmath429 bilinear forms @xmath430 are symmetric , nonnegative , and continuous with constant @xmath431 . hence , bilinear forms ( [ eq_pii_93 ] ) satisfy properties ( [ eq_pii_72 ] ) ( [ eq_pii_74 ] ) , where @xmath379 , @xmath380 . it is obvious to see that condition ( [ eq_pii_75 ] ) in * theorem 5 * for bilinear forms ( [ eq_pii_93 ] ) is equivalent to the following condition @xmath432 which by - turn is equivalent to the condition @xmath433 therefore , from * theorem 5 * we obtain the next proposition about convergence of nonstationary domain decomposition scheme ( [ eq_pii_91 ] ) ( [ eq_pii_92 ] ) . * theorem 7 . * _ suppose that @xmath349 _ , @xmath434 , _ and functions @xmath435 satisfy property ( [ eq_pii_95 ] ) . then the sequence @xmath385 _ , _ obtained by nonstationary domain decomposition scheme ( [ eq_pii_91 ] ) ( [ eq_pii_92 ] ) , converges strongly in @xmath70 to the exact solution of nonlinear penalty variational equation ( [ eq_pii_41 ] ) for unilateral multibody contact problem , i.e. @xmath386_. now let us consider a particular case of domain decomposition method ( [ eq_pii_91 ] ) ( [ eq_pii_92 ] ) . in each iteration @xmath248 let us choose functions @xmath435 as follows @xcite : @xmath436 where @xmath43 , @xmath437 . then , taking into consideration that @xmath438 , we obtain the method @xcite : @xmath439 @xmath440 in each step @xmath248 of this method we have to solve in parallel @xmath2 variational equations ( [ eq_pii_97 ] ) , which correspond to elasticity problems in subdomains with prescribed displacements @xmath441 through penalty on some subareas of possible contact zones @xmath31 . therefore , we can conventionally name this method as nonstationary parallel dirichlet dirichlet domain decomposition scheme . note , that in some particular cases of penalty robin robin domain decomposition method ( [ eq_pii_91 ] ) ( [ eq_pii_92 ] ) ( for example @xmath442 , @xmath443 or @xmath444 , @xmath412 ) we can pass the limit to @xmath193 , and obtain domain decomposition schemes without penalty @xcite . the advantages of proposed domain decomposition schemes are their simplicity , and the regularization of the original contact problem because of the use of penalty term . these domain decomposition schemes have only one iteration loop , which deals with domain decomposition and nonlinearity of unilateral contact conditions . the disadvantage of proposed methods is that we have to choose a penalty parameter . finally , let us say that iterative methods ( [ eq_pii_46 ] ) and ( [ eq_pii_71 ] ) for the solution of nonlinear variational equations are rather general . from these methods , besides parallel robin robin type domain decomposition schemes ( [ eq_pii_85 ] ) ( [ eq_pii_86 ] ) and ( [ eq_pii_91 ] ) ( [ eq_pii_92 ] ) , we also can obtain other different particular iteratve methods for the solution of penalty variational equation of unilateral multibody contact problems , which do not lead to domain decomposition . thus , taking bilinear form @xmath445 , + @xmath446 , + and iterative parameter @xmath447 in ( [ eq_pii_46 ] ) , we obtain iterative method for the solution of multibody unilateral contact problems , which can be viewed as a generalization of penalty iteration method , proposed in @xcite for the solution of crack problems with nonpenetration conditions . taking in method ( [ eq_pii_71 ] ) bilinear forms @xmath448 in each step @xmath248 as follows @xmath449 where @xmath450 @xmath451 and iterative parameter @xmath447 , we obtain iterative method , which can be viewed as active set method , i.e. semi - smooth newton method , for unilateral multibody contact problems . for one - body contact problems active set method was introduced in @xcite . the convergence theorem for active set method , as a variant of semi - smooth newton method , was proved in @xcite . note , that bilinear forms @xmath452 and @xmath453 are symmetric , nonnegative and coercive . however , these two particular cases of methods ( [ eq_pii_46 ] ) and ( [ eq_pii_71 ] ) do not lead to the domain decomposition . in the next section we investigate numerical efficiency of proposed by us penalty parallel robin robin type domain decomposition schemes ( [ eq_pii_85 ] ) ( [ eq_pii_86 ] ) and ( [ eq_pii_91 ] ) ( [ eq_pii_92 ] ) . we provide numerical investigation of proposed domain decomposition schemes for plane unilateral contact problems of two elastic bodies @xmath454 , @xmath455 . for the numerical solution of linear variational problems in subdomains , we use finite element method ( fem ) with linear and quadratic triangular elements . at first let us compare the convergence rates of different particular domain decomposition schemes . consider contact problem of two transversally isotropic elastic bodies @xmath456 , @xmath455 with the plane of isotropy , parallel to the plane @xmath457 ( fig . 2 ) . the material properties of the bodies are : @xmath458 , @xmath459 , @xmath460 , where @xmath461 , @xmath462 , and @xmath463 are the elasticity modulus , poisson s ratio , and the shear modulus for the body @xmath8 in the plane of isotropy , and @xmath464 , @xmath465 , @xmath466 are these constants in the orthogonal direction , @xmath455 . the length and height of each body is the same , and is equal to @xmath467 . the distance between bodies before the deformation is @xmath468 , the compression of the bodies is @xmath469 , @xmath470 , and the possible contact area is @xmath471,\ , \ , x_{2 } = 4\ , b\right\}$ ] . the problem is solved by parallel robin robin domain decomposition schemes , using fem with 1595 quadratic triangular elements in each body ( 15 elements on each side of possible contact area @xmath472 ) . we take the penalty parameter in the form @xmath473 , @xmath474 , where @xmath475 is dimensionless penalty coefficient . the initial guesses for displacements @xmath476 , @xmath455 are chosen using the bar model @xcite : @xmath477^{- } } { \theta e'_{1 } \left(1+c\right)},\,\ , { \bf x } \in s_{12},\,\,\ , u_{2\ , n}^{0 } ( { \bf x})=\frac{4\ , bc\left[d_{12 } ( { \bf x})-\delta \right]^{- } } { \theta e'_{2 } \left(1+c\right ) } + \delta , \,\ , { \bf x } \in s_{21 } .\ ] ] we use the following stopping criterion for domain decomposition schemes @xmath478 where @xmath479^{2 } } $ ] is the discrete norm , @xmath480 are the finite element nodes on the possible contact area , and @xmath481 is the relative accuracy for displacements . 3 shows the approximations to dimensionless normal contact stress @xmath482 , @xmath483 , @xmath484 , obtained by parallel neumann neumann scheme ( [ eq_pii_88 ] ) ( [ eq_pii_89 ] ) ( @xmath485 ) at iterations @xmath486 ( curves 14 ) for optimal iterative parameter @xmath487 and accuracy @xmath488 . the dashed curve represents the exact solution for two half - spaces , obtained in @xcite . hence , the real contact area is @xmath489\kern-0.15em\right]$ ] , where @xmath490\kern-0.15em\right]=\left\{{\bf x}=\left(x_{1 } , x_{2 } \right)^{{\rm t } } : \,\,\,\ , x_{1 } \in [ y;\ , \ , z],\ , \ , x_{2 } = 4\ , b\right\}$ ] . at fig . 4 and fig . 5 the convergence rates of different particular domain decomposition schemes are compared . the total iteration number @xmath491 dependence on iteration parameter @xmath309 for accuracy @xmath488 is shown at fig . 4 , and its dependence on logarithmic accuracy @xmath492 for optimal iteration parameter @xmath493 is shown at fig . the first curve at these figures represents the parallel neumann neumann scheme ( @xmath494 , @xmath495 ) , curves 2 , 3 , 4 and 5 correspond to the parallel robin robin schemes ( [ eq_pii_85 ] ) ( [ eq_pii_86 ] ) with @xmath496 equal to @xmath497\kern-0.15em\right]$ ] , @xmath498\kern-0.15em\right]$ ] , @xmath499\kern-0.15em\right]$ ] , and @xmath500\kern-0.15em\right]$ ] ( @xmath501 , @xmath502 ) respectively . curve 3 also represents the nonstationary parallel dirichlet dirichlet scheme ( [ eq_pii_97 ] ) ( [ eq_pii_98 ] ) . the optimal iterative parameters @xmath503 for the schemes represented at curves 15 are @xmath487 , 0.39 , 0.72 , 0.85 and 0.92 respectively . for @xmath319 and accuracy @xmath488 these schemes converge in 21 , 11 , 5 , 11 and 14 iterations . thus , the convergence rate of stationary robin robin domain decomposition schemes is linear . the parallel robin robin scheme ( [ eq_pii_85 ] ) ( [ eq_pii_86 ] ) with the surfaces @xmath504 , @xmath505 most closed to the real contact area ( @xmath506\kern-0.15em\right]$ ] ) , and the nonstationary parallel dirichlet dirichlet scheme ( [ eq_pii_97 ] ) ( [ eq_pii_98 ] ) ( @xmath507 ) , which are represented at curve 3 , have the highest convergence rates . these two schemes also have the widest range from which the iterative parameter @xmath508 can be chosen . the convergence rate of parallel neumann neumann scheme ( @xmath509 ) , which is represented at curve 1 , is the most slow . now let us investigate the convergence of penalty method and its dependence on finite element discretization . consider the unilateral contact problem of two isotropic bodies @xmath510 and @xmath511 , one of which has a groove ( fig . 6 ) . the bodies are uniformly loaded by normal stress with intensity @xmath318 . each body has length @xmath512 and height @xmath513 , and the grove has length @xmath514 . the material properties of the bodies are the same : @xmath515 , @xmath516 . the distance between bodies before the deformation is @xmath517^{+ } \right\}^{3/2 } $ ] , where @xmath518 , @xmath519 . the possible contact area is @xmath520,\ , \ , x_{2 } = h\right\}$ ] . the problem is solved by nonstationary parallel dirichlet dirichlet domain decomposition scheme ( [ eq_pii_97 ] ) ( [ eq_pii_98 ] ) with finite element approximations on triangles . the penalty parameter is taken as follows @xmath521 where @xmath475 is dimensionless penalty coefficient . initial guesses for displacements are chosen using the bar model @xcite : @xmath522^{- } } { \left(1+c\right ) } , \,\,\ , { \bf x } \in s_{12 } , \ ] ] @xmath523^{- } } { \left(1+c\right ) } + \delta , \,\,\ , { \bf x } \in s_{21 } , \ ] ] where @xmath524 . we use ( [ eq_pii_99 ] ) as stopping criterion for iterative process . for iterative parameter @xmath525 $ ] , accuracy @xmath526 , penalty coefficients @xmath475 and finite element meshes considered below , parallel dirichlet dirichlet scheme , applied to solve this problem , converges in 215 iterations . we investigate the dependence of the quality of numerical solution , obtained by this scheme , on penalty parameter and finite element mesh . plots at fig . 7 represent approximations to dimensionless contact stress @xmath527 , @xmath528 for bodies with size @xmath529 and external load @xmath530 , obtained by dirichlet dirichlet scheme for different dimensionless penalty coefficients @xmath475 at fixed finite element mesh with 64 linear triangular finite elements on each side of possible contact area @xmath472 . curves 14 correspond to @xmath531 , @xmath532 , @xmath533 and @xmath534 respectively . plots at fig . 8 represent approximations to @xmath535 , @xmath536 for bodies with length @xmath537 , height @xmath538 , and external load @xmath539 , obtained for differrent dimensionless penalty coefficients @xmath475 and different finite element meshes . curves 1 and 2 at this figure correspond to @xmath540 for dimensionless penalty coefficients @xmath531 and @xmath541 respectively at finite element mesh with 32 linear triangular elements on each side of possible unilateral contact area @xmath472 . curves 3 and 4 correspond to @xmath540 for @xmath531 and @xmath541 respectively , but for finite element mesh with 64 linear triangular elements on each side of @xmath542 . dashed curve at this figure and at fig . 7 represents the exact solution , obtained in @xcite for the contact of two half - spaces . here we see that in spite of the solution , obtained for penalty coefficient @xmath531 at mesh with 32 finite elements on each side of @xmath472 ( curve 1 at fig . 8) , the solution obtained for lower penalty coefficient @xmath541 at the same mesh become instable . but if we refine finite element mesh twice for the penalty coefficient @xmath541 , then the influence of errors on the perturbation of initial data will decrease , and we shall obtain much better approximation to the exact solution ( curve 4 at fig . hence , we conclude that for obtaining a nice approximation to the solution , we need to decrease penalty parameter and to refine finite element mesh simultaneously . for the solution of unilateral multibody contact problems of elasticity we have proposed on continuous level a class of parallel robin robin type domain decomposition schemes , which are based on penalty method for variational inequalities and some stationary or nonstationary iterative methods for nonlinear variational equations . in each step of these schemes we have to solve in parallel linear variational equations in subdomains , which correspond to some elasticity problems with robin boundary conditions on possible contact areas . numerical analysis of proposed domain decomposition schemes has been made for plane two - body contact problems using linear and quadratic finite element approximations on triangles . convergence rates of different particular domain decomposition schemes have been compared and their dependence on iterative parameter @xmath309 has been investigated . the penalty parameter and mesh refinement influence on the numerical solution have been examined . numerical experiments have confirmed the theoretical results about convergence of these domain decomposition schemes . the advantages of proposed domain decomposition schemes are their simplicity , and the regularization of the original contact problem because of the use of penalty term . these domain decomposition schemes have only one iteration loop , which deals with domain decomposition and nonlinearity of unilateral contact conditions . in addition , these methods are obtained on continuous level , and they do not depend on any discretization technique . p. avery , c. farhat , the feti family of domain decomposition methods for inequality - constrained quadratic programming : application to contact problems with conforming and nonconforming interfaces , comput . methods appl . * 198 * ( 2009 ) 16731683 . i. i. dyyak , i. i. prokopyshyn , convergence of the neumann parallel scheme of the domain decomposition method for problems of frictionless contact between several elastic bodies , journal of mathematical sciences * 171*(4 ) ( 2010 ) 516533 . i. i. dyyak , i. i. prokopyshyn , domain decomposition schemes for frictionless multibody contact problems of elasticity , in : g. kreiss et al ( eds . ) , numerical mathematics and advanced applications 2009 . proceedings of enumath 2009 , the 8th european conference on numerical mathematics and advanced applications , uppsala , july 2009 , springer , berlin heidelberg , 2010 , pp . 297305 . i. i. prokopyshyn , penalty method based domain decomposition schemes for contact problems of elastic solids , dissertation for a candidate s degree in physics and mathematics by specialty 01.05.02 mathematical modeling and computational methods , national academy of sciences of ukraine , pidstryhach institute for applied problems of mechanics and mathematics , lviv , 2010 , url:194.44.15.230:8080/v25/resources / thesis / prokopyshyn.pdf [ in ukrainian ] . i. i. prokopyshyn , r. m. martynyak , numerical investigation of contact interaction of two solids with a groove by domain decomposition method , problems of computational mechanics and strength of structures * 16 * ( 2011 ) 240251 [ in ukrainian ] . grigorenko , i. i. dyyak , s. i. matysyak , i. i. prokopyshyn , domain decomposition methods applied to solve frictionless - contact problems for multilayer elastic bodies , int . . mech . * 46*(4 ) ( 2010 ) 388399 .

we study the boundary value problem for asymptotically flat stationary black ring solutions to the five - dimensional vacuum einstein equations . assuming the existence of two additional commuting axial killing vector fields and the horizon topology of , we show that the only asymptotically flat black ring solution with a regular horizon is the pomeransky - senkov black ring solution .

0710.4600

in four dimensions , stationary black hole spactimes have been studied by many authors and these studies are known as _ uniqueness theorems of black holes _ @xcite . israel showed that the only static , asymptotically flat vacuum solution with a regular event horizon is the schwarzschild solution specified only by its mass @xcite . shortly afterward , he generalized the theorem to the situation with an electromagnetic field @xcite , i.e , he presented the proof that the only static , asymptotically flat electrovac solution with a regular non - degenerate event horizon is the reissner - nordstrm solution with two parameters , the mass @xmath1 and the electric charge @xmath2 which are subject to the inequality @xmath3 . bunting and masood - ul - alam succeeded in proving these two theorems without the assumption of a single connected component of a black hole @xcite by using the positive energy theorem @xcite . on the other hand , for non - static and stationary space - times , the einstein - maxwell system can be reduced to two - dimensional boundary value problem . the field equation is derived by the ernst potential associated with the axial killing vector field . the essential part in showing the uniqueness theorem for stationary black holes is that two solutions with the same asymptotic condition are isometric to each other . carter @xcite showed that the infinitesimally neighboring vacuum solutions with the same asymptotic conditions are equal , and robinson generalized its theorem to the stationary electrovac system @xcite . thereafter , using so - called _ robinson identity _ , robinson also succeeded in proving the uniqueness of the vacuum kerr family @xcite with @xmath4 among all asymptotically flat , stationary and axisymmetric black hole solutions with a non - degenerate event horizon , i.e. , showing that two arbitrary , not necessarily infinitesimally neighboring , solutions with the same boundary conditions are equal to each other . in electromagnetic system , mazur derived the divergence identity ( _ mazur identity _ ) @xcite , which is based on that the ernst equation describe a non - linear sigma model on the symmetric space @xmath5 , and showed that the only possible and axisymmetric black hole solutions to the einstein - maxwell equations is the kerr - newman solution specified by the mass @xmath1 , the angular momentum @xmath6 and the electric charge @xmath2 with the constraint @xmath7 . in recent years , studies of black holes in higher dimensions have attracted much attention in the context of string theory and the brane world scenario . in fact , it has been predicted that higher - dimensional black holes would be produced in a future linear collider @xcite . such physical phenomena are expected not only to give us a piece of evidence for the existence of extra dimensions but also to help us to draw some information toward quantum gravity . studies on stationary black hole solutions are important since we may detect the hawking radiation after the formation of stationary black holes in a collider . a striking feature of asymptotically flat stationary black hole solutions in five dimensions is that they admit event horizons with non - spherical topologies in contrast to four dimensions . for instance , the topology of the event horizon in higher dimensions can not be uniquely determined @xcite in contrast to four - dimensional ones , which is restricted only to the two sphere @xcite . in five dimensions , however , the possible geometric types of the horizon topology are @xmath8 and @xmath9 @xcite , and in dimensions higher than five , more complicated @xcite . the black ring solutions with the horizon topology @xmath0 , which rotate along the @xmath10 direction , were found by emparan and reall as solutions to the five - dimensional vacuum einstein equations @xcite . this is the first example of black hole solution with non - spherical topology . in addition to the black ring solution , the rotating black hole solution with @xmath8 horizon topology had been already found by myers and perry @xcite . remarkably , within some range of the parameters , there are one black hole and two black rings with the same values of the mass and the angular momentum , which means the violation of the uniqueness known in four dimensions . subsequently , other black ring solutions were found . the black ring solutions with a rotating two sphere were found by mishima and iguchi @xcite , and moreover , one with two angular momenta was constructed by pomeransky and senkov @xcite by using the inverse scattering method @xcite . for the asymptotically flat , static solutions of higher - dimensional vacuum einstein equations , the schwarzschild - tangherlini solution @xcite is the unique solution @xcite , and moreover , which is stable against linear perturbations @xcite . it has been shown that the five - dimensional myers - perry solution is unique if the topology is restricted to @xmath8 and the spacetime admits three commuting killing vectors @xcite . hence it is natural to ask whether the pomeransky - senkov black ring solution is also unique under the assumptions of the existence of three commuting killing vector field and the horizon topology of @xmath11 . as mentioned above , however , there are two different black ring solutions for the same mass and the same angular momenta . therefore , we must add some additional information to consider the bondary value problem for black ring solutions . one of the examples is the _ rod structure _ introduced by harmark @xcite . by introducing the rod structure , hollands and yazadjiev @xcite applied the discussion of morisawa and ida to the case of non - spherical horizon topology and showed that two asymptotically flat and five - dimensional black hole solutions with the same topology , the same mass , the same angular momenta and the same rod structure are isometric to each other . in this article , we study the boundary value problem for stationary black ring solutions to the five - dimensional vacuum einstein equations . assuming the existence of two additional commuting axial killing vector fields and the horizon topology of @xmath0 , we show that the only asymptotically flat black ring solution with a regular horizon is the pomeransky - senkov black ring solution . our proof consists of two steps : first , we present a more general black ring solution than the pomeransky - senkov black ring solution in the sense that the solution , in general , has a conical singularity and is characterized by independent four parameters , i.e. , the mass , two angular momenta and an additional parameter . by the requirement of the absence of a conical singularity on the solution , it coincides with the pomeransky - senkov black ring solution . second , following the discussion in ref . @xcite , two arbitrary asymptotically flat black ring solutions with the same mass , the same two angular momenta and the same ratio of the radius of @xmath12 to the radius of @xmath13 as the solution are isometric . hence we can conclude that the only asymptotically flat black ring solution without a conical singularity to the five - dimensional vacuum einstein equations is the pomeransky - senkov black ring solution . the remainder of this article is organized as follows : in sec.[sec : solution ] we present general black ring solutions with four parameters . in sec.[sec : boundary ] , we study the rod structure of the solutions . in sec.[sec : mazur ] , we give a short explanation of the mazur identity . in sec.[sec : coincidence ] , using the mazur identity , we show that an arbitrary asymptotically black ring solution with the same mass , the same two angular momenta and the same rod structure as our solution are isometric to it . in sec.[sec : conclusion ] , we state the final results and the theorem . # 1s_#1 the metric of general black ring solution , which in general has a conical singularity , is given by @xmath14 where the c - metric coordinates @xmath15 run the ranges of @xmath16 and @xmath17 or @xmath18 , respectively . the solution has four independent parameters satisfying the inequalities @xmath19 , @xmath20 , @xmath21 and @xmath22 with @xmath23 the function @xmath24 appearing in the metric is defined as : @xmath25 since the other functions @xmath26 and the one - form @xmath27 have considerably complicated forms , we do not write it here . the explicit expressions of them are given in appendix [ app : metric ] . as will be mentioned later , under the choice of the parameters : @xmath28 which is the condition for a conical singularity inside the black ring to vanish , the metric reduces to that of the pomeransky - senkov black ring solution . in this section , we give the rod structure of the black ring solution obtained in sec.[sec : solution ] . to investigate the rod structure , we introduce the canonical coordinates defined by @xmath29 then , the metric can be written in the form @xmath30 where @xmath31 run @xmath32 . the metric functions @xmath33 and @xmath34 depend only on the coordinates @xmath35 and @xmath36 . \(i ) the semi - infinite spacelike rod @xmath37 $ ] and the finite rod @xmath38 $ ] have the direction @xmath39 , i.e. , for @xmath40 , @xmath41 $ ] and @xmath40 , @xmath42 $ ] , @xmath43 holds . since these give @xmath44 , @xmath40 , @xmath41 $ ] and @xmath40 , @xmath42 $ ] denote @xmath45-axis , i.e. , the plane invariant under the rotation associated with the killing vector @xmath46 . for @xmath40 and @xmath41 $ ] , the periodicity of the angular variable @xmath45 becomes @xmath47 to cure conical singularities in the region @xmath42 $ ] , one must impose the following condition on the parameters in the solutions : @xmath48 where the constant @xmath49 is defined by @xmath50 with @xmath51 the periodicities ( [ eq : period ] ) and ( [ eq : period2 ] ) of @xmath45 require putting the parameters as @xmath52 @xmath53 it should be noted that the solution with the parameters ( [ eq : pom_parameter ] ) exactly coincides with the solutions without a conical singularity obtained by pomeransky and senkov , which describe the rotating black ring in two orthogonal planes independently , although the choice of the remainder parameters ( [ eq : singular_parameter ] ) yields singular solutions . \(ii ) the finite timelike rod @xmath54 $ ] corresponds to an event horizon with topology @xmath0 since @xmath46 vanishes on both side of this rod . one see that @xmath43 for @xmath40 and @xmath55 $ ] . @xmath56 denotes the eigenvector with respect to the eigenvalue of zero and can be written in the form of @xmath57 , where @xmath58[2b(1-c^2)-(1-b)c\alphazero ] } { 2(1-b)b(1-c)^2[2b(1+c)^2-(1+b)c\alphazero][2b(1-c^2)-(1+b)c\alphazero ] k^2 } , \\ { { \omega}_{2}}^{2}&= & \frac{(1+b)[2b(1-c^2)-(1-b)c\alphazero ] } { \alphazero(1-b)[2b(1-c^2)-(1+b)c\alphazero ] } \nonumber\\&\times & \frac{(1+c)^2[2b(1-c^2)-c\alpha]^2[2b(1-c)^2-(1-b)c\alpha]^2 } { c^2[4b^2(1-c^2)^2-(1-b^2)c^2\alphazero^2]^2 k^2}.\end{aligned}\ ] ] here the two constants @xmath59 and @xmath60 denote the angular velocities of the horizon along the directions @xmath61 and @xmath46 , respectively . \(iii ) the semi - infinite spacelike rod @xmath62 $ ] has the direction @xmath63 . since these give @xmath64 , @xmath40 , @xmath65 $ ] means the @xmath66-axis , i.e. , the plane under a rotation with respect to the killing vector field @xmath61 . the periodicity of the angular variable @xmath66 is computed as @xmath67 -plane with the three - dimensional @xmath68 part suppressed . ] next , we introduce the coordinates @xmath69 defined by @xmath70 and @xmath71 . in the asymptotic region @xmath72 , the metric behaves as @xmath73 where the adm mass @xmath74 and adm angular momenta @xmath75 , @xmath76 are given by @xmath77 k^2 } { 2(1+b)[2b(1-c)^2-(1-b)c\alpha][2b(1-c^2)-(1-b)c\alpha]},\end{aligned}\ ] ] @xmath78 } { [ ( 1+b)c\alphazero-2b(1-c^2 ) ] } \nonumber\\&\times & \frac{[(1+b)c\alphazero-2b(1-c)^2]^2 [ ( 1-b^2)c^2\alphazero^2 - 4b^2(1-c^2)^2]^2 k^6 } { [ ( 1-b)^2(1+b)c^2\alphazero^2 - 4b(1-b^2)(1-c)c\alphazero+4b^2(1+b)(1-c)^3(1+c)]^3 } , \\ { j_{\psi}}^{2}&= & \frac{16 \pi^2 b^2 ( 1-b ) c^4 ( 1-c)^2 \alphazero } { ( 1+b)[(1-b)c\alphazero-2b(1-c)^2]^2 } \nonumber\\&\times & \frac{[(1-b^2 ) c^2 \alphazero^2 - 4 b^2 ( 1-c^2)^2]^2 k^6 } { [ ( 1-b)c\alphazero-2b(1-c^2)]^3[(1+b)c\alphazero-2b(1-c^2)]}.\end{aligned}\ ] ] here we give the brief review on the formalism developed in ref . @xcite , where it is shown that the myers - perry solution is unique within a class of the five - dimensional asymptotically flat solutions with the horizon topology of @xmath8 and additional two commuting spacelike killing vectors . we consider the five - dimensional space - times admitting two commuting killing vector fields @xmath79 . then , the metric can be written in the form @xmath80 where @xmath81 run @xmath82 and @xmath83 . the rescaled three - dimensional metric @xmath84 , the metric functions @xmath85 and @xmath86 are independent of @xmath66 and @xmath45 . the twist potentials @xmath87 are defined by @xmath88 then , the vacuum einstein equations reduce to the system of the five scalar fields @xmath86 and @xmath87 on the three - dimensional space : @xmath89 and the einstein equations for the three - dimensional space : @xmath90 where @xmath91 is the covariant derivative with respect to the metric @xmath84 and @xmath92 denotes the inner product by @xmath84 . here we assume the existence of another killing vector field , a timelike killing vector field @xmath93 which commutes with the other killing vector fields @xmath94 . here we consider the case where two space orthogonal to all killing vector fields @xmath95 and @xmath96 is integrable . from the frobenius conditions , @xmath97 . hence the metric can be written in weyl - papapetrou - type form : @xmath98 all the metric functions depend only on @xmath35 and @xmath36 . the differential equations of the scalar fields are given by the axisymmetric solutions of eqs.([eq : f ] ) and ( [ eq : o ] ) on the abstract flat three surface with metric @xmath99 which is written in the cylindrical coordinates . namely , @xmath100 and @xmath101 are replaced with @xmath102 and @xmath103 , respectively . once the five potentials @xmath86 and @xmath87 are obtained from eqs.([eq : f ] ) and ( [ eq : o ] ) , eq.([eq : gamma ] ) reduce to the equations with respect to the gradient of the metric function @xmath104 : @xmath105 + { 1 \over 4 } f^{ij}f^{mn}(f_{im,\rho}f_{jn,\rho}-f_{im , z}f_{jn , z } ) \nonumber\\ & & + { 1 \over 2 } f^{-1}f^{ij } ( \omega_{i,\rho}\omega_{j,\rho}-\omega_{i , z}\omega_{j , z } ) , \\ { 1 \over \rho}\sigma_{,z } & = & { 1 \over 4 } f^{-2 } f_{,\rho } f_{,z } + { 1 \over 4 } f^{ij}f^{mn } f_{im,\rho } f_{jn , z } + { 1 \over 2 } f^{-1}f^{ij } \omega_{i,\rho } \omega_{j , z } , \\ w^i_{3,\rho } & = & \rho f^{-1 } f^{ij } \omega_{j , z } , \\ w^i_{3,z } & = & -\rho f^{-1 } f^{ij } \omega_{j,\rho}.\end{aligned}\ ] ] this system is described by the following action which is invariant under the global @xmath106 transformation : @xmath107.\label{eq : action}\end{aligned}\ ] ] here , we introduce the @xmath106 matrix @xmath108 defined by @xmath109 where it is noted that this matrix is symmetric @xmath110 and unimodular @xmath111 . since we choose the killing vector fields @xmath112 and @xmath113 to be spacelike , all the eigenvalues of @xmath108 are real and positive . therefore , there exists an @xmath106 matrix @xmath114 such that @xmath115 the square root matrix @xmath114 is determined up to the global @xmath116 transformation . in fact , under the rotation @xmath117 for any @xmath118 , @xmath108 is invariant . hence , the action describes a non - linear sigma model on the symmetric space @xmath119 . we define a current matrix as @xmath120 which is conserved if the scalar fields are the solutions of the equation of motion derived by the action ( [ eq : action ] ) . then , the action ( [ eq : action ] ) can be written in terms of @xmath121 and @xmath108 as follows @xmath122 let us consider two sets of the field configuration @xmath123}$ ] and @xmath124}$ ] satisfying the equations eq.([eq : f ] ) and eq.([eq : o ] ) . we denote the difference between the value of the functional obtained from the field configuration @xmath124}$ ] and the value obtained from @xmath123}$ ] as a bull s eye @xmath125 , e.g. , @xmath126}-j^i_{[0]},\end{aligned}\ ] ] where the subscripts @xmath127}$ ] and @xmath128}$ ] denote the quantities associated with the field configurations @xmath123}$ ] and @xmath124}$ ] , respectively . the deviation matrix @xmath129 is defined by @xmath130}=\phi_{[1]}\phi^{-1}_{[0]}-{\bf 1},\end{aligned}\ ] ] where @xmath131 is the unit matrix . using the relation between the derivative of the deviation matrix and the @xmath132 : @xmath133}\stackrel{\odot}j{}^i\phi_{[0]}^{-1},\label{eq : deriv}\end{aligned}\ ] ] where @xmath91 is a covariant derivative associated with the abstract three - metric @xmath134 . the mazur identity as the integration over the region @xmath135 is given by @xmath136 where @xmath137 run @xmath138 , and @xmath139 . the matrix @xmath140 is defined by @xmath141}^{-1}{\ } ^t\stackrel{\odot}j{}^ag_{[1]}.\end{aligned}\ ] ] it should be noted that the right hand side of the identity ( [ eq : id ] ) is positive except the case of @xmath142 . therefore , we must have @xmath142 if the boundary conditions under which the left hand side of eq.([eq : id ] ) vanishes are imposed at @xmath143 . then , from eq.([eq : deriv ] ) , @xmath129 is a constant matrix over the region @xmath144 . to show that the limiting value of @xmath129 is zero on at least one part of the boundary is sufficient to obtain the coincidence of two solutions @xmath123}$ ] and @xmath124}$ ] . the boundary integral in the left hand side of the mazur identity ( [ eq : id ] ) is decomposed into the integrals over the segments of the rod and the integral over the infinity as follows @xmath145 where @xmath146 in the last term denotes the infinity . now consider the line integrals of the twist one - forms @xmath147 over the @xmath36-axis in the canonical coordinates @xmath148 . the twist one - forms vanish on the @xmath45-invariant planes and the @xmath66-invariant plane by its definition , i.e. , the values of twist potentials are constant on the @xmath36-axis . hence this can be expressed in the form @xmath149^{z = ck^2}_{z =- ck^2}\label{eq : line}\end{aligned}\ ] ] on the @xmath36-axis . on the other hand , the left hand side of eq.([eq : line ] ) can be written in terms of the komar integral as follows @xmath150 where @xmath151 denotes the spatial cross section of the event horizon and @xmath152 are angular momenta associated with the spacelike killing vectors @xmath153 . using the ambiguity of the twist potentials @xmath87 in addition of a constant , without loss of generality , we always put their values on the @xmath36-axis such that @xmath154 for @xmath155 $ ] , and @xmath156 for @xmath157 $ ] . we estimate the integrals over the five boundaries and show that they vanish under the preferable boundary conditions . \(i ) @xmath45-invariant planes : @xmath158 and @xmath159 . the boundary integral over these regions are given by @xmath160^{-ck^2}_{-\infty},\\ & & \int^{k^2}_{ck^2}\rho\frac{\partial { \rm tr}\psi } { \partial z}\biggl|_{\rho=0}dz=\biggl[\rho{\rm tr}\psi|_{\rho=0 } \biggr]_{ck^2}^{k^2},\end{aligned}\ ] ] respectively . we assume that for @xmath161 , the five scalar fields behave as @xmath162}\simeq f_{\psi\psi[a]}^{(2)}\rho^2+o(\rho^4),\label{eq : a1}\\ & & f_{\phi\phi[a]}\simeq f_{\phi\phi[a]}^{(0)}+o(\rho^2),\label{eq : a2}\\ & & f_{\phi\psi[a]}\simeq f_{\phi\psi[a]}^{(2)}\rho^2+o(\rho^4),\label{eq : a3}\\ & & \omega_{\phi[a]}\simeq \omega_{\phi[a]}^{(0)}+o(\rho^2),\label{eq : a4}\\ & & \omega_{\psi[a]}\simeq \omega_{\psi[a]}^{(0)}+o(\rho^2),\label{eq : a5}\end{aligned}\ ] ] where @xmath163 run @xmath164 and the coefficients @xmath165}^{(2k)}$ ] and @xmath166}^{(2k)}$ ] defined by @xmath165}=\sum_{k=0}^{\infty}f_{ij[a]}^{(2k)}\rho^{2k}$ ] , @xmath166}=\sum_{k=0}^\infty\omega_{i[a]}^{(2k)}\rho^{2k}$ ] are independent of @xmath35 . the boundary condition ( [ eq : a1 ] ) comes from the requirement that @xmath40 , @xmath41 $ ] and @xmath40 , @xmath42 $ ] are the @xmath45-invariant plane , i.e. , the plane invariant under the rotation with respect to the axial killing vector @xmath46 . the regularity on the invariant plane requires the other conditions ( [ eq : a2])-([eq : a5 ] ) . hence , for @xmath161 , @xmath167 behaves as @xmath168}^{(2)}f_{\psi\psi[1]}^{(2)}f_{\psi\psi[0]}^{(2)}\rho^3 } + \frac{1}{f_{\psi\psi[0]}^{(2)2}f_{\psi\psi[1]}^{(2)2}f_{\phi\phi[0]}^{(0)2}f_{\phi\phi[1]}^{(0)}\rho}\nonumber\\ & & \times\biggl [ \left(\stackrel{\odot}{{\omega_{\psi}}^{(0)}}\right)^2\left ( -f_{\psi\psi[1]}^{(2)}f_{\phi\phi[1]}^{(0)}f_{[0]}^{(4)}-f_{\psi\psi[0]}^{(2)}f_{\phi\phi[0]}^{(0)}f_{[1]}^{(4)}-f_{\psi\psi[0]}^{(2)}f_{\psi\psi[1]}^{(2)}f_{\phi\phi[0]}^{(2)}f_{\phi\phi[1]}^{(0)}\right)\nonumber\\ & & + \left(\stackrel{\odot}{{\omega_{\phi}}^{(0)}}\right)^2f_{\psi\psi[0]}^{(2)}f_{\psi\psi[1]}^{(2)2}f_{\phi\phi[0]}^{(0)}+2\stackrel{\odot}\omega_\psi^{(0)}f_{\psi\psi[0]}^{(2)}f_{\psi\psi[1]}^{(2)}f_{\phi\phi[0]}^{(0)}(\stackrel{\odot}\omega_\psi^{(2)}f_{\phi\phi[1]}^{(0)}-\stackrel{\odot}\omega_\phi^{(0)}f_{\phi\psi[1]}^{(2 ) } ) \biggr]+o(\rho)\nonumber \\ & = & o(\rho),\end{aligned}\ ] ] where @xmath169}:=f_{\phi\phi[a]}^{(2)}f_{\psi\psi[a]}^{(2)}-f_{\phi\psi[a]}^{(2)2}$ ] , and in the last equality we used the eq.([eq : twist ] ) for the two solutions @xmath123}$ ] and @xmath124}$ ] with the same angular momenta . \(ii ) @xmath66-invariant plane : @xmath170 . the boundary integral over the region is @xmath171_{ck^2}^{\infty}.\end{aligned}\ ] ] we assume that for @xmath161 , the five scalar fields behave as @xmath172}\simeq\rho^2f_{\phi\phi[a]}^{(2)}+o(\rho^4),\label{eq : b1}\\ & & f_{\psi\psi[a]}\simeq f_{\psi\psi[a]}^{(0)}+o(\rho^2),\label{eq : b2}\\ & & f_{\psi\phi[a]}\simeq\rho^2f_{\psi\phi[a]}^{(2)}+o(\rho^4),\label{eq : b3}\\ & & \omega_{\phi[a]}\simeq \omega_{\phi[a]}^{(0)}+o(\rho^2),\label{eq : b4}\\ & & \omega_{\psi[a]}\simeq \omega_{\psi[a]}^{(0)}+o(\rho^2),\label{eq : b5}\end{aligned}\ ] ] where the boundary condition ( [ eq : b1 ] ) comes from the requirement that @xmath40,@xmath173 $ ] is the @xmath66-invariant plane . the regularity on the invariant plane , i.e. , the finiteness of the scalar fields , requires the other conditions ( [ eq : b2])-([eq : b5 ] ) . hence , for @xmath161 , @xmath167 behaves as @xmath174}^{(2)}f_{\phi\phi[1]}^{(2)}f_{\phi\phi[0]}^{(2)}\rho^3 } + \frac{1}{f_{\phi\phi[0]}^{(2)2}f_{\phi\phi[1]}^{(2)2}f_{\psi\psi[0]}^{(0)2}f_{\psi\psi[1]}^{(0)}\rho}\nonumber\\ & & \times\biggl [ \left(\stackrel{\odot}{{\omega_{\phi}}^{(0)}}\right)^2\left ( -f_{\phi\phi[1]}^{(2)}f_{\psi\psi[1]}^{(0)}f_{[0]}^{(4)}-f_{\phi\phi[0]}^{(2)}f_{\psi\psi[0]}^{(0)}f_{[1]}^{(4)}-f_{\phi\phi[0]}^{(2)}f_{\phi\phi[1]}^{(2)}f_{\psi\psi[0]}^{(2)}f_{\psi\psi[1]}^{(0)}\right)\nonumber\\ & & + \left(\stackrel{\odot}{{\omega_{\psi}}^{(0)}}\right)^2f_{\phi\phi[0]}^{(2)}f_{\phi\phi[1]}^{(2)2}f_{\psi\psi[0]}^{(0)}+2\stackrel{\odot}\omega_\phi^{(0)}f_{\phi\phi[0]}^{(2)}f_{\phi\phi[1]}^{(2)}f_{\psi\psi[0]}^{(0)}(\stackrel{\odot}\omega_\phi^{(2)}f_{\psi\psi[1]}^{(0)}-\stackrel{\odot}\omega_\psi^{(0)}f_{\psi\phi[1]}^{(2 ) } ) \biggr]+o(\rho)\nonumber \\ & = & o(\rho),\end{aligned}\ ] ] where we also used eq.([eq : twist ] ) for the two solutions @xmath123}$ ] and @xmath124}$ ] with the same angular momenta . \(iii ) horizon : @xmath175 , @xmath176 . the regularity on the horizon requires the following behavior of the five - scalar fields for @xmath177 : @xmath172}\simeq f_{\phi\phi[a]}^{(0)}+o(\rho^2),\\ & & f_{\psi\psi[a]}\simeq f_{\psi\psi[a]}^{(0)}+o(\rho^2),\\ & & f_{\psi\phi[a]}\simeq f_{\psi\phi[a]}^{(0)}+o(\rho^2),\\ & & \omega_{\phi[a]}\simeq \omega_{\phi[a]}^{(0)}+o(\rho^2),\\ & & \omega_{\psi[a]}\simeq \omega_{\psi[a]}^{(0)}+o(\rho^2).\end{aligned}\ ] ] therefore , @xmath167 on the horizon behaves as @xmath178 \(iv ) infinity : @xmath179 with @xmath180 finite @xmath181 . from the analysis in ref . @xcite , the asymptotic flatness requires that the metric in the canonical coordinates behaves as @xmath172}\simeq ( \sqrt{\rho^2+z^2}-z)\left(1+\frac{\tilde f_{\phi\phi[a]}^{(2)}}{2\sqrt{\rho^2+z^2}}\right)+o\left(\frac{1}{\rho^2+z^2}\right),\\ & & f_{\psi\psi[a]}\simeq ( \sqrt{\rho^2+z^2}+z)\left(1+\frac{\tilde f_{\psi\psi[a]}^{(2)}}{2\sqrt{\rho^2+z^2}}\right)+o\left(\frac{1}{\rho^2+z^2}\right),\\ & & f_{\psi\phi[a]}\simeq \zeta\frac{\rho^2}{(\rho^2+z^2)^{3/2}}+o\left(\frac{1}{\rho^2+z^2}\right),\\ & & \omega_{\phi[a]}\simeq \tilde \omega_{\phi[a]}^{(0)}+o\left(\frac{1}{\sqrt{\rho^2+z^2}}\right),\\ & & \omega_{\psi[a]}\simeq \tilde \omega_{\psi[a]}^{(0)}+o\left(\frac{1}{\sqrt{\rho^2+z^2}}\right)\end{aligned}\ ] ] for @xmath182 with @xmath180 finite , where @xmath183 is a gauge - invariant constant . for two solutions with the same masses @xmath74 , the functions @xmath184}^{(2)}$ ] and @xmath185}^{(2)}$ ] can be written in the form @xmath186}^{(2)}=\frac{4}{3\pi}(m+\eta),\quad \tilde f_{\psi\psi[a]}^{(2)}=\frac{4}{3\pi}(m-\eta),\end{aligned}\ ] ] where the constant @xmath187 is not gauge - invariant . the ambiguity in the choice of the parameter @xmath187 means that the coordinate @xmath36 is uniquely determined up to the translation @xmath188 as far as the coordinate @xmath35 conjugate to @xmath36 is fixed at the infinity . since in our proof we choose the coordinate @xmath36 such that the horizons are located on @xmath54 $ ] for two configurations @xmath123}$ ] and @xmath124}$ ] , we choose the same values of @xmath187 for the two solutions . under the choice of this gauge , @xmath167 behaves as @xmath189\frac{\rho}{\rho^2+z^2}+o\left(\frac{1}{\rho^2+z^2}\right)\\ & = & o\left(\frac{1}{\rho^2+z^2}\right).\end{aligned}\ ] ] in the neighborhood of the infinity . hence in the neighborhood of @xmath146 , @xmath190 from ( i)-(iv ) , the boundary integral ( [ eq : integral ] ) vanishes on each segment of the rod and the infinity . the deviation matrix @xmath129 is constant and has asymptotic behavior as @xmath191 . hence @xmath129 vanishes over @xmath143 . thus the two configurations @xmath123}$ ] and @xmath124}$ ] coincides with each other . the rigidity theorem of the four - dimensional black hole space - times states that asymptotically flat , stationary , analytic spacetimes assure the existence of a axial killing vector field @xcite . recently , the rigidity theorem was generalized to higher - dimensional space - times @xcite . in particular , the rigidity theorem in five - dimensions guarantees at least one axial killing vector field . as conjectured by reall @xcite , there may exist black hole solutions admitting only two commuting killing vector fields , although all of five - dimensional stationary black hole solutions found so far have three mutually commuting killing vector fields . hence , as this stage , it is natural to concentrate on asymptotically flat black hole solutions to the five - dimensional vacuum einstein equations admitting three commuting killing vector fields . one of authors and ida @xcite showed that the only black hole solution with a regular event horizon homeomorphic to @xmath8 is the five - dimensional myers - perry black hole solution in this class . however , it is impossible to generalize this theorem to the solutions with horizon topology @xmath0 since there are two black ring solutions with different shapes for the same mass and the same angular momenta @xcite . hence , we must introduce some additional geometrical information in order to consider the uniqueness of black rings as the boundary value problem . for instance , one of the candidates is rod structure introduced by harmark @xcite . as mentioned in sec.[sec : boundary ] , the black ring solution have the following rod structure : @xmath192 , \quad v=(0,0,1),\nonumber\\ & & ( ii ) [ z_1,z_2 ] , \quad v=(1,\omega_1,\omega_2)\nonumber\\ & & ( iii ) [ z_2,z_3 ] , \quad v=(0,0,1),\nonumber\\ & & ( iv ) [ z_3,\infty],\quad v=(0,1,0),\label{eq : rod}\end{aligned}\ ] ] which is characterized by four segments @xmath193\ ( i=0,1,\cdots,3 ; z_0=-\infty , z_4=\infty)$ ] and eigenvectors @xmath56 with respect to a zero eigenvalue of the three - dimensional matrix @xmath33 for each segments . the constants @xmath59 and @xmath60 mean the angular velocities of the horizon in the directions of @xmath61 and @xmath46 , respectively . by introducing the rod structure @xcite , hollands and yazadjiev applied the discussion in ref . @xcite to the case of non - spherical horizon topology and showed the following theorem in ref.@xcite . [ eq : thm ] consider two stationary , asymptotically flat , vacuum black objects spacetimes of the five - dimensions with commuting two axial killing vector fields and a timelike killing vector field . then , if both solutions have the same topology , the same rod structure and the values of the mass @xmath74 and angular momenta @xmath194 , @xmath195 , they are isometric . however , even if we restrict the horizon topology to @xmath0 , this theorem does not imply the uniqueness of the pomeransky - senkov black ring solution within the class of these solutions since there may exist another black ring solution admitting different rod structures without a conical singularity for the same asymptotic charges , mass and two angular momenta . our end of this article is to prove the uniqueness of the pomeransky - senkov black ring solution by showing that the black rings with the rod structures different from that of the pomeransky - senkov black ring solution have conical singularities . our proof is composed of two steps : first , we show the existence of asymptotically flat black ring solution with conical singularities and without curvature singularities such that under the condition of no conical singularity , they coincides with the pomeransky - senkov black ring solution ; next , once these black ring solution is given , using the theorem [ eq : thm ] obtained by hollands and yazadjiev , we can show the uniqueness of the black ring solution ( [ eq : metric ] ) in this class of the solutions admitting three mutually commuting killing vector fields , i.e. , a timelike killing vector field and two axial killing vector fields . however , we should note the following point . if we apply the hollands - yazadjiev s theorem to this black ring solution ( [ eq : metric ] ) , it seems to be specified by the asymptotic charges @xmath196 and the four additional parameters @xmath197 appearing in the rod data , although all of these parameters are not independent . in the proof in sec.[sec : coincidence ] the only four parameters @xmath196 and @xmath198 appear , where @xmath199 denote the mass and angular momenta , respectively , and the constant @xmath198 has the geometrical meaning of the ratio of the radius of @xmath12 to the radius of @xmath13 . in terms of these parameters , we obtain the following result : consider asymptotically flat black ring solutions to the five - dimensional vaccum einstein equations admitting three commuting killing vector fields , i.e. , two axial killing vector fields and a timelike killing vector field . then , in this class of solutions , the only solution with the horizon topology of @xmath200 is the black ring solution ( [ eq : metric ] ) specified by a mass @xmath74 , two angular momenta @xmath201 and the ratio @xmath198 of the radius of @xmath202 to the radius of @xmath203 . in particular , if we impose that the black ring solutions do not admit a conical singularity , we obtain the main result in this article : the only asymptotically flat , five - dimensional black ring solution with commuting two axial killing vector fields and a timelike killing vector field and without a conical singularity is the pomeransky - senkov solution . we thank daisuke ida and hideo kodama for useful comments . we also thank a. a. pomeransky and r. a. senkov for informing us of the seed solution generating the black ring solution with two angular momenta . the work of yy is supported by the grant - in aid for scientific research ( no . 19540304 and no . 19540098 ) from japan ministry of education . the one - form @xmath27 appeared in the general black ring metric ( [ eq : metric ] ) is defined by @xmath204 where @xmath205 { \beta}^{2 } \right . \nonumber\\ & & \left . + \left [ 8 \ty { \tx}^{2}{\sq}^{3}{q}^{3 } \left ( b-1 \right ) \left ( 1+b \right ) ^{2 } \left ( 1+y \right ) { \alphazero}^{3 } -4 b\sq{3 } { q}^{2}{\sq}^{2 } \left ( 1+b \right ) \times \right.\right . \nonumber\\ & & \quad \left.\left.\left ( 8 \ty q\am \sq{1 } \left ( b-1 \right ) \left ( x+1 \right ) + 2 \bpy { \tx}^{2}\ty { \sq{2}}^{2 } -2 \ap \am { \sq{1}}^{2 } \left ( b-1 \right)+\sq{3 } \ty { \tx}^{2}{\sq{2}}^{2 } \left ( 1+b \right ) \left ( 1+y \right ) \right ) { \alphazero}^{2 } \right.\right . \nonumber\\ & & \quad \left.\left . + 2 { b}^{2}q\sq { \sq{1}}^{2}\sq{2 } { \sq{3}}^{3 } \left ( 1+b \right ) \left ( -{\sq{1}}^{2}\sy { \sx}^{2 } \left ( y-1 \right ) + 2 \ap \am \sq{2}+8 \ap \sy q \left ( x+1 \right ) \right ) \alphazero \right.\right . \nonumber\\ & & \quad \left.\left . + { b}^{3}{\sx}^{2}{\sq{1}}^{4}\sy { \sq{3}}^{5}{\sq{2}}^{2}\sq{0 } \left ( 1+y \right ) \right ] \beta \right . \nonumber\\ & & \left . -8 { \sq{3}}^{3}b{q}^{3}\sq{2 } { \sq}^{2}\tx \left ( 1+b \right ) ^{2 } \left ( 1+y \right ) ^{2 } \left ( x+1 \right ) { \alphazero}^{2}-4 { b}^{2}{\sq{3}}^{5}{q}^{2}{\sq{2}}^{2}\sq \sx \sq{1 } \left ( 1+y \right ) ^{2 } \left ( 1+b \right ) \left ( x+1 \right ) \alphazero \right\ } \nonumber , \\ & & \omega_{\psi}= -\frac{b \sq \left ( 1+b \right ) ^{2 } { q}^{2 } { \sq{2}}^{2}\sq{3 } k \left ( 1-x \right ) \left ( 1+x \right ) } { 4 \sq{1 } { h(y , x ) } } \sqrt{\frac { \alphazero}{\left ( 1+b \right ) { \cy}^{3}\beta } } \times \nonumber\\ & & \frac { \left ( 4 { q}^{2}{\sq}^{2 } \left ( b-1 \right ) \left ( 1+b \right ) { \alphazero}^{2}+{b}^{2}{\sq{2}}^{2}{\sq{0}}^{2}{\sq{1}}^{2}{\sq{3}}^{2 } \right ) } { \left ( 2 \left ( 1+b \right ) q \sq \alphazero+b{\sq{0}}^{2}{\sq{1}}^{2 } \right ) ^{2 } } \left ( 4 { q}^{2}{\sq}^{2 } \left ( b-1 \right ) \left ( 1+b \right ) \left ( -q+2 y+\lambda \right ) \ty { \alphazero}^{2 } \right . \nonumber\\ & & \left . + 4 bq\sq{1 } \sq{3 } \sq \left ( y{\lambda}^{2}+2 \lambda+2 { y}^{2}\lambda+4 y-{q}^{2}y \right ) \left ( \sq+4 bq \right ) \alphazero + { b}^{2}\sq{0 } \sq{2 } { \sq{3}}^{3}{\sq{1}}^{3 } \left ( \lambda+q+2 y \right ) \sy \right)\nonumber .\end{aligned}\ ] ] the functions @xmath26 are defined as : @xmath206 \beta -4 \cmy b{q}^{2}\sq { \sq{3}}^{4 } \left ( x+1 \right ) ^{2 } \left ( 1+b \right ) \left ( 1+y \right ) \alphazero \right\},\nonumber \\ j(x , y)&= & \frac{-1}{8 \left ( x - y \right ) { \sq{1}}^{2}{\cy}^{2 } k } \sqrt { \frac { \sq \left ( 1-b \right ) } { 2 b \left ( b\sq+4 q \right ) \left ( 1+b \right ) \alphazero \beta } } \left\ { \left ( b\sq+4 q \right ) { k}^{3 } \left ( 1+b \right ) \alphazero q\sq{3 } \left ( 1+y \right ) \left ( x+1 \right ) \right . \nonumber\\ & & \left . \left ( 8 b{q}^{2}\sq { \sq{1}}^{2}\alphazero \left ( b-1 \right ) \left ( y-1 \right ) \left ( x-1 \right ) \left ( 2 q \left ( b-1 \right ) \sq \left ( \sq{3 } \dxy \left ( 1+b \right ) -\sq{1 } { \sq{0}}^{2}\ty \tx \right ) \alphazero \right.\right.\right . \nonumber\\ & & \left.\left.\left . -\sq{3 } b{\sq{1}}^{2}\sq{0 } \left ( \sq{0 } \dxy \left ( b-1 \right ) + \sq{2 } { \sq{3}}^{2}\sy \sx \right ) \right ) { \beta}^{2 } \right.\right . \nonumber\\ & & \left.\left . + \left ( -8 { q}^{3}\ty \tx { \sq}^{3 } \left ( 1+b \right ) \left ( b-1 \right ) ^{2 } \left ( \sq{3 } \am \left ( 1+b \right ) -4 q \left ( y-1 \right ) \left ( x-1 \right ) \sq{1 } \right ) { \alphazero}^{3 } \right.\right.\right . \nonumber\\ & & \left.\left.\left . -4 { q}^{2}{\sq{3}}^{3}\sx { \sq}^{2}b\sq{1 } \sy \left ( b-1 \right ) \left ( 1+b \right ) \left ( \sq{1 } \ap \left ( b-1 \right ) -2 \sq{2 } \ty \tx \right ) { \alphazero}^{2 } \right.\right.\right . \nonumber\\ & & \left.\left.\left . + 2 q \left ( b-1 \right ) \ty { \sq{3}}^{3}{b}^{2}\sq { \sq{1}}^{2}\sq{2 } { \sq{0}}^{2}\tx \left ( \sq{2 } \am \left ( 1+b \right ) + 2 \sq{1 } \sy \sx \right ) \alphazero \right.\right.\right . \nonumber\\ & & \left.\left.\left . + { \sq{3}}^{5}\sx { b}^{3}{\sq{2}}^{2}{\sq{1}}^{4}\sy \sq{0 } \left ( \sq{0 } \ap \left ( b-1 \right ) -4 q \left ( y-1 \right ) \left ( x-1 \right ) \sq{2 } \right ) \right ) \beta \right.\right . \nonumber\\ & & \left.\left . + \left ( 1+b \right ) \sq{3 } \left ( 2 q\sq \left ( b-1 \right ) \alphazero - b{\sq{2}}^{2}{\sq{3}}^{2 } \right ) \left ( 4 \ty { q}^{2}\am { \sq}^{2}\tx \left ( b-1 \right ) \left ( 1+b \right ) { \alphazero}^{2 } \right.\right.\right . \nonumber\\ & & \left.\left.\left . -4 bq \sq \sq{1 } \sq{2 } { \sq{3}}^{2 } \sx \sy \ty \tx \alphazero -{b}^{2 } { \sq{1}}^{2}{\sq{2}}^{2}{\sq{3}}^{4 } \sx \sy \ap \right ) \right ) { \sq{2}}^{2}b \right\}\nonumber\end{aligned}\ ] ] @xmath207 \right\ } \nonumber\\ & & \times \left ( 4 { q}^{2 } \sq \left ( y-1 \right ) \ty \left ( b-1 \right ) \left ( x-1 \right ) \bpx \alphazero \beta + b{\sq{3}}^{4 } \left ( x+1 \right ) \sx \sy \cmy \right ) \nonumber\\ & & -\frac{b{\sq{2}}^{2}\am \tx { k}^{2 } } { 16{\cy}^{2}\beta { \sq{1}}^{2}{\sx}^{2 } \left ( x - y \right ) { \bpx}^{2}{\ty}^{2 } } \times \nonumber\\ & & \left ( -b{\sq{1}}^{4 } \left ( x-1 \right ) \sx \sy \cpy \beta+4 { q}^{2}\sq \left ( 1+y \right ) \ty \left ( 1+b \right ) \left ( x+1 \right ) \bpx \alphazero \right ) \times \nonumber\\ & & \left\ { -128 \bpx \bpy { q}^{4}{\alphazero}^{2}{\sq}^{2}\ty \left ( b-1 \right ) ^{2 } \left ( x-1 \right ) ^{2 } \left ( b\sq+4 q \right ) { \beta}^{2 } \right . \nonumber\\ & & \left . + 16 \sq { \sq{3}}^{2}{q}^{2}\sx \alphazero \left ( b-1 \right ) \left ( x-1 \right ) \left ( b\sq+4 q \right ) \left [ 2 q\sq \ty \bpx \left ( b-1 \right ) \left ( 1+b \right ) \left ( 1+y \right ) \alphazero \right.\right . \nonumber\\ & & \left.\left . -{\sq{3}}^{2}b \left ( { \sq{2}}^{2}\bpx \left ( 1+b \right ) \left ( 1+y \right ) \ty+b{\sy}^{2}{\sq{1}}^{2 } \left ( b-1 \right ) \left ( x+1 \right ) \right ) \right ] \beta \right . \nonumber\\ & & \left . -{\sq{3}}^{4}\bpx \cmy { \sx}^{2 } \left ( 1+b \right ) \left ( 2 q\sq \left ( b-1 \right ) \alphazero - b{\sq{2}}^{2}{\sq{3}}^{2 } \right ) ^{2 } \left ( 1+y \right ) \right\ } \nonumber\\ & & + \frac{{k}^{2 } \left ( 1+b \right ) \am \ap \tx q { \sq{2}}^{2}b \left ( b-1 \right ) } { 4 \left ( x - y \right ) { \bpx}^{2}{\ty}^{2}{\sx}^{2}{\sq{1}}^{2}{\cy}^{2 } } \left ( 2 \alphazero q\sq \ty \bpx - b{\sq{1}}^{2}{\sq{3}}^{2}\sy \sx \right ) \times \nonumber\\ & & \left\ { 4 \sq \alphazero q \left ( b-1 \right ) \left ( x-1 \right ) \left ( b\sq+4 q \right ) \left [ 2 \sq q\ty \bpx \left ( -8 \bpy q \left ( x+1 \right ) + { \sq{1}}^{2 } \left ( b-1 \right ) \left ( y-1 \right ) \sx \right ) \alphazero \right.\right . \nonumber\\ & & \quad \left.\left . -b\sx { \sq{3}}^{2}{\sq{1}}^{2 } \left ( { \sq{2}}^{2}\ty \left ( y-1 \right ) \bpx+2 { \sy}^{2}\sq{1 } \sq{3 } b \left ( x+1 \right ) \right ) \right ] \beta \right . \nonumber\\ & & \left . + { \sq{3}}^{2}\bpx \sx \left ( 1+b \right ) \left ( 1+y \right ) \left ( 2 q\sq \left ( b-1 \right ) \alphazero - b{\sq{2}}^{2}{\sq{3}}^{2 } \right ) \right . \times \nonumber\\ & & \quad \left . \left [ 2 q\sq \left ( 2 \left ( b\sq+4 q \right ) \left ( x+1 \right ) \ty+{\sq{1}}^{2 } \left ( b-1 \right ) \left ( y-1 \right ) \sx \right ) \alphazero - b\sx { \sq{1}}^{2}{\sq{2}}^{2}{\sq{3}}^{2 } \left ( y-1 \right ) \right ] \right\ } \nonumber\\ & & -\frac{2 { k}^{2}\sy { \tx}^{2 } { \it h(x , y)}}{\sq \cy \left ( x - y \right ) ^{2}\ty \bpx \sx } \times \nonumber\\ & & \left\ { 32 { q}^{2}\left ( b\sq+4 q \right ) \bpy \left ( x-1 \right ) \left ( x+1 \right ) \sy b\sq \left ( b-1 \right ) \alphazero \beta \right . \nonumber\\ & & \quad \left . + \bpx \sx \left ( y-1 \right ) \left ( 1+y \right ) \left ( 1+b \right ) \left ( 2 q\sq \left ( b-1 \right ) \alphazero - b{\sq{2}}^{2}{\sq{3}}^{2 } \right ) ^{2 } \right\ } , \nonumber\end{aligned}\ ] ] where the several polynomials are defined by @xmath208 and the constants are defined by @xmath209 apparently , here are seven parameters @xmath210 . these parameters obey the following three relations : @xmath211 thus the four of them are independent . for review , m. heusler , _ black hole uniqueness theorems _ , ( cambridge university press , cambridge , 1996 ) . w. israel , phys . rev . * 164 * , 1776 ( 1967 ) . w. israel , commun . * 8 * , 245 ( 1968 ) . g. l. bunting and a. k. m. masood - ul - alam , gen . rel . grav . * 19 * , 147 ( 1987 ) . a. k. m. masood - ul - alam , class . * 9 * , l53 ( 1992 ) . r. schoen and s .- t . yau , commun . * 65 * , 45 ( 1979 ) . e. witten , commun . * 80 * , 381 ( 1981 ) . b. carter , phys . * 26 * , 331 ( 1971 ) . d. c. robinson , phys . rev . * 10 * , 458 ( 1974 ) . d. c. robinson , phys . * 34 * , 905 ( 1975 ) . p. o. mazur , j. phys . * 15 * , 3173 ( 1982 ) . t. banks and w. fischler , arxiv : hep - th/9906038 ; + s. b. giddings and s. d. thomas , phys . d * 65 * , 056010 ( 2002 ) ; + s. dimopoulos and g. landsberg , phys . lett . * 87 * , 161602 ( 2001 ) . see for review , p. kanti , int . j. mod . a*19 * , 4899 ( 2004 ) . c. helfgott , y. oz and y. yanay , jhep 0602 , 025 ( 2006 ) . g. j. galloway and r. schoen , to appear in commun . arxiv : gr - qc/0509107 . s. w. hawking , commun . phys . * 25 * , 152 ( 1972 ) . s. w. hawking and g. f. r. ellis , _ the large scale structure of space - time _ ( cambridge university press , 1973 ) . r. emparan and h. s. reall , phys . lett . * 88 * , 101101 ( 2002 ) . r. c. myers and m. j. perry , annals phys . * 172 * , 304 ( 1986 ) . t. mishima and h. iguchi , phys . d * 73 * , 044030 ( 2006 ) . pomeransky , r.a . senkov , arxiv : hep - th/0612005 . v. a. belinskii and v. e. zakharov , sov . phys . jetp * 48 * , 985 ( 1978 ) ; + v. a. belinskii and v. e. zakharov , sov . jetp * 50 * , 1 ( 1979 ) . v. a. belinski and e. verdaguer , _ gravitational solitons _ , ( cambridge university press , 2001 ) . s. tomizawa , y. morisawa and y. yasui , phys . d * 73 * , 064009 ( 2006 ) . s. tomizawa and m. nozawa , phys . d * 73 * , 124034 ( 2006 ) . s. tomizawa , h. iguchi and t. mishima , phys . d * 74*,104004 ( 2006 ) . s. tomizawa , h. iguchi and t. mishima , arxiv : hep - th/0702207 . h. iguchi and t. mishima and s. tomizawa , arxiv:0705.2520[hep - th ] . a. pomeransky , phys . rev d * 73 * , 044004 ( 2006 ) . t. koikawa , prog . * 114 * , 793 ( 2005 ) . t. azuma and t. koikawa , arxiv : hep - th/0512350 . h. iguchi and t. mishima , arxiv : hep - th/0604050 . j. castejon - amenedo and v. s. manko , phys . d * 41 * , 2018 ( 1990 ) . g. w. gibbons , d. ida and t. shiromizu , prog . . suppl . * 148 * , 284 ( 2003 ) ; + g. w. gibbons , d. ida and t. shiromizu , phys . * 89 * , 041101 ( 2002 ) . a. ishibashi and h. kodama , prog . phys . * 110 * , 901 ( 2003 ) . y. morisawa and d. ida , phys . d * 69 * , 124005 ( 2004 ) . t. harmark , phys . d * 70 * , 124002 ( 2004 ) ; t. harmark and p.olesen , phys . d * 72 * , 124017 ( 2005 ) . s. w. hawking , commun . phys . * 25 * , 152 ( 1972 ) . s. hollands , a. ishibashi and r. m. wald , commun . phys . * 271 * , 699 ( 2007 ) . h. s. reall , phys . d * 68 * , 024024 ( 2003 ) . s. hollands and s. yazadjiev , arxiv:0707.2775[gr - qc ] .

in this paper we introduce a new schwarz framework and theory , based on the well - known idea of space decomposition , for nonsymmetric and indefinite linear systems arising from continuous and discontinuous galerkin approximations of general nonsymmetric and indefinite elliptic partial differential equations . the proposed schwarz framework and theory are presented in a variational setting in banach spaces instead of hilbert spaces which is the case for the well - known symmetric and positive definite ( spd ) schwarz framework and theory . condition number estimates for the additive and hybrid schwarz preconditioners are established . the main idea of our nonsymmetric and indefinite schwarz framework and theory is to use weak coercivity ( satisfied by the nonsymmetric and indefinite bilinear form ) induced norms to replace the standard bilinear form induced norm in the spd schwarz framework and theory . applications of the proposed nonsymmetric and indefinite schwarz framework to solutions of discontinuous galerkin approximations of convection - diffusion problems are also discussed . extensive 1-d numerical experiments are also provided to gauge the performance of the proposed schwarz methods . schwarz methods and preconditioners , domain decomposition , space decomposition , inf - sup condition , strong and weak coercivity , condition number estimates . 65n55 , 65f10

1308.3211

the original schwarz method , proposed and analyzed by hermann schwarz in 1870 @xcite , is an iterative method to find the solution of a partial differential equation ( pde ) on a complicated domain which is the union of two overlapping simpler subdomains . the method solves the equation on each of the two subdomains by using the latest values of the approximate solution as the boundary conditions on the parts of the subdomain boundaries which are inside of the given domain . the idea of splitting a given problem posed on a large ( and possibly complicated ) domain into several subproblems posed on smaller subdomains and then solving the subdomain problems either sequentially or in parallel is a very appealing idea . divide - and - conquer " idea is at the heart of every domain decomposition or schwarz method . it is well - known that @xcite the domain decomposition strategy can be introduced at the following three different levels : the continuous level for pde analysis as proposed and analyzed by hermann schwarz in 1870 , the discretization level for constructing ( hybrid and composite ) discretization methods , and the algebraic level for solving algebraic systems arising from the numerical approximations of pde problems . these three levels are often interconnected , and each of them has its own merit to be studied . most of the recent efforts and attentions have been focused on the algebraic level . the field of domain decomposition methods has blossomed and undergone intensive and phenomenal development during the last thirty years ( cf . @xcite and the references therein ) . the phenomenal development has largely been driven by the ever increasing demands for fast solvers for solving important and complicated scientific , engineering and industrial application problems which are often governed mathematically by a pde or a system of pdes . it has also been infused and facilitated by the rapid advances in computer hardware and the emergence of parallel computing technologies . at the algebraic level , domain decomposition methods or schwarz methods have been well developed and studied for various numerical approximations ( discretizations ) of many types of pde problems including finite element methods ( cf . @xcite ) , mixed finite element methods and spectral methods ( cf . @xcite ) , and discontinuous galerkin methods ( cf . @xcite ) . a general abstract framework , backed by an elegant convergence theory , was well established many years ago for symmetric and positive definite ( spd ) pde problems and their numerical approximations ( cf . @xcite and the references therein ) . despite the tremendous advances in domain decomposition ( schwarz ) methods over the past thirty years , the current framework and convergence theory are mainly confined to spd problems in hilbert spaces . because the framework and especially the convergence theory indispensably rely on the spd properties of the underlying problem and the hilbert space structures , they do not apply to genuinely nonsymmetric and/or indefinite problems . as a result , the spd framework and theory leave many important and interesting problems uncovered as pointed out in @xcite . this paper attempts to address this important issue in schwarz methods . the goal of this paper is to introduce a new schwarz framework and theory , based on the well - known idea of space decomposition as in the spd case , for nonsymmetric and indefinite linear systems arising from continuous and discontinuous galerkin approximations of general nonsymmetric and indefinite elliptic partial differential equations under some minimum " structure assumptions . unlike the spd framework and theory , our new framework and theory are presented in a variational setting in banach spaces instead of hilbert spaces . such a general framework allows broader applications of schwarz methods . both additive schwarz and multiplicative as weel as hybrid schwarz methods are developed . a comprehensive convergence theory is provided which includes condition number estimates for the additive schwarz preconditioners and hybrid schwarz preconditioners . the main idea of our nonsymmetric and indefinite schwarz framework and theory is to use weak coercivity ( satisfied by the nonsymmetric and indefinite bilinear form ) induced norms to replace the standard bilinear form induced norm in the spd schwarz framework and theory ( see sections [ sec-2][sec-4 ] for a detailed exposition ) . as expected , working with such weak coercivity induced norms and nonsymmetric and indefinite bilinear forms is quite delicate . it requires new and different technical tools in order to establish our convergence theory . the remainder of this paper is organized in the following way . in section [ sec-2 ] , we introduce notation , the functional setting , and the variational problems which we aim to solve . section 2 also contains some further discussions on the main idea of the paper . section [ sec-3 ] is devoted to establishing an abstract additive schwarz , multiplicative schwarz , and hybrid schwarz framework for general nonsymmetric and indefinite algebraic problems in a variational setting in general banach spaces . in section [ sec-4 ] , we present an abstract convergence theory for the additive and hybrid schwarz methods proposed in section [ sec-3 ] . in section [ sec-5 ] , we present some applications of the proposed nonsymmetric and indefinite schwarz framework to discontinuous galerkin approximations of convection - diffusion ( in particular , convection - dominated ) problems . we also provide extensive 1-d numerical experiments to gauge the performance of the proposed nonsymmetric and indefinite schwarz methods . let @xmath0 be a real hilbert space with the inner product @xmath1 and the induced norm @xmath2 . let @xmath3 be two reflexive banach spaces endowed with the norms @xmath4 and @xmath5 respectively . let @xmath6 be a real bilinear form defined on the product space @xmath7 and @xmath8 be a real linear functional defined on @xmath9 . we consider the following variational problem : find @xmath10 such that @xmath11 the well - posedness of the above variational problem has been extensively studied . one of such results is summarized in the following theorem : ( cf . @xcite ) [ babuska ] suppose that @xmath8 is a bounded linear functional on @xmath9 . assume that @xmath6 is continuous and _ weakly coercive _ in the sense that there exist constants @xmath12 such that @xmath13 then problem has a unique solution @xmath10 . moreover , @xmath14 [ rem2.1 ] ( a ) theorem [ babuska ] is called lax - milgram - babuka theorem in the literature ( cf . it was first introduced to the finite element context in @xcite ( also see @xcite ) . an earlier version of the theorem can also be found in @xcite . \(b ) as pointed out in @xcite , condition can be replaced by the following more restrictive condition : there exists a constant @xmath15 such that @xmath16 the above condition can be viewed as a _ weak coercivity _ condition for the adjoint bilinear form @xmath17 of @xmath6 . ( c ) weak coercivity condition is often called the _ inf - sup _ or _ babuka brezzi _ condition in the finite element literature @xcite for a different reason . it appears and plays a vital role for saddle point problems and their ( mixed ) finite element approximations ( cf . @xcite ) . \(d ) theorem [ babuska ] is certainly valid when @xmath18 . since condition is weaker than the strong coercivity , then theorem [ babuska ] is a stronger result than the classical lax - milgram theorem for the case @xmath18 . indeed , for most convection - dominated convection - diffusion problems , @xmath18 . however , there are situations where condition holds but strong coercivity fails . ( e ) there are also situations where one prefers to use different norms for the trial space @xmath19 and the test space @xmath9 even if @xmath18 . theorem [ babuska ] also provides a convenient framework to handle such a situation . as problem is posed on infinite dimensional spaces @xmath19 and @xmath9 , to solve it numerically , one must approximate @xmath19 and @xmath9 by some finite dimensional spaces @xmath20 . here @xmath21 is a positive integer which denotes the dimension of @xmath22 and @xmath23 . if one of ( or both ) @xmath22 and @xmath23 is not a subspace of its corresponding infinite dimensional space , then one also needs to provide an approximate bilinear form @xmath24 for @xmath6 so that @xmath24 is well defined on @xmath25 . in addition , if @xmath23 is not a subspace of @xmath9 one also needs to provide an approximate linear functional @xmath26 for @xmath8 so that @xmath26 is well defined on @xmath23 . once @xmath27 and @xmath26 are constructed , the galerkin method for problem is defined as seeking @xmath28 such that @xmath29 pick a basis @xmath30 of @xmath19 and a basis @xmath31 of @xmath9 . it is trivial to check that the discrete variational problem can be rewritten as the following linear system problem : @xmath32 where @xmath33_{j=1}^n$ ] is the coefficient vector of the representation of @xmath34 in terms of the basis @xmath30 and @xmath35_{i , j=1}^n , & \qquad a_{ij } & = a(\phi^{(j)},\psi^{(i)}),\\ { \mathbf{f}}&=\bigl[f^{(i)}\bigr]_{i=1}^n , & \qquad f^{(i ) } & = f(\psi^{(i)}).\end{aligned}\ ] ] the properties of matrix @xmath36 ( called a stiffness matrix ) are obviously determined by the properties of the discrete bilinear form @xmath24 and the approximate spaces @xmath22 and @xmath23 . when @xmath37 it is well known that @xcite @xmath36 is symmetric if and only if @xmath24 is symmetric and @xmath36 is positive definite provided that @xmath24 is strongly coercive on @xmath38 . in general , @xmath36 is just an @xmath39 nonsymmetric real matrix if @xmath24 is not symmetric . @xmath36 also can be indefinite ( i.e. , @xmath36 has at least one negative and one positive eigenvalue ) if @xmath24 fails to be coercive . as is a square linear system , by a well - known algebraic fact we know that has a unique solution @xmath40 provided that the stiffness matrix @xmath36 is nonsingular . this nonsingular condition on @xmath36 becomes necessary if one wants to be uniquely solvable for _ arbitrary _ vector @xmath41 . for most application problems ( such as boundary value problems for elliptic pdes ) , one needs to consider various choices of the load " functional @xmath8 , so the vector @xmath41 is practically arbitrary " in . hence , besides some deeper mathematical and algorithmic considerations , asking for the stiffness matrix @xmath36 to be _ nonsingular _ is a minimum " requirement for the discretization method to be practically useful . sufficient conditions on the discrete bilinear form @xmath24 and the approximate spaces @xmath22 and @xmath23 which infer the unique solvability of the linear system have been well studied and understood in the past thirty years . in particular , for the spd type ( algebraic ) problems arising from various discretizations of boundary value problems for elliptic pdes @xcite . in the following we shall quote some of these well - known results in a theorem which is a counterpart of theorem [ babuska ] . @xcite ) [ discrete_babuska ] suppose that @xmath26 is a bounded linear functional on @xmath23 . assume that @xmath24 is continuous and weakly coercive in the sense that there exist constants @xmath42 such that @xmath43 then problem has a unique solution @xmath28 . moreover , @xmath44 where @xmath45 is a positive constant . a few remarks are in order about the above well - posedness theorem . [ rem2.2 ] ( a ) condition is equivalent to requiring that the adjoint @xmath46 of @xmath24 is weakly coercive . \(b ) conditions are analogies of their continuous counterparts . discrete weak coercivity condition is often called the _ inf - sup _ or _ babuka brezzi _ condition in the finite element literature @xcite for a different reason . it is the most important one in a set of sufficient conditions for a mixed finite element to be stable ( cf . @xcite ) . \(c ) a numerical method which fulfills conditions is guaranteed to be uniquely solvable and stable . hence , these conditions can be used as a test stone to determine whether a numerical method is a good " method . for this reason , we shall call the numerical method an _ inf - sup preserving method _ or a _ weak coercivity preserving method _ if it satisfies . \(d ) theorem [ discrete_babuska ] focuses on the unique solvability and the stability of the numerical method not on the accuracy of the method . we like to note that method indeed is an accurate numerical method provided that approximate spaces @xmath22 and @xmath23 are accurate approximations of @xmath19 and @xmath9 ( cf . @xcite ) . as we briefly explained above , approximating the variational problem by a galerkin method certainly results in solving the linear system . it is well known that the common dimension @xmath47 of the approximation spaces @xmath22 and @xmath23 has to be sufficiently large in order for the galerkin method to be accurate . as a result , the size of the linear system ( i.e. , the size of the matrix @xmath36 ) is expected to be very large in applications . moreover , if is a variational formulation of some elliptic boundary value problem , then the stiffness matrix @xmath36 is certainly ill - conditioned in the sense that the condition number @xmath48 is very large . here @xmath49 denotes a matrix norm of @xmath36 . for example , in the case of second and fourth order elliptic boundary value problems , @xmath50 and @xmath51 , respectively , where d is the spatial dimension of the domain . consequently , it is not efficient to solve linear system directly using classical iterative methods even if they converge . furthermore , unlike in the spd case , classical iterative methods often do not converge for general nonsymmetric and indefinite linear system ( cf . @xcite ) . as a first step toward developing better iterative solvers for nonsymmetric and indefinite linear system , it is natural to design a good " preconditioner ( i.e. , an @xmath39 real matrix @xmath52 ) such that @xmath53 is well - conditioned ( i.e. , @xmath54 is relatively small , say , significantly smaller than @xmath55 ) . then one can try classical iterative methods . in particular , the generalized minimal residual ( gmres ) method can be used on the preconditioned system @xmath56 one can also develop some new ( and hopefully better ) iterative methods if classical iterative methods still do not work as well on as one had hoped . as was already mentioned in section [ sec-1 ] , the focus of this paper is exactly what is described above . our goal is to develop a new schwarz framework and theory , based on the well - known idea of space decomposition , for solving nonsymmetric and indefinite linear system which arises from the galerkin method as an approximation of the variational problem . as expected , our nonsymmetric and indefinite schwarz framework and theory are natural extensions of the well - known spd schwarz framework and theory which were nicely described in @xcite . for the sake of notational brevity , throughout the remainder of this paper we shall suppress the sub - index @xmath47 in the discrete spaces @xmath22 and @xmath23 and in discrete functions @xmath34 , @xmath57 and @xmath58 . in other words , @xmath19 and @xmath9 are used to denote @xmath22 and @xmath23 , and @xmath59 , @xmath60 and @xmath61 are used to denote @xmath34 , @xmath57 and @xmath58 . in addition , we shall make an effort below to use the same or similar terminologies , as well as space and norm notation as those in @xcite for the symmetric and positive definite ( spd ) schwarz framework and theory . we shall also make comments about notation and terminologies which have no spd counterparts and try to make links between the well known spd schwarz framework and theory and our nonsymmetric and indefinite schwarz framework and theory . to motivate , we recall that in the spd schwarz framework and theory @xcite , since @xmath18 and the discrete bilinear form @xmath24 is symmetric and strongly coercive , @xmath62 defines a convenient norm ( which is also equivalent to the @xmath63-norm ) on the space @xmath19 ( as well as on its subspaces ) . this bilinear form induced norm plays a vital role in the spd schwarz framework and theory . unfortunately , without the symmetry and strong coercivity assumptions on @xmath24 , @xmath62 is not a norm anymore when @xmath18 . it is not even well defined if @xmath64 ! to overcome this difficulty , the existing nonsymmetric and indefinite schwarz framework and theory ( cf . @xcite ) , which only deal with the case @xmath18 , assume that @xmath24 has a decomposition @xmath65 where @xmath66 is assumed to be symmetric and strongly coercive ( i.e. , it is spd ) and @xmath67 is a perturbation of @xmath66 . in this setting @xmath66 then induces an equivalent ( to @xmath63 ) norm @xmath68 and one then works with this norm as in the spd case . unfortunately and understandably , such a setting requires that @xmath67 is a _ small _ perturbation of @xmath66 , which is why the existing nonsymmetric and indefinite schwarz framework and theory only apply to nearly " spd problems . hence , it leaves more interesting and more difficult nonsymmetric and indefinite problems unresolved . to develop a new schwarz framework and theory for general nonsymmetric and indefinite problems , our only assumptions on the discrete problem are those stated in the well - posedness theorem [ discrete_babuska ] . we now restate those assumptions on the discrete bilinear form @xmath24 and its adjoint @xmath46 using the new function and space notation ( i.e. , after suppressing the sub - index @xmath47 ) as follows : * main assumptions * * _ continuity _ there exists a positive constant @xmath69 such that @xmath70 , @xmath71 such that @xmath72 [ rem3.1 ] ( a ) since @xmath73 , then the continuity condition is equivalent to @xmath74 \(b ) assumptions ( ma@xmath75 ) and ( ma@xmath76 ) impose some restrictions on the underlying galerkin method . but as we noted in remark [ rem2.2 ] , these are some minimum " conditions for the galerkin method to be practically useful . from that point of view , ( ma@xmath75 ) and ( ma@xmath76 ) are not restrictions at all . as it was pointed out in the previous subsection , for a general nonsymmetric and indefinite problem , since the discrete bilinear form @xmath24 is not strongly coercive , then @xmath77 is not a norm anymore . in fact , @xmath77 may not even be defined if @xmath64 . so a crucial question is what norms ( if any ) would @xmath24 induce on @xmath19 and @xmath9 which are equivalent to @xmath4 and @xmath5 . it turns out that @xmath24 does induce equivalent norms on both @xmath19 and @xmath9 , and these norms are hidden in the weak coercivity conditions and ! this key observation leads to the main idea of this paper ; that is , we define the following weak coercivity induced norms : @xmath78 assumptions ( ma@xmath75 ) and ( ma@xmath76 ) immediately infer the following norm equivalence result . since its proof is trivial , we omit it . [ lem3.1 ] the following inequalities hold : @xmath79 we conclude this subsection by noting that the variational setting laid out so far is a banach space setting . no hilbert space structure is required for the space @xmath19 and @xmath9 . this is not only mathematically interesting but also practically valuable because for some pde application problems it is imperative to work in a banach space setting . we also note that if @xmath18 and @xmath24 is spd ( i.e. , it is symmetric and strongly coercive ) , then @xmath80 . hence , we recover the standard bilinear form induced norm ! it is well known @xcite that schwarz domain decomposition methods can be presented abstractly in the framework of the space decomposition method . in particular , the physical domain decomposition provides a practical and effective way to construct the required space decomposition and local solvers in the method . to some extent , the space decomposition method to the schwarz domain decomposition method is what the lu factorization is to the classical gaussian elimination method . like in the spd schwarz framework @xcite ) , there are two essential ingredients in our nonsymmetric and indefinite schwarz framework , namely , ( i ) _ construction of a pair of compatible " space decompositions for @xmath19 and @xmath9 _ and ( ii ) _ construction of a local solver ( or local discrete bilinear form ) on each pair of local spaces_. however , there is an obvious and crucial difference between the spd schwarz framework and our nonsymmetric and indefinite schwarz framework . when @xmath64 , our framework requires space decompositions for both spaces @xmath19 and @xmath9 , and these two space decompositions must be chosen compatibly in the sense to be described below let @xmath81 be two sets of reflexive banach spaces with norms @xmath82 and @xmath83 respectively . we note that @xmath84 and @xmath85 are used to denote the so - called _ coarse spaces _ in the domain decomposition context . for @xmath86 , let @xmath87 denote some _ prolongation operators_. [ rem3.3 ] in the schwarz method literature ( cf . @xcite ) , @xmath88 is often used to denote both the prolongation operator from @xmath89 to @xmath19 and its matrix representation . such a choice of notation is due to the fact that the matrix representation of the _ not - explicitly - defined _ restriction operator @xmath90 from @xmath19 to @xmath89 is always chosen to be the transpose of the matrix representation of the prolongation operator . as expected , such a dual role notation may be confusing to some readers . to avoid such a potential confusion we use different notations for operators and their matrix representations throughout this paper . we also like to note that in the construction of all schwarz methods the restriction operators / matrices are not primary " operators / matrices but derivative " operators / matrices in the sense that they are not chosen independently . instead , they are determined by the prolongation operators / matrices . one often first defines the matrix representation of the ( desired ) restriction operator as the transpose of the the matrix representation of the prolongation operator and then defines the restriction operator to be the unique linear operator which has the chosen matrix representation ( under the same bases in which the prolongation matrix is obtained ) . this will also be the approach adopted in this paper for defining our restriction operators ( see definition [ def3.2 ] ) . clearly , such a definition of the restriction operators is not only abstract but also depends on the choices of the bases of the underlying function spaces . however , its simplicity and convenience at the matrix level , which are what really matter in practice , make the definition appealing and favorable so far . suppose that the following relations hold : @xmath91 where @xmath92 and @xmath93 stand for the ranges of the linear operators @xmath94 and @xmath95 respectively . associated with each pair of local spaces @xmath96 for @xmath97 , we introduce a local discrete bilinear form @xmath98 defined on @xmath99 , which can be taken either as the restriction of global discrete bilinear form @xmath100 on @xmath99 or as some approximation of the restriction of @xmath100 on @xmath99 . we call these two choices of local discrete bilinear form @xmath98 an _ exact local solver _ and an _ inexact local solver _ respectively . after the local discrete bilinear forms are chosen , we can define what constitutes as a _ compatible space decomposition_. [ def3.1 ] ( i ) a pair of spaces @xmath89 and @xmath101 are said to be compatible with respect to @xmath102 if they satisfy the following conditions : * _ local continuity_. there exists a positive constant @xmath103 such that @xmath104 * _ local weak coercivity_. there exist positive constants @xmath105 and @xmath106 such that @xmath107 \(ii ) a pair of space decompositions @xmath108 and @xmath109 of @xmath19 and @xmath9 satisfying are said to be compatible if each pair of @xmath89 and @xmath101 is compatible with respect to @xmath102 for @xmath110 . obviously conditions ( la@xmath75 ) and ( la@xmath76 ) on @xmath98 are the analogies of ( ma@xmath75 ) and ( ma@xmath76 ) on @xmath100 . by theorem [ discrete_babuska ] , these conditions guarantee that the local problem of seeking @xmath111 such that @xmath112 is uniquely solvable for any given bounded linear functional @xmath113 on @xmath101 . moreover , ( la@xmath75 ) and ( la@xmath76 ) are minimum " conditions for achieving such a guaranteed unique solvability ( cf . remark [ rem2.2 ] ) . furthermore , like its global counterpart , the local weak coercivity condition ( la@xmath76 ) induces the following two equivalent norms on @xmath89 and @xmath101 : @xmath114 where @xmath115 for any @xmath116 . trivially , we have [ lem3.2 ] suppose that @xmath89 and @xmath101 are compatible with respect to @xmath98 . then the following inequalities hold : @xmath117 throughout this section , we assume that we are given a global discrete problem , and the global discrete bilinear form @xmath100 fulfills the main assumptions ( ma@xmath75 ) and ( ma@xmath76 ) so that problem has a unique solution @xmath10 . in addition , we assume we are given a pair of space decompositions @xmath108 and @xmath109 of @xmath19 and @xmath9 , the prolongation operators @xmath118 and @xmath119 , and the local discrete bilinear forms @xmath120 such that the given space decompositions are compatible with respect to the given local discrete bilinear forms in the sense of definition [ def3.1 ] . our goal in this subsection is to construct the additive schwarz method for problem using the given information . to continue , we now introduce two sets of _ projection - like _ operators @xmath121 and @xmath122 for @xmath123 . these projection - like - operators will serve as the building blocks for the constructions of both our additive and multiplicative schwarz methods . for any fixed @xmath124 and @xmath125 , define @xmath126 and @xmath127 by @xmath128 we recall that @xmath129 for all @xmath130 and @xmath131 . we also note that since @xmath89 and @xmath101 are assumed to be compatible , theorem [ discrete_babuska ] then ensures both @xmath132 and @xmath133 are well defined for @xmath134 . since @xmath89 and @xmath101 may not be subspaces of @xmath19 and @xmath9 , @xmath135 and @xmath136 may not belong to @xmath19 and @xmath9 . to pull them back to the global discrete spaces @xmath19 and @xmath9 , we appeal to the _ prolongation operators _ @xmath94 and @xmath95 for help . define the composite operators @xmath137 trivially , we have @xmath138 and @xmath139 for @xmath140 . we now are ready to define the following additive schwarz operators . following @xcite we define @xmath141 the matrix interpretation of the additive operator @xmath142 is similar to but slightly more complicated than the one in the spd schwarz framework . in particular , the additive operator @xmath143 does not exist in the the spd framework . for the reader s convenience , we give below a brief matrix interpretation for both @xmath142 and @xmath143 . fixing a basis for each of @xmath144 and @xmath101 , let @xmath36 and @xmath145 denote respectively the global and local stiffness matrices of the bilinear forms @xmath24 and @xmath102 with respect to the given bases . let @xmath146 and @xmath147 denote the matrix representations of the linear operators @xmath148 and @xmath143 with respect to the given bases . lastly , let @xmath149 and @xmath150 denote the matrix transposes of @xmath151 and @xmath152 . using the above notation and the well - known fact that composite linear operators are represented by matrix multiplications , we obtain from and that @xmath153 thus , @xmath154 where @xmath155 and @xmath156 denote the inverse matrices of @xmath145 and @xmath157 , respectively . we also note that the compatibility assumptions ( la@xmath75 ) and ( la@xmath76 ) imply that @xmath155 and @xmath156 do exist . finally , it follows from , , and that @xmath158 from the above expressions we obtain the following two additive schwarz preconditioners for both @xmath36 and its transpose @xmath159 : @xmath160 it is interesting to note that @xmath161 which means that the nonsymmetric schwarz preconditioner @xmath52 can be used to precondition both the linear system and its adjoint system without any additional cost . as it was already alluded to in remark [ rem3.3 ] , we now formally define our restriction operators @xmath162 and @xmath163 . [ def3.2 ] for @xmath110 , let @xmath164 ( resp . @xmath165 ) be the unique linear operator whose matrix representation is given by @xmath150 ( resp . @xmath166 ) under the same bases of @xmath167 and @xmath101 in which @xmath166 and @xmath150 are obtained . by the design , the matrix representations @xmath90 and @xmath168 of @xmath169 and @xmath170 satisfy @xmath171 and @xmath172 . the multiplicative schwarz methods for solving problem refer to various generalizations of the original schwarz alternating iterative method ( cf . however , they also can be formulated as linear iterations on some preconditioned systems ( cf . @xcite ) . in this paper we adopt the latter point of view to present our nonsymmetric and indefinite multiplicative schwarz methods . we shall use the same notation as in section [ sec-3.4 ] . we first introduce the following two so - called _ error propagation operators _ : @xmath173 where @xmath174 denotes the identity operator on @xmath19 or on @xmath9 . we then define the following two preconditioned " operators : @xmath175 it is easy to check that the algebraic matrix representations of the above operators are , respectively , @xmath176 then our multiplicative schwarz iterative methods are defined as @xmath177 where @xmath178 are either @xmath179 or @xmath180 , and @xmath181 takes either @xmath182 or @xmath183 which are easily computable from @xmath41 in . [ rem3.2 ] ( a ) clearly , the case with the triple @xmath184 corresponds to the classical multiplicative schwarz method for ( cf . @xcite ) . \(b ) the case with the triple @xmath185 can be regarded as a symmetrized " multiplicative schwarz method for nonsymmetric and indefinite problems . however , we note that the operator @xmath186 and matrix @xmath187 are not symmetric in general because @xmath188 and @xmath189 may not be symmetric . \(c ) unlike in the spd case , the norm @xmath190 could be larger than @xmath191 for convection - dominant problems as shown by the numerical tests given in section [ sec-5 ] although the multiplicative schwarz method appears to be convergent in all those tests . consequently , the convergent behavior of the multiplicative schwarz method presented above is more complicated than its spd counterpart . in this subsection , we consider a hybrid schwarz method which combines the additive schwarz idea ( between subdomains ) and the multiplicative schwarz idea ( between levels ) . the hybrid method is expected to take advantage of both additive and multiplicative schwarz methods . the iteration operator of our hybrid schwarz method is given by @xmath192 thus , the preconditioned " hybrid schwarz operator has the following form : @xmath193 where @xmath194 , called a relaxation parameter , is an undetermined positive constant . since the corresponding matrix representations of @xmath195 , and @xmath196 are easy to write down , we omit them to save space . in this section we shall first establish condition number estimates for additive schwarz operator @xmath142 and for its matrix representation @xmath197 . we then present a condition number estimate for the hybrid operator @xmath198 . our convergence theory rests on the following structure assumptions : * _ compatibility assumption_. assume that @xmath199 is a compatible decomposition of @xmath200 in the sense of definition [ def3.1 ] . * _ energy stable decomposition assumption_. there exist positive constants @xmath201 and @xmath202 such that every pair @xmath203 admits a decomposition @xmath204 with @xmath130 and @xmath131 such that @xmath205 * _ strengthened generalized cauchy - schwarz inequality assumption_. there exist constants @xmath206 $ ] for @xmath207 such that @xmath208 * _ local stability assumption_. there exist positive constants @xmath209 and @xmath210 such that for @xmath86 @xmath211 * _ approximability assumption_. there exist ( small ) positive constants @xmath212 and @xmath213 such that for @xmath214 and @xmath215 @xmath216 where @xmath217 are the same as in ( sa@xmath76 ) . @xmath218 and @xmath219 . we now explain the rationale and motivation of each assumption listed above . [ rem4.1 ] ( a ) we note that @xmath220 and @xmath221 are defined in and , and @xmath222 and @xmath223 are defined in and . \(b ) in the spd schwarz theory ( cf . @xcite ) , the local bilinear forms are always assumed to be strongly coercive which then infers the invertibilty of the local stiffness matrices . however , such an assumption is often not listed explicitly . on the other hand , in our nonsymmetric and indefinite schwarz theory , the invertibilty of the local stiffness matrices may not hold . we explicitly list it as an assumption in ( sa@xmath224 ) . \(c ) for a given compatible pair of space decompositions @xmath199 , decompositions of each function @xmath124 and @xmath125 may not be unique . assumption ( sa@xmath75 ) assumes that there exists at least one decomposition which is energy stable for every function in @xmath19 and @xmath9 . it imposes a constraint on both the choice of the space decompositions @xmath199 and on the choice of the local bilinear forms @xmath225 . \(d ) we note that different norms are used for two functions on the right - hand side of , and @xmath217 is defined for @xmath207 . we set @xmath226_{i , j=0}^j$ ] and note that @xmath227 is a @xmath228 matrix . we shall also use the submatrix @xmath229 in our convergence analysis to be given in section [ sec-4 ] . since the bilinear form @xmath24 is not an inner product , the standard cauchy - schwarz inequality does not hold in general . but it does hold in this generalized sense with @xmath230 , see lemma [ lem4.1 ] . moreover , we expect that each pair @xmath231 for @xmath232 only interacts with very few remaining pairs in the space decomposition @xmath233 . hence , the matrix @xmath234 , which is symmetric , is expected to be sparse and nearly diagonal in most applications . on the other hand , we expect that @xmath235 for @xmath236 . \(e ) local stability assumption ( sa@xmath237 ) imposes a condition on the choice of the prolongation operators @xmath94 and @xmath95 . it requires that these operators are bounded operators . \(f ) assumption ( sa@xmath238 ) , which does not appear in the spd theory , imposes a local approximation condition on the projection - like operators @xmath239 and @xmath240 and on the restriction operators @xmath162 and @xmath163 . ( g ) because of the norm equivalence properties , , and , one can easily replace the weak coercivity induced norms by their equivalent underlying space norms or vice versa in all assumptions ( sa@xmath75)(sa@xmath238 ) . however , one must track all the constants resulting from the changes . the main reason for using the current forms of the assumptions is that they allow us to give a cleaner presentation of our nonsymmetric and indefinite schwarz convergence theory to be described below . first , we state the following simple lemma . [ lem4.1 ] the following generalized cauchy - schwarz inequalities hold : @xmath241 are immediate consequences of the definitions of the norms @xmath242 , @xmath222 and @xmath223 . [ lem4.2 ] under assumptions ( sa@xmath224 ) and ( sa@xmath237 ) , the following estimates hold : @xmath243 for any @xmath124 , by assumption ( sa@xmath237 ) and lemma [ lem4.1 ] we get for @xmath244 , @xmath245 hence , holds . follows immediately from and . by assumption ( sa@xmath237 ) and lemma [ lem4.1 ] we obtain @xmath246 hence , holds . follows from , , , and . from the proof for we can obtain @xmath247 this result along with and yields . the proof is complete . we now are ready to give an upper bound estimate for the additive schwarz operator @xmath142 . [ prop4.1 ] under assumptions ( sa@xmath224)(sa@xmath237 ) the following estimate holds : @xmath248 \|v\|_a \qquad\forall v\in v,\end{aligned}\ ] ] where @xmath226_{i , j=0}^j$ ] , @xmath249 and @xmath250 denotes the number of nonzero entries in the vector @xmath251 , i.e. , the number of nonzeros among the last @xmath252 entries of the @xmath253th column of the matrix @xmath227 . for any @xmath125 , let @xmath254 be an energy stable decomposition of @xmath61 as defined in ( sa@xmath75 ) . by the definition of @xmath142 , , , , , and we get for any @xmath124 @xmath255 \|v\|_a { \vertw\vert_{\mbox{\tiny w}}}. \nonumber \end{aligned}\ ] ] hence , holds . the proof is complete . as expected , it is harder to get a lower bound estimate for the additive schwarz operator @xmath142 . such a bound then readily provides an upper bound for @xmath256 . to this end , we first establish the following key lemma . [ lem4.3 ] ( i ) suppose that for every @xmath124 , @xmath257 forms a stable decomposition of @xmath258 . then under assumptions ( sa@xmath224 ) and ( sa@xmath75 ) the following inequality holds : @xmath259 \(ii ) if the condition of ( i ) does not hold , then under assumptions ( sa@xmath224)(sa@xmath238 ) we have @xmath260 \| v \|_a,\end{aligned}\ ] ] where @xmath261 is the same as in proposition [ prop4.1 ] . \(i ) for any @xmath124 , let @xmath262 , @xmath263 for @xmath110 . since @xmath264 @xmath265 is indeed a decomposition of @xmath59 which is assumed to be stable . by assumption ( sa@xmath75 ) we conclude that holds for @xmath59 , which gives . \(ii ) let @xmath59 be same as in part ( i ) . recall that @xmath266 . using the identities @xmath267 , \\ { \widetilde{{\mathcal{p}}}}_j v & = \frac{1}{j+1 } \bigl [ { \widetilde{{\mathcal{p}}}}_j u + \sum_{i=0}^j { \widetilde{{\mathcal{p}}}}_j ( v-{\mathcal{p}}_i v ) \bigr ] \qquad \mbox{for } j = 1 , 2 , \cdots , j,\end{aligned}\ ] ] the triangle inequality , @xmath268 and we get @xmath269 \\ & \leq \frac12 \bigl [ { { \omega}_{\mbox{\tiny w}}}\| u \|_a + \bigl({\delta_{\mbox{\tiny v}}}+ \widehat{\delta}_v\bigr ) \|v\|_a \bigr ] , \\ \| { \widetilde{{\mathcal{p}}}}_j v \|_{a_j } & \leq \frac{1}{j+1 } \bigl [ \|{\widetilde{{\mathcal{p}}}}_j u\|_{a_j } + \sum_{i=0}^j \|{\widetilde{{\mathcal{p}}}}_j ( v-{\mathcal{p}}_i v)\|_{a_j } \bigr ] \\ & \leq \frac{1}{j+1 } \bigl [ { { \omega}_{\mbox{\tiny w}}}\|u\|_{a } + \sum_{i=0}^j \theta_{ij } { \delta_{\mbox{\tiny v}}}\|v\|_{a } \bigr ] \\ & \leq \frac{1}{j+1 } \bigl [ { { \omega}_{\mbox{\tiny w}}}\|u\|_{a } + n_j(\theta ) { \delta_{\mbox{\tiny v}}}\| v \|_{a } \bigr ] \qquad\mbox{for } j = 1,2 , \cdots , j.\end{aligned}\ ] ] then summing the above inequality we obtain @xmath270 \| v \|_a.\end{aligned}\ ] ] hence , holds . the proof is complete . we now are ready to establish a lower bound estimate for the additive schwarz operator @xmath142 . [ prop4.2 ] ( i ) under assumptions of ( i ) of lemma [ lem4.3 ] , the following estimate holds : @xmath271 \(ii ) under assumptions of ( ii ) of lemma [ lem4.3 ] , the following estimate holds : @xmath272 provided that @xmath273<1 $ ] where @xmath274}.\end{aligned}\ ] ] consequently , operator @xmath142 is invertible . for any @xmath125 , let @xmath254 be an energy stable decomposition of @xmath61 , that is , @xmath275 and holds . then we have @xmath276 the desired estimates and follow from substituting and into , respectively . the proof is complete . [ rem4.2 ] we note that the argument used in the proof of lower bound estimate is in the spirit of the so - called schatz argument ( cf . @xcite ) which is often used to derive finite element error estimates for nonsymmetric and indefinite problems . it is interesting to see that a similar argument also plays an important role in our schwarz convergence theory . combining propositions [ prop4.1 ] and [ prop4.2 ] we obtain our first main theorem of this paper . [ thm4.1 ] ( i ) if every @xmath124 has a unique decomposition @xmath277 with @xmath130 , then under assumptions ( sa@xmath224)(sa@xmath237 ) the following condition number estimate holds : @xmath278 . \end{aligned}\ ] ] \(ii ) if the above unique decomposition assumption does not hold , then under assumptions ( sa@xmath224)(sa@xmath238 ) the following condition number estimate holds : @xmath279 k_0 . \end{aligned}\ ] ] where @xmath280 the above condition number estimates for the operator @xmath142 also translates to its matrix representation . [ thm4.2 ] ( i ) under assumptions of ( i ) of theorem [ thm4.1 ] the following condition number estimate holds : @xmath281 . \end{aligned}\ ] ] \(ii ) under assumptions of ( ii ) of theorem [ thm4.1 ] the following condition number estimate holds : @xmath282 k_0,\end{aligned}\ ] ] where @xmath283 as in the case of spd problems @xcite , we replace the structure assumption ( sa@xmath75 ) by the following one : * _ energy stable decomposition assumption . _ there exist positive constants @xmath284 and @xmath285 such that every pair @xmath286 admits a decomposition @xmath287 with @xmath288 and @xmath289 such that @xmath290 we remark that the new _ energy stable decomposition assumption _ ( @xmath291 ) implies that any pair @xmath203 has a stable decomposition ( in the sense of ( sa@xmath75 ) ) of the following form : @xmath292 where @xmath293 is a stable decomposition ( in the sense of ( @xmath291 ) ) for @xmath294 . next lemma shows that @xmath295 ( resp . @xmath142 ) and @xmath296 ( resp . @xmath143 ) are mutually conjugate with respect to the bilinear form @xmath24 . [ lem4.8 ] the following identities hold : @xmath297 since the proof is trivial , we omit it to save space . the following proposition is the analogue to proposition [ prop4.1 ] for the hybrid operator @xmath198 . [ prop4.3 ] under assumptions ( sa@xmath224 ) , ( @xmath291 ) , ( sa@xmath76 ) and ( sa@xmath237 ) the following estimate holds : @xmath298 \|v\|_a \end{aligned}\ ] ] for all @xmath124 . where @xmath299 . let @xmath300 and @xmath301 . for any @xmath124 and @xmath125 , let @xmath302 and @xmath303 . obviously , @xmath304 and @xmath305 . by assumption ( @xmath291 ) , @xmath306 admits an energy stable decomposition @xmath293 . thus , @xmath307 where we have used to obtain the last inequality . the above inequality in turn implies that @xmath308 and @xmath309 hence , holds and the proof is complete . next , we derive a lower bound estimate for @xmath310 . the following proposition is an analogue of proposition [ prop4.2 ] . [ prop4.4 ] under assumptions ( sa@xmath224 ) , ( @xmath291 ) , ( sa@xmath76)(sa@xmath238 ) , along with the assumption @xmath311 the following estimate holds : @xmath312 provided that @xmath313 . where @xmath314 consequently , operator @xmath198 is invertible . for any @xmath124 and @xmath125 . let @xmath315 and @xmath316 . assumption ( @xmath291 ) ensures that @xmath317 has an energy stable decompositions @xmath318 with @xmath289 , that is , @xmath319 using the following identity @xmath320,\end{aligned}\ ] ] ( sa@xmath238 ) , and we get @xmath321 \\ & \leq \frac{1}{j } \bigl [ \| u \|_a { \vert\psi\vert_{\mbox{\tiny w } } } + \sum^j_{j = 1 } \sum^{j}_{i = 1}a \bigl(v - { \mathcal{p}}_i v , { \mathcal{s}}_j^\dag \psi_j \bigr ) + \alpha\bigl\| { \mathcal{p}}_0 \bigl(\widehat{{\mathcal{p}}}v -v\bigr)\bigr\|_a { \vert\psi \vert_{\mbox{\tiny w}}}\bigr]\\ & \leq \frac{1}{j } \bigl [ \| u \|_a { \vert\psi\vert_{\mbox{\tiny w } } } + \sum^j_{j = 1 } \sum^{j}_{i = 1}a_j \bigl ( { \widetilde{{\mathcal{p}}}}_j \bigl(v - { \mathcal{p}}_i v \bigr ) , \psi_j \bigr ) + \alpha { { \omega}_{\mbox{\tiny v}}}\bigl \| { \widetilde{{\mathcal{p}}}}_0 \bigl(\widehat{{\mathcal{p}}}v - v\bigr ) \bigr \|_{a_0}{\vert\psi\vert_{\mbox{\tiny w } } } \bigr ] \\ & \leq \frac{1}{j } \bigl[\| u \|_a { \vert\psi\vert_{\mbox{\tiny w } } } + \sum^j_{j = 1 } \sum^{j}_{i = 1}\bigl \| { \widetilde{{\mathcal{p}}}}_j \bigl(v-{\mathcal{p}}_i v\bigr ) \bigr\|_{a_j}{\vert\psi_j\vert_{\mbox{\tiny w$_j$ } } } + \alpha{{\omega}_{\mbox{\tiny v}}}\widehat{\delta}_v \| v \|_a { \vert\psi\vert_{\mbox{\tiny w } } } \bigr ] \\ & \leq \frac{1}{j } \bigl [ \| u \|_a { \vert\psi\vert_{\mbox{\tiny w } } } + \sum^j_{j = 1 } \sum^{j}_{i = 1 } \theta_{ij } { \delta_{\mbox{\tiny v}}}\|v \|_{a } { \vert\psi_j\vert_{\mbox{\tiny w$_j$ } } } + \alpha{{\omega}_{\mbox{\tiny v}}}\widehat{\delta}_v \| v \|_a { \vert\psi\vert_{\mbox{\tiny w } } } \bigr ] \\ & \leq \frac{1}{j } \bigl [ \| u \|_a { \vert\psi\vert_{\mbox{\tiny w } } } + \sum^j_{j = 1 } n_j \bigl(\widehat{\theta } \bigr ) { \delta_{\mbox{\tiny v}}}\|v \|_{a } { \vert\psi_j\vert_{\mbox{\tiny w$_j$ } } } + \alpha{{\omega}_{\mbox{\tiny v}}}\widehat{\delta}_v \| v \|_a{\vert\psi\vert_{\mbox{\tiny w } } } \bigr ] \\ & \leq \frac{1}{j } \bigl [ \| u \|_a { \vert\psi\vert_{\mbox{\tiny w } } } + n\bigl(\widehat{\theta}\bigr ) { \delta_{\mbox{\tiny v}}}\|v \|_{a } \sum^j_{j = 1 } { \vert\psi_j\vert_{\mbox{\tiny w$_j$ } } } + \alpha{{\omega}_{\mbox{\tiny v}}}\widehat{\delta}_v \| v \|_a { \vert\psi\vert_{\mbox{\tiny w } } } \bigr ] \\ & \leq \frac{1}{j } \bigl [ \| u \|_a { \vert\psi\vert_{\mbox{\tiny w } } } + { \widetilde{c}_{\mbox{\tiny w}}}n \bigl(\widehat{\theta } \bigr ) { \delta_{\mbox{\tiny v}}}\|v \|_{a } { \vert\psi\vert_{\mbox{\tiny w } } } + \alpha{{\omega}_{\mbox{\tiny v}}}\widehat{\delta}_v \| v \|_a { \vert\psi\vert_{\mbox{\tiny w } } } \bigr].\end{aligned}\ ] ] the desired estimate follows from the assumption @xmath322 . we note that the assumption @xmath323 is equivalent to asking @xmath324 to be invertible , which holds for sufficiently small or large relaxation parameter @xmath194 . combining propositions [ prop4.3 ] and [ prop4.4 ] we obtain our third main theorem of this paper . [ thm4.3 ] under assumptions ( sa@xmath224)(sa@xmath238 ) and @xmath311 the following condition number estimate holds : @xmath325\ , k_1.\end{aligned}\ ] ] in this section we shall use our abstract framework and the abstract convergence theory developed in sections [ sec-3 ] and [ sec-4 ] to construct three types of schwarz methods for discontinuous galerkin approximations of the following general diffusion - convection problem : @xmath326 where @xmath327 is a bounded domain with lipschitz continuous boundary @xmath328 . @xmath329 satisfies @xmath330 @xmath331 for some positive constants @xmath332 and @xmath333 . so is uniformly elliptic in @xmath334 ( * ? ? ? * chapter 8) . assume that @xmath335 or @xmath336^d$ ] , @xmath337 and @xmath338 . let @xmath339 , then the variational formulation of is defined as @xcite @xmath340 where @xmath341 it is easy to check that all the conditions of the classical lax - milgram theorem hold in the _ positive definite case_. it also can be shown @xcite that in the _ indefinite case _ all the conditions of theorem [ babuska ] are satisfied provided that problem and its adjoint problem are uniquely solvable for arbitrary source terms . it is also well known @xcite that in _ indefinite case _ the bilinear form @xmath343 satisfies a grding - type inequality instead of the strong coercivity . consider a special case of where @xmath348 , @xmath349^d$ ] , and @xmath350 where @xmath351 . to discretize this problem , we shall use an interior penalty discontinuous galerkin ( ipdg ) scheme developed in @xcite . for this scheme we require a shape - regular triangulation @xmath352 of the domain @xmath334 . the scheme can then be written in the form where @xmath353 \cdot [ w]\ , ds \label{eq5.9 } \\ & \quad+ \sum_{e \in \mathcal{e}^{\circ}_h } \int_e \{\mathbf{b}u\}_{upw } \cdot [ w]\ , ds - \sum_{e \in \gamma } \int_{e } \{\epsilon \nabla_hu\ } \cdot [ w ] \ , ds + \sum_{e \in \gamma^+ } \int_{e } \mathbf{b \cdot n}uw\ , ds , \notag \\ & f(w ) : = \sum_{k \in \mathcal{t}_h } \int_k f w\ , dx . \label{eq5.10}\end{aligned}\ ] ] where @xmath354 , @xmath355 is the unit outward normal vector to @xmath356 , and @xmath357 indicates the outflow portion of @xmath356 defined as @xmath358 @xmath359 $ ] and @xmath360 are the standard jump and average operators , respectively , and @xmath361 is the upwind flux . to define this flux , we consider a vector valued function @xmath362 defined on two neighboring elements @xmath363 and @xmath364 of @xmath352 with common edge @xmath365 . suppose that @xmath366 for @xmath367 . then @xmath368 is defined on the edge @xmath365 as follows : @xmath369 where @xmath370 is the unit outward normal vector of @xmath371 on @xmath365 for @xmath367 . the choice of this scheme was made because it was shown @xcite that in the positive definite case ( i.e. when holds ) this scheme satisfies ( ma1 ) and ( ma2 ) ( cf . section [ sec-3.2 ] ) . once a discretization scheme is chosen we can begin to develop our space decomposition and local solvers . in this example , we will obtain the space decomposition by using a nonoverlapping domain decomposition . let @xmath372 be a coarse mesh of @xmath334 and @xmath373 a nonoverlapping partition @xmath374 of @xmath334 such that @xmath375 . then we define @xmath376 for @xmath377 and @xmath378 . for the prolongation operator @xmath379 we use the polynomial interpolation on each element @xmath380 . @xmath381 for each @xmath382 and @xmath380 . for the prolongation operators @xmath383 we use the following natural injection into @xmath19 : @xmath384 for the local bilinear forms @xmath102 we use the exact local solvers defined by @xmath385 for @xmath386 . note that in this example we only have one set of subspaces @xmath387 and one set of prolongation operators @xmath388 so we shall only have one set of projection - like operators @xmath389 defined in and . using these projection - like operators we can then build the schwarz operators @xmath390 , @xmath391 , and @xmath392 defined in , , and respectively . in this section we present several @xmath191-d numerical experiments to gauge the theoretical results proved in the previous section . for these experiments we concentrated on equation in the domain @xmath393 with the following choices of constant coefficient : note that these choices of coefficients put us in the convection dominated regime and fit the criteria of the positive definite case characterized by . for this reason we are able to use the discretization scheme and domain decomposition techniques described in section [ sec-5.1 ] . in these experiments we use a uniform fine mesh size @xmath400 and a uniform coarse mesh size @xmath401 . the equations are solved numerically using standard gmres , gmres after using @xmath390 preconditioning , the multiplicative schwarz iterative method , and gmres after using @xmath392 preconditioning . to verify the dependence of @xmath402 and @xmath403 , we use a varying number of subdomains @xmath404 . our first goal in these experiments is to compare the performance of the schwarz methods to that of standard gmres in order to verify the usefulness of such methods . we would like to verify numerically that the estimates given in previous sections are tight . in particular , we would like to find an example that shows that @xmath405 does in fact depend linearly on the number of subdomains @xmath252 as predicted in theorem [ thm4.2 ] . for multiplicative schwarz iteration we would like to estimate @xmath406 , noting that if this norm is less than @xmath191 it guarantees convergence of the method . if not , we shall need to rely on the spectral radius @xmath407 to guarantee this convergence . tables [ tab1][tab4 ] collect the test results on the additive , multiplcative , and hybrid schwarz methods proposed in section [ sec-3 ] . where @xmath408 represents the original system with no preconditioning . from these numerical results we can make the following observations : 1 . any of these methods offers an improvement in terms of the cpu time needed to solve the system when compared to solving the system using standard gmres . gmres after using @xmath390 or @xmath392 for preconditioning performs better when the number of subdomains @xmath252 is not too large . 3 . in all of these tests @xmath405 and @xmath409 depend on the number of subdomains @xmath252 . particularly in * test 2 * , we see an example that exhibits approximate linear dependence . see figure [ fig6.1 ] . 4 . for @xmath406 we do not observe such a strong dependence on the number of subdomains @xmath252 . 5 . in these tests @xmath406 is greater than @xmath191 ; thus , we can not rely upon this as an indicator for convergence of the multiplicative schwarz iterative method . @xmath410 is not a unique metric in judging the convergence of gmres after preconditioning with @xmath411 and @xmath412 . for instance , in * test 4 * @xmath405 decreases while the number of iterations necessary for gmres increases as @xmath252 increases . this is opposite of the behavior that is observed in the previous tests . our numerical experiments verify that @xmath410 is not a unique metric for the convergence of gmres . therefore , we must rely on other metrics to predict the convergence behavior of gmres . the following theorem is used in @xcite to test convergence of gmres after schwarz preconditioning in the indefinite case : [ thm6.1 ] consider the linear system @xmath413 where @xmath414 and @xmath415 . if the symmetric part of @xmath36 is positive definite , then after @xmath416 step of gmres , the norm of the residual @xmath417 is bounded by @xmath418 where @xmath419 is the minimal eigenvalue of the symmetric part of @xmath36 and @xmath420 is the operator norm of @xmath36 given by @xmath421 unfortunately , this theorem can not be applied in our case because we are not guaranteed that @xmath411 and @xmath412 have positive definite symmetric part ( i.e. , @xmath419 ) . in the tests previously done , we find that these operators can be indefinite ( i.e. , @xmath422 ) . another result that could be of help in this area is the following theorem ( cf . @xcite ) . [ thm6.2 ] consider the linear system @xmath413 where @xmath414 and @xmath415 . further suppose that @xmath36 is diagonalizable . then after @xmath416 steps of gmres , the residual @xmath423 satisfies @xmath424 where @xmath19 is a nonsingular matrix of eigenvectors of @xmath36 and @xmath425 denotes the spectrum of @xmath36 . the above theorem says that the spread of the spectrum is a metric to judge the performance of gmres with gmres performing better when the spectrum of @xmath36 is clustered . with this theorem in mind , let us examine the spectrum of the matrix @xmath36 and @xmath411 for @xmath404 obtained in * test 2 * and * test 4*. note that in figure [ fig6.2](a ) and figure [ fig6.3](a ) the spectrum has a large spread which is consistent with the fact that gmres performed poorly on the original system without preconditioning . we also see that after preconditioning , the spectrum of @xmath411 is clustered which corresponds to improved performance of gmres after preconditioning with @xmath411 . lastly , we note that as the number of subdomains @xmath252 increases , the spread of the spectrum of @xmath411 increases . this corresponds to a decreased performance in gmres after preconditioning with @xmath411 as @xmath252 increases . this result leads us to believe that to accurately judge the behavior of gmres after preconditioning one needs to analyze the spectrum of the preconditioned system . similarly , we find that to accurately predict the performance of the multiplicative schwarz iterative method one needs to analyze the spectral radius of @xmath426 . another property of @xmath405 that is of interest is its dependence on the fine mesh parameter @xmath427 and the coarse mesh parameter @xmath428 . it was shown in @xcite that @xmath429 using two - level non - overlapping domain decomposition for an ipdg approximation of the equation with @xmath430 and @xmath431 . this work uses the existing framework that is laid down for the symmetric and positive definite case . we would like to test our new framework to see if this dependence on @xmath428 and @xmath427 is still true . for this reason in * test 1 - 4 * we have calculated @xmath405 with @xmath432 and @xmath433 when @xmath434 . we also calculated @xmath405 with @xmath435 and @xmath436 when @xmath437 . from table [ table 5.5 ] and table [ table5.6 ] it seems that we can not expect @xmath429 but instead we might expect the dependence to be one of the type @xmath438 where @xmath439 and @xmath440 are positive real numbers with @xmath441 or some other more complicated dependence on @xmath428 and @xmath427 . this work was initiated while the first autthor was a long term visitor of the i m a at university of minnesota in fall 2010 . the first author would like to thank the i m a for its financial support and its hospitality . the first author would also like to thank blanca ayuso and ohannes karakashian for their many stimulating and critical discussions which have substantially helped to shape the structure and results of this paper . p. f. antonietti and b. ayuso , _ schwarz domain decomposition preconditioners for discontinuous galerkin approximations of elliptic problems : non - overlapping case _ , m2an math ( * 41 * ) , 2007 , pp . 2154 . i. babuka and a. k. aziz , _ survey lectures on the mathematical foundations of finite element method _ , aziz ( ed . ) , the mathematical foundations of the fem with application to pde , academic press , 1972 , pp . 5359 . m. dryja and o. b. widlund , _ towards a unified theory of domain decomposition algorithms for elliptic problems _ , proceedings of third international symposium on domain decomposition methods for partial differential equations ( t. chan etc . editors ) , siam , philadelphia , 1990 . pp . 321 . c. lasser and a. toselli , _ an overlapping domain decomposition preconditioner for a class of discontinuous galerkin approximations of advection - diffusion problems _ , math . comp . , ( * 72 * ) , 2003 , pp . 12151238 .

we investigate numerically the ability of three models ( jet , structured outflow , energy injection ) to accommodate the optical light - curve breaks observed in 10 grb afterglows ( 980519 , 990123 , 990510 , 991216 , 000301c , 000926 , 010222 , 011211 , 020813 , and 030226 ) , as well as the relative intensities of the radio , optical , and-ray emissions of these afterglows . we find that the jet and structured outflow models fare much better than energy injection model in accommodating the multiwavelength data of the above 10 afterglows . for the first two models , a uniform circumburst medium provides a better fit to the optical light - curve break than a wind - like medium with a stratification . however , in the only two cases where the energy injection model may be at work , a wind medium is favoured ( an energy injection is also possible in a third case , the afterglow 970508 , whose optical emission exhibited a sharp rise but not a steepening decay ) . the best fit parameters obtained with the jet model indicate an outflow energy of ergs and a jet opening of . structured outflows with a quasi - uniform core have a core angular size of and an energy per solid angle of , surrounded by an envelope where this energy falls - off roughly as with angle from the outflow axis , requiring thus the same energy budget as jets . circumburst densities are found to be typically in the range , for either model . we also find that the reverse shock emission resulting from the injection of ejecta into the decelerating blast wave at about 1 day after the burst can explain the slowly decaying radio light - curves observed for the afterglows 990123 , 991216 , and 010222 . 2nu^2 _ 0_e__0 3cm^-3 5310 ^ 53 gamma - rays : bursts - ism : jets and outflows - radiation mechanisms : non - thermal - shock waves

astro-ph0508426

since the prediction of radio ( paczyski & rhoads 1993 ) and optical transients ( mszros & rees 1997 ) associated with grbs , more than 100 afterglows have been observed ( including @xmath0-ray transients ) . good monitoring in all three frequency domains has been achieved for about 20 afterglows ; the radio , optical , and @xmath0-ray flux was observed to decay as a power - law , @xmath8 ( @xmath9 ) , confirming the expectations . for 10 of these well - observed afterglows 980519 , 990123 , 990510 , 991216 , 000301c , 000926 , 010222 , 012111 , 020813 , 030226 the optical light - curve decay exhibits a steepening at about 1 day after the burst , with an increase @xmath10 in the temporal index as low as 0.4 ( afterglow 991216 ) and as high as 1.8 ( afterglow 000301c ) . optical light - curve breaks have or may have been observed in other afterglows ; they are not included in the sample of afterglows modelled in this work because those afterglows have been adequately monitored at only one optical frequency . with the exception of the afterglow 010222 , for which one , late ( 10 days ) measurement suggests that a break occurred also in the @xmath0-ray light - curve , currently available @xmath0-ray observations do not extend sufficiently before and after the optical break time to prove that the light - curve break is achromatic . within the measurement uncertainties , a single power - law fits well the decay of all adequately monitored @xmath0-ray afterglows : 990123 , 990510 , 991216 , 000926 ( and 010222 until the last measurement ) . breaks , in the form of peaks , have been observed in the radio emission of all 10 afterglows above , usually at @xmath11 days after the burst , often occurring after the optical light - curve break . given that the radio and optical breaks are not simultaneous , they can not both arise from the dynamics of the afterglow , the structure of the grb ejecta , or some property of the circumburst medium ( cbm ) , as these mechanisms should yield achromatic breaks . furthermore , given that , before their respective breaks , the radio emission rises while the optical falls - off , these temporal features can not be due to the passage of the same afterglow continuum break through the observing band . with the possible exception of the afterglow 990123 , there is no evidence for an evolution ( softening ) of the optical continuum across the light - curve break for the above 10 afterglows ( panaitescu 2005 ) . furthermore , the only spectral feature whose passage could yield the large steepening @xmath10 observed in some cases the peak of the forward shock ( fs ) continuum should yields a rising or flat optical light - curve before the break , contrary to what is observed . therefore , it is the radio peak which should be attributed to a spectral break crossing the observing domain . that spectral break should be the injection frequency @xmath12 , which , for reasonable shock microphysical parameters , should reach the radio at around 10 days . indeed , for the afterglow 991208 , there is observational evidence ( galama 2000 ) that the peak of the afterglow continuum crosses the 10100 ghz domain at that time . further evidence for the passage of a spectral break through radio is provided by that the peak time of the radio flux of the afterglow 030329 increases with decreasing observing frequency ( frail 2005 ) . today , the generally accepted reason for the optical light - curve break is the narrow collimation of the grb ejecta . as predicted by rhoads ( 1999 ) , if the grb ejecta are collimated , then the afterglow light - curve should exhibit a steepening when the jet begins to expand sideways . more than half of the steepening @xmath10 is due by the finite angular opening of the jet : as the grb remnant is decelerated ( by sweeping - up of the cbm ) and the relativistic doppler beaming of the afterglow emission decreases , an ever increasing fraction of the emitting surface becomes visible to the observer ; when the jet edge is seen , that fraction can not increase any longer and the afterglow emission exhibits a faster decay ( panaitescu , mszros & rees 1998 ) . because it arises from the blast wave dynamics , a light - curve break should also be present at radio wavelengths at the same time as the optical break . however , radio observations before 1 day are very scarce and strongly affected by galactic interstellar scintillation ( goodman 1997 ) until after 10 days , thus they can not disentangle the jet - break from that arising from the passage of the fs peak frequency . another mechanism for the optical light - curve breaks has been proposed by rossi , lazzati & rees ( 2002 ) and zhang & mszros ( 2002 ) : if grb outflows are endowed with an angular structure ( non - uniform distribution of the ejecta kinetic energy with direction ) , as first proposed by mszros , rees & wijers ( 1998 ) , then a steepening of the afterglow decay would arise when the brighter , outflow symmetry axis becomes visible to the observer . in this model , the stronger the angular structure is , the larger the break magnitude @xmath10 should be . a third mechanism for breaks rests on the proposal of paczyski ( 1998 ) and rees & mszros ( 1998 ) that the fs energizing the cbm could be refreshed by the injection of a substantial energy through some delayed ejecta which were released at the same time with the grb - producing ejecta , but had a smaller lorentz factor , or were ejected sometime later , and which catch up with the decelerating fs during the afterglow phase . fox ( 2003 ) have proposed that the early ( 0.0030.1 day ) , slow decay of the optical emission of the afterglow 021004 is caused by such an injection process . in this scenario , when the energy injection episode ends , the fs deceleration becomes faster and the afterglow emission should exhibit a steepening . note the various origins of the afterglow light - curve break in each model . in the * jet * model , the break is caused by the changing outflow dynamics when the jet starts to spread and by the outflow s geometry . in the * structured outflow ( so ) * model the origin is , evidently , the outflow s anisotropic surface brightness . for both these models , special relativity effects play an important part . in the * energy injection ( ei ) * model , the break originates in the altered outflow dynamics at the time when the energy injection subsides . the purpose of this work is to compare the ability of these three models in accommodating + @xmath13 the shape of the light - curve breaks observed in the optical emission of the afterglows 980519 , 990123 , 990510 , 991216 , 000301c , 000926 , 010222 , 012111 , 020813 , 030226 and + @xmath14 the relative intensity of the radio , optical , and @xmath0-ray emissions of these afterglows , + for either a uniform ( homogeneous ) cbm , or one with a @xmath1 density radial stratification , as expected if grbs progenitors are massive stars ( woosley 1993 , paczyski 1998 ) . for the first task above , we performed an analytical test of the models ( see table 2 of panaitescu 2005 ) , based on comparing + @xmath13 the pre- and post - break optical light - curve decay indices , @xmath15 and @xmath16 , + @xmath14 those in the @xmath0-rays , + @xmath17 the slopes @xmath18 and @xmath19 of the power - law optical and @xmath0-ray continua ( @xmath20 and @xmath21 ) , and + @xmath22 the optical to@xmath0-ray spectral energy distribution ( sed ) slope , @xmath23 , + with the relations among them expected for each model . we note that , in the framework of the relativistic fireballs ( mszros & rees 1997 ) , the optical and @xmath0-ray decay indices and sed slopes are tightly connected , as only one continuum feature , the cooling frequency @xmath24 ( sari , narayan & piran 1998 ) , can be between these domains at the times when observations were usually made ( 0.1100 days ) , and that the sed slope increases by a fixed amount , @xmath25 , across this spectral break . including the radio afterglow emission in the analytical test of the three break models is less feasible and often unconstraining . first , that the afterglow radio flux is modulated by diffractive and refractive interstellar scintillation makes it difficult to determine accurately the radio sed slope ( see figs . 4 and 5 of frail , waxman & kulkarni 2000a for the best monitored radio afterglow 970508 ) . second , after the injection frequency has fallen below the radio ( during the decay phase of the radio light - curve ) , we do not expect , in general , any spectral break to be between radio and optical , hence the radio and optical light - curve indices should be the same . this is , indeed , the case for most afterglows ; nevertheless , there are a few troubling exceptions : over 12 decades in time , the radio emission of the afterglows 991208 , 991216 , 000926 , and 010222 exhibits a much shallower decay than that observed at optical wavelengths at the same time or prior to the radio decay . an analytical investigation of the various possible ways to decouple the radio and optical light - curves ( panaitescu & kumar 2004 ) has led to the conclusion that , the anomalous radio decay is due to a contribution from the reverse shock to the radio emission . however , radio observations provide an indirect constraint on them because the flux and epoch of the radio peak determine the fs synchrotron peak flux , injection frequency , and self - absorption frequency . these three spectral properties constrain the afterglow parameters pertaining to the outflow dynamics ( energy per solid angle , jet initial opening , medium density ) and emission ( magnetic field strength , typical post - shock electron energy ) , parameters which determine the shape and epoch of the optical light - curve break , as well as the location of the cooling frequency . the best way to take into account the constraints arising from the radio emission and to test fully the three break models is to calculate numerically the afterglow dynamics and emission , and to fit all the available measurements . data fitting also allows the determination of the various model parameters . the basic equations employed in our numerical calculations of the outflow dynamics are presented in panaitescu & kumar ( 2000 ) for spherical outflows , panaitescu & kumar ( 2001 ) for the jet model , panaitescu & kumar ( 2003 ) for the so model ( structured outflows ) , and panaitescu , mszros & rees ( 1998 ) for the ei model ( energy injection ) . the basic equations for the calculation of the afterglow spectral features ( synchrotron and inverse compton peak fluxes , absorption , injection , and cooling frequencies ) , and of the afterglow emission at any wavelength are given in panaitescu & kumar ( 2000 , 2001 ) . equations for the spectral characteristics can be also found in sari ( 1998 ) , for the dynamics and emission from spherical blast - waves interacting with a uniform medium in waxman , kulkarni & frail ( 1998 ) , granot , piran & sari ( 1999 ) , wijers & galama ( 1999 ) and for a wind medium in chevalier & li ( 2000 ) . the dynamics of blast - waves with energy injection is also treated in sari & mszros ( 2000 ) . rhoads ( 1999 ) provided a detailed treatment of the dynamics and emission from jets interacting with uniform media ( see also sari , piran & halpern 1999 ) . the jet dynamics is determined by the initial jet opening @xmath26 , the ejecta initial kinetic energy per solid angle @xmath27 ( or , equivalently , by the jet energy @xmath28 ) , and the cbm particle density @xmath29 or , in the case of a wind of a massive star , by the ratio of the mass - loss rate to the wind speed , @xmath30 . for convenience , the latter parameter is given normalized to a mass - loss rate of @xmath31 at a speed of @xmath32 , for a wind @xmath33 where @xmath34 is the fs radius , and we used the usual notation @xmath35 . the deceleration of the jet , as it sweeps - up the cbm , is calculated assuming that the post - shock gas has the same internal energy per mass ( temperature ) as that immediately behind the fs , which is equal to the bulk lorentz factor of the shocked cbm , @xmath36 . the lateral size of the jet is assumed to increase at the co - moving sound speed , and the kinetic energy per solid angle is approximated as uniform during the spreading . radiative ( synchrotron and inverse compton ) losses are calculated from the electron distribution and magnetic field strength . in the jet model , the light - curve break is seen when the jet edge becomes visible to the observer . for an observer located close to the jet axis , this time can be approximated by @xmath37 , using the initial jet opening and ignoring the lateral spreading that occurred until the jet - break time . if the fs dynamics were adiabatic , then energy conservation would lead to @xmath38)^{-3/8 } \label{gs0}\ ] ] for a uniform medium and @xmath39)^{-1/4}\ ] ] for a wind , where @xmath40 is the observer time in days . then the jet - break time is given by @xmath41 for a uniform cbm and @xmath42 for a wind . given that observer locations off the jet axis ( but within @xmath26 ) have a small effect on the resulting afterglow light - curves ( granot 2002 ) , we consider that , in the jet model , the observer is always on the jet axis . furthermore , we ignore the possible existence of a counter - jet whose emission would become visible to the observer when the semi - relativistic dynamics sets , it would affect the radio afterglow emission beyond 100 days after the burst . at the jet - break time , the afterglow decay index at observing frequency @xmath43 steepens from that corresponding to a spherical outflow ( sari 1998 ) : @xmath44 to @xmath45 ( rhoads 1999 ) . in equation ( [ a1sph ] ) , @xmath46 is the power - law index of the post - shock electron distribution with energy ( equation [ [ p ] ] ) . the dynamics of structured outflows is calculated similarly to that of jets . to track the lateral fluid flow and the change of the kinetic energy per solid angle , the outflow surface is divided in infinitesimal rings and we consider that each ring spreads at a rate proportional to the local sound speed and to the ring width . this prescription for lateral flow reduces to that given above for a jet if the outflow is uniform and collimated . in this work we consider only axially symmetric power - law outflows , whose angular distribution of the ejecta kinetic energy per solid angle @xmath47 is given by @xmath48 with @xmath49 . the uniform core is used to avoid a diverging outflow energy for @xmath50 . this power - law structure is sufficiently complex , given the constraining power of the available data . given that , in this model , the light - curve break is due to the brighter core becoming visible , the observer location @xmath51 relative to the symmetry axis is a crucial parameter in determining the break time ( rossi 2002 ) , which is the same as given in equations [ [ tjs0 ] ] and [ [ tjs2 ] ] but with @xmath51 in place of @xmath26 . analytical results for the light - curve pre- and post - break decay indices can be obtained if the observer is located within the core ( @xmath52 ) , but only numerically for outer locations . in the former case , the light - curve break occurs when the edge of the uniform core becomes visible to the observer , and the light - curve decay index at a frequency @xmath53 increases from that given in equation ( [ a1sph ] ) to @xmath54 for a homogeneous medium and @xmath55 for a wind medium ( panaitescu & kumar 2003 ) , provided that the structural parameter @xmath56 , where , for a uniform medium , @xmath57 if @xmath58 and @xmath59 if @xmath60 . for a wind , the two values of @xmath61 are swapped . for @xmath56 , the emission from the outflow envelope ( @xmath62 ) is brighter than that from the core , and sets the light - curve post - break decay index . for @xmath63 , the core dominates the afterglow emission and the post - break decay index is @xmath64 @xmath10 is just the steepening produced by the finite angular extent of the emitting outflow . in this case the structure outflow model reduces to a that of a jet with sharp edges and no sideways expansion . for observers located outside the core , the break magnitude @xmath10 can be larger ( see figs . 2 , 3 , and 4 in panaitescu & kumar 2003 ) . the injection of energy in the fs has two effects : it mitigates the fs deceleration and generates a reverse shock ( rs ) which energizes the incoming ejecta and contributes to the radio afterglow emission . for ease of interpretation , we consider an energy injection which is a power - law in the observer time : @xmath65 where @xmath66 is the rate of the influx of energy per solid angle . evidently , this injection has an effect on the afterglow dynamics only if the total added energy , @xmath67 , is comparable or larger with that existing in the afterglow after the grb phase , @xmath27 . in this case , the light - curve decay index during the injection process is @xmath68 ( panaitescu 2005 ) . after the energy injection subsides , the post - break index @xmath16 has the value given in equation ( [ a1sph ] ) ( for an adiabatic , spherical blast - wave ) . numerically , the energy injection is modelled by first calculating the dissipated energy and that added as kinetic energy during the collision between an infinitesimal shell of delayed ejecta and the fs . the partition of the incoming ejecta energy is determined from energy and momentum conservation , and depends only on the ratio @xmath69 , with @xmath70 being the lorentz factor of the incoming ejecta . numerically , this factor is obtained from the kinematics of the catching - up between the freely flowing delayed ejecta and the decelerating fs . it can also be calculated analytically , if it is assumed that all the ejecta were released on a timescale much shorter than the observer time when the catching up occurs . with this assumption , it can be shown that the ejecta fs contrast lorentz factor is constant : @xmath71 once the dissipated fraction is known , we track the adiabatic conversion of the internal energy into kinetic , as described in panaitescu ( 1998 ) . although the concept of adiabatic losses implies that there will be a dispersion in the lorentz factor of the swept - up cbm , we ignore it and , just as for the other two models , we assume that all the fluid behind the fs moves at the same lorentz factor @xmath36 . even if the injected energy is negligible , the rs crossing the delayed ejecta can be of relevance for the lower frequency afterglow emission at days after the burst . to calculate this emission , we set the typical electron energy behind the rs by assuming that all the dissipated energy is in the shocked ejecta , which would be strictly correct only if the density of the shocked cbm were much larger than that of the delayed ejecta . otherwise , this assumption leads to an overestimation of the post - rs electron energy . after the energy injection ceases , we track the evolution of the rs electron distribution , subject to adiabatic and radiative cooling . the tracked electron distribution is used for the calculation of the rs continuum break - frequencies . in contrast , for all three models , the fs electron distribution is set using the current lorentz factor @xmath36 of the fs ( see below ) . this last approximation is more appropriate for a uniform medium than for a wind , as in the former case @xmath72 while for the latter @xmath73 constant . as presented above , the jet model has two dynamical parameters : the initial kinetic energy per solid angle @xmath27 and the initial jet opening @xmath26 . the so model has three such parameters : the energy per solid angle @xmath27 on the axis ( and in the uniform core ) , the core angular size @xmath74 , and the structural parameter @xmath75 for the outflow envelope . the ei model has three dynamical parameters : the total injected energy @xmath67 , the temporal index @xmath76 of the injection law , and the time @xmath77 when the injection episode ends , the initial @xmath27 being less relevant if @xmath78 . in addition , all models have another parameter which determines the outflow dynamics : the external medium density @xmath29 ( or the wind parameter @xmath79 ) . the observer location @xmath51 relative to the outflow symmetry axis is relevant only for the so model , but not for the jet model , as long as @xmath80 , or for the ei model , where the outflow is spherically symmetric . the calculation of the synchrotron and inverse compton emissions ( and of the radiative losses ) , requires three more parameters : one for the magnetic field strength , parameterized by the fraction @xmath81 of the post - shock energy in it : @xmath82 @xmath29 being the co - moving frame particle density in the unshocked fluid , @xmath83 the proton mass , one for the minimal electron energy behind the shock , parameterized by the fraction @xmath84 of the post - shock energy in electrons if all had the same energy @xmath85 : @xmath86 @xmath87 being the electron random lorentz factor and @xmath88 the electron mass , and the index @xmath89 of the electron power - law distribution with energy : @xmath90 equations ( [ b ] ) and ( [ gammai ] ) apply also to the rs if @xmath36 is replaced by the lorentz factor of the shocked ejecta as measured in the frame of the incoming ( unshocked ) ejecta . the parameters @xmath84 and @xmath81 determine the synchrotron characteristic frequency @xmath12 corresponding to the typical electron energy , as well as the self - absorption and cooling frequencies . we assume that the three microphysical parameters @xmath81 , @xmath84 , and @xmath89 , have the same value behind both the forward and reverse shock . summarizing , the jet model has six free parameters , the ei model seven , and the so model eight . the @xmath91-band extinction due to dust in the host galaxy , @xmath92 , which affects the observed slope of the optical sed and the overall optical flux , is an extra parameter for all models . to determine the host frame @xmath92 extinction , we assume an smc - like reddening curvefor the host galaxy and use the parameters inferred by pei ( 1992 ) for smc . all these parameters are constrained by the multiwavelength afterglow data in the following way . the pre - break decay index @xmath15 of the optical light - curve determines the electron distribution index @xmath89 for the jet and so models ( , [ [ a2s0 ] ] , [ [ a2s2 ] ] ) , and the energy injection index @xmath76 for the ei model ( ) . the post - break decay index @xmath16 overconstrains the electron index @xmath89 for the jet model , determines the structural parameter @xmath75 for the so model , and the electron index @xmath89 for the ei model . the jet opening @xmath26 for the jet model and the observer location @xmath51 for the so model ( if the observer is outside the core ) or the core angular size @xmath74 ( if the observer is within the core ) are set by the epoch of the optical light - curve break ( and [ [ tjs2 ] ] ) . the same observable determines the time @xmath77 when the energy injection ceases in the ei model . for the so model , whenever the observer is outside the outflow core , the @xmath74 is not well - constrained because it has a weak effect on the resulting afterglow emission . the outflow energy @xmath27 ( or the injected one @xmath67 ) , the medium density @xmath29 ( or the wind parameter @xmath79 ) , and the two microphysical parameters @xmath84 and @xmath81 are constrained by the afterglow spectral parameters flux at the peak of the spectrum , self - absorption , injection , and cooling frequencies which in turn are constrained by the relative intensities of the radio , optical and @xmath0-ray emissions ( wijers & galama 1999 , granot 1999 ) . that the self - absorption frequency is not always constrained by the radio observations accounts for part of the resulting parameter uncertainties , but note that the parameters @xmath27 , @xmath29 , @xmath81 , and @xmath84 are also constrained by matching the radio emission at late times , when the blast - wave is only mildly relativistic , after the self - similar relativistic dynamics phase employed in analytical calculations of afterglow parameters . for the ei model , the initial outflow energy @xmath27 is irrelevant , as the blast - wave kinetic energy is that injected starting at a time before the first observation . finally , the host extinction @xmath92 is constrained by the ratio of the observed optical flux to the intrinsic optical flux which is consistent with the @xmath0-ray emission ( for the other afterglow parameters ) . table [ tchi ] lists the reduced chi - square , @xmath93 , for the best fits obtained with the jet , so , and ei models , for the two types of circumburst media : homogeneous and wind - like . for a uniform cbm , the best fits with the ei model are very poor and , for almost all afterglows , they are substantially worse than those for a wind . this is so because , for a uniform cbm , the observed post - break optical decay index @xmath16 ( given by equation [ [ a1sph ] ] ) requires a higher electron index @xmath89 than for a wind , leading to an optical sed slope ( @xmath18 ) and an optical to@xmath0-ray spectral index ( @xmath94 ) larger than those observed . also , the ei model has difficulties in accommodating both the radio and @xmath0-ray fluxes , with the fs emission either overestimating the observed radio flux or underestimating the @xmath0-ray flux . [ cols="<,^,^,^,^,^,^,^,^ " , ] figs . [ 0519jo][0226jo ] display the best fits obtained with the jet model for the afterglows 010222 , 011211 , 020813 , and 030226 , which we did not present previously , and comparable or better fits obtained with the so and ei models . with the exception of the jet model and a uniform cbm for the afterglow 990510 , all best fits presented are not acceptable in a statistical sense , as they have @xmath95 . in general , fits with @xmath96 appear adequate upon visual inspection , a significant fraction of the @xmath97 arising often from optical measurements ( which have the smallest uncertainties ) , indicating either a small , short - lived afterglow fluctuation or , perhaps , an occasional outlier resulting from underestimated observational errors . in some cases there are also systematic discrepancies between the model fluxes and observations , most often in the radio . as a general rule , if the fits are poorer than @xmath98 for both types of circumburst media then the fits are not shown , with the exception of those obtained with the jet and so models for the afterglows 011211 and 030226 , for which we want to show that the latter model is much better than the former in accommodating the 1 day , sharp optical light - curve break . also as a general rule , if the fits obtained with the jet and so models are comparable , only one is shown in the figures . for clarity , the figures contain only the radio optical frequencies where observations span the longest time , but the fitted data set included few or several other bands in these two domains ( see figure captions ) . with the exception of the afterglow 000926 , the available @xmath0-ray measurements are in only one band , which we use to infer the @xmath0-ray flux at the mid energy of that band . the amplitude of interstellar scintillation , whose calculation is based on the treatment and maps given by walker ( 1998 ) , is indicated with vertical bars at the time of the radio observations . in all figures , the best fit for a uniform cbm is shown with continuous lines while that for a wind medium with dotted lines . the best fit obtained with the jet model is discussed in panaitescu & kumar ( 2002 ) . the addition of more radio data and some optical outliers ( which were previously excluded ) lead now to a worse fit ( @xmath97-wise ) but to the same afterglow parameters , within their uncertainty . the best fit for the so model , shown in fig . [ 0519jo ] , has a @xmath93 comparable to that for the jet model ( table 1 ) but is qualitatively poorer , as it overestimates the observed early radio . although the measured radio fluxes are within the amplitude of the fluctuations caused by the interstellar galactic gas , such a constant offset can not be caused by scintillation ( unfortunately , @xmath97-statistics does not reflect the smaller probability of systematic discrepancies ) . the best fit obtained with the jet model and a uniform medium is presented in panaitescu & kumar ( 2001 ) . there we assumed that the radio flare seen at 1 day ( kulkarni 1999 ) arises in the grb ejecta energized by the rs , and included in the modelling only the radio measurements at 330 days , when the 8 ghz flux of this afterglow is less than @xmath99 . this sets an upper limit on the fs peak flux , @xmath100 , which , together with that @xmath101 before the jet - break and @xmath102 after that , requires a small cbm density : @xmath103 . including the radio measurements before a few days , as well as the millimeter data , the @xmath104-band fluxes after 10 days ( with which the @xmath104-band light - curve appears to be a single power - law , unlike the @xmath105-band emission , which exhibits a steepening at @xmath106 days ) , and other optical measurements previously left out of our modelling , we find now a poorer fit ( @xmath107 for a uniform cbm , @xmath108 for a wind ) , but with the same afterglow parameters as before . we have also tested the ability of the jet model including the emission from the rs crossing the grb ejecta to explain the optical flash observed at 20600 seconds ( akerlof 1999 ) , which peaked 50 s after the burst at about 0.8 jy , and the 1 day radio flare , as proposed by sari & piran ( 1999 ) . given that the observer injection timescale ( @xmath109 s ) is comparable to the burst duration , the approximation of instantaneous ejecta release and its consequence given in equation ( [ gi ] ) may be invalid . for this reason , we let the ejecta lorentz factor , @xmath70 , to be a free parameter of the fit . together with the injected energy , @xmath67 , it determines the number of the radiating electrons in the grb ejecta and the lorentz factor of the rs , the peak flux and frequency of the synchrotron rs emission . the best fits obtained with the jet model interacting with a tenuous medium ( so that the fs emission accommodates the late radio measurements ) and emission from the grb ejecta are shown in fig . [ 0123jet+rs ] . we find that , for the same microphysical parameters @xmath84 and @xmath81 behind both shocks , the peak rs optical emission is dimmer than observed by a factor 50 for a uniform cbm and a factor 5 for a wind , the latter yielding a good fit to the remainder of the early optical measurements ( but not as good as a uniform medium to the 1 day optical break ) . we also find that , under the same assumption of equal microphysical parameters , the rs emission peaks at 8 ghz at 0.1 day , a factor 10 too early . in other words , taking into account the cooling of the rs electrons accelerated at @xmath109 s and the decrease on the synchrotron self - absorption optical thickness of the rs at radio frequencies , we do not find a set of afterglow parameters for which the rs emission can peak in the radio as late a 1 day while , for the same microphysical parameters , the fs emission can accommodate the rest of afterglow observations . to explain the optical flash of the afterglow 990123 , the grb ejecta must have a larger parameter @xmath81 than for the forward shock , indicating that the ejecta is magnetized ( zhang , kobayashi & mszros 2003 , panaitescu & kumar 2004 ) , a feature that may be required for by the early optical emission of the afterglow 021211 as well ( kumar & panaitescu 2003 , zhang 2003 ) . however , for the low density jet model , an injection of ejecta in the fs at about the same time when the radio flare is seen allows this model to accommodate that flare , as shown in fig . [ 0123jet+ei ] . note that the injected energy is less than that required for the fs to explain the rest of the observations , the incoming ejecta provide only the fresh electrons to radiate at radio frequencies , but do not alter the dynamics of the fs . after including the 1 day radio flare measurements , as well as the millimetre measurements at 110 days , we also find with the jet model a higher density solution ( fig . [ 0123jet ] ) , @xmath110 , which is close to that obtained for other afterglows . this higher densityis required by the larger fs peak flux necessary to accommodate the 1 day flare , when @xmath111 mjy . for a uniform cbm , the high density fit is slightly better than the low density solution , the improvement being more substantial for a wind . note , however , that the former underestimates the radio measurements at 30 days . the so model yields fits of the same quality as the jet model ( table 1 ) . the best fit structural parameter satisfies @xmath112 for a uniform medium and @xmath113 for a wind ( table [ tso ] ) , thus the post - break afterglow emission arises mostly from the outflow core in the former case and from the envelope for the latter . the best fit obtained with the ei model for a wind medium is shown in fig . [ 0123eis2 ] . @xmath97-wise , this low density model provides a better fit than the jet and so models , although it overestimates the early radio measurements . also , this solution has an extremely low wind density ( table [ tei ] ) , about 100 times smaller than known for galactic wr stars . for such a tenuous wind , equation ( [ tjs2 ] ) implies that , if the outflow opening @xmath26 is wider than @xmath114 , then the jet - break would appear later than about 30 days ( after the latest measurement ) , and the corresponding outflow energy would be larger than @xmath115 ergs , which is less than that of the jet shown in fig . [ 0123jet ] . the best fit obtained for a uniform medium is very poor ( it either overestimates the radio emission or underproduces @xmath0-rays ) and is not shown . the best with the jet model for a uniform cbm is shown in panaitescu & kumar ( 2001 ) . a wind provides a poorer fit ( table 1 ) , as it fails to accommodate the strong break of the optical light - curves of this afterglow . structured outflows yield poorer fits for either type of medium , the best fit structural parameter being @xmath116 for a uniform cbm and @xmath117 for a wind . the best fit obtained with the jet model and a uniform cbm is shown in panaitescu & kumar ( 2001 ) . because the radio decay of this afterglow , @xmath118 , is significantly slower than that measured at optical wavelengths after the break , @xmath119 , in our previous modelling we have employed a double power - law electron distribution , harder ( smaller index @xmath89 ) at low energies , to accommodate the shallower radio decay , and a softer ( larger index @xmath89 ) at high energies , to explain the post - break optical light - curve decay . here we use a single power - law electron distribution , to assess the ability of each model to explain all the data without recourse to a light - curve decay steepening originating in a break in the electron distribution . consequently , the jet model ( as well as any other single emission component model ) will not accommodate the decays of both the radio and optical emissions of the afterglow 991216 , as illustrated by the higher @xmath93 given in table 1 and the best fit shown in fig . [ 1216jet ] . the addition of emission from the rs which energizes the ejecta catching up with the fs at about 1 day improves the radio fit , as shown in fig . [ 1216jet+ei ] . a slower decay radio light - curve is obtained as the fs emission overtakes that from the rs at about 10 days . as for the afterglow 990123 ( fig . [ 0123jet+ei ] ) , the energy injected is small enough that it does not alter the dynamics of the fs . however , the afterglow parameters of the jet+ei model for 991216 are different than those obtained with the jet model ( fig . [ 1216jet ] ) because we require now that the fs peaks in the radio at a later time and at a lower flux ( @xmath120 mjy ) . the so model yields a best fit which is very similar to that shown in fig . [ 1216jet ] for the jet model . the smaller intensity - averaged source size resulting in the so model leads to a larger amplitude for the interstellar scintillation and , thus , to a better @xmath97-wise fit ( table 1 ) , but this model also fails to accommodate the slow radio decay . the best fit structural parameter satisfies @xmath121 ( table [ tso ] ) , the outflow envelope emission dominates over that from the core after the optical break . the best fit for the jet model is discussed in panaitescu ( 2001 ) . as for the afterglow 991216 , the slower radio decay of 000301c , @xmath122 , and its faster optical fall - off , @xmath123 , prompted us to consider a double power - law electron distribution . furthermore , the large break magnitude @xmath124 exhibited by the optical light - curve 000301c ( the largest in the entire sample - table 1 ) exceeds that allowed by the jet model for the electron distribution index @xmath89 required by the pre - break decay index and optical sed slope ( panaitescu 2005 ) . such a strong break is also better explained if there is a contribution from the passage of a spectral break . however , since we want to test the three models for light - curve break without any contribution from another mechanism , we list in table 1 the best fit obtained previously with a soft electron distribution . [ 0301jo ] shows the best fit obtained with the so model , which is only slightly better than that of the jet model , has similar parameters , and shares the same deficiencies : it fails to explain the sharpness of the optical light - curve break and yields a radio emission decaying faster than observed . we note that optical measurements between 3.0 and 4.5 days , when a light - curve bump is seen , have been excluded from the fit . the best fit obtained with the jet model is shown in panaitescu & kumar ( 2002 ) . after including in the data set the near - infrared measurements , the noisy @xmath125-band , and some optical outliers which were previously excluded , the best fit is @xmath97-wise poorer but the fit parameters are the same as before ( within their uncertainty ) . an equally good fit can be obtained with the so model ( fig . [ 0926jo ] ) . we note that for both the jet and s0 models , the @xmath0-ray emission is mostly inverse compton scatterings . the best fit obtained with the jet model is shown in fig . [ 0222jet ] . note that the model radio light - curve decays after 1 day faster than observed . for a uniform cbm , the best fit obtained with a structured outflow is slightly better in the radio , but it too fails to explain the slow radio decay . for a wind , the best fit obtained with the so model overestimates the millimeter emission [ @xmath126 mjy , @xmath127 mjy ] attributed to the host galaxy by frail ( 2002 ) , who also suggest a possible host synchrotron emission at 8.5 ghz of @xmath128 . this is only marginally consistent with the @xmath129 reported by galama ( 2003 ) and the host synchrotron sed , @xmath130 , adopted by frail ( 2002 ) , which imply a host flux @xmath131 . if we subtract a radio host contribution of @xmath132 from the 44 radio measurements , then , within the jet model , the contribution to @xmath97 of the radio data decreases from 107 to 72 for a uniform medium , and from 103 to 53 for a wind , with similar changes for the so model . though these are significant improvements , the best fit still underestimates the radio flux measured after 10 days , because the optical measurements have a smaller uncertainty and determine the electron index @xmath89 and , implicitly , the decay of the model radio emission . in general , one - component models can not accommodate both the host - subtracted radio emission , @xmath133 , of the afterglow 010222 at 1200 days , and the optical decay , @xmath134 , observed at 10100 days . a better fit to the radio emission of the afterglow 010222 may be obtained with the jet model if , in addition to the fs emission , there is a rs contribution to the early radio afterglow , as illustrated in fig . [ 0222jet+ei ] for a uniform cbm . just as for the afterglows 990123 and 991216 , the incoming ejecta carry less energy than that in the fs ( and do not alter its dynamics ) and the medium density is lower than for the jet model because the fs peak flux must be smaller , to match the radio flux measured after 10 days , when the fs synchrotron peak frequency crosses the radio domain . as illustrated in fig . [ 1211jet ] , even for a uniform cbm , the jet model has difficulty in accommodating the sharp break exhibited by the optical light - curve of this afterglow , assuming that the 12 day optical emission is not a fluctuation . such fluctuations have been seen in the afterglows 000301c , 021004 , and 030329 , but the lack of a continuous monitoring for 011211 prevents us to determine if its optical emission at 1 day is indeed a fluctuation . a stronger break can be obtained with the so model when the brighter , outflow core becomes visible to an observer located outside it , hence the better fit that the so model yields ( fig . [ 1211jo ] ) . the best fit obtained with the jet model is shown in fig . [ 0813jet ] . structured outflows yield fits with @xmath112 for a uniform cbm and @xmath113 for a wind , with the post - break emission arising mostly in the outflow core and envelope , respectively . a significant improvement over the jet model is obtained only for a uniform cbm ( table 1 ) . a fit of comparable quality is also obtained with the ei model for a wind medium ( fig . [ 0813eis2 ] ) , but not for a uniform cbm , which overproduces radio emission and largely underestimates the observed @xmath0-ray fluxes . for the best fit parameters given in table [ tei ] , equation ( [ tjs2 ] ) shows that a jet - break would occur after the last measurement ( 40 days ) if the outflow opening were larger than @xmath135 , which implies a minimum collimated energy of @xmath136 ergs , which is 3 times larger than that of the jet shown in fig . [ 0813jet ] . therefore , just as for 990123 , the ei model may not involve significantly larger energies than the jet model , if allowance is made for a possible outflow collimation . as for 011211 , the strong optical light - curve break of this afterglow is better accommodated by the so model than the jet model ( fig . [ 0226jet ] and [ 0226jo ] ) , though an acceptable fit to the optical break is obtained only for a uniform cbm ( in addition , for a wind medium , the so model overproduces radio emission ) . however , given the poor coverage before 0.3 days , it is possible that the measured optical emission at that time is a fluctuation below that expected in the jet model . although the decay of the optical emission of this afterglow does not exhibit a steepening , being so far the longest monitored power - law fall - off ( 2300 days ) , it can be explained with the jet model if its unusual rise at 12 days is interpreted as a jet becoming visible to an observer located outside the initial jet opening ( see fig . 1 in panaitescu & kumar 2002 ) . this requires a rather wide jet , with @xmath137 , and yields a poor fit to the @xmath0-ray emission ( @xmath138 ) . a less energetic envelope surrounding this jet would produce the prompt @xmath139-ray emission and would account for the flat optical emission prior to the 2 day optical rise , hence this model is similar to a structured outflow with a wide uniform core and an envelope whose structure is poorly constrained by observations . an other possible reason for the 2 day sharp rise of this afterglow is a powerful energy injection episode in which the incoming ejecta carry an energy larger than that already existing in the fs ( panaitescu , mszros & rees 1998 ) , the blast - wave dynamics is altered drastically by the energy injection . [ 0508ei ] shows the best with the ei model . note that the given that the light - curve rise is achromatic , as it arises from the blast - wave dynamics , which explains why the @xmath0-ray light - curve can not be accommodated . in this work we expand our previous investigation ( panaitescu & kumar 2003 ) of the * jet model * to four other grb afterglows for which an optical light - curve break was observed : 010222 ( for which radio observations were not available at the time of our previous modelling ) , 011211 , 020813 , and 030226 . we also note that , here , we have not employed a broken power - law electron distribution to explain both the slower decaying radio emission and the faster falling - off optical light - curve observed for some afterglows ( 991216 and 000301c ) . our prior conclusion that a homogeneous circumburst medium provides a better fit to the broadband afterglow emission than environments with a wind - like stratification stands ( table 1 ) . for afterglows with a sharp steepening @xmath10 of the optical light - curve decay ( 990510 , 000301c , 011211 , and 030226 ) , this is due to that the transition between the jet asymptotic dynamical regimes ( spherical expansion at early times and lateral spreading afterward ) occurs faster for a homogeneous medium than for a wind ( kumar & panaitescu 2000 ) . for the afterglows with shallower breaks , a wind medium ( as expected if the grb progenitor is a massive star ) provides an equally good fit as ( or , at least , not much worse than ) a homogeneous medium . for the acceptable fits ( table 2 ) obtained with the jet model and for a uniform medium , the jet initial kinetic energy , @xmath140 , is between 2 and 6 @xmath141 ergs , the initial jet opening angle , @xmath26 , is between @xmath114 and @xmath142 ( with 000926 an outlier at @xmath143 ) , and the medium density , @xmath29 , between 0.05 and 1 @xmath144 ( with 000926 an outlier , again , at @xmath145 ) . for a wind , @xmath140 has about the same range , @xmath26 ranges from @xmath114 to @xmath146 , while the wind parameter @xmath79 is between 0.1 and 2 . 70% of the 64 galactic wr stars analyzed by nugis & lamers ( 2000 ) have a @xmath79 parameter in this range ; for the rest @xmath147 . we note that the @xmath140 resulting from our numerical fits is uncertain by a factor 2 , @xmath26 by about 30% , @xmath29 by almost one order of magnitude , and @xmath79 by a factor of 2 . although the inferred wind density parameter @xmath79 is in accord with the measurements for wr stars a uniform medium is instead favoured by the fits to the afterglow data . this is an interesting issue for the established origin of long bursts in the collapse of massive stars . wijers ( 2001 ) has proposed that the termination shock resulting from the interaction between the wind and the circumstellar medium could homogenize the wind . the numerical hydrodynamical calculations of garcia - segura , langer & mac low ( 1996 ) show that , for a negligible pressure @xmath148 of the circumstellar medium ( about @xmath149 ) , the termination shock of the rsg wind has a radius of 10 pc , outside a shell of uniform density is form . however , this radius is much larger than the distance of about 1 pc where the afterglow emission is produced . chevalier , li & fransson ( 2004 ) made the case that , if the burst occurred in an intense starburst region , where the interstellar pressure is about @xmath150 , then the shocked rsg wind would form bubble of uniform density of about @xmath151 extending from 0.4 to 1.6 pc , in accord with the fit densities and the blast - wave radius @xmath152^{1/4}$ ] pc , resulting from equation ( [ gs0 ] ) . the * structured outflow model * involving an energetic , uniform core , surrounded by a power - law envelope ( equation [ [ eq ] ] ) which impedes the lateral spreading of the core , retains most of the ability of the jet model to yield a steeply decaying optical light - curve after the core ( or its boundary ) becomes visible to an observer located outside the core opening ( or within it ) , while accommodating , at the same time , the observed slope of the optical spectral energy distribution ( sed ) . this is so because , in both models , more than half of the increase @xmath10 in the light - curve power - law decay index is due to the finite opening of the bright(er ) ejecta . for a jet , the remainder of the steepening @xmath10 is caused by the lateral spreading . for a structured outflow , an extra contribution to the steepening , which can lead to a @xmath10 even larger than that produced by a jet , results if the observer is located outside the core opening . for a uniform medium , we find that the structured outflow model provides a better fit than the jet model for 6 of the 10 afterglows analyzed here ( table 1 ) . for a wind , the former model works better for three afterglows , fits of comparable quality being obtained for the other six . only the afterglow 990510 is better explained with a jet , for either type of circumburst medium . within the framework of structured outflows , 8 out of 10 afterglows are better fit with a uniform medium than with a wind , which is partly due to that , just as for a jet , the light - curve steepening is sharper if the medium is uniform ( panaitescu & kumar 2003 ) . for the `` acceptable '' fits ( @xmath96 ) obtained with the so model , the ejecta kinetic energy per solid angle in the core , @xmath27 , ranges from 0.5 to 3 @xmath153 for either type of medium , with an uncertainty of a factor 3 . the angular opening of the core , @xmath74 , is between @xmath154 and @xmath155 for a homogeneous medium , and about @xmath155 for a wind ( 991216 being an outlier at @xmath154 ) , with an uncertainty less than 50% . the structural parameter @xmath75 ( equation [ [ eq ] ] ) , which characterizes the energy distribution in the outflow envelope , is found to be between 1.5 and 2.7 for a uniform medium , and between 1.0 and 2.3 for a wind ( with 000926 being an outlier at @xmath156 ) . these values are consistent with that proposed by rossi ( 2002 ) , @xmath157 , to explain the quasi - universal jet energy resulting in the jet model . for structured outflows , the ejecta kinetic energy contained within an opening of twice the location of the observer ( relative to the outflow symmetry axis ) , @xmath51 , ranges from 1 to 4 @xmath141 ergs , for either type of medium . hence , the energy budget required by the structured outflow model is very similar to that of the jet model . the best fit medium densities obtained with the structured outflow model are similar to those resulting for a jet . we find that the * energy injection model * , where a light - curve decay steepening is attributed to the cessation of energy injection in the forward shock , can be at work only in two afterglows , 990123 and 020813 . the major shortcoming of the ei model is that , in order to explain the steep fall - off of the optical light - curve after the break with a spherical outflow , an electron distribution that is too soft ( an index @xmath89 too large ) is required to accommodate the relative intensities of the radio and @xmath0-ray emissions . low density solutions ( @xmath158 , @xmath159 ) , for which the @xmath0-ray emission is mostly synchrotron , yield a cooling frequency that is too low , resulting in model @xmath0-ray fluxes underestimating the observations by a factor of at least 10 . high density solutions ( @xmath160 , @xmath161 ) , for which the @xmath0-ray is mostly inverse compton scatterings , produce a forward shock peak flux that is too large , which leads to model radio fluxes overestimating the data by a factor of 10 . we note that , due to adiabatic cooling , the grb ejecta electrons which were energized by the reverse shock during the burst , do not contribute significantly to the afterglow radio emission at days after the burst . numerically , we find that , if the microphysical parameters are the same for both the reverse and forward shocks , then the 1 day radio flare of the afterglow 990123 can not be explained with the emission from the grb ejecta electrons , as proposed by sari & piran ( 1999 ) , as the synchrotron emission from these electrons peaks in the radio earlier than 0.1 day , regardless of how relativistic is the reverse shock . for the reverse shock emission to be significant at 1 day after the burst , there must be some other ejecta catching up with the decelerating outflow at that time or the assumption of equal microphysical parameters for both shocks must be invalid ( panaitescu & kumar 2004 ) . furthermore , the optical emission from the ejecta electrons is below that observed at 100 s ( akerlof 1999 ) by a factor 5 for a wind and a factor 50 for a uniform medium . a delayed injection of ejecta , carrying less kinetic energy than that already existing in the forward shock ( not altering the forward shock dynamics ) , may be at work in the afterglows 990123 , 991216 , and 010222 , whose radio decay is substantially slower than that observed in the optical after the break . as shown by us ( panaitescu & kumar 2004 ) , such a decoupling of the radio and optical emissions can not be explained by models based on a single emission component ( forward shock ) and require the contribution of the reverse shock [ 0123jet+ei ] , [ 1216jet+ei ] , and [ 0222jet+ei ] , illustrate that the sum of the reverse and forward shock emissions yields a slower decaying radio emission , consistent with the observations . however , for the forward shock peak flux to be as dim as observed in the radio after 10 days , when the forward shock peak frequency decreases to 10 ghz , an extremely tenuous medium is required : @xmath162 or @xmath163 . the former might be compatible with a massive star progenitor if grbs occur in the `` superbubble '' blown by preceding supernovae exploding in the same molecular cloud ( scalo & wheeler 2001 ) , while the latter points to a progenitor with a low mass and low metallicity or , otherwise , to a small mass - loss rate and high wind speed during the last few thousand years before the collapse , when the environment within 1 pc is shaped by the stellar wind . as a final conclusion , we note that the structured outflow model is a serious contender to the jet model in accommodating the broadband emission of grb afterglows with optical light - curve breaks . in both models , the best fit parameters describing the ejecta kinetic energy jet opening & energy or core opening & energy density along the outflow symmetry axis have narrow distributions , hinting to a possible universality of these parameters . this work was supported by nsf grant ast-0406878 . + some or all of the radio data for the afterglows 011211 , 020813 , and 030226 referred to in this paper were drawn from the grb large program at the vla + ( http://www.vla.nrao.edu/astro/prop/largeprop ) and can be found at dale frail s website + http://www.aoc.nrao.edu/ dfrail / allgrb_table.shtml . akerlof c. , 1999 , nature , 398 , 400 berger e. , 2000 , apj , 5454 , 56 bertoldi f. , 2003 , gcn circ . 1497 + ( http://gcn.gsfc.nasa.gov/gcn3/1497.gcn3 ) borozdin k. , trodolybov s. , 2003 , apj , 583 , l57 bremer m. , 1998 , a&a , 332 , l13 butler n. , 2003 , apj , 597 , 1010 chary r. , 1998 , 498 , l9 chevalier r. , li z. , 2000 , apj , 536 , 195 chevalier r. , li z. , fransson c. , 2004 , apj , 606 , 369 corbet r. , 1999 , gcn circ . 506 + ( http://gcn.gsfc.nasa.gov/gcn3/506.gcn3 ) covino s. , 2003 , a&a , 404 , l5 fox d. , 2003 , nature , 422 , 284 frail d. , waxman e. , kulkarni s. , 2000a , apj , 537 , 191 frail d. , 2000b , apj , 534 , 559 frail d. , 2000c , apj , 538 , l129 frail d. , 2002 , apj , 565 , 829 frail d. , 2003 , aj , 125 , 2299 frail d. , 2005 , apj , 619 , 994 fynbo j. , 2001 , a&a , 373 , 796 galama t. , 1998 , apj , 497 , l13 galama t. , 1999 , nature , 398 , 394 galama t. , 2000 , apj , 541 , l45 galama t. , 2003 , apj , 587 , 135 garcia m. , 1998 , apj , 500 , l105 garcia - segura g. , langer n. , mac low m. , 1996 , a&a , 316 , 133 garnavich p. , 2000 , apj , 543 , 61 goodman j. , 1997 , new astronomy , 2 , 449 gorosabel j. , 2004 , a&a , 422 , 113 granot j. , piran t. , sari r. , 1999 , apj , 527 , 236 granot j. , panaitescu a. , kumar p. , woosley s. , 2002 , apj , 570 , l61 halpern j. , 1999 , apj , 517 , l105 halpern j. , 2000 , apj , 543 , 697 harrison f. , 2001 , apj , 559 , 123 holland s. , 2002 , apj , 124 , 639 int zand j. , 2001 , apj , 571 , 876 jakobsson p. , 2001 , a&a , 408 , 941 jaunsen , a. 2001 , apj , 546 , 127 jensen b. , 2001 , a&a 370 , 909 klose s. , 2004 , aj , 128 , 1942 kulkarni s. , 1999 , nature , 398 , 394 kumar p. , panaitescu a. , 2000 , apj , 541 , l9 kumar p. , panaitescu a. , 2003 , mnras , 346 , 905 masetti n. , 2001 , a&a , 374 , 382 mszros p. , rees m.j . , 1997 , apj , 476 , 232 mszros p. , rees m.j . , wijers r. , 1998 , apj , 499 , 301 nicastro l. , 1999 , a&a , 138 , s437 nugis t. , lamers h. , 2000 , a&a , 360 , 227 paczyski b. , 1998 , apj , 494 , l45 paczyski b. , rhoads j. , 1993 , apj , 418 , l5 pandey s. , 2004 , a&a , 417 , 919 panaitescu a. , 2001 , apj , 556 , 1002 panaitescu a. , 2004 , mnras , 350 , 213 panaitescu a. , 2005 , mnras , in press + ( http://xxx.lanl.gov/ps/astro-ph/0506577 ) panaitescu a. , kumar p. , 2000 , apj , 543 , 66 panaitescu a. , kumar p. , 2001 , apj , 554 , 667 ( pk01 ) panaitescu a. , kumar p. , 2002 , apj , 571 , 779 ( pk02 ) panaitescu a. , kumar p. , 2003 , apj , 592 , 390 panaitescu a. , kumar p. , 2004 , mnras , 353 , 511 panaitescu a. , mszros p. , rees m. , 1998 , apj , 503 , 314 pedersen h. , 1998 , apj , 496 , 311 pei y. , 1992 , apj , 395 , 130 piro l. , 2000 , preprint + ( http://xxx.lanl.gov/ps/astro-ph/0001436 ) piro l. , 1998 , a&a , 331 , l41 piro l. , 1999 , gcn circ . 500 + ( http://gcn.gsfc.nasa.gov/gcn3/500.gcn3 ) piro l. , 2001 , apj , 558 , 442 price p. , 2001 , apj , 549 , l7 rees m.j . , mszros p. , 1998 , apj , 496 , l1 rhoads j. , 1999 , apj , 525 , 737 rhoads j. , fruchter a. , 2001 , apj , 546 , 117 rossi e. , lazzati d. , rees , m.j . , 2002 , mnras , 332 , 945 sahu k. , 1997 , apj , 489 , l127 sari p. , piran t. , 1999 , apj , 517 , l109 sari p. , mszros p. , 2000 , apj , 535 , l33 sari r. , narayan r. , piran t. , 1998 , apj , 497 , l1 sari , r. , piran , t. , halpern j. , 1999 , apj , 519 , l17 scalo j. , wheeler j , 2001 , apj , 562 , 664 smith i. , 2001 , a&a , 380 , 81 sokolov v. , 1998 , a&a , 334 , 117 stanek k. , 2001 , apj , 563 , 592 takeshima t. , 1999 , gcn circ . 478 + ( http://gcn.gsfc.nasa.gov/gcn3/478.gcn3 ) vrba f. , 1999 , apj , 528 , 254 walker m. , 1998 , mnras , 294 , 307 waxman e. , kulkarni s. , frail d. , 1998 , apj , 497 , 288 wijers r. , galama t. , 1999 , apj , 523 , 177 wijers r. , 2001 , proc . of second rome workshop `` grbs in the afterglow era '' , eds . e. costa , f. frontera , j. hjorth ( berlin : springer ) , 306 woosley , s. 1993 , apj , 405 , 273 zhang b. , mszros p. , 2002 , apj , 571 , 876 zhang b. , kobayashi s. , mszros p. , 2003 , apj , 595 , 950

the approximability of a convex body is a number which measures the difficulty to approximate that body by polytopes . we prove that twice the approximability is equal to the volume entropy for a hilbert geometry in dimension two end three and that in higher dimension it is a lower bound of the entropy . as a corollary we solve the entropy upper bound conjecture in dimension three and give a new proof in dimension two from the one found in .

1207.1342

hilbert geometries are a family of metric spaces defined in the interior of an open and bounded convex set using cross - ratios following the construction of the hyperbolic geometry s projective model @xcite . they are actually length space with an underlying finsler structure which is riemannian only if the convex set is an ellipsoid @xcite . the present paper focuses on the volume growth of these geometries and more specifically on the volume entropy . this immediately raises the problem of the definition of the volume one uses , as in finsler geometry there is nt uniqueness of the notion of volume . in this paper we will mostly work with two volumes : the _ busemann volume _ which also appears as the hausdorff measure associated to the distance , and the _ holmes - thompson _ volume which is built thanks to the natural symplectic structure of the cotangent bundle . the first one is more convenient for computational purposes , but the second one possesses better and crucial properties , such as minimality property of totally geodesic hypersurface @xcite . both happen to be equivalent and belong to a larger family of measures over the hilbert geometries that we call _ normalised density measures _ and for which all the results presented in this paper coincide . therefore in a open convex set @xmath0 , we will consider a member @xmath1 of this family of measures and if we denote by @xmath2 the metric ball of radius @xmath3 centred at the point @xmath4 , then the volume entropy of @xmath0 will be defined as @xmath5 let us stress out that thus defined , the entropy of @xmath0 does not depend on either the base point @xmath6 nor on the normalised density measure chosen and is actually a projective invariant attached to @xmath0 ( one can also replace the `` @xmath7 '' by a `` @xmath8 '' in all the definitions of this paper , and get the same results for the corresponding numbers ) . the question we address in this essay is the _ entropy upper bound conjecture _ which states that if @xmath0 is an open and bounded convex subset of @xmath9 , then @xmath10 . let us recall the main related results . the first one is a complete answer to the the conjecture in the two dimensional case by g. berck & a. bernig & c. vernicos @xcite ] , where the authors actually obtained an upper bound as a function of @xmath11 , the upper minkowski dimension ( or _ ball - box _ dimension ) of the set of extremal points of @xmath0 , as follows : @xmath12 the second results is a more precise statement with respect to the asymptotic volume growth of balls and involves another projective invariant introduced by g. berck & a. bernig & c. vernicos @xcite called the _ centro - projective _ area of @xmath0 defined by @xmath13 where for any @xmath14 , @xmath15 is the gauss curvature , @xmath16 the normal and @xmath17 the antipodal map defined by @xmath18 . let us remind the reader that both @xmath19 and @xmath20 are defined almost everywhere as alexandroff s theorem states @xcite . now we can state the second main result of g. berck & a. bernig & c. vernicos @xcite which encloses former results by b. colbois and p. verovic @xcite . if @xmath21 is @xmath22 , then @xmath23 and @xmath24 . moreover , without any assumption on @xmath25 , if @xmath26 then @xmath27 . the third one which is also a rigidity results requires stronger assumptions on the hilbert geometries : they have to be divisible , which means that they admit a compact quotient , and they have to be hyperbolic in the sense of gromov , which implies by y. benoist @xcite that their boundary is @xmath28 and strictly convex . let us stress out that among them only the hyperbolic geometry has a @xmath22 boundary , and as hyperbolicity implies non - nullity of the cheeger constant(see b. colbois and c. vernicos @xcite ) , their entropy is strictly positive . m. crampon @xcite results states that for a divisible open bounded convex set @xmath0 in @xmath9 whose boundary is @xmath28 , * @xmath10 ; * and equality occurs if and only if @xmath0 is an ellipsoid . in the present paper we relate the entropy to another invariant associated to a convex set , its _ approximability _ whose name was introduced by schneider and wieacker @xcite and which somewhat measure how well a convex set can be approximated by polytopes . more precisely , let @xmath29 be the smallest number of vertices of a polytope whose hausdorff distance to @xmath0 is less than @xmath30 . then the approximability of @xmath0 is @xmath31 the main result which is of interest for this essay is the following upper bound , obtained by fejes - toth @xcite in dimension @xmath32 and by bronshteyn - ivanov @xcite in the general case : if @xmath0 is a bounded open convex set in @xmath9 , then @xmath33 our main result it the following one let @xmath34 be a bounded open convex set , then * @xmath35 ; * if @xmath36 or @xmath37 then @xmath38 . the first important corollary is that it gives a proof of the entropy upper bound conjecture in dimension @xmath32 and @xmath37 . let @xmath0 be a bounded open convex set in @xmath9 , with @xmath39 , then @xmath10 . the second one tells us that in dimension @xmath32 there are hilbert geometries with intermediate volume growth . let @xmath40 be an increasing function such that @xmath41 then there exist an open bounded convex set @xmath42 in @xmath43 , @xmath44 , and three constants @xmath45 , @xmath46 and @xmath47 , such that for @xmath48 one has @xmath49 therefore there are convex sets @xmath0 with maximal entropy and zero centro - projective area , and there are convex sets with zero entropy which are not polytopes . this corollary directly follows from our proof of the main theorem ( see section [ lesinegalites ] ) and schneider and wieacker @xcite results on the approximability in dimension @xmath32 . the last statement follows from our work @xcite , where we showed that polytopes have polynomial growth of order @xmath50 in dimension two . the equalities and inequalities also implies the following four new results , let @xmath51 be a convex body , * @xmath52 , where @xmath53 is the hausdorff dimension of the farthest points of @xmath0 . * if @xmath36 or @xmath37 , * * @xmath54 is a projective invariant of @xmath0 . * * let @xmath55 be the polar dual of @xmath0 , then @xmath56 . * if @xmath36 , then @xmath57 . a _ proper _ open set in @xmath9 is a set not containing a whole line . a hilbert geometry @xmath58 is a non empty _ proper _ open convex set @xmath0 in @xmath9 ( that we shall call _ convex domain _ ) with the hilbert distance @xmath59 defined as follows : for any distinct points @xmath6 and @xmath60 in @xmath0 , the line passing through @xmath6 and @xmath60 meets the boundary @xmath61 of @xmath0 at two points @xmath62 and @xmath63 , such that one walking on the line goes consecutively by @xmath62 , @xmath6 , @xmath60 @xmath63 . then we define @xmath64,\ ] ] where @xmath65 $ ] is the cross ratio of @xmath66 , i.e. , @xmath67 = \frac{\| q - a \|}{\| p - a \| } \times \frac{\| p - b \|}{\| q - b\| } > 1,\ ] ] with @xmath68 the canonical euclidean norm in @xmath69 . if either @xmath62 or @xmath63 is at infinity the corresponding ratio will be taken equal to @xmath70 . note that the invariance of the cross ratio by a projective map implies the invariance of @xmath71 by such a map . these geometries are naturally endowed with a @xmath72 finsler metric @xmath73 as follows : if @xmath74 and @xmath75 with @xmath76 , the straight line passing by @xmath6 and directed by @xmath77 meets @xmath61 at two points @xmath78 and @xmath79 . then let @xmath80 and @xmath81 be two positive numbers such that @xmath82 and @xmath83 , in other words these numbers corresponds to the time necessary to reach the boundary starting at @xmath6 with the speed @xmath77 and @xmath84 . then we define @xmath85 should @xmath78 or @xmath79 be at infinity , then corresponding ratio will be taken equal to @xmath86 . the hilbert distance @xmath59 is the length distance associated to @xmath73 . we shall denote by @xmath2 the metric ball of radius @xmath3 centred at the point @xmath4 and by @xmath87 the corresponding metric sphere . thanks to that finsler metric , we can built two important borel measures @xmath0 . the first one is called the _ busemann _ volume , will be denoted by @xmath1 ( it is actually the hausdorff measure associated to the metric space @xmath88 , see @xcite , exemple 5.5.13 ) , and is defined as follows . to any @xmath74 , let @xmath89 be the open unit ball in @xmath90 of the norm @xmath91 and @xmath92 the euclidean volume of the open unit ball of the standard euclidean space @xmath9 . consider the ( density ) function @xmath93 given by @xmath94 where @xmath95 is the canonical lebesgue measure of @xmath9 equal to @xmath70 on the unit `` hypercube '' . @xmath96 for any borel set @xmath97 of @xmath0 . the second one , called the _ holmes - thompson _ volume will be denoted by @xmath98 , and is defined as follows . let @xmath99 be the polar dual of @xmath100 and @xmath101 the density defined by @xmath102 . then @xmath98 is the measure associated to that density . we can actually consider a wider family of measure as follows let @xmath103 be the set of pointed properly open convex sets in @xmath9 . these are the pairs @xmath104 , such that @xmath105 is a properly open convex set and @xmath106 a point inside @xmath105 . we shall say that a function @xmath107 is a _ proper density _ if it is continuous : : with respect to the hausdorff pointed topology on @xmath103 ; monotone decreasing : : with respect to inclusion of the convex sets , i.e. , if @xmath108 then @xmath109 . chain rule compatible : : if for any projective transformation @xmath110 one has @xmath111 we will say that @xmath112 is a _ normalised proper density _ if in addition @xmath112 coincides with the standard riemannian volume on the hyperbolic geometry of ellipsoids . let us denote by @xmath113 the set of proper densities over @xmath103 . let us now recall a result of benzecri @xcite which states that the action of the group of projective transformations on @xmath103 is co - compact . then , as remarked by l. marquis , for any pair @xmath114 of proper densities , there exists a constant @xmath115 ( @xmath116 for the normalised ones ) such that that for any @xmath117 one has @xmath118 in the same way we defined the busemann and the holmes - thompson volumes , to any proper density @xmath112 one can associate a borel measure on @xmath0 @xmath119 . integrating the equivalence ( [ eqmeasures ] ) we obtain that for any pair @xmath114 of densities , there exists a constant @xmath115 such that for any borel set @xmath120 we will have @xmath121 we shall call _ proper measures with density _ the family of measures obtain this way . to a proper density @xmath122 we can also associate a @xmath123-dimensional measure , denoted by @xmath124 , on hypersurfaces in @xmath0 as follows . let @xmath125 be smooth a hypersurface , and consider for a point @xmath6 in the hypersurface @xmath125 its tangent hyperplane @xmath126 , then the measure will be given by @xmath127 in section [ lesinegalites ] we will simply denote by @xmath128 the @xmath123-dimensional measure associated with the holmes - thompson measure . let now @xmath119 be a proper measure with density over @xmath0 , then the volume entropy of @xmath0 is defined by @xmath129 this number does not depend on either @xmath112 or @xmath6 and is furthermore equal to the spherical entropy ( see @xcite ) : @xmath130 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ we show the equality between the critical exponent of any discretisation with the volume entropy , which therefore also holds for any discrete co - compact subgroup of isometries of a divisible hilbert geometry . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ * a subset @xmath131 of a hilbert geometry @xmath0 is _ separated _ , if there exists @xmath30 such that the distance between any two distinct points of @xmath131 is greater than or equal to @xmath132 . * a discrete subset @xmath131 of a hilbert geometry @xmath0 is a _ net _ if there exists @xmath133 , such that any point of @xmath0 is at distance less or equal to @xmath134 from a point of @xmath131 . such an @xmath134 is called a covering radius of the net . * a separated net will be called a _ discretisation_. our present definition of a discretisation is slightly different from our paper @xcite . in the later we called discretisation the graph structure associated to a separated net . notice also that a maximal separated set is a net and a minimal net is separated . let @xmath131 be a discrete subset of a convex set @xmath58 endowed with its hilbert geometry and @xmath135 some fixed point . * we call _ poincar series _ of @xmath131 , the following series @xmath136 * we call _ critical exponent _ of @xmath131 , denoted by @xmath137 , the number @xmath138 * if @xmath139 is a discrete sub - group of isometries of @xmath58 , its critical exponent @xmath140 , will be the critical exponent of @xmath141 , the orbit of the point @xmath142 . notice that the definition of the critical exponent does not depend on the base point @xmath142 . the following property is classical . we nevertheless give its proof by sake of completeness . [ ppt1 ] let @xmath131 be a discrete subset of a convex set @xmath58 endowed with its hilbert geometry and @xmath135 some fixed point . then we have the following equality @xmath143 let us define @xmath144 by @xmath145 and @xmath146 by @xmath147 in the annulus @xmath148 there are @xmath149 points of @xmath131 , thus @xmath150 suppose @xmath151 , and fix some @xmath132 such that @xmath152 . then for @xmath153 large enough there is a constant @xmath154 such that @xmath155 . hence the partial sums @xmath156 are bounded by the term of a converging series and thus the poincar series of @xmath131 converges and we deduce that @xmath157 . now suppose @xmath158 , and let @xmath159 . then by the left inequality of equation [ eqmajmin ] we have @xmath160 as a consequence if we fix some @xmath132 such that @xmath161 , then for @xmath153 large enough there is a constant @xmath154 such that @xmath162 , which implies that the poincar series of @xmath131 diverges . hence @xmath163 . [ netsep ] let @xmath58 be a convex set endowed with its hilbert geometry 1 . the critical exponent of a separated set is less than or equal to the volume entropy of @xmath0 . the critical exponent of a net is greater than or equal to the volume entropy of @xmath0 . the critical exponent of any discretisation is equal to the volume entropy . let @xmath131 be an @xmath132-separated set in @xmath58 . let @xmath164 the lower bound on the volume of balls of radius @xmath165 as given in @xcite , then in the ball of radius @xmath166 centred at @xmath167 , denoted by @xmath168 , there are at most @xmath169 points of @xmath131 . thus @xmath170 and our first claim follows from property [ ppt1 ] . let @xmath131 be a net in @xmath58 with covering radius @xmath134 . let @xmath171 be the upper bound of the volume of balls of radius @xmath134 as given in @xcite . then we have @xmath172 and again our second claims follows from property [ ppt1 ] . the following is thus a straightforward consequence . let @xmath58 be a convex set endowed with its hilbert geometry . the critical exponent of any discrete co - compact subgroup of isometries is equal to the volume entropy . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ we compare the hausdorff topology induced by an euclidean metric with the hausdorff topology induced by the hilbert metric on compact subset of an open convex set . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ we recall that the lowner ellipsoid of a compact set , is the ellipsoid with least volume containing that set . in this section we will suppose , without loss of generality , that @xmath0 is a bounded open convex set , whose lowner ellipsoid @xmath173 is the euclidean unit ball and @xmath174 is the center of that ball . it is a standard result that @xmath175 is then contained in @xmath0 , i.e. , we have the following sequence of inclusions @xmath176 we call _ asymptotic ball _ of radius @xmath166 centred at @xmath174 the image of @xmath0 by the dilation of ratio @xmath177 centred at @xmath174 , and we denote it by @xmath178 . let us denote by _ hausdorff - euclidean _ distance the usual hausdorff distance between compact subset of the @xmath20-dimensional euclidean space , and by _ hausdorff - hilbert _ the metric between compact subsets of @xmath0 , defined in the same way as the usual hausdorff metric , but by replacing the euclidean distance by the hilbert distance in the definition . we would like to relate the hausdorff - hilbert neighborhoods of the asymptotic ball @xmath178 with its hausdorff - euclidean neighborhoods . [ keyprop ] let @xmath0 be a convex domain and let @xmath174 be the centre of its lowner ellipsoid , which is supposed to be the unit euclidean ball . 1 . the @xmath179-hausdorff - euclidean neighborhood of the asymptotic ball @xmath178 is contained in its @xmath180-hausdorff - hilbert neighborhood . 2 . for any @xmath181 , the @xmath153-hausdorff - hilbert neighborhood of the asymptotic ball @xmath178 is contained in its @xmath182-hausdorff - euclidean neighborhood . for any point @xmath183 on the boundary of @xmath0 , and for @xmath184 let @xmath185 . this map sends @xmath186 bijectively on the boundary of @xmath187 , which we shall denote by @xmath188 and call the _ asymptotic sphere _ centred at @xmath174 with radius @xmath189 . _ first claim : _ any point of a compact set in the @xmath179-hausdorff - euclidean neighborhood of @xmath178 , either lies inside @xmath178 , or is contained in an euclidean ball of radius @xmath179 centred on a point of @xmath178 . we recall that the ball of radius @xmath190 , i.e. , @xmath191 is inside @xmath0 , and thus so is the ball of radius @xmath192 . let @xmath183 be a point of the boundary , by convexity the interior of the convex closure of @xmath6 and @xmath191 is a subset of @xmath0 . it is the projection of a cone of basis @xmath191 . hence the image of @xmath191 by the dilation of ratio @xmath193 centred at @xmath6 lies in that `` cone '' , therefore stays inside @xmath0 , and it is an euclidean ball of radius @xmath194 centred at @xmath195 , that we shall denote by @xmath196 . for the hilbert distance of @xmath196 a point in the euclidean ball of radius @xmath197 centred at @xmath195 is at a distance less or equal to @xmath198 from @xmath195 . now a standard comparison arguments states that for any two points @xmath106 and @xmath199 in @xmath200 the following inequality occurs @xmath201 from this inequality it follows that any point in the euclidean ball of radius @xmath197 centred at @xmath195 , is inside @xmath202 , the hilbert metric ball centred at @xmath195 . now for any @xmath203 , the ball of radius @xmath197 contains the ball of radius @xmath204 . this implies that for any point in @xmath178 , the euclidean ball of radius @xmath204 centred at that point is contained in the hilbert ball of radius @xmath198 centred at the same point , which allows us to obtain the first part of our claim . s proof ] _ second claim : _ this follows from the fact that under our assumptions , @xmath0 itself is in the @xmath205 hausdorff - euclidean neighborhood of @xmath178 . [ keycor ] let @xmath0 be a convex domain and let @xmath174 be the centre of its lowner ellipsoid , which is supposed to be the unit euclidean ball . 1 . the @xmath206-hausdorff - euclidean neighborhood of @xmath207 is contained in its @xmath208-hausdorff - hilbert neighborhood . 2 . for any @xmath181 , the @xmath153-hausdorff - hilbert neighborhood of @xmath207 is contained in its @xmath209-hausdorff - euclidean neighborhood . this follows from the fact that for any hilbert geometry , under our assumptions , the following inclusions are satisfied : @xmath210 this is a refinement of a result of @xcite in our case . let @xmath106 be a point on the boundary and let @xmath211 be the other intersection of the straight line @xmath212 with @xmath186 . then following our assumption we have @xmath213 actually the first inclusion is always true . indeed suppose @xmath199 is on the half line @xmath214 such that @xmath215 which in other words implies that we have @xmath216 therefore @xmath217 which implies in turn that @xmath218 now regarding the second inclusion , let again @xmath199 be on the half line @xmath214 but this time such that @xmath219 , then on one hand we have @xmath220 and on the second hand thanks to the inequalities ( [ insidelowner ] ) we get @xmath221 which implies that @xmath222 [ keyrmk ] notice that proposition [ keyprop ] is still valid if we only assume @xmath0 to contain a ball of radius @xmath190 centred at @xmath174 . in the same way , corollary [ keycor ] and the inclusions ( [ inclusions ] ) will still be true if @xmath0 contains the ball of radius @xmath192 centred at @xmath174 and is included in the ball of radius @xmath70 centred at the same point @xmath174 , because one needs only to replace @xmath20 by @xmath223 in the inequalities ( [ eqincone ] ) and ( [ eqinctwo ] ) and remark that @xmath224 . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ we define the polytopal dimension of a convex body and investigate it properties . noticeably we recall that an upper bound exists which is attained by the convex @xmath225 bodies . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ let @xmath0 be an open convex set in @xmath9 . for any @xmath226 , let @xmath29 be the smallest number of vertices of a polytope whose hausdorff distance to @xmath0 is less than @xmath132 . we call _ polytopal dimension _ of @xmath0 , denoted by @xmath227 , the number @xmath228 the following result is due to r. schneider and j. a. wieacker @xcite [ thmsw ] let @xmath229 , then @xmath230 admits a critical value @xmath231 called _ approximability number of _ @xmath0 , such that , if @xmath232 then @xmath233 and if @xmath234 then @xmath235 . it is quite obvious that the following occurs the polytopal dimension is twice the approximability number , i.e. , @xmath236 . let us take a sequence of real numbers @xmath237 converging to @xmath86 such that , @xmath238 let us begin by assuming that @xmath239 . according to theorem [ thmsw ] , for any @xmath240 and @xmath241 large enough @xmath242 . in other words for @xmath241 large enough @xmath243 taking the logarithm on both sides and then dividing by @xmath244 , which is negative , implies that @xmath245 and as @xmath246 this gives @xmath247 which implies that @xmath248 now let us assume that @xmath249 . once more , following theorem [ thmsw ] , for any @xmath250 and @xmath241 large enough @xmath251 . in other words for @xmath241 large enough @xmath252 taking the logarithm on both sides and then dividing by @xmath244 which is negative , implies that @xmath253 and as @xmath246 this gives @xmath254 which in turn implies @xmath255 equations [ eqappegalpd1 ] and [ eqappegalpd2 ] imply our statement . now the main result by e. m. bronshteyn and l. d. ivanov @xcite asserts that for any convex set @xmath0 inscribed in the unit euclidean ball , there are no more than @xmath256 points whose convex hull is no more than @xmath132 away from @xmath0 in the hausdorff topology . this easily induces the following upper bound for any convex set @xmath0 : @xmath257 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ we bound from below the volume entropy by the approximability in all dimensions . in dimension two and three we prove the inverse inequality and in doing so prove the upper bound entropy conjecture and the equality between these two invariants in these two dimensions . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ [ pdmajorant ] let @xmath0 be a bounded convex domain in @xmath43 or @xmath258 . the polytopal dimension of @xmath0 is bigger than the volume entropy , _ i.e. _ , @xmath259 to prove that theorem , we will actually prove the following stronger statement , valid in dimension @xmath32 and @xmath37 , which should be true in higher dimension as well . [ growthpolytope ] let us choose a family of proper measures with density denoted by @xmath260 . then for any @xmath36 or @xmath37 there are affine maps @xmath261 from @xmath262 and polynomials @xmath263 of degree @xmath20 and @xmath123 such that for any open convex polytope @xmath264 with @xmath265 vertices inside the unit euclidean ball of @xmath9 containing the ball of radius @xmath192 , one has @xmath266 the same result holds for the asymptotic balls . we will consider the holmes - thompson measure for the proof and work with the asymptotic balls . a change of family of measures will only change the constants . the passage from the asymptotic balls to the metric balls is done thanks to the inclusions ( [ inclusions ] ) . let @xmath267 be the asymptotic ball of radius @xmath166 centred at @xmath174 , and us denote by @xmath268 . _ two dimensional case : _ the length of each edge of @xmath267 in @xmath264 is less than @xmath269 thanks to the triangular inequality and because by the proof of corollary [ keycor ] , @xmath267 is inside the hilbert ball of radius @xmath270 centred at @xmath174 of @xmath271 . therefore the length of the boundary of @xmath267 is less than @xmath272 . now using the co - area inequality ( lemma 2.13 ) of @xcite we conclude that the area of @xmath267 is less than @xmath273 , where @xmath274 is a constant depending on the dimension . _ three dimensional case : _ consider one of the faces of @xmath267 , then by minimality of the holmes - thompson volume , the area of that face is less than the sum of the areas of the triangles obtained thanks to the convex closure of @xmath174 and an edge of the given face of @xmath267 . * claim : * _ the area of such a triangle is less than @xmath275 , for @xmath154 some constant independent of @xmath166 . _ thanks to that claim , whose proof we postpone , if @xmath276 is the number of edges of @xmath277 , the area of @xmath278 is less than @xmath279 . let @xmath280 be the number of faces of @xmath277 and let us recall euler s formula : @xmath281 besides , as each face is surrounded by at least three edges and each edge belongs to two faces , one has the classical inequality ( where equality is obtained in a simplex ) , @xmath282 combining the previous two inequalities we get a linear upper bound of the number of edges by the number of vertexes as follows : @xmath283 hence the area of @xmath278 is less than @xmath284 , and by using once again the co - area inequality @xcite we conclude that the volume of @xmath267 is less than @xmath285 , where @xmath286 is a constant independent of the three dimensional convex domain . _ proof of the claim : _ in substance we claim that in a polygon @xmath287 , if @xmath6 and @xmath60 are two consecutive vertexes , and @xmath288 , @xmath289 are two points at a distance less than @xmath290 from a given point @xmath174 , on the lines @xmath291 and @xmath292 respectively , then the affine triangle @xmath293 has an area less than @xmath294 , for some universal constant @xmath154 . in aim to convince the reader , let @xmath295 and @xmath296 be , respectively , the second intersection of the lines @xmath291 and @xmath292 with the convex @xmath297 . is bounded by @xmath294 . ] then the volume of our triangle for @xmath287 is less that its volume in the quadrilateral @xmath298 , while the distances @xmath299 and @xmath300 remain the same by construction . up to a change of chart , we can suppose that this quadrilateral is actually a square . this allows us to use the result in vernicos @xcite which states that the hilbert geometry of the square is bi - lipschitz to the product of the hilbert geometries of its sides , using the identity as a map . therefore our affine triangle is inside the disc of radius @xmath301 for this product geometry , for some constant @xmath154 independent of our initial conditions , which implies that its area for @xmath0 is less than @xmath302 for some universal constant . le us remark that if we link this to our study of the asymptotic volume of the hilbert geometry of polytopes @xcite we obtain the following corollary let @xmath264 be an open convex polytope with @xmath265 vertices in @xmath9 , for @xmath36 or @xmath37 , then there are three constants @xmath303 , @xmath304 and @xmath305 such that for any point @xmath306 one has @xmath307 now let us come back to our initial problem an see how theorem [ growthpolytope ] implies theorem [ pdmajorant ] . we remind the reader that @xmath128 stands for the @xmath123-dimensional holmes - thompson measure . let @xmath174 be the centre of the lowner ellipsoid of @xmath0 which is supposed to be the unit euclidean ball . we consider @xmath166 large enough such that the euclidean ball of radius @xmath192 is inside all the convex studied in the sequel . consider the asymptotic ball @xmath178 and let @xmath267 the polygon inside @xmath178 at euclidean - hausdorff distance less than @xmath308 , with the minimum number @xmath309 of vertices . by proposition [ keyprop ] it lies in its @xmath310 hilbert - hausdorff neighborhood , and therefore contains the asymptotic ball of radius @xmath311 , for @xmath154 a constant independent of @xmath166 ( _ e.g. _ one can take @xmath312 following ( i ) and ( ii ) of the proof of corollary [ keycor ] ) . thanks to the crofton formula ( see @xcite ) we know that the area of @xmath267 is less than the area of the asymptotic ball @xmath178 , but bigger than the area of the asymptotic ball of radius @xmath311 . therefore one has @xmath313 which implies that the logarithm of the areas of @xmath267 and @xmath178 are asymptotically the same in the following sense @xmath314 if we denote by @xmath271 the image of @xmath267 by the dilation of ratio @xmath315 , then by construction , @xmath271 is inside @xmath0 and therefore we have @xmath316 now let us use the result from theorem [ growthpolytope ] , for @xmath166 such that @xmath317 we have the existence of two constants @xmath318,@xmath319 and a polynomial @xmath320 of degree @xmath20 such that @xmath321 now @xmath322 which implies that @xmath323 let @xmath0 be an open bounded convex set in @xmath9 , for @xmath36 or @xmath37 , then @xmath324 we are now going to study the inverse inequality . [ pdminoreent ] let @xmath0 be an bounded convex domain in @xmath9 . the volume entropy is bigger or equal to the polytopal dimension of @xmath0 , i.e. , @xmath325 [ lessegments ] consider four points @xmath62 , @xmath63 , @xmath326 and @xmath11 such that the straight lines @xmath327 , @xmath328 and @xmath329 are distinct and the scalar products @xmath330 and @xmath331 are strictly positive . let also denote by @xmath60 the intersection point between the straight lines @xmath332 and @xmath333 . we also suppose that @xmath0 is a convex domain such that the segments @xmath334 , @xmath335 and @xmath336 belong to its boundary . let @xmath6 be a point inside the convex @xmath0 and denote by @xmath295 the intersection between the straight line @xmath337 and the segment @xmath335 . finally we denote by @xmath338 the image under the dilation centred at @xmath6 with ratio @xmath339 of the segment @xmath335 . then one has @xmath340 where @xmath341 is the image of @xmath335 under the dilation centred at @xmath60 sending @xmath295 on @xmath6 and @xmath342 is the ratio @xmath343 . straightforward computation , using the fact that the convex domain @xmath0 is inside the convex obtained as the intersection of the half planes defined by the line @xmath332 , @xmath344 and @xmath333 . without loss of generality we suppose that the euclidean unit ball is the lowner ellipsoid of @xmath0 , and that @xmath174 is the centre of that ball . we will do two proofs , one in the two dimensional case , which gives more informations and leads to the left hand side of the intermediate growth corollary , and one for all dimension which consists in building an appropriate separated net . _ the two dimensional case : _ we will consider a family @xmath345 of polytopes as follows . * for each @xmath166 , @xmath346 contains @xmath178 and is in the @xmath347-hausdorff - euclidean neighborhood of the asymptotic ball @xmath178 , which is included in the @xmath348-hausdorff - hilbert neighborhood by proposition [ keyprop ] . therefore , @xmath346 will lie inside @xmath0 . * all but one edge of @xmath346 are tangent to @xmath178 and all its vertexes belong to the boundary of its @xmath349-hausdorff neighborhood ( let us denote it @xmath350 ) . to see this , start from any point @xmath106 in @xmath350 and follow this algorithm : 1 . draw one of the tangent to @xmath178 , it will meet the boundary of its @xmath351 at two points @xmath352 and @xmath353 , where @xmath354 , @xmath355 are positively oriented . we begin again at @xmath353 drawing the other tangent to @xmath356 passing by @xmath353 , which will meet the boundary at @xmath357 . 3 . for @xmath358 , if the second tangent from @xmath359 has its second intersection with @xmath350 on the arc of from @xmath352 to @xmath359 ( in the orientation of the construction ) , we stop and consider for @xmath346 the convex hull of @xmath360 , otherwise we take for @xmath361 this other side of this tangent and start again that step . now this algorithm will necessarily finish , because the arclength of @xmath362 on @xmath363 built this way is bigger than @xmath364 , by convexity . at the end of this algorithm we obtain , by minimality , a polygon which has at least @xmath365 . by the inclusion of @xcite we obtain @xmath366 which implies that the behaviour of @xmath367 is the same as the behaviour of @xmath368 as @xmath369 now let @xmath370 be the image of @xmath346 under the dilation of ratio @xmath371 centred at @xmath174 . by construction @xmath370 contains @xmath0 , which implies @xmath372 * claim : * let @xmath373 and @xmath374 be the two points of tangency of two incident edges of @xmath346 at @xmath375 , tangent to @xmath356 . then there is a constant @xmath240 and @xmath376 depending only on @xmath377 such that for any @xmath378 , @xmath379 let us assume the claim is satisfied and for @xmath378 consider a vertex @xmath77 of @xmath346 whose incident edges are tangent to @xmath178 . let @xmath63 and @xmath326 the two points of tangency , then by the triangle inequality , @xmath380 . therefore the length of @xmath346 is bigger than @xmath381 , where @xmath144 is number of edges of @xmath346 . ( because of the possible exception at @xmath352 and the last point of the construction above ) . hence the `` volume entropy '' of the family @xmath346 is bigger than @xmath382 and as we have @xmath383 we deduce from that the desired result by taking the infimum limit as @xmath369 let us now prove the claim . to do so , let @xmath384 ( resp . @xmath385 ) be opposite vertex to @xmath375 on the edge containing @xmath373 ( resp . @xmath374 ) . now let us consider the image @xmath375 , @xmath62,@xmath63 , @xmath326 and @xmath11 of the five points @xmath386 , @xmath384 , @xmath373 , @xmath374 and @xmath385 by the dilation of ratio @xmath387 centred at @xmath174 . then we are in the same configuration as in lemma [ lessegments ] , with @xmath370 instead of @xmath0 . let @xmath388 , then following ( [ minorationlg ] ) we have @xmath389 therefore we need to obtain a lower bound for @xmath390 . to do this , let @xmath6 be the intersection of the line @xmath391 with the lines @xmath344 . then thanks to thales s theorem we have @xmath392 concerning the distance @xmath393 , recall that the unit ball centred at @xmath174 is the lowner ellipsoid of @xmath0 and therefore we get @xmath394 , because by convexity @xmath6 is in @xmath0 . regarding the distance @xmath395 , as @xmath386 is on the boundary of the @xmath347 euclidean neighborhood of @xmath178 , @xmath375 is on the boundary of the @xmath396 neighborhood of @xmath0 . hence we obtain @xmath397 , because the segment @xmath398 $ ] intersects @xmath0 . this way we obtain @xmath399 which in turn implies that @xmath400 hence @xmath401 therefore if @xmath402 then for all @xmath378 we get @xmath403 . finally using the fact that @xmath404 and taking @xmath378 we get @xmath405 _ general case : _ we suppose now that our convex is @xmath20-dimensional . consider @xmath406 the @xmath348-hilbert neighborhood of the metric ball @xmath207 , and take a maximal @xmath407-separated net @xmath408 on its boundary . this set contains @xmath409 points . now let us take the convex hull @xmath410 of these points . this is a polytope with @xmath411 vertices . let us prove that it is included in the @xmath412-hilbert neighborhood of @xmath207 and contains @xmath207 . if this hold we will have for some real constant @xmath326 independent of @xmath166 ( see corollary [ keycor ] once again ) , @xmath413 first notice that @xmath414 is a convex set ( see busemann @xcite , chapter ii , section 18 , page 105 ) . therefore the convex hull is inside @xmath414 . now let us suppose by contradiction that @xmath410 does not contain @xmath207 . hence there exists some points @xmath60 in @xmath207 which is not in @xmath410 . therefore , by the hahn - banach separation theorem , there exists a linear form @xmath62 , some constant @xmath106 and hyperplane @xmath415 which separates @xmath60 and @xmath410 , i.e. , @xmath416 and @xmath417 for all @xmath418 . consider then @xmath419 the parallel hyperplane to @xmath420 containing @xmath60 . let us say that a point @xmath106 such that @xmath421 is _ above _ the hyperplane @xmath422 . then let us define by @xmath423 the part of the boundary of @xmath414 which is above @xmath422 . now we want to metrically project each point of @xmath424 onto @xmath422 , that is to say that to each point of @xmath424 we associate its closest point on @xmath422 . however if @xmath0 is not strictly convex , the projection might not be unique ( see the appendix [ aprojection ] ) , that is why we are going to distinguish two cases . _ first case : _ the convex set @xmath0 is strictly convex , then the metric projection is a map from @xmath424 to @xmath422 and it is continuous , furthermore the point on @xmath425 are fixed and by convexity @xmath425 is homeomorphic to a @xmath426-dimensional sphere . therefore by borsuk - ulam s theorem ( or its version known as the _ antipodal map theorem _ ) , there is a point @xmath6 on @xmath424 whose metric projection is @xmath60 . now as @xmath6 is on the boundary of @xmath414 necessarily @xmath427 hence for all points @xmath106 in @xmath425 , we have @xmath428 _ second case : _ the convex set @xmath0 is not strictly convex . then let us approximate it by a smooth and strictly convex set @xmath429 such that @xmath430 , and for all pair of points @xmath431 , @xmath432 then metrically project @xmath424 onto @xmath422 with respect to @xmath429 . by the same argument from the first case , we obtain a point @xmath6 such that for all @xmath106 in @xmath425 we have @xmath433 which also implies by [ smoothapprox ] that for all @xmath106 in @xmath425 we have @xmath434 in either cases , using the lemma [ spheremonotonic ] of the appendix [ adistanboule ] , we deduce that all points on @xmath435 at distance less or equal to @xmath436 from @xmath6 are above @xmath422 and are therefore contained in @xmath424 . we then infer that there are no points of @xmath437 at distance less or equal to @xmath436 from @xmath6 , which contradicts the maximality of the set @xmath437 . we can now conclude , by considering the discrete family @xmath438 which is by our construction an @xmath439 separated set , and therefore its critical exponent is a lower bound of the entropy following [ netsep ] . for any @xmath440 , there exists @xmath441 such that @xmath442 , we thus get the following inequality : @xmath443 which implies that @xmath444 the last inequality comes from the same computations done during two dimensional proof . a point @xmath106 of a convex body @xmath153 is called a _ farthest point _ of @xmath153 if and only if , for some point @xmath445 , @xmath106 is farthest from @xmath199 among the points of @xmath153 . the set of farthest points of @xmath153 , which are special exposed points , will be denoted by @xmath446 . thus a point @xmath447 belongs to @xmath446 if and only if there exists a ball which circumscribes @xmath153 and contains @xmath106 in its boundary . we thus have the following corollary in dimension @xmath32 , let @xmath0 be a plane hilbert geometry , and let @xmath448 be the minkowski dimension of extremal points and @xmath53 the hausdorff dimension of the set @xmath449 of farthest points then we have the following inequalities @xmath450 the left hand side inequality remains valid for higher dimensional hilbert geometries . the left hand side of inequality ( [ eqentfarthestpoints ] ) comes from r. schneider and j. a. wieacker @xcite , whereas the right hand one from g. berck , a. bernig and c. vernicos @xcite . inequality ( [ eqentfarthestpoints ] ) induces a new result concerning the approximability in dimension @xmath32 , as it implies that @xmath451 lastly we are also able to prove the following result which relates the entropy of a convex set and the entropy of its polar body . [ cvxeandpolarsameentropy ] let @xmath0 be a hilbert geometry of dimension @xmath32 or @xmath37 , then @xmath452 it suffices to prove that the approximability of a convex body @xmath0 containing the origin and its polar @xmath55 are equal . however , notice that for @xmath132 small enough , if @xmath453 is a polytope with @xmath19 vertexes inside the @xmath132-hausdorff neighborhood of @xmath0 , then its polar @xmath454 is a polytope with @xmath19 faces containing @xmath55 and contained in its @xmath132-hausdorff neighborhood . a known fact ( see gruber @xcite ) states that the approximability can be computed either by minimising the vertexes or the faces . hence @xmath455 and our result follows from the main theorem . the following is a reformulation and a detailed proof of a statement found in section 21 and 28 of busemann - kelly s book @xcite in any dimension . let @xmath456 be a hilbert geometry in @xmath9 . let @xmath6 be a point of @xmath0 and @xmath420 an hyperplane intersecting @xmath0 . then @xmath457 is a _ metric projection _ of @xmath6 onto @xmath420 , i.e. , @xmath458 if and only if @xmath186 has , at its intersection with the straight line @xmath337 , supporting hyperplanes concurrent with @xmath420 ( the intersection of these three hyperplanes is an @xmath426-dimensional affine space ) . let us suppose first that such concurrent support hyperplanes exists . let @xmath106 and @xmath199 be the intersections of the line @xmath337 with @xmath186 . assume that @xmath459 and @xmath134 are supporting hyperplanes of @xmath186 respectively at @xmath106 and @xmath199 whose intersection with @xmath420 is the @xmath426-affine space @xmath460 . let us show that for any @xmath461 and any @xmath462 we have @xmath463 let us suppose that @xmath106 is on the half line @xmath464 and @xmath199 on the half line @xmath465 and denote by @xmath466 and @xmath467 the intersection of @xmath61 with the half line @xmath468 and @xmath469 respectively . then let @xmath142 be the intersection of @xmath459 with the line @xmath470 and @xmath471 the intersection of @xmath470 with @xmath134 . by thales theorem , the cross - ratio of @xmath472 $ ] is equal to the cross ratio of @xmath473 $ ] and standard computation shows that @xmath472\leq [ x',p',q',y']$ ] , with equality if an only if @xmath474 and @xmath475 . hence the inequality ( [ eqmetproj ] ) holds , and if the convex set is strictly convex , this inequality is always strict , for @xmath476 . on @xmath420 . ] reciprocally : recall that when a point @xmath296 of @xmath0 goes to the boundary , its distance to @xmath6 goes to infinity . hence by continuity of the distance and compactness there exists a point @xmath60 on @xmath477 such that @xmath478 . now consider the hilbert ball @xmath2 of radius @xmath479 centred at @xmath6 . let once more @xmath106 , @xmath199 , @xmath459 and @xmath134 be defined as before , and let @xmath480 be the hyperplane passing by @xmath60 and @xmath481 . then this hyperplane has to be tangent to the ball @xmath2 , otherwise one can find a point @xmath296 on @xmath480 inside the open ball ( i.e. @xmath482 , however by the reasoning done in our first step we would conclude that this point is at a distance bigger or equal to @xmath3 , which would be a contradiction . by minimality of the point @xmath60 , @xmath420 is also a supporting hyperplane of @xmath2 at @xmath60 . hence we have to distinguish between two cases . if @xmath0 is @xmath28 , then by uniqueness of the tangent hyperplanes at every point @xmath483 . otherwise , @xmath0 is not @xmath28 at @xmath106 or @xmath199 . in that case it is possible to change one of the hyperplane , say @xmath459 , with @xmath484 passing by @xmath106 and @xmath485 ( which might be at infinity , which would mean that we consider parallel hyperplanes ) . notice that there is no uniqueness of the metric projections ( also called `` foot '' by busemann ) . however if the convex set is strictly convex , then we will have a unique projection , if furthermore the convex is @xmath28 , this projection will be given by a unique pair of supporting hyperplanes . this appendix is an adaptation of a proof in a minkowski space provided to the author by a. thompson @xcite [ peakless ] let @xmath58 be a hilbert geometry . the distance function of a geodesic to a point is a peakless function , i.e. , if @xmath486\to \omega$ ] is a geodesic segment , then for any @xmath487 and @xmath488 one has @xmath489 [ spheremonotonic ] let @xmath58 be a two dimensional hilbert geometry . suppose @xmath174 is point of @xmath0 , and @xmath6 and @xmath60 are two points on the intersection of the metric sphere @xmath490 centred at @xmath174 and radius @xmath166 with a line passing by @xmath174 . if @xmath154 denotes one of the arcs of the sphere @xmath490 from @xmath6 to @xmath60 , then for any point @xmath295 on the half line @xmath491 , the function @xmath492 is monotonic on @xmath154 . suppose now that @xmath496 $ ] and @xmath497 $ ] do not intersect , which implies that @xmath295 is outside the ball @xmath207 . then the line @xmath501 intersects @xmath291 at @xmath498 . because @xmath106 and @xmath199 lie on the sphere of radius @xmath166 , @xmath502 . also , as @xmath6 is one of the nearest points to @xmath295 on @xmath154 , we have @xmath503 hence if apply the proposition [ peakless ] to the segment @xmath504 $ ] and @xmath295 , as @xmath505 $ ] we get @xmath506 . extremum problems with inequalities as subsidiary conditions . in _ studies and essays presented to r. courant on his 60th birthday , january 8 , 1948 _ , pages 187204 . interscience publishers , inc . , new york , 1948 .

the covariogram of a convex body is the function which associates to each the volume of the intersection of with matheron asked whether determines , up to translations and reflections in a point . positive answers to matheron s question have been obtained for large classes of planar convex bodies , while for there are both positive and negative results . one of the purposes of this paper is to sharpen some of the known results on matheron s conjecture indicating how much of the covariogram information is needed to get the uniqueness of determination . we indicate some subsets of the support of the covariogram , with arbitrarily small lebesgue measure , such that the covariogram , restricted to those subsets , identifies certain geometric properties of the body . these results are more precise in the planar case , but some of them , both positive and negative ones , are proved for bodies of any dimension . moreover some results regard most convex bodies , in the baire category sense . another purpose is to extend the class of convex bodies for which matheron s conjecture is confirmed by including all planar convex bodies possessing two non - degenerate boundary arcs being reflections of each other . 52a20 , 52a22 , 52a38 , 60d05 . convex body , convex polytope , covariogram , genericity result , geometric tomography , set covariance , quasicrystal .

math0702892

let @xmath3 be a convex body in @xmath7 . the _ covariogram _ @xmath5 of @xmath3 is the function @xmath8 where @xmath9 and @xmath10 denotes volume in @xmath7 . this functional , which was introduced by matheron in his book @xcite on random sets , is also sometimes called the _ set covariance _ , and it coincides with the _ autocorrelation _ of the characteristic function of @xmath3 : @xmath11 the covariogram @xmath5 is clearly unchanged by a translation or a reflection of @xmath3 . ( the term _ reflection _ will always means reflection in a point . ) matheron @xcite and , independently , adler and pyke @xcite asked the following question . does the covariogram determine a convex body , among all convex bodies , up to translations and reflections ? matheron conjectured a positive answer for the case @xmath12 but this conjecture has not been completely settled . matheron @xcite observed that , for @xmath13 and for all @xmath14 , the derivatives @xmath15 give the distribution of the lengths of the chords of @xmath3 parallel to @xmath16 . such information is common in _ stereology , _ _ statistical shape recognition _ and _ image analysis , _ when properties of an unknown body are to be inferred from chord length measurements ; see @xcite , @xcite and @xcite , for example . blaschke ( cf . @xcite ) asked whether the distribution of the lengths of chords ( in all directions ) of a convex body characterizes the body , up to rigid motions , but mallows and clark @xcite proved that this is false even for convex polygons . in fact ( see @xcite ) the covariogram problem is equivalent to the problem of determining a convex body from all its separate chord length distributions , one for each direction @xmath13 . adler and pyke @xcite asked matheron s question in probabilistic terms . does the distribution of the difference @xmath17 of independent random variables @xmath18 and @xmath19 uniformly distributed over @xmath3 determine @xmath3 , up to translations and reflections ? since the convolution in is , up to a multiplicative factor , the probability density of @xmath17 , this problem is equivalent to the covariogram one . matheron s problem is also relevant in _ x - ray crystallography , _ where the atomic structure of a _ crystal _ ( or _ quasicrystal _ ) is to be found from _ diffraction images . _ a convenient way of describing many important examples of quasicrystals is via the _ cut and project scheme_. here to the atomic structure , represented by a discrete set @xmath20 contained in a space @xmath21 , is associated a _ lattice _ @xmath22 in a higher dimensional space @xmath23 and a _ window _ @xmath24 ( which in many cases is a convex set ) . in this setting @xmath20 coincides with the projection on @xmath21 of the points of the lattice @xmath22 which belong to @xmath25 . in many examples the lattice @xmath22 can be determined by the diffraction image . to determine @xmath20 it is however necessary to know @xmath26 : the covariogram problem enters at this point , since the covariogram of @xmath26 can be obtained by the diffraction image ; see @xcite . enns and ehlers @xcite express in terms of the covariogram the distributions of random line segments in a convex body , under different types of randomness with which they are generated . the monograph @xcite contains an extensive discussion of retrieval problems for convex bodies , while the survey @xcite deals with algorithmic aspects of reconstruction problems in convex geometry . the first contribution to matheron s question was made in 1993 by nagel @xcite , who gave a positive answer when @xmath3 is a planar convex polygon ; see also schmitt @xcite . matheron s conjecture is still unsettled for general planar convex bodies , but it has been confirmed for @xmath27 convex bodies , non - strictly convex bodies , and convex bodies that are not @xmath28 ; see @xcite . it has been recently shown that every convex polytope in @xmath29 is determined by its covariogram , up to translations and reflections ( cf . @xcite ) . for @xmath30 there exist examples of convex polytopes that are not determined by their covariogram ( cf . however @xcite proves that , if @xmath31 is a @xmath32-dimensional simplicial convex polytope in general relative position with respect to @xmath33 , the determination by the covariogram data is unique for every @xmath34 ( see next section for all unexplained definitions ) . the paper @xcite discusses various open retrieval problems related to the covariogram . one of the purposes of this paper is to sharpen some of the known results on matheron s conjecture , indicating how much of the covariogram information is needed to get the uniqueness of determination . we indicate some subsets of the support of the covariogram , with arbitrarily small lebesgue measure , such that the covariogram , restricted to those subsets , identifies certain geometric properties of the body . these results are more precise in the planar case , but some of them , both positive and negative ones , are proved for bodies of any dimension . moreover some results regard _ most _ convex bodies , in the _ baire category sense . _ another purpose is to extend the class of convex bodies for which matheron s conjecture is confirmed by including all planar convex bodies possessing two non - degenerate boundary arcs being reflections of each other . given two convex bodies @xmath3 and @xmath35 in @xmath7 and a closed set @xmath36 we introduce the following property involving @xmath37 and @xmath38 @xmath39 is a shorthand notation standing for `` covariogram coincidence '' ( where covariogram is traditionally referred to by the letter @xmath40 ) . @xmath41 : : the equality @xmath42 holds for all @xmath43 in some neighbourhood of @xmath44 the following theorem presents two choices of the set @xmath18 for which @xmath41 implies the coincidence of @xmath3 and @xmath35 up to translations and reflections , under the assumption @xmath45 before stating the theorem we need to introduce the notion of local symmetry and give some related explanations . a pair of closed boundary arcs of a planar convex body @xmath3 is said to be a _ local symmetry _ of @xmath3 if they are reflections of each other in a point , have disjoint and nonempty relative interiors , and they are not properly contained in a pair of boundary arcs with the same properties . a planar convex body @xmath3 is called _ locally symmetric _ if it possesses a local symmetry . planar convex bodies without local symmetries are called _ globally non - symmetric . _ it is known that the support of @xmath5 is the _ difference body _ of @xmath3 , @xmath46 , and that @xmath47 is @xmath48-symmetric . if @xmath49 and @xmath50 are arcs of @xmath51 which compose a local symmetry , then the set @xmath52 , translated in such a way to be @xmath48-symmetric , is the union of two arcs @xmath53 and @xmath54 of @xmath55 . we say that these arcs of @xmath55 correspond to the local symmetry @xmath56 ( see figs . [ 06.05.23,21:46 ] and [ 05.12.19,11:36 ] ) . a convex body @xmath3 is said to belong to the class @xmath57 if its boundary is a two - times continuously differentiable manifold and all its principal curvatures are non - zero ( for detailed information see ( * ? ? ? * section 2.5 ) ) . [ 05.08.10,15:35 ] let @xmath3 and @xmath35 be planar convex bodies and let @xmath3 be @xmath57 regular . let @xmath58 be the collection of all the arcs of @xmath55 which correspond to local symmetries of @xmath3 . by @xmath59 we denote the midpoint of the segment joining the endpoints of the arc @xmath60 let @xmath61 and let @xmath62 or @xmath63 ( see figs . [ 06.05.23,21:46 ] and [ 05.12.19,11:36 ] ) . then @xmath41 implies the coincidence of @xmath3 and @xmath64 up to translations and reflections . @xmath65 cc ( 40,30 ) ( 12,8 ) ( 11,20)@xmath3 & ( 40,30 ) ( 4,4 ) ( 21,16)@xmath48 ( 3,25)@xmath47 + & we remark that the two choices of @xmath18 defined in the statement of theorem [ 05.08.10,15:35 ] are in some sense minimal for the assertion of the theorem to hold ( for further details see remark [ 06.05.16,16:17 ] below ) . moreover the set @xmath66 depends only on @xmath5 and not on @xmath3 , in the sense that if @xmath35 and @xmath3 satisfy @xmath67 or @xmath68 then the same set @xmath66 corresponds to @xmath35 and @xmath3 . this is the content of the second part of theorem [ 06.05.16,15:21 ] . the following corollary is a direct consequence of theorem [ 05.08.10,15:35 ] . [ 06.05.16,17:58 ] let @xmath3 and @xmath35 be planar convex bodies and let @xmath3 be @xmath57 regular . then the following statements hold . a. there exists a closed and at most countable subset @xmath18 of @xmath69 with no accumulation points in @xmath70 such that @xmath71 implies the coincidence of @xmath3 and @xmath64 up to translations and reflections . furthermore , @xmath18 can be chosen lying on a strictly convex curve . b. if @xmath3 is globally non - symmetric and @xmath18 is either @xmath72 or @xmath73 then @xmath41 implies the coincidence of @xmath3 and @xmath64 up to translations and reflections . @xmath65 it is an open question whether part ii of the above theorem holds for all strictly convex @xmath3 not necessarily @xmath74 the positive answer to this question would imply the confirmation of matheron s conjecture for all planar convex bodies . the following theorem presents determination results involving locally symmetric and symmetric convex bodies . [ 06.05.16,15:21 ] let @xmath3 and @xmath35 be planar convex bodies with @xmath3 strictly convex . then the following statements hold true a. if @xmath3 is @xmath48-symmetric and @xmath53 is a closed simple curve in @xmath70 bounding an open region in @xmath47 that contains the origin , then the conditions @xmath75 and @xmath76 imply the coincidence of @xmath3 and @xmath64 up to translations and reflections . b. if @xmath18 is either @xmath72 or @xmath73 then @xmath41 implies the coincidence of all local symmetries of @xmath3 and @xmath64 up to translations . @xmath65 the statement of the following corollary follows directly from the second part of theorem [ 06.05.16,15:21 ] and two results from ( * ? ? ? * theorem 1.1 and proposition 1.4 ) . [ 06.05.24,09:59 ] every locally symmetric convex body in @xmath77 is determined uniquely , up to translations and reflections , by its covariogram function . @xmath65 let @xmath78 denote the closed unit ball in @xmath7 centered at the origin . further on , we introduce the condition @xmath79 a relaxation of @xmath80 and the `` local coincidence '' condition @xmath81 @xmath82 : : the equality @xmath83 holds for all @xmath43 in some neighbourhood of @xmath18 and a suitable constant @xmath84 @xmath85 : : for every boundary point @xmath86 of @xmath3 there exists a boundary point @xmath87 of @xmath35 such that for some @xmath88 the bodies @xmath89 and @xmath90 coincide , up to translations and reflections ; the same statement also holds with the roles of @xmath3 and @xmath35 interchanged . in the following theorem the relationship between the conditions @xmath91 @xmath92 @xmath93 and @xmath85 is discussed . [ 06.05.15,12:36 ] let @xmath3 and @xmath35 be convex bodies in @xmath94 then the following statements hold true . a. if @xmath3 is strictly convex , then @xmath85 implies @xmath95 b. if @xmath96 and @xmath3 is @xmath97 then @xmath85 is equivalent to @xmath95 c. if @xmath96 and @xmath3 is strictly convex , then @xmath98 is equivalent to @xmath95 d. there exist planar convex bodies @xmath3 and @xmath35 belonging to the class @xmath57 such that @xmath67 holds , while @xmath99 does not . e. there exist @xmath3 and @xmath64 which are convex @xmath32-polytopes , such that the conditions @xmath75 and @xmath99 hold , while @xmath67 does not . @xmath65 it is an open problem whether for @xmath57 convex bodies in @xmath100 @xmath101 the condition @xmath67 implies @xmath81 the space @xmath102 endowed with the hausdorff metric is locally compact and by this a baire space ( see @xcite and @xcite ) . thus , we may speak about statements that hold for most convex bodies , i.e. , for all convex bodies with at most a meager set of exceptions . we recall that a set is said to be _ meager _ if it is a countable union of nowhere dense sets and _ residual _ if it is a complement of a meager set . trivially , a finite intersection of residuals is a again a residual . furthermore , every set possessing a residual subset is also a residual . [ 06.05.08,17:28 ] in the baire category sense , for most convex bodies @xmath3 and all convex bodies @xmath35 in @xmath7 the following statements hold true . a. if @xmath3 and @xmath35 satisfy @xmath93 then @xmath3 and @xmath35 coincide , up to translations and reflections . b. if @xmath96 and @xmath3 and @xmath35 satisfy @xmath91 then @xmath3 and @xmath35 coincide , up to translations and reflections . furthermore , the above two statements can not be extended to all pairs of convex bodies @xmath3 and @xmath64 since there exist bodies @xmath3 and @xmath35 not coinciding , up to translations and reflections , and such that @xmath103 holds . in addition , the bodies @xmath3 and @xmath35 satisfying the above conditions can be chosen belonging to the class @xmath74 @xmath65 it is an open question whether part ii of the statement can be carried over to convex bodies of higher dimensions . the previous theorem is related to ( * ? ? ? * theorem 2 ) and ( * ? ? ? * theorem 6.2 ) . theorem 2 from @xcite states that most convex bodies @xmath3 in @xmath7 , for any @xmath6 , are determined by the combined knowledge of the _ width _ of @xmath3 in direction @xmath16 and of the @xmath104-dimensional volume of @xmath105 , for each @xmath13 . here @xmath106 denotes the orthogonal complement to @xmath107 and @xmath105 stands for the orthogonal projection of @xmath3 onto the hyperplane @xmath108 ( see also @xcite , where `` local versions '' of some results from @xcite are obtained , as well as ( * ? ? ? * theorems 3.3.17 and 3.3.18 ) . ) the relation with the covariogram comes from the fact that knowing its support @xmath109 is equivalent to knowing the width of @xmath3 in all directions , and that the knowledge of @xmath5 in a neighbourhood of @xmath48 gives the volumes of all @xmath110-dimensional projections of @xmath3 . this follows from the formula @xmath111 proved in @xcite and @xcite . here @xmath112 stands for right derivative . theorem 6.2 from @xcite is another result related to theorem [ 06.05.08,17:28 ] , which states that most convex planar bodies are determined by the covariogram function over its entire support . in section 3 we prove theorem [ 06.05.15,12:36 ] . sections 4 and 5 are independent of each other and present proofs of theorem [ 05.08.10,15:35 ] and theorem [ 06.05.16,15:21 ] , respectively . in section 6 we prove theorem [ 06.05.08,17:28 ] . the euclidean @xmath32-dimensional space with the origin @xmath113 scalar product @xmath114 and the norm @xmath115 is denoted by @xmath116 the unit sphere and the unit ball in @xmath7 are denoted by @xmath117 and @xmath118 respectively . the orthogonal projection of a set @xmath119 onto an affine space @xmath120 is denoted by @xmath121 if @xmath122 then @xmath106 stands for the orthogonal complement to @xmath123 by @xmath124 we denote the @xmath125-dimensional volume of a convex set in @xmath7 of dimension at most @xmath126 if @xmath127 we omit the subscript and write simply @xmath128 we write @xmath129 for the @xmath125-dimensional hausdorff measure in @xmath116 in what follows , in integrals on @xmath125-dimensional spheres in @xmath7 ( with @xmath130 ) we abbreviate @xmath131 by @xmath132 the abbreviations @xmath133 @xmath134 @xmath135 @xmath136 and @xmath137 stand for boundary , interior , relative interior , closure and affine hull , respectively . following the monograph @xcite we denote by @xmath102 and @xmath138 the classes of non - empty , compact , convex sets and compact , convex sets with non - empty interior , respectively . elements of @xmath138 are said to be _ convex bodies . _ if @xmath139 then @xmath140 is called the _ radius function _ of @xmath141 we also introduce the _ support function _ of @xmath3 by @xmath142 the _ difference body _ of a convex body @xmath3 is the set @xmath143 the function @xmath144 is called the _ width function _ of @xmath141 observe that for @xmath145 the quantity @xmath146 is equal to the distance between the two distinct supporting hyperplanes of @xmath3 orthogonal to @xmath123 the face of a convex body @xmath147 in the direction @xmath148 is denoted by @xmath149 elements of @xmath138 representable as intersection of a finite number of closed halfspaces are called _ convex @xmath32-polytopes . _ two convex @xmath32-polytopes @xmath150 are said to be in a _ general relative position _ if for every @xmath145 and @xmath151 @xmath152 is either empty or a singleton . if @xmath3 is a convex body in @xmath7 and @xmath153 then , for @xmath154 the normal cone @xmath155 does not depend on the choice of @xmath86 in @xmath156 we denote @xmath155 by @xmath157 and call it the _ normal cone of @xmath3 in direction @xmath123 _ if @xmath18 is a subset of @xmath158 then the set of all outward unit normals of @xmath3 at points of the set @xmath18 is called the _ spherical image _ of @xmath18 with respect to @xmath141 two boundary arcs @xmath53 and @xmath159 of @xmath160 are said to be _ antipodal _ if their spherical images with respect to @xmath3 are reflections of each other . for @xmath88 the _ @xmath161-neighbourhood of a set _ @xmath162 is the set @xmath163 that is , the set consisting of all those points @xmath43 whose distance to some point of @xmath18 is strictly less than @xmath164 the hausdorff distance @xmath165 between non - empty compact sets @xmath3 and @xmath35 in @xmath7 is the least possible @xmath166 such that @xmath167 and @xmath168 information on the hausdorff distance in the class of convex bodies is collected in @xcite . we introduce the distance function @xmath169 for sets @xmath170 as the minimum of @xmath171 where @xmath172 ranges over all translations and reflections . the _ area measure of order @xmath104 _ of a convex body @xmath173 ( see @xcite ) is given by @xmath174 where @xmath175 is a borel set in @xmath176 if @xmath12 then @xmath177 is said to be the _ length measure _ of @xmath141 given a strictly convex body @xmath178 @xmath179 denotes the boundary point of @xmath3 with outward normal @xmath123 if @xmath180 is absolutely continuous with respect to the measure @xmath181 on @xmath182 then we denote by @xmath183 the _ radon - nikodym derivative _ of @xmath184 i.e. the function obeying the equality @xmath185 for all borel subsets @xmath175 of @xmath186 ( see also formula ( 4.2.20 ) from @xcite ) . if the function @xmath183 is continuous , @xmath187 is the radius of curvature of @xmath3 at the point @xmath188 we remark that , according to our definition , @xmath189 may be zero for some values of @xmath123 if @xmath190 then @xmath183 is continuous and strictly positive , and @xmath191 is said to be the _ curvature _ of @xmath3 at @xmath188 we parametrize the unit circle @xmath186 in a standard manner by the vector function @xmath192 where @xmath193 given a @xmath57 planar convex body @xmath194 we put @xmath195 @xmath196 and @xmath197 for @xmath198 with @xmath199 we introduce the notation @xmath200 for @xmath57 convex bodies @xmath201 one can show that @xmath202 it is known that for @xmath57 planar convex bodies @xmath3 and @xmath35 one also has @xmath203 the knowledge of @xmath204 and the curvature @xmath205 for @xmath206 $ ] allows to determine the arc @xmath207 more formally , the parametrization @xmath208 of @xmath209 is determined from the representation @xmath210 equality can be found in @xcite and @xcite ( cf . also a more general result stated in theorem 4.3.2 from @xcite ) . by we see that two antipodal arcs @xmath209 and @xmath211 with @xmath212 and @xmath213 are reflections of each other if and only if @xmath214 for every @xmath215.$ ] in this section we prove parts i , iii , iv and v of theorem [ 06.05.15,12:36 ] . the proof of part ii is postponed to the next section . using @xmath85 and strict convexity of @xmath194 we see that for every @xmath216 there exists a relatively open subset @xmath217 of @xmath117 such that @xmath218 @xmath219 and the equality @xmath220 holds for some @xmath221 depending on @xmath123 then , since @xmath222 for every @xmath223 and @xmath224 from we deduce @xmath225 since @xmath117 is compact , there is a finite sub - family @xmath226 of @xmath227 covering @xmath176 by we have @xmath228 for every borel set @xmath175 being a subset of @xmath229 or @xmath230 for some @xmath231 furthermore , can be derived for any borel subset @xmath175 of @xmath117 using the decomposition @xmath232 and the inclusion - exclusion principle . hence @xmath233 and by this @xmath75 ( and @xmath234 for each @xmath145 ) . using the latter equality together with , we deduce that for every @xmath145 the functions @xmath5 and @xmath235 coincide in some neighbourhood @xmath236 of @xmath237 consequently , @xmath5 and @xmath235 coincide in the open set @xmath238 which encloses @xmath239 for @xmath240 let @xmath86 and @xmath87 be the endpoints of the arc @xmath241 . then @xmath242 and @xmath243 belong to @xmath51 and @xmath244 is a parallelogram . following @xcite we define @xmath245 to be @xmath246 with the sign determined by the condition @xmath247 see figs . [ 05.12.30,01:24 ] and [ 06.09.22,09:56 ] . it is known that @xmath248 where @xmath249 denotes , throughout the paper , the counterclockwise rotation about the origin by the angle @xmath250 we also have the equality @xmath251 cc ( 34,25 ) ( -1,-0.7 ) ( 32,19)@xmath3 ( 26,1)@xmath86 ( 26,22)@xmath87 ( -1,-2)@xmath242 ( -1.5,18)@xmath243 & ( 34,25 ) ( 3,0 ) ( 31,6)@xmath43 ( 9,17)@xmath245 ( 6,-0.5)@xmath48 + & first we prove that @xmath98 implies @xmath95 in the plane @xmath252 for every @xmath253 hence , from we get that @xmath98 implies @xmath254 it is easy to prove that the mapping @xmath255 is continuous and , by , also its inverse is continuous . moreover it maps a punctured neighbourhood @xmath40 of @xmath48 in a set @xmath256 with @xmath257 . let @xmath40 be a punctured neighbourhood of @xmath48 in which @xmath5 and @xmath235 coincide , up to an additive constant . then , in view of , @xmath258 for all @xmath259 and we get that @xmath260 and @xmath261 are translates of each other ( cf . the related lemma 1.5 from @xcite ) . the set @xmath262 is the union of the four lunettes outside @xmath31 ( where a _ lunette _ of @xmath3 is a compact set bounded by a chord of @xmath3 and a boundary arc of @xmath3 joining the endpoints of this chord ) . let @xmath86 and @xmath87 be as in the definition of @xmath263 the sum of the areas of the two lunettes adjacent to @xmath264 $ ] and @xmath264-x$ ] equals @xmath0 , the sum of the areas of the other two lunettes equals @xmath265 . similar considerations hold for @xmath64 too , and thus @xmath266 since @xmath267 for @xmath268 we obtain from and that @xmath42 for @xmath269 the proof of the converse implication is similar . we argue backwards : an open subset @xmath256 of @xmath47 with @xmath270 and @xmath271 for @xmath272 is mapped by @xmath273 onto a punctured neighbourhood @xmath40 of the origin . thus , using and , we obtain @xmath274 for @xmath275 in view of theorem [ 06.05.15,12:36 ] ( parts i and ii ) it suffices to construct @xmath3 and @xmath35 satisfying @xmath85 and @xmath276 let @xmath277 be a regular triangle of unitary edge length with center at the origin . if @xmath278 denote the unit outer normals to the edges , the area measure of @xmath277 is @xmath279 , where @xmath280 is the dirac delta distribution on the manifold @xmath186 centered in @xmath281 . for each @xmath282 let @xmath283 be a continuous non - negative function on @xmath186 supported in a small arc centered at @xmath281 , whose integral on @xmath186 is @xmath284 . moreover choose @xmath283 s in such a way that @xmath285 let @xmath286 be a convex body in @xmath77 whose length measure has density @xmath287 it is clear that the measure @xmath288 approximates in some sense the measure @xmath289 in fact , it can be shown that the _ prohorov metric _ ( see @xcite for the definition ) of @xmath288 and @xmath290 can be made arbitrarily small . let @xmath291 be a slight rotation of @xmath286 such that the supports of the area measures of @xmath286 and @xmath291 are disjoint . by a stability result for the minkowski problem with respect to the prohorov metric proved in ( * theorem 3.1 ) @xmath292 is close to @xmath293 , while @xmath294 is close to @xmath295 in the metric @xmath296 ( theorem 3.1 from @xcite is a strengthening of theorem 7.2.2 from @xcite , see also related theorems 4.3.5 and 7.2.6 from this monograph ) . consequently , the area of @xmath292 is close to @xmath297 while the area of @xmath294 is close to @xmath298 . let @xmath299 see figs . [ 06.01.04,09:52 ] and [ 06.01.04,09:53 ] depicting possible choices of bodies @xmath3 and @xmath64 respectively . by construction , @xmath3 and @xmath35 are @xmath57 and they satisfy @xmath81 thus , by theorem [ 06.05.15,12:36 ] ( parts i and ii ) , they have equal covariograms in a neighbourhood of @xmath239 furthermore , @xmath3 has smaller area then @xmath64 since by the two - dimensional version of the _ steiner formula _ ( see ( * ? ? ? * section 4.1 ) ) we have @xmath300 and @xmath301 cc ( 30,25 ) ( -20,-20 ) ( 3,20)@xmath3 & ( 30,25 ) ( -20,-30 ) ( 3,20)@xmath35 + & in the proof of theorem [ 06.05.15,12:36 ] ( part v ) we shall need the following lemma , presenting a formula which is also related to a formula given in @xcite . [ 06.01.06,01:19 ] let @xmath31 be a convex polygon in @xmath302 and @xmath303 be given by @xmath304 then @xmath305 and for @xmath306 we have @xmath307 where @xmath308 depends only on @xmath309 let @xmath310 and @xmath311 be antipodal vertices of @xmath31 such that @xmath312 and @xmath313 let @xmath314 be the union of the two edges of @xmath31 adjacent to @xmath315 in view of the assumption @xmath316 we have @xmath317 . the closure of the set @xmath318 ) \setminus ( p \cup(p+u))$ ] consists of two triangles @xmath319 , possibly empty , where one edge of @xmath320 is @xmath321 $ ] and the other two edges are parallel to the two edges of @xmath314 ( see fig . [ 06.01.06,11:10 ] ) . therefore @xmath322 where @xmath308 depends only on the directions of the edges of @xmath31 adjacent to @xmath310 and to @xmath323 since @xmath324)\setminus p ) + v(\delta_1 ) + v(\delta_2)\ ] ] and @xmath325 we get the desired formula . c ( 40,44 ) ( 4,4 ) ( 7,30)@xmath31 ( 30,30)@xmath326 ( 17,5)@xmath327 ( 16,33)@xmath328 ( 13.8,16.7)@xmath329 + if @xmath330 and @xmath331 is a convex body in @xmath332 then the covariogram function of @xmath333 is given by @xmath334 where @xmath335 and @xmath336 we introduce convex polygons @xmath337 which are obtained from the square @xmath338 ^ 2 $ ] by `` cutting off '' isosceles triangles at the vertices of @xmath339 the polygon @xmath340 is constructed by cutting off the isosceles triangle with lateral sides having length @xmath341 at the vertex @xmath342 of @xmath343 for each @xmath344 where the constants @xmath341 are defined as follows : @xmath345 see figs . [ 05.12.27,11:43 ] and [ 05.12.27,11:44 ] depicting @xmath346 and @xmath347 cc ( 24,25 ) ( 2,2 ) ( 1,23)@xmath346 & ( 24,25 ) ( 2,2 ) ( 1,23)@xmath348 + & no translation or reflection of @xmath346 coincides with @xmath349 and moreover , for each @xmath350 , @xmath351 furthermore , it is easy to see that @xmath352 lemma 3.1 from @xcite proves that for each @xmath353 the knowledge of the covariogram of a convex polygon @xmath354 near the boundary of its support determines the set @xmath355 but for @xmath356 we have @xmath357 hence @xmath358 and @xmath359 do not coincide in some neighbourhood of @xmath360 for @xmath361 , let @xmath362 be defined as in the statement of lemma [ 06.01.06,01:19 ] . since for each @xmath363 @xmath364 lemma [ 06.01.06,01:19 ] implies @xmath365 for each @xmath366 . since this set is a neighbourhood of @xmath48 the proof for @xmath96 is concluded by putting @xmath367 and @xmath368 for @xmath369 we define @xmath370^{d-2 } , \ h = p_2 \times [ -1,1]^{d-2},$ ] and the property proves the assertion . in ( * ? ? ? * lemma 2.1 ) it is shown that for planar @xmath57 convex bodies the asymptotic behaviour of the covariogram function near the boundary of its support allows to determine the non - ordered pair @xmath371 for every @xmath372 thus , the following lemma holds . [ 05.08.10,10:47 ] let @xmath3 be a planar @xmath57 convex body . then the covariogram of @xmath3 over any neighbourhood of @xmath55 determines the mapping @xmath373 where @xmath16 ranges over @xmath374 @xmath65 suppose that @xmath53 and @xmath159 are two disjoint antipodal boundary arcs of @xmath141 let @xmath375 be an endpoint of @xmath376 let us denote by @xmath377 the convex curve obtained by joining @xmath159 and the appropriate half of the line which is tangent to @xmath159 at @xmath378 we say that the translated arc @xmath379 _ captures _ the endpoint @xmath375 of @xmath159 if @xmath53 intersects @xmath377 at two points which bound an arc of @xmath377 containing @xmath375 in its relative interior ( see fig . [ 05.12.16,17:43 ] ) . cc ( 28,20 ) ( 4,4 ) ( 6,17)@xmath53 ( 13,2)@xmath380 ( 18,6)@xmath375 ( 25,12)@xmath159 & ( 30,24 ) ( 0,-2 ) ( 12,-2)@xmath381 ( 21,22)@xmath382 ( 29,8)@xmath383 + & the following lemma on capturing arcs improves slightly lemma 4.2 from @xcite , since it also indicates which translation vector can be chosen for making a capture . [ 05.08.10,14:15 ] let @xmath3 be a planar @xmath384 convex body . assume that the antipodal arcs @xmath209 and @xmath211 ( where @xmath212 and @xmath213 ) are not reflections of each other . let @xmath385 be equal to @xmath386 for the case @xmath387 and be equal to the maximal value in @xmath388 $ ] such that for every @xmath389 $ ] the equality @xmath214 holds , otherwise . we put @xmath390 then there exists a vector @xmath16 arbitrarily close to @xmath391 such that either @xmath204 is captured by @xmath392 or @xmath393 is captured by @xmath394 without loss of generality we assume that @xmath395 and @xmath396 the mapping @xmath397 is a reflection with respect to the midpoint of the segment @xmath398.$ ] by the choice of @xmath385 and by we see that the arcs @xmath399 and @xmath400 coincide for @xmath401 $ ] and differ in any neighbourhood of @xmath402 consequently , there exists a point @xmath403 on @xmath404 which is arbitrarily close to @xmath405 and is not in @xmath406 we pick the point @xmath383 on @xmath407 which has the same ordinate as the point @xmath403 on @xmath209 ( see also fig . [ 06.09.22,10:43 ] ) . in the case @xmath408 the arc @xmath409 contains the point @xmath403 and the point @xmath410 since @xmath411 and @xmath412 we see that the endpoint @xmath204 of @xmath209 is captured by a translate of the arc @xmath413 the corresponding translation vector @xmath414 can be chosen arbitrarily close to the vector @xmath415 using the symmetry of the arcs @xmath416 and @xmath417 we get that @xmath418 hence @xmath419 in view of the invariance of the statement of the lemma with respect to interchanging @xmath386 and @xmath420 with @xmath421 and @xmath422 respectively , the opposite case @xmath423 is settled analogously . the following lemma is a strengthening of proposition 5.1 from @xcite . it states that @xmath57 regularity of a planar convex body @xmath3 can be recognized from the covariogram of @xmath3 over every neighbourhood of @xmath239 [ 05.08.10,14:25 ] let @xmath3 and @xmath35 be plane convex bodies with @xmath45 then the condition @xmath67 implies that also @xmath35 belongs to the class @xmath74 clearly , under the given assumptions we get the equality @xmath254 strict convexity of @xmath3 is equivalent to strict convexity of @xmath424 thus , since @xmath3 is strictly convex , we get that @xmath35 is strictly convex , as well . it can be seen that @xmath35 belongs to @xmath28 by examining the asymptotic behaviour of @xmath425 ( restricted to @xmath426 ) at boundary points of @xmath35 ( see @xcite ) . further on , in order to get that @xmath35 is from the class @xmath57 we can argue in the same way as in the proof of lemma 6.2 from @xcite , where the equality of covariograms is used only at points close to their support . assume that @xmath427 first , by lemma [ 05.08.10,14:25 ] we deduce that @xmath428 if @xmath3 is centrally symmetric , then the knowledge of the mapping @xmath429 determines @xmath141 thus , in view of lemma [ 05.08.10,10:47 ] , we get the assertion . now let us assume that @xmath3 is not centrally symmetric . further on , let @xmath430 be an arbitrary open set with @xmath431 let us prove by contradiction that @xmath3 is determined within the class @xmath57 by its covariogram over @xmath432 assume the contrary , i.e. , there exists a planar convex body @xmath35 from the class @xmath57 such that @xmath35 can not be obtained from @xmath3 by reflection or translation and @xmath42 for all @xmath433 let @xmath434 be such that @xmath435 in @xcite it is shown that replacing @xmath35 by an appropriate translation or reflection there exist arcs @xmath49 and @xmath50 containing @xmath436 and @xmath437 respectively , in their relative interiors and contained in the set @xmath438 furthermore , in @xcite it is also noticed that if we assume additionally that @xmath49 and @xmath50 are maximal arcs with the above properties , then @xmath49 and @xmath50 are antipodal to each other , i.e. , @xmath439 and @xmath440 for @xmath441 with @xmath442 and @xmath443 we shall get a contradiction by showing that @xmath49 or @xmath50 is not maximal , i.e. , there exists an arc which strictly contains @xmath49 or @xmath50 and is contained in @xmath444 obviously , @xmath445 for @xmath215 \cup [ t_1+\pi , t_2+\pi].$ ] by lemma [ 05.08.10,10:47 ] we get the equality @xmath446 for each @xmath447 if @xmath448 then using the equalities @xmath449 @xmath450 and the continuity of the functions @xmath205 and @xmath451 we get that there exists an @xmath452 such that the equality @xmath445 holds for @xmath453.$ ] consequently , by we have @xmath454 a contradiction to the maximality of @xmath455 thus , in the sequel we assume that @xmath456 if there exists an @xmath452 such that @xmath214 for @xmath457,$ ] then in view of we have @xmath445 for @xmath453,$ ] which , by , implies the equality of the arcs @xmath458 and @xmath459 a contradiction to the maximality of @xmath455 now let us switch to the case when for every @xmath452 the functions @xmath205 and @xmath460 restricted to @xmath461 $ ] are not identically equal , i.e. , there exists a @xmath453 $ ] with @xmath462 let @xmath385 be the maximal scalar such that @xmath463 and @xmath464 for @xmath465.$ ] if @xmath466 we put @xmath467 if @xmath468 then for some @xmath469 the arc @xmath416 is a translate of @xmath470 or @xmath471 in this case we put @xmath472 by lemma [ 05.08.10,14:15 ] we see that either the endpoint @xmath204 of @xmath49 can be captured by @xmath50 or the endpoint @xmath393 of @xmath50 can be captured by @xmath455 furthermore , the corresponding translation vector can be chosen arbitrarily close to @xmath473 or @xmath474 without loss of generality , we assume that @xmath204 is captured by @xmath475 in @xcite it is shown that in this case a small arc @xmath476 is determined by the knowledge of @xmath477 and the values of the covariogram functions at points arbitrarily close to @xmath478 this means that we have the equality @xmath479 a contradiction to the maximality of @xmath475 theorem [ 06.05.15,12:36 ] ( part iii ) and the statement of theorem [ 05.08.10,15:35 ] for the case @xmath480 trivially imply the statement of theorem [ 05.08.10,15:35 ] for the case @xmath481 we only need to verify the implication @xmath482 for planar @xmath57 convex bodies @xmath194 since the reverse implication is covered by part i of the theorem . we borrow the notations from the statement of theorem [ 05.08.10,15:35 ] . let @xmath430 be an arbitrary open set with @xmath483 the set @xmath66 does not have accumulation points in @xmath484 because for any @xmath88 finitely many local symmetries of @xmath3 have length greater than @xmath485 therefore only finitely many points of @xmath66 lie outside @xmath430 . let @xmath486 , for @xmath487 and some @xmath488 be all the local symmetries of @xmath3 corresponding to points of @xmath66 outside @xmath430 . let @xmath489 and @xmath490 be some antipodal connected components of @xmath491 choose @xmath492 such that @xmath493 is adjacent to @xmath489 and @xmath494 is adjacent to @xmath490 ( or vice versa ) . let @xmath86 and @xmath87 be antipodal points of @xmath51 . either @xmath495 and @xmath496 , for some @xmath497 , or @xmath498 and @xmath499 , for a suitable choice of @xmath489 and @xmath490 , or @xmath500 and @xmath501 , for a suitable choice of @xmath489 and @xmath490 and @xmath502 . in the first case , by lemma [ 05.08.10,10:47 ] , @xmath51 and @xmath503 , suitably translated , coincide in a neighbourhood of @xmath86 and @xmath87 . to deal with the second case we observe that the points of @xmath66 , which correspond to local symmetries contained in @xmath504 , belong to @xmath430 . therefore arguments similar to those of the proof of theorem [ 05.08.10,15:35 ] ( for the case when @xmath18 from that theorem is given by @xmath505 ) imply that if @xmath42 for each @xmath506 , then @xmath504 is contained in a translate or a reflection of @xmath503 . the third case follows from the previous two . [ 06.05.16,16:17 ] it is natural to look for minimal ( with respect to inclusion ) sets @xmath18 such that @xmath41 implies coincidence of @xmath3 and @xmath35 , up to translations and reflections . since the covariogram is @xmath48-symmetric , we limit our discussion to @xmath48-symmetric sets @xmath44 we claim that the set @xmath18 defined in theorem [ 05.08.10,15:35 ] is minimal in the following sense . for certain @xmath507 sets @xmath3 it suffices to remove from @xmath66 two pairs of opposite points to violate the conclusion of the theorem . let us construct a corresponding counterexample . assume that two local symmetries of @xmath3 have the same center of symmetry , say @xmath48 . let @xmath508 and @xmath509 be the arcs that constitute these local symmetries , and @xmath510 be the midpoints defined as in the statement of theorem [ 05.08.10,15:35 ] , which correspond to the local symmetries @xmath508 and @xmath511 let @xmath512 , be the connected components ( in counterclockwise order ) of @xmath513 . we claim that there exist a convex body @xmath35 which is not a translation or a reflection of @xmath3 and such that @xmath514 holds . it suffices to define @xmath35 as the body obtained from @xmath3 by flipping the boundary arcs @xmath515 and @xmath516 that is , the boundary of @xmath35 is composed of the arcs @xmath517 and @xmath518 the bodies @xmath3 and @xmath35 satisfy @xmath85 and thus their covariograms coincide in a neighbourhood of @xmath519 , by theorem [ 06.05.15,12:36 ] ( part iii ) . moreover if @xmath308 and @xmath520 are the arcs which constitute a local symmetry of @xmath3 , different from @xmath508 and @xmath509 , then the boundaries of @xmath3 and @xmath35 ( properly translated and , possibly , reflected ) coincide in a neighbourhood of @xmath308 and @xmath520 . therefore @xmath5 and @xmath235 coincide in a neighbourhood of the midpoints corresponding to the local symmetry . we emphasize that the example constructed here is similar in nature to the example from the proof of theorem [ 06.05.08,17:28 ] ( for the case @xmath96 ) . if @xmath3 is a @xmath57 globally non - symmetric planar convex body , then we see that the set @xmath66 introduced in the proof of theorem [ 05.08.10,15:35 ] is empty . this remark obviously yields part ii of corollary [ 06.05.16,17:58 ] . [ 06.05.08;14:43 ] let @xmath3 and @xmath35 be strictly convex bodies in @xmath521 let @xmath3 be @xmath48-symmetric and let @xmath254 then for scalars @xmath522 with @xmath523 and @xmath524 the subset @xmath525 of @xmath503 is centrally symmetric if and only if for every @xmath215 $ ] we have @xmath526 where @xmath527 let us prove the sufficiency . the equality @xmath528 is equivalent to the condition that @xmath529 is a translate of @xmath530 ( see for the relation among the gradient and @xmath531 ) . but , since @xmath3 is @xmath48-symmetric , @xmath532 consequently , the diagonals of @xmath529 are translates of @xmath533 $ ] and @xmath534.$ ] the chord @xmath535 $ ] is the only chord of @xmath35 being a translate of @xmath536 $ ] ( because that chord is an _ affine diameter , _ that is , @xmath537 and @xmath538 are antipodal ) . since @xmath539 @xmath535 $ ] is a diagonal of @xmath540 a similar argument implies that @xmath541 $ ] is the other diagonal of @xmath540 hence , for every @xmath215,$ ] the point @xmath542 is a reflection of @xmath543 with respect to the midpoint @xmath544 of the diagonal @xmath535 $ ] and the sufficiency is verified . now let us show the necessity . if @xmath545 is centrally symmetric , then a translate of this set is contained in @xmath546 since @xmath75 and @xmath3 is @xmath48-symmetric , a translate of the same set coincides with @xmath547 the latter implies that for every @xmath215 $ ] the parallelogram @xmath529 is a translate of @xmath530 and , in view of , shows the sufficiency . _ part i. _ the equality @xmath75 implies that @xmath35 is strictly convex . let @xmath40 be any open set containing @xmath548 we pick an arbitrary @xmath549 and show that for a sufficiently small @xmath88 the boundary arcs @xmath550 and @xmath551 around the antipodal points @xmath552 and @xmath553 respectively , are symmetric with respect to a reflection in a point . if @xmath554 is ranging from @xmath555 to @xmath556 then the midpoint of the chord @xmath557 $ ] of @xmath426 traverses a path starting at @xmath558 and terminating at the origin . thus , for some @xmath559 $ ] the midpoint @xmath560 of @xmath561 $ ] lies in @xmath548 c ( 40,30 ) ( 4,4 ) ( 36,14)@xmath426 ( 21,16)@xmath48 ( 9,28)@xmath562 ( -1,8)@xmath563 + clearly , for some @xmath88 the midpoint of @xmath564 $ ] lies in @xmath40 for all @xmath565 with @xmath566 let @xmath567 be scalars such that @xmath568 the assumption of the theorem and imply that @xmath569 where @xmath570 and @xmath215.$ ] hence , in view of lemma [ 06.05.08;14:43 ] , @xmath571 is a reflection of @xmath211 in a point . thus , each pair of antipodal points of @xmath35 can be enclosed in the relative interior of symmetric boundary arcs of @xmath64 which implies the central symmetry of @xmath572 _ part ii . _ the case @xmath573 can be transformed to the case @xmath574 using part iii of theorem [ 06.05.15,12:36 ] . thus , we assume that @xmath575 the statement of the theorem follows then from lemma [ 06.05.08;14:43 ] applied for arbitrary @xmath523 with @xmath576 and @xmath420 sufficiently close to @xmath577 let @xmath3 be an arbitrary locally symmetric convex body in @xmath521 let us show that @xmath5 determines @xmath194 up to translations and reflections . we assume that @xmath3 is strictly convex , since for non - strictly convex bodies the determination was verified in ( * ? ? ? * theorem 1.1 ) . by part ii of theorem [ 06.05.16,15:21 ] we can determine all local symmetries of @xmath194 up to translations . but then the theorem follows from proposition 1.4 in @xcite , stating that if additionally to the knowledge of @xmath5 a non - degenerate boundary arc of @xmath3 is known , then @xmath3 can be determined uniquely , up to translations and reflections . the _ nikodym distance _ @xmath578 ( also known as the _ symmetric difference metric _ ) between convex bodies @xmath3 and @xmath35 in @xmath7 is given by @xmath579 it is known that @xmath578 generates the same topology in the class of convex bodies as the hausdorff distance @xmath580 @xcite . furthermore the inequality ( see @xcite ) @xmath581 for all @xmath9 shows that the operator @xmath582 is continuous provided the class of convex bodies is endowed with the nikodym distance , and the distance between covariograms is measured with respect to the maximum norm . let @xmath583 it suffices to approximate a @xmath507 convex body @xmath3 by a totally non - symmetric @xmath507 one . in order to do this we approximate , for each @xmath452 , @xmath584 by a continuous positive function @xmath585 with the property that the set @xmath586 has empty relative interior and @xmath587 by minkowski s theorem and the relative stability result ( theorem 7.2.2 from @xcite ) there exists a convex body @xmath588 whose radius of curvature is @xmath585 and @xmath589 let us construct @xmath590 . the relative interior of @xmath591 , @xmath592 , is open and therefore it is the disjoint union of denumerably many open intervals . it is thus possible to construct an odd continuous function @xmath593 in @xmath186 with the property that @xmath594 for each @xmath16 , @xmath593 vanishes outside @xmath592 , and @xmath593 never vanishes in @xmath592 ( except for the case @xmath595 , since in this case it has to vanish in at least two antipodal points ) . the function @xmath596 , for @xmath597 , satisfies the required properties . let us denote by @xmath598 the class of all planar convex bodies @xmath3 which are not determined , up to translations and reflections , by their covariogram over every neighbourhood of @xmath239 given @xmath599 we introduce the class @xmath600 such that @xmath601 if and only if there exists a planar convex body @xmath35 with the properties @xmath602 clearly @xmath603 . let us prove now that for every @xmath599 the class @xmath604 is closed . let @xmath605 be a sequence of convex bodies belonging to @xmath604 and converging to @xmath606 and for each @xmath502 let @xmath607 be the convex body associated to @xmath608 . by the blaschke selection theorem there exists a subsequence @xmath609 converging to some @xmath610 obviously , @xmath611 , @xmath612 and @xmath613 let @xmath43 be a point from @xmath614 then @xmath43 belongs to @xmath615 definitely , i.e. for all @xmath125 sufficiently large . since @xmath616 and passing to the limit ( recall ) one obtains @xmath617 . therefore @xmath618 is indeed in @xmath619 thus , @xmath598 is a countable union of closed sets . consequently , the complement @xmath620 of @xmath598 is a countable intersection of open sets . using corollary [ 06.05.16,17:58 ] ( part i ) , we get that @xmath620 contains all @xmath621 totally non - symmetric planar convex bodies . further on , applying lemma [ 05.08.12,11:37 ] , wee see that @xmath620 is dense in the class of all convex bodies in @xmath521 let @xmath622 be the class of all planar convex bodies @xmath3 which are not determined , up to translations and reflections , by their covariogram over every neighbourhood of the origin . let @xmath598 be as in the previous proof , and @xmath623 be the class of all planar strictly convex bodies . we have @xmath624 it is well known that the class @xmath625 of all non - strictly convex bodies is meager . therefore , its subclass @xmath626 is meager , as well . the class @xmath627 is meager because , by theorem [ 06.05.15,12:36 ] ( part iii ) , it is a subclass of @xmath598 . then @xmath622 is meager , since it is the union of two meager classes . given convex bodies @xmath628 the _ cross covariogram function _ of @xmath3 and @xmath35 is defined by @xmath629 where @xmath43 ranges over @xmath116 the support of @xmath630 is obviously equal to @xmath631 for the proof of theorem [ 06.05.08,17:28 ] ( part i ) for @xmath369 we need the following lemma . a. if @xmath86 is a vertex of @xmath277 and @xmath633 is in the relative interior of the support cone of @xmath277 at @xmath634 then for sufficiently small @xmath635 @xmath636 where @xmath637 , i.e. @xmath308 does not depend on @xmath638 . b. if @xmath639 is an edge of @xmath640 @xmath86 is the midpoint of @xmath641 and @xmath633 is in the relative interior of the support cone of @xmath277 at @xmath634 then for sufficiently small @xmath642 @xmath643 where @xmath644 _ i. _ let @xmath645 t \cap [ -\lambda t+p - tu].$ ] since @xmath646 is a vertex of @xmath647 it is also a vertex of @xmath648 provided @xmath649 the latter condition is equivalent to @xmath650 analogously , the vertex @xmath651 of @xmath652 is also a vertex of @xmath648 provided @xmath653 let @xmath654 be such that point @xmath655 lies in @xmath277 . then in view of the above remarks for @xmath656 both @xmath646 and @xmath651 are vertices of @xmath657 moreover , it can be seen that @xmath31 is a parallelotope whose facets are parallel to facets of @xmath277 incident to @xmath634 and @xmath658 $ ] is a diagonal of @xmath657 then @xmath659 is a vector joining the endpoints of this diagonal . consequently , for @xmath660 $ ] the polytopes @xmath661 and @xmath662 are homothetic with homothety ratio @xmath663 and @xmath648 does not depend on the choice of @xmath664 the above facts easily imply the statement of part i. _ ii . _ let @xmath648 be introduced in the same way as above , and @xmath665 be a hyperplane through the origin orthogonal to the line @xmath666 then the set @xmath648 can be approximated by the cylinder @xmath667 so that we have @xmath668 clearly , for small @xmath554 the polytope @xmath669 is the intersection of the @xmath110-dimensional simplices @xmath670 with @xmath671 which means that we can apply the statement of part i for the cross covariogram function of these @xmath110-dimensional simplices . by this we get the statement of part ii . as was mentioned in @xcite , it is sufficient to prove the determination property for @xmath672 in the case when @xmath31 is a simplicial polytope such that @xmath31 and @xmath33 are in general relative position . it is known ( see @xcite and ( * ? ? ? * theorem 2.1 ) ) that if a polytope @xmath35 has the difference body @xmath673 ( for @xmath31 as above ) , then @xmath674 thus , clearly for the determination of @xmath64 up to translations and reflections , it is sufficient to retrieve the set @xmath675 let @xmath16 be an outward facet normal of @xmath31 and let @xmath676 since @xmath31 is a simplicial polytope , @xmath277 is a @xmath110-dimensional simplex . since @xmath31 and @xmath33 are in general relative position we have that @xmath677 is a singleton consisting of some vertex @xmath86 of @xmath678 clearly , @xmath679 and @xmath680 . let us pick a point @xmath681 from @xmath682 and a point @xmath683 from @xmath684 consider the point @xmath685 from @xmath686 ( which is the facet of @xmath673 parallel to @xmath277 ) . let us consider an arbitrary vector @xmath687 then , for small @xmath635 @xmath688 polytopes @xmath689 and @xmath690 involved in have facets @xmath691 and @xmath692 respectively , both lying in @xmath693 thus , it can be seen that the polytope @xmath694 from can be approximated by the cylinder @xmath695,$ ] where @xmath696 and @xmath697 clearly @xmath698 if and only if either @xmath682 or @xmath699 is a point . this is equivalent to @xmath700 , and in view of , to @xmath701 for each @xmath702 . when @xmath703 then the function @xmath704 is determined by . in view of lemma [ 05.10.24,13:17 ] ( part ii ) , this function determines the set @xmath705 . given a convex body @xmath706 and a vector @xmath151 we introduce the body @xmath707 which is the union of all those chords of @xmath3 that are parallel to @xmath43 and are not shorter than @xmath708.$ ] it can easily be shown that @xmath709 consequently , @xmath710 let @xmath711 clearly , @xmath712 is the orthogonal projection of @xmath713 onto @xmath714 for @xmath145 we have @xmath715 formula is presented in @xcite and @xcite . now let us come to the proof of the next theorem . the counterexample constructed below is strongly related to some counterexample from @xcite . let @xmath716 and @xmath717 be relatively open subsets of @xmath117 bounded by @xmath718-dimensional spheres and such that the sets @xmath719 are mutually disjoint . let @xmath3 be a @xmath57 convex body satisfying the conditions @xmath720 for @xmath721 and @xmath722 for @xmath723 then we introduce the body @xmath35 defined by @xmath724 for @xmath725 and @xmath726 otherwise ( see figs . [ 05.08.30,16:50]-[05.08.30,16:51 ] for the illustration in the case @xmath96 ) . it can be seen that @xmath254 by standard compactness arguments , there exists an @xmath731 such that for every @xmath2 with @xmath732 and for every two - dimensional linear space @xmath665 containing @xmath43 the endpoints of the two chords of @xmath733 which are translates of @xmath708 $ ] either all belong to @xmath734 or all belong to @xmath735 . hence , for every @xmath736 , with @xmath737 and for @xmath738 we have @xmath739 we recall that the volume of a convex body @xmath3 in @xmath7 with @xmath740 can be written as @xmath741 let @xmath145 and @xmath554 be such that @xmath742 and @xmath743 . up to translations of @xmath194 we may assume that @xmath744 and we have @xmath745 in view of the equalities , @xmath746 the latter implying the coincidence of @xmath5 and @xmath235 for @xmath2 in a neighbourhood of @xmath48 . d. hug and r. schneider , _ stability results involving surface area measures of convex bodies _ , palermo ( 2 ) suppl . ( 2002 ) , no . 70 , part ii , 2151 , iv international conference in `` stochastic geometry , convex bodies , empirical measures @xmath748 applications to engineering science '' , vol . ii ( tropea , 2001 ) . , _ polytopes and brunn - minkowski theory _ , polytopes : abstract , convex and computational ( scarborough , on , 1993 ) , nato adv . 440 , kluwer acad . , dordrecht , 1994 , pp . 273299 . s. s. skiena , _ geometric reconstruction problems _ , handbook of discrete and computational geometry . 2nd ed . , discrete mathematics and its applications . boca raton , fl : chapman & hall / crc , 2004 , pp .

we use first - principle calculations for transition metal impurities v , cr , mn , fe , co and ni in gaas as well as cr and mn in gan , gap and gasb to identify the basic features of the electronic structure of these systems . the microscopic details of the hole state such as the symmetry and the orbital character , as well as the nature of the coupling between the hole and the transition metal impurity are determined . this could help in the construction of model hamiltonians to obtain a description of various properties beyond what current first - principle methods are capable of . we find that the introduction of a transition metal impurity in iii - v semiconductor introduces a pair of levels with symmetry - one localized primarily on the transition metal atom referred to as crystal - field resonance ( cfr ) and the other localized primarily on the neighboring anions referred to as the dangling bond hybrid ( dbh ) . in addition , a set of nonbonding states with symmetry , localized on the transition metal atom are also introduced . each of the levels is also spin - split . considering mn in the host crystal series gan gap gaas gasb , we find that while in gan the hole resides in the level deep in the band gap , in gaas and gasb it resides in the level located just above the valence - band maximum . thus , a dbh - cfr level anticrossing exists along this host - crystal series . a similar anticrossing occurs for a fixed host crystal ( _ e.g. _ gaas ) and changing the 3 impurity along the 3 series : v in gaas represents a dbh - below - cfr limit , whereas mn corresponds to the dbh - above - cfr case . consequently , the identity of the hole carrying orbital changes . the symmetry ( vs. ) , the character ( dbh vs. cfr ) as well as the occupancy of the gap level , determine the magnetic ground state favored by the transition metal impurity . lda+u calculations are used to model the effect of lowering the energy of the mn state by varying u. we find that this makes the dbh state more host - like , and at the same time , diminishes ferromagnetism . while the spin - splitting of the host valence band in the presence of the impurity has been used to estimate the exchange coupling between the hole and the transition metal impurity , we show how using this would result in a gross underestimation of the coupling .

cond-mat0309509

the prospect of manipulating the electron spin to store and transport information in semiconductor devices has led to renewed interest in the physics of transition metal ( tm ) impurities in semiconductors - an area which was intensively studied in the eighties @xcite . current interest @xcite in achievement of ferromagnetism ( fm ) at ambient temperatures has led to the investigation of the mechanism that stabilizes fm in transition - metal doped semiconductors . one useful approach to obtain an understanding of the electronic properties of these systems is the _ first principles electronic structure approach _ where one focuses on the explicit electronic and spin wavefunctions of the system . variational minimization of the total energy , determines within the underlying approximations of the spin density functional theory some of the basic features of the states involved , such as the extent of localization , the magnitude of the spin interactions as well as the identity of disorder and compensating defects ( antisites ; interstitials ) . however , the approach does have the drawback of underestimating the extent of electron correlations in addition to being a zero - temperature approach . while comparison with experiment ( _ e.g. _ ferromagnetic temperature vs. alloy composition @xcite ) can be used to assess the extent to which electron correlations are underestimated , the first - principle results have to be generally mapped onto a model hamiltonian to calculate finite temperature properties . while model hamiltonians @xcite have been widely used to describe the properties of these systems , the underlying assumptions in choosing a particular form for the hamiltonian are rarely justified in their own right . generally @xcite , one renormalizes away the electronic degree of freedom and retains only the spin degree of freedom for the transition metal impurity . one then assumes a local interaction between the transition metal impurity and the free carrier ( usually rkky - like ) , and then solves for various physical properties of these systems . in the present work we use first - principles calculations to examine whether the assumptions made in model hamiltonian treatments are consistent with an _ ab - initio _ description of the electronic structure of these systems . our detailed results are then cast in the language of a simple electronic structure model , which could be used in an informed construction of model hamiltonians . we start by identifying the main physical quantities that come into play in determining the electronic structure of these systems . when a trivalent cation site such as ga of a iii - v semiconductor is replaced by divalent mn , an acceptor level ( denoted as e(0/- ) ) is generally created in the band gap . if the fermi level @xmath7 lies below this e(0/- ) level , then mn is charge neutral , _ i.e. _ , its formal oxidation + 3 equals that of the ga atom being replaced . in this case there is a hole in the mn - related orbital . if , on the other hand , the fermi level is above e(0/- ) , then the mn - related orbital captures an electron from the fermi sea ( _ i.e. _ creating a hole there ) , becoming negatively charged ( _ i.e. _ oxidation state mn@xmath8 ) . in this case the hole resides in the fermi sea . the mn - induced hole for @xmath9 e(0/- ) features prominantly in contemporary theories of ferromagnetism the model hamiltonians involve three entities - the host crystal states , the transition metal atom , and the impurity - induced hole state . there are approximations made at various levels which involve decoupling various degrees of freedom . at the first level , one decouples the orbital degrees of freedom associated with the transition metal atom , describing it with a localized spin - only part . the spin is interacting with a hole system through a local exchange interaction . at the next level of approximation , one reduces the problem to that of the transition metal spin interacting with the hole spin , assuming that the host crystal is unperturbed . the main assumptions made in such approaches , which we wish to examine , are : \(i ) _ the hole resides in a bulk - like , hydrogenic , delocalized state . _ this picture is based on the assumption that the perturbing potential @xmath10-@xmath11 generated by the impurity is dominated by a long - ranged coulomb part , as a result of which only a small percentage of the charge resides in the wigner - seitz cell and the rest is distributed over a large portion of the host crystal . in this `` host - like hole '' picture , one reduces the problem to a quasi - hydrogenic form in which the acceptor state is designated via quantum numbers ( @xmath12 , @xmath13 ... ) of the host lattice . in such cases the wavefunction of the acceptor level is delocalized , and can essentially be constructed from the host crystal @xmath14 states . this picture is motivated by the fact that divalent , post transition metal atom elements such as zn@xmath8 form in iii - v semiconductors quasi - hydrogenic acceptor states @xcite with small binding energies . similarly , extrinsic @xmath13-type doping of ii - vi dilute magnetic semiconductor cdmnte @xcite also form hydrogenic hole states . however , unlike zn in gaas or extrinsic @xmath13-type cdmnte , the mn atom introduced into iii - v s has chemically active @xmath5 orbitals @xcite , so it is not obvious that the acceptor state it forms in gan , gap or gaas would qualify as a delocalized host - like hydrogenic state . indeed , the microscopic features determining the localization of the hole wavefunction , such as the @xmath5-character of the acceptor level must be considered . such interactions could change the symmetry ( @xmath0 vs. @xmath1 ) of the hole state , hence its coupling to the host . the pertinent quantum designation of the hole state is impurity - like ( @xmath0 , @xmath1 ) not host - like effective - mass @xcite . \(ii ) _ the host valence band maximum ( vbm ) levels are unperturbed by the transition metal impurity . _ in this view , the host band structure represented in the model hamiltonian could be described by a @xmath15.@xmath13 model , valid for the pure host crystal and is decoupled from the part of the hamiltonian involving the host + hole system . however , since one of the symmetry representations of the mn @xmath5 orbitals in tetrahedral sites ( @xmath0 , @xmath1 ) is the same as that of the zincblende vbm ( @xmath0 ) , such states could couple , hence become mutually perturbed . \(iii ) _ the spin of the hole couples to the spin of the transition metal impurity via an interatomic local exchange interaction @xmath16 . _ as only the spin degree of freedom of the transition metal atom is considered , while the orbital degree of freedom is ignored , the free carriers feel the effective magnetic field produced by the transition metal impurity spin . this is modelled as a _ local _ exchange interaction , @xmath16 , between the transition metal impurity and the spin of the free carrier . hence , the magnitude of @xmath16 determines the energy scale of ferromagnetic ordering . areas visited by the free carrier are rendered ferromagnetic . however certain materials @xcite are found to show activated behavior in their transport implying no free charge carriers , yet they exhibit ferromagnetism . the current model which requires delocalized carriers can not explain ferromagnetism in such systems . in what follows , we use first principles calculations to examine the validity of assumptions ( i)-(iii ) for 3@xmath5 impurities v , cr , mn , fe , co and ni in gaas , as well as for mn and cr in gan , gap and gasb . we then construct a qualitative model that explains our numerical results . we find that \(i ) the mn - induced hole could have significant 3@xmath5 character . the assumption of a `` delocalized hydrogenic hole '' is not supported by first - principle calculations . the depth of the acceptor level ( reflecting its localization ) and the coupling of the 3@xmath5 impurity orbitals to the hole change markedly with the host crystal in the series gan @xmath2 gap @xmath2 gaas @xmath2 gasb . the hole generated by introducing mn in gan is found to have significant 3@xmath5 character , while in gasb the hole is found to have primarily host character . further , the symmetry of the hole depends on the combination of the host crystal with the impurity atom . for example while in gaas : co the hole has dominantly @xmath0 character , the corresponding isoelectronic impurity znse : fe has a hole with @xmath1 symmetry . \(ii ) the presence of the transition metal impurity perturbs the valence band of the host semiconductor . the extent of the perturbation depends on the relative position of the impurity generated levels ( referenced to the valence band maximum of the host ) which have the same symmetry as the valence band maximum . in gaas : v which has levels with @xmath1 symmetry in the bandgap , the perturbation is small , while for gaas : mn the perturbation is large . \(iii ) the interaction between the spin of the tm atom and the spin of the host - like hole has a predominantly non - local part . this is evidenced by the strong stabilization of the ferromagnetic state for mn and cr pairs in gaas at @xmath17 8 @xmath18 separation . this interaction induces a spin - polarization of the host - like states . the direction of the spin - polarization depends on the relative energy position of the cation vacancy generated ( host - like ) states with respect to the impurity states . furthermore , the band - theoretic description of cr in gap shows a partially occupied mid - gap band , and the wavefunctions associated with this mid - gap state are localized . yet , even in the absence of free carriers , our total energy calculations predict a ferromagnetic ground state to be strongly stabilized , while no long range magnetic order is expected . there have been considerable earlier first - principles @xcite work on the electronic structure of these systems . one of the most well - studied systems is mn in gaas , which is found to be half - metallic @xcite . the gaas cell with one mn atom in it has a net magnetic moment of 4 @xmath19 @xcite , with part of the moment residing on the as neighbors of the mn atom . the hole state is found to be strongly hybridized with the transition metal state , and has been referred to as a hybridized band of holes @xcite . the mn atom and its four nearest neighbors are found to account for most of the density of states at the valence band edge @xcite . since only the first shell of as atoms surrounding mn are affected by the spin polarization of the mn atom , the interactions are believed to be short - ranged @xcite . the first principle results have been interpreted by ref . @xcite as suggestive of a @xmath6/@xmath20 electron configuration on mn @xcite . lda+u calculations @xcite have been used to obtain a description of the electronic structure consistent with photoemission . first - principle calculations have been used earlier to elucidate the magnetism in these systems . mahadevan and zunger @xcite developed a simple model of interaction of the cation - vacancy generated states with the transition metal states to understand how ferromagnetism results when mn is doped into the chalcopyrite semiconductor cdgep@xmath21 . sato and katayama - yoshida @xcite have calculated the energy difference between ferromagnetic and the random alloy to determine which impurity could give rise to ferromagnetism . they found that at low concentrations v , cr , and mn doping in iii - v stabilized the ferromagnetic state , while , fe , co and ni doping stabilized the magnetically disordered state . mirbt , sanyal and mohn @xcite showed that the interaction of the transition metal impurity with the as dangling bond states could result in a spin polarization of the hybridized dangling bond states . the partial occupancy of these spin - polarized levels results in ferromagnetism . sanvito , ordejon and hill @xcite found an decrease of the spin - splitting of the valence band maximum of the gaas host with impurity concentration . this is in contrast with what is expected from the mean - field kondo hamiltonian traditionally used to describe these systems @xcite . they attribute the deviation to a breakdown of the mean - field approximation , while they say that the kondo hamiltonian is good enough to provide a description of the ferromagnetism . schilfgaarde and mryasov @xcite have used the total energies obtained from first - principle calculations for different materials to extract exchange interaction strengths . they find a decrease of the exchange interactions with concentration which prompts them to suggest that the picture @xcite of carrier - mediated ferromagnetism is not valid for these systems . we build on the current understanding of the electronic and magnetic properties that exists in the literature . however , we focus our calculations specifically on the examination of the features ( i)-(iii ) assumed as `` input '' to most model hamiltonian theories . we have carried out first - principle electronic structure calculations using density functional theory , within the momentum space total energy pseudopotential method @xcite , using ultra - soft pseudopotentials @xcite as implemented in the vasp @xcite code . the ga pseudopotentials that we used for gaas and gap did not include the ga 3@xmath5 states in the valence . while this is usually a good approximation for gaas and gap , it has been found that for gan not retaining ga 3@xmath5 states in the valence leads to erroneous results for some physical properties such as optimized lattice constant , cohesive energy @xcite . we therefore used ultra - soft pseudopotentials which included ga @xmath5 states in the basis for gan . we studied transition metal impurities - v , cr , mn , fe , co and ni in 64 atom supercells of zincblende gasb , gaas , gap and gan . in table i we compare the calculated lattice constants of pure iii - v using pw91 gga exchange functional @xcite with the experimental values @xcite . we have fixed the equilibrium lattice constant of the supercells at the calculated values of the pure host given in table i. the basis sets had a cutoff energy for plane waves equal to 13.3 ry for gasb , gaas and gap and 29.4 ry for gan . we used a monkhorst pack grid of 4x4x4 k - points which includes @xmath14 . the cell - internal positions of the atoms were allowed to relax to minimize the forces . the equilibrium transition metal - to - as bond lengths in gaas were 2.47 , 2.47 , 2.48 , 2.44 , 2.36 and 2.34 @xmath18 for v , cr , mn , fe , co and ni respectively . the @xmath5 partial density of states as well as the local moment at the transition metal were calculated within a sphere of radius of 1.2 @xmath18 about the atoms and have been broadened with a gaussian of 0.2 ev full width at half maximum . the total energy differences between the ferromagnetic and the antiferromagnetic spin arrangements were computed for tm pairs at first and fourth neighbor separations for a parallel and an anti - parallel arrangement of their spins to determine whether a specific transition metal impurity resulted in a ferromagnetic state or not . _ lda vs. gga : _ in order to compare lda @xcite and gga @xcite exchange functionals , we consider the case of co impurity in gaas , where earlier lda work @xcite suggests a nonmagnetic ground state . using the experimental lattice constant of 5.65 @xmath18 for gaas , we find that the gga calculations lead to a magnetic ground state with a moment of 2 @xmath19 . the energy of this state is strongly stabilized by @xmath17 150 mev compared to the nonmagnetic state . using lda exchange functional we find that while the nonmagnetic state is stabilized for a 2x2x2 monkhorst pack grid as used in the earlier work @xcite , the magnetic state with the moment of 2 @xmath19 is stabilized by @xmath17 40 mev for a 4x4x4 monkhorst pack k - points grid . these observations are consistent with the fact that gga calculations have a greater ability to stabilize a magnetic ground state than lda calculation . for other impurities - cr and mn in gaas , the lda and gga results are found to give the same ground state . we use the gga exchange functional throughout this work . the introduction of various transition metal impurities lead to defect levels in the band gap of the semiconducting host . we compute the formation energies of the transition metal impurities in various charge states @xmath22 . the formation energy for a defect comprising of atoms @xmath23 in the charge state @xmath22 was computed using the expression @xcite @xmath24 where @xmath25 and @xmath26 are , respectively the total energies of a supercell with and without the defect @xmath23 . @xmath27 denotes the number of atoms of defect @xmath23 transferred in or out of the reservoir , while @xmath28 denotes their chemical potentials . _ total energies : _ the total energies of the charged supercells were computed by compensating any additional charge on the impurity atom by a uniform jellium background and have been corrected for interactions between charges in neighboring cells using the makov and payne correction @xcite . we use the static dielectric constant values - 15.69 for gasb , 12.4 for gaas , 11.11 for gap and 10.4 for gan @xcite . the quadrupole moment of the isolated defects was calculated as the difference between the moments of the supercell with the charged defect and that with the neutral defect . _ transition energies : _ the defect transition energy @xmath29 is the value of the fermi energy @xmath7 at which @xmath30=@xmath31 . the zero of the fermi energy is chosen as the valence band maximum @xmath32 of the pure host . _ chemical potential limits : _ as the reservoir supplying the atoms could be elemental solids , or compounds formed from the elements , we express @xmath28 as the sum of the energy of the element in its most stable structure @xmath33 , and an additional energy @xmath34 _ i.e _ @xmath35 = @xmath33 + @xmath34 . the stable structures we considered for the elements were nonmagnetic body - centred cubic ( bcc ) for v , antiferromagnetic bcc for cr , the antiferromagnetic face - centred cubic ( fcc ) for mn , ferromagnetic bcc for fe , ferromagnetic hexagonal for co , ferromagnetic fcc for ni and the nonmagnetic base - centred orthorhombic structure for ga . the required ranges of @xmath34 are determined by @xmath36 @xmath37 0 ; @xmath38 @xmath37 0 ; @xmath39 @xmath37 0 ( no precipitation of solid elements ) and by the formation energies of the semiconducting host and competing binary phases formed between the elements of the semiconductor and the transition metal impurity . this could be the most stable nias phase of mnas in the case of gaas : mn and the mnp phase of cras for gaas : cr . the energies e(@xmath23 ) , e(0 ) , and @xmath34 entering eq.(1 ) are calculated within the density functional formalism discussed earlier . no correction for the band gap underestimation was made . changing the k - point mesh from 2x2x2 to 4x4x4 changed the formation energies by @xmath17 20 mev . we used a plane wave cutoff of 13.3 ry for the calculations involving mn in gaas . increasing the cutoff to 29.4 ry , changed the formation energies by @xmath17 10 mev . we now divide the main features of the first principle calculations into three main entities introduced in section i : a. the nature of the impurity - induced level in the gap b. the impurity - induced valence - band resonances c. the perturbed host vbm then , in section v , we will provide a simple model that explains all of our numerical results qualitatively . figure 1 shows the transition - metal local dos for v , cr , mn , fe , co and ni in gaas , projected into irreducible representations @xmath0 and @xmath1 and spin directions + and - . the vbm is at the zero of the energy . the gga band gap of pure gaas is found to be 0.3 ev ; all the impurities v - ni introduce levels into this band gap . we first discuss the nature of these gap levels , and then the circumstances how and when a hole is present in them . for spin - up @xmath40 ( shaded region ) , spin down @xmath41 ( solid black line ) and @xmath1 ( dashed line ) symmetries . the zero of energy represents the valence band maximum of the host . the number of k - points used is 64.,width=528 ] from fig . 1 we see that the sequence of levels occupied for cr in gaas are @xmath40 , @xmath42 , @xmath41 and @xmath40 in the order of increasing energy . for a free atom one would expect levels of one spin channel to be filled up before levels of the other spin channel ; the deviation that one observes here reflects solid state effects . the two sets of @xmath40 and @xmath41 levels that we find for each impurity are suggestive of bonding / antibonding combinations arising from hybridization . we therefore determined the atoms on which each of the @xmath0 states are localized by computing atom - projected dos . bonding states with a large wavefunction amplitude on the tm site are referred to as `` crystal field resonances '' ( cfr ) @xcite , whereas antibonding @xmath0 states with low contribution on the tm which are localized instead on the four nearest as atoms are referred to as the dangling bond hybrid ( dbh ) . the full explanation of the genesis of these states will be provided in sec . v. we see that , \1 . _ symmetry of gap levels and lowest unoccupied levels _ : substituting cr , mn and co in gaas introduce levels with up - spin character and @xmath0 symmetry in the band gap . these levels are partially occupied by 1 , 2 and 2 electrons for neutral cr , mn and co , respectively : cr@xmath43 ( @xmath44 ) , mn@xmath43 ( @xmath45 ) and co@xmath43 ( @xmath45 ) . the levels introduced by v@xmath43 ( @xmath46 ) and fe@xmath43 ( @xmath47 ) are fully occupied . the first unoccupied levels have @xmath40 and @xmath48 symmetry for v and fe respectively . _ d character of gap levels _ : the transition metal projected partial density of states for different transition metal impurities in gaas given in fig . 1 indicates that the gap level / first unoccupied level is strongly @xmath5-like for the early transition metal impurities v and cr , while for the heavier 3@xmath5 elements _ e.g. _ mn these levels have less @xmath5 character . an increased @xmath5 character of the gap level would imply increased spatial localization of the wavefunction in the vicinity of the impurity . and @xmath49+@xmath50 about the tm impurity for mn in gan ( filled squares ) , cr in gaas ( open inverted triangles ) and mn in gaas ( filled circles ) compared with the result for an electron gas ( solid line ) . @xmath51 is the lattice constant of the host supercell.,width=528 ] \3 . _ degree of localization of gap levels _ : we quantify the degree of localization by plotting in fig . 2 the charge q(r@xmath52)=@xmath53 enclosed between concentric spheres with radius @xmath49 and @xmath54 centered about the impurity atom . the integrated charge between the spheres is plotted as a function of r@xmath52 . for comparison , we show also the result expected for a homogeneous charge distribution ( electron gas ) , where the charge density at any point in the cell is given by reciprocal volume @xmath55 . we see that q(r@xmath52 ) for the tm impurities has little similarity to the results for an electron gas . changing the impurity from mn to cr in gaas , we see an increase in the charge density localized in the vicinity of the impurity . we find that till a radius which includes second neighbors of the tm atom , the integrated charge for cr is higher than for mn . further , we find that the enclosed charge in the vicinity of the impurity atom is higher in gan : mn than in gaas : mn and the decay of the wavefunction is faster . point for @xmath3 , @xmath56 ( upper panel ) and @xmath4 ( lower panel ) states for v - co impurities in gaas.,width=528 ] \4 . _ the negative exchange splitting of the gap levels _ : having established the identity of the gap levels , we now investigate their spin splittings . in fig . 3 we have plotted the spin splittings of the cfr and dbh levels at @xmath14 point for the impurities v - co in gaas obtained from an analysis of their eigenvalues / eigenfunctions . for v and cr the spin splitting of the dbh levels is positive _ i.e. _ @xmath57 states are at lower energies compared to @xmath58 . however , for mn , fe and co the splitting is negative with the @xmath58 states at lower energies compared to @xmath57 . a similar negative exchange splitting was observed earlier for the te states in mnte @xcite , the ce states in cefe@xmath21 @xcite , and the mo states in sr@xmath21femoo@xmath59 @xcite . the explanation was that the @xmath13 states of te in cdte , the @xmath5 states of ce in cefe@xmath21 and the @xmath5 states of mo in sr@xmath21femoo@xmath59 are sandwitched in between the 3@xmath5 states of the transition metal atom . the @xmath13-@xmath5 hybridization results in an exchange splitting of these states opposite in direction to that on the transition metal atom . indeed we see from fig . 1 that for the cases where the dbh states are bracketed by the spin split cfr states , the spin splitting is negative . _ enhanced exchange splitting for @xmath3 states _ : from fig . 3 , we see that the exchange splitting of the @xmath3 states is larger than that of the @xmath56 states for mn , fe and co impurities . having discussed the existence of impurity - induced levels in the band gap of the host semiconductor , we next discuss the location of these levels . _ acceptor transitions for gap levels _ : single particle lda or gga levels do not have any rigorous meaning . we thus calculate _ transition _ energies , @xmath60(@xmath22,@xmath61 ) which correspond to the value of the fermi energy @xmath7 at which the defect changes from a charge state @xmath22 to @xmath61 . table ii provides the calculated and measured @xcite acceptor / donor transition energies for various transition metal impurities in gaas . the calculated acceptor levels for mn and cr in gasb , gaas , gap and gan are plotted in fig . 4 where the host band edges are aligned according to their calculated unstrained valence band offsets @xcite . we see that as the electronegativity of the host crystal increases in the sequence gasb @xmath2 gaas @xmath2 gap @xmath2 gan , its bulk ionization energy (= position of vbm with respect to vacuum ) increases . the acceptor level is thus farther away from the vbm of gan than it is from the vbm of gaas . thus , gan : mn and gap : cr have more localized hole states whereas gasb : mn has more delocalized holes . this behavior , whereby the acceptor energy level does not follow the host valence band energy ( as in the case of hydrogenic impurities ) characterizes localized states @xcite . \7 . _ multiplet states and violation of isovalency rule _ : we use the level occupancies ( fig . 1 ) as well as the net magnetic moments that we obtain for different transition metal impurities in gaas ( table iii ) to obtain the multiplet configuration describing the ground state . these are given in table iii . we also provide the multiplet configuration observed from experiment @xcite and find that there is agreement in all cases . it is interesting to compare the ground state multiplets of two isoelectronic cases - znse : fe@xmath8 and gaas : co@xmath62 in their neutral charge states . in both cases we expect an electron configuration of @xmath20 . normally , one would expect to find equal multiplets for isoelectronic cases @xcite ( isovalency rule ) . this expectation is based on the fact that we are looking at a low transition metal impurity concentration regime where basic crystal field theory ideas are expected to be sufficient to explain the observed ordering of energy levels . however , we find in gaas : co@xmath62 the configuration @xmath63t@xmath21 , i.e the hole is in the @xmath4 level ( @xmath64 ) , whereas in znse : fe@xmath8 we find @xmath65e , i.e the hole is in the @xmath66 level ( @xmath67 ) . the reason for the difference is that the stronger @xmath13-@xmath5 hybridization for gaas : co pushes the @xmath57 levels to higher energies , so that the @xmath68 levels are occupied first and the hole resides in the @xmath57 level . we thus conclude that the isovalency rule is not applicable , and one can not assume that the hole is in a `` generic '' @xmath5 state . _ fm vs. afm ground state and their relation to the symmetry of the gap levels _ : having summarized the nature of the level induced in the gap by the introduction of the transition metal impurity , we now analyze when a ferromagnetic state is favored . in table iii we provide the energy difference between the ferromagnetic and antiferromagnetic energies for two tm atoms at nearest @xmath69e@xmath70 and fourth neighbor @xmath69e@xmath71 fcc positions in a 64 atom supercell of gaas . we find that ( a ) when the level in the gap is fully occupied as in v@xmath43 and fe@xmath43 , the favored ground state is antiferromagnetic . ( b ) when the level in the gap is partially occupied and has @xmath0 symmetry as in cr@xmath43 , mn@xmath43 , the ferromagnetic state is lower in energy . this is also the case for electron doped v@xmath72 in gaas which is strongly ferromagnetic . although co@xmath43 ( @xmath63t@xmath21 ) also has a hole in the @xmath0 level , the system is at the brink of a ferromagnetic - to - nonmagnetic transition . ( c ) when the level in the gap has @xmath1 symmetry , as in the case of electron doped fe@xmath72 , the stability of the fm state is weaker . evidently the _ symmetry _ of the hole carrying state strongly determines the magnetic order . _ we conclude that fm is stabilized strongly only when the hole resides in the level with @xmath0 symmetry . _ ( note , viz . v , that in @xmath73 symmetry @xmath0 states are strongly bonded to their neighboring atoms , whereas the lobes of the @xmath1 orbitals point in between the nearest - neighbor atoms . ) figure 1 shows that in addition to the gap levels , the introduction of a transition metal atom gives rise to resonance levels that lie deep within the valence band of the host semiconductor . in most model hamiltonian theories @xcite , one usually ignores the orbital degree of freedom of the transition metal impurity , and the presence of the impurity is included only as a localized spin of value 5/2 . in our calculations we find that the degree of localization of such deep resonances ( thus , the possibility of depicting them as local point - like spin ) varies sharply with the position of the impurity in the periodic table . for heavier tm such as fe and mn , the deeper resonance level has significant tm @xmath5 character ( being crystal - field resonances ) , while for the early tm impurities in gaas , one finds that the deeper @xmath0 levels have significantly less tm character ( being dangling bond hybrids ) . this is discussed next . _ anticrossing of the two @xmath0 levels in different host materials : _ level anticrossing is evident when keeping the impurity atom fixed , and , changing the host semiconductor . considering the example of mn , we find that by changing the host from gasb to gan , the dbh and the cfr exhibit anticrossing . this is not the only difference : we find that the exchange splitting of the dbh levels is in the same direction as the cfr levels ( positive ) in gan : mn , in contrast to gaas : mn . further , in gan : mn the @xmath74 levels lie above the @xmath75 levels , unlike the case in gaas : mn . the reason is evident from fig . 4 which shows that the vbm of gan is much deeper than the vbm of gaas . since the free mn@xmath8 ion has its @xmath5 orbitals _ above _ the gan vbm , but _ below _ the vbm of gasb or gaas , an anticrossing occurs along the gan @xmath2 gap @xmath2 gaas @xmath2 gasb series . this is illustrated in fig . 5 which shows that in gan : mn for the up spin channel , the upper @xmath0 is more localized than the lower @xmath0 , whereas in gaas : mn the localization sequence is reversed . this clarifies a confusion that existed in the literature @xcite regarding the question of whether the gap level is localized or not . our result shows that the answer depends on the host . these results also clarify the nature of the acceptor transition for mn in different materials . gan : mn can be viewed as a @xmath76-like case since its configuration is @xmath77 ( hole in @xmath78 ) , and the ( 0/- ) acceptor transition is from a mn configuration @xmath76 to @xmath6 . on the other hand for mn in gap , gaas and gasb we have the configuration ( @xmath6 + hole ) , _ i.e. _ @xmath79 and the acceptor transition is from a mn configuration of ( @xmath6+hole ) to @xmath6 . projected partial density of states evaluated within a sphere of radius 1.2 @xmath18 for a mn impurity in gan ( upper panel ) and gaas ( lower panel ) . the number of k - points used is 64 . , width=528 ] \2 . _ occupancy of the valence band resonances and comparison with photoemission _ : table iii gives the calculated occupancy of the crystal field resonances of the 3@xmath5 impurities in gaas ( in square brackets ) . these levels are found to have a configuration of `` @xmath6 '' for cr , mn and fe . experimentally the position of these levels can be detected by valence band photoemission @xcite . by suitably tuning the photon energy so that the photo - ionization cross - section is maximum for the tm - related states , an electron can be ionized from these deep cfr levels . @xcite used resonant valence band photoemission and showed that the the cfr levels for mn in gaas are located at @xmath80 - 4 ev . a direct comparison of the position of these levels with the single - particle density of states calculated for gaas : mn places these energies at @xmath80 - 2 - 3 ev . the lda error in the position of these states is because of the self - interaction correction ( sic ) that places these energies too high @xcite . as pointed out earlier @xcite for the 3@xmath5 states in ii - vi s , the experimental result should be compared with the total energy difference between the configurations @xmath76 and @xmath6 and not with the bare single particle eigenvalues . alternatively , the lda error can be empirically corrected by using the simplified lda+u version of sic . in fig . 6 we plot the mn @xmath81 partial density of states as a function of u for gaas : mn . as @xmath82 increases , the position of the mn @xmath3 level is pushed deeper into the gaas valence band . agreement with xps for the @xmath81 being at @xmath80 - 4 ev occurs for u @xmath17 2 ev . at the same time , the mn character of the dbh state at @xmath7 ( not shown ) decreases with increasing @xmath82 . as the dbh hole becomes more delocalized , the e@xmath83-e@xmath84 stabilization energy ( insert to fig . 6 ) is reduced ; a `` host - like hole '' obtained for unphysically large u leads to nearly vanishing fm stabilization energy . clearly , the picture of `` host like hole '' is invalid for gaas : mn , since for the @xmath82 that leads to agreement with xps the dbh hole is still localized to some extent , whereas for very large @xmath82 , when the hole is delocalized , there is no ferromagnetism . projected partial density of states evaluated within a sphere of radius 1.2 @xmath18 for on - site coulomb interaction strengths u=0 , 6 and 10 and 15 for mn in gaas . the number of k - points used is 4x4x4 . the inset shows the variation in e(fm)-e(afm ) for two mn atoms at nearest neighbor positions.,width=528 ] having studied the impurity - induced levels in the gap and deep in the host valence band , we next examine the perturbation of the host states , especially the host valence band maximum by the presence of the impurity atom . figure 7 shows the up and down spin band dispersions for 3 @xmath85 cr doped gap supercell ( panels ( a ) and ( b ) ) . the band dispersion of the gap host without the impurity has been provided in panel ( c ) for comparison . the thickness of the lines depicting these bands has been made proportional to the cr @xmath5 character of the states . we see that cr introduces a new band within the band gap of gap . in a band - theoretic picture , this system is metallic , with the fermi energy within the impurity band . interestingly , ( 1 ) the host band dispersions are significantly altered by the presence of the impurity . in particular the vbm is found to have significant tm @xmath5 character for the 3@xmath85 cr concentration represented by the supercell . ( 2 ) a cr - induced spin - splitting of the valence band maximum is observed . effects ( 1 ) and ( 2 ) suggest that the host vbm is sufficiently perturbed by the transition metal . using 64 k - points . the positions of the as atoms ( filled circles ) with respect to the mn atoms is shown in the inset.,width=528 ] another way of detecting perturbations in the host bands is to examine the host projected dos of the system containing the impurity . in fig . 8 we have plotted the as @xmath13 partial density of states projected onto different as atoms labelled 1 - 4 for a gaas supercell containing 2 mn atoms . the as atom labelled 1 has one mn nearest neighbor , while the as atom labelled 2 has two mn nearest neighbors . the as atoms show a strong polarization which increases with the number of mn neighbors . the as atoms labelled 3 and 4 which are far away from the mn atoms show a reduced polarization . to pictorially see the perturbation in the vbm states , we compare in fig . 9 the wavefunction squared along two chains in the ( 110 ) plane for the valence band maximum of the pure gaas host ( panel ( a ) ) as well as the vbm ( _ i.e. _ state below dbh ) of the system with the mn impurity in the up ( panel ( b ) ) and down ( panel ( c ) ) spin channels . the upper chain in panels ( b ) and ( c ) contains the perturbing mn impurity . the perturbation of the vbm in the presence of the impurity can be assessed by comparing the perturbed charge density for each spin channel with the unperturbed charge density of the host lattice . we find that the perturbations are significant in the chain containing the mn atom , while in adjoining chains , the perturbation is limited in extent . further the perturbations are stronger in the up spin channel than in the down spin channel . and each contour is 1.6 times larger . , width=336 ] to evaluate the 3@xmath5-induced spin - splitting in the vbm , we reference the up and down spin vbm eigenvalues of the impure system to the corresponding vbm of the pure host semiconductor . this is done by aligning the average potentials on ga atoms far away from the impurity for the two systems . the presence of the impurity band with @xmath0 symmetry above the vbm for mn and cr impurities complicates the identification of the valence band maximum . we associate the highest occupied triply degenerate eigenvalues at @xmath14 point with the impurity band , and the next deeper set as @xmath86 . the shift with respect to the pure host @xmath87=@xmath88(gaas : mn)-@xmath32 ( gaas ) and @xmath89=@xmath90(gaas : mn)-@xmath32 ( gaas ) is given in table iv for the impurities v , cr and mn in different host semiconductors . we find that the perturbation of the host vbm is smaller in the down spin channel compared to the up spin channel . this is consistent with what we find from the charge density plotted in fig . the spin splitting of the valence band maximum depends strongly on the transition metal impurity and host semiconductor combination . considering the case of impurities in gaas , we find that while the spin splitting associated with the introduction of v is only 0.06 ev , it increases to 0.39 ev for mn . keeping the impurity fixed ( mn ) , and varying the semiconductor host ( gaas to gan ) , we find the splitting decreases from 0.39 ev to 0.1 ev . the small valence band splittings in the case of v in gaas as well as mn in gan compared with that for mn in gaas is because of the larger energy separation between the interacting @xmath0 states in the former cases compared to the latter . we are now in a position to examine whether the physical picture of the electronic structure of 3@xmath5 impurities as assumed in model hamiltonian theories ( reviewed in sec . i ) is consistent with first - principles calculations ( outlined in sec . \(i ) _ the nature of the tm - induced hole state _ : a 3@xmath5 impurity in a iii - v semiconductor generates two sets of states with @xmath0 symmetry , and one set of states with @xmath1 symmetry in each spin channel . while one set of @xmath0 states are localized on the tm atom ( cfr ) , the other are localized on the host anion atoms next to the impurity ( dbh ) . these states cfr and dbh exhibit anticrossing for a fixed tm as a function of the host anion gan @xmath2 gap @xmath2 gaas @xmath2 gasb , or for a fixed host as a function of the impurity v @xmath2 mn . the localization of the hole state decreases as we move from mn in gan to mn in gap , and then to mn in gasb . not all impurities introduce holes . in gaas , v@xmath43 and fe@xmath43 have no hole ; cr@xmath43 , mn@xmath43 and v@xmath72 have @xmath0 holes ; and fe@xmath72 has an @xmath1 hole . in all cases , however , the hole is non - hydrogenic , manifesting significant admixture of 3@xmath5 character and showing deep acceptor levels whose energies do not follow the host vbm . this implies that the neglect of the short - range part of the impurity potential and the consequent expansion of the acceptor wavefunction in terms of a single host wavefunction are questionable . the effective mass of the hole state is therefore different from that of the host , as observed in recent experiments @xcite . the exchange splitting of the cfr states is different for the @xmath0 states from that for the @xmath1 states . while the splitting for the @xmath1 states is larger than that for the @xmath0 states for v and cr in gaas , the order is reversed for mn , fe and co. this reversal in the order of the spin splitting of the cfr states is accompanied by a reversal in the sign of the spin splitting of the dbh states . the identity of the hole state - both the symmetry as well as the character depends on the impurity - host combination . while the hole carrying orbital for fe in znse has @xmath1 symmetry , the hole is found to be located in an orbital with @xmath0 symmetry for the isovalent doping of co in gaas . \(ii ) _ the nature of the host vbm _ : the introduction of the transition metal perturbs the valence band of the host crystal . we find this perturbation to be large when the state in the gap has @xmath4 character . this is because the effective coupling is larger since the dbh states have strong host character . we find that the vbm is spin - split in the presence of 3@xmath5 impurity , and that the vbm in the up - spin channel is perturbed more strongly than in the spin down channel in the presence of the impurity . \(iii ) _ ferromagnetism and symmetry _ : impurities with fully occupied dbh - like @xmath0 gap states such as v@xmath43 , fe@xmath43 show antiferromagnetism . partial occupation of the @xmath4 as in cr@xmath43 , mn@xmath43 or v@xmath72 show ferromagnetism . partially occupied @xmath1-like level like in fe@xmath72 show weak or no ferromagnetism . \(iv ) _ ferromagnetism and hole localization _ : using lda+u as an artificial device to explore the consequences of delocalized host - like - hole states we find ( insert to fig . 8) that in this limit there is reduced ferromagnetism . we find that despite the well known gga - lda band gap error , as well as the underestimation of the location of deep cfr states due to sic , these first principle calculations provide us with the correct spin multiplets . lda+u changes some details ( cfr locations ) , but does not alter the basic picture emerging from gga / lda when the hole is dbh - like as in gaas : mn . most of the results of the density functional study of the electronic structure of 3@xmath5 impurities in iii - v s ( sec iii ) can be captured by a simple model . the electronic structure of substitutional 3@xmath5 in iii - v semiconductors can be understood as arising from the interaction of the host cation vacancy (= anion dangling bonds ) with the crystal - field and exchange split orbitals of a 3@xmath5 ion . \(a ) _ the dangling bonds for a column iii cation vacancy @xmath91 _ : a cation vacancy for a column iii element gives rise to a fully - occupied @xmath12-like @xmath92 level located deep in the host valence band , and a partially - occupied @xmath13-like @xmath0 level located just above the host valence band maximum , with wavefunction amplitude localized primarily on the neighboring atoms @xcite . this is evident from the wavefunction squared of the ga vacancy dangling bond state shown in the ( 110 ) plane in fig . the neutral vacancy @xmath93 has a deficiency of 3 electrons , _ i.e _ the orbital configuration is @xmath94 , where @xmath13 denotes its major orbital character . spin - polarization splits this @xmath95 vacancy level into spin - up [ @xmath96 and spin - down [ @xmath97 states , but the splitting is small ( 90 mev at the @xmath14 point ) on account of the rather delocalized nature of these pure host dangling bond orbitals . symmetry in pure gaas . the lowest contour corresponds to 0.0017 e/@xmath98 and each contour is 1.6 times larger . , width=528 ] \(b ) _ the crystal - field split tm 3@xmath5 orbitals _ : the tetrahedral crystal - field of the zincblende host splits the tm @xmath5 levels into @xmath99 and @xmath100 , with @xmath1 below @xmath0 in the point - ion limit @xcite ; the crystal - field ( cf ) splitting of the ion is denoted by @xmath101(@xmath0-@xmath1 ) . exchange interactions further split these levels into spin - up ( + ) and spin - down ( - ) , with exchange - splittings @xmath102[@xmath48(d)-@xmath42(d ) ] and @xmath103[@xmath41(d)-@xmath40(d ) ] . transition metal ion ( left panel ) with the anion dangling bond levels ( right panel ) , when the tm @xmath5 levels are energetically deeper than the dangling bond levels.,width=528 ] transition metal ion ( left panel ) with the anion dangling bond levels ( right panel ) , when the tm @xmath5 levels are energetically shallower than the dangling bond levels.,width=528 ] the energy levels of a cation - substituted tm in a iii - v semiconductor can be thought of @xcite as the result of hybridization between the anion dangling bonds generated by a column iii cation vacancy @xmath91 [ ( i ) above ] , and the crystal - field and exchange split @xmath5 levels of a tm ion placed at the vacant site [ ( ii ) above ] . there are two limiting cases : when the 3@xmath5 levels are well below the host cation dangling bonds ( e.g. mn in gaas , fig . 11 ) , or when the 3@xmath5 levels are well above the host cation dangling bonds ( e.g. v in gaas , fig . the dangling bond states are shown on the right hand side of fig . 11 and 12 , while the crystal field and exchange split tm @xmath5 levels are shown on the left hand side of fig . 11 and 12 . the levels generated after hybridization are shown in the central panel . the @xmath95 levels of the anion dangling bond hybridize with the @xmath100 levels of the transition metal . in contrast , the @xmath99 level of the tm ion remains largely unperturbed since the host does not have localized @xmath1 states in this energy range , available for significant coupling . considering the examples of mn in gaas and gan , we plot the charge density of the @xmath1 and the @xmath0 states in the ( 110 ) plane in fig . 13 . it is evident that the @xmath1 states for mn in gaas are essentially nonbonding , while in gan , as a result of the reduced mn - mn separation , there is weak interaction between the mn atoms . the hybridization in the @xmath0-channel creates bonding , transition - metal localized `` crystal - field resonances '' ( cfr ) : @xmath74 , @xmath104 , as well as the host - anion localized antibonding `` dangling bond hybrids '' ( dbh ) : @xmath105 , @xmath106 , whereas the @xmath1-channel creates the non - bonding @xmath75 and @xmath68 states . this model explains the existence of two sets of @xmath40 and @xmath41 levels that we found in fig . 1 . the available electrons for occupation of these levels are n=(n-1)+6 for a @xmath107s@xmath108 transition metal atom ( three electrons are used to complete the anion dangling bond state to @xmath109(@xmath13 ) , leaving @xmath110 at the transition metal ion . ) for gaas : v ( fig . 12 ) the ordering of levels after hybridization is @xmath111 with increasing energy . hence for v we have n=(n-1)+6=8 electrons occupy the dbh and cfr levels . thus , v has the configuration ( @xmath112 ) , as seen in fig . 1 and table iii . in gaas : mn , ( b ) @xmath56 in zinc blende gan : mn , ( c ) @xmath3 in gaas : mn , ( d ) @xmath3 in zinc blende gan : mn , ( e ) @xmath4 in gaas : mn and ( f ) @xmath4 in zinc blende gan : mn . the lowest contour corresponds to 0.015 e/@xmath98 and each contour is 1.6 times larger.,width=528 ] the order of levels for @xmath113 @xmath114 @xmath115 ( fig . the number of electrons ( n-1)+6 is 10 , 11 and 12 for mn , fe and co respectively . this agrees with fig . 1 showing that mn and fe in gaas have the ordering of levels shown in fig . 11 , with fully - filled @xmath81 and @xmath116 levels and 2 , 1 and 0 holes in the @xmath57 level . by an analysis of the density of states obtained within our first principle calculations , we have determined ( table iii ) the energy minimizing orbital configurations for the transition metal impurities v , cr , mn , fe and co in gaas in fully relaxed configurations . the first unoccupied orbital for each impurity has been indicated in boldface in table iii . the simple model of figs . 11 , 12 gives the same result . -like levels in gaas for different 3@xmath5 impurities . , width=528 ] \1 . _ level anticrossing _ : the model explains how the hopping interaction between the @xmath0 states on the transition metal impurity with the cation - vacancy states generates a pair of @xmath0 states in each spin channel . the bonding - antibonding character of these states is determined by the relative separation of the interacting levels as well as their interaction strengths . hence as depicted in fig . 14 , one could by a suitable choice of the tm impurity change the character of the gap levels . when the orbital energy of the 3@xmath5 ion lies below the host dangling bond , we have a `` cfr - below - dbh '' situation , illustrated in fig . 11 . in this case one has cfr states in the valence band of the semiconductor while the gap levels are more delocalized with dominant weight in the dangling bonds . this is the case for gaas : fe , mn , and co. conversely , when the orbital energy of the 3@xmath5 ion lies _ above _ the host dangling bond , we have the `` cfr - above - dbh '' situation , illustrated in fig . 12 . in this case the gap level is cfr - like . while fig . 1 illustrates anticrossing when changing the 3@xmath5 atom , but keeping the host fixed , _ e.g. _ gaas , fig . 5 suggests that there is also anticrossing when keeping the 3@xmath5 atom fixed _ e.g. _ mn , but changing the host crystal gan @xmath2 gap @xmath2 gasb . indeed , as the anion dangling bond becomes deeper ( gasb @xmath2 gaas @xmath2 gap @xmath2 gan ) or the tm @xmath5 level becomes shallower ( mn @xmath2 v ) , there will be a dbh - cfr anticrossing , resulting in the level ordering shown in fig . the cfr - above - dbh is exemplified by gaas : v ( see fig . 1 ) which has a shallow @xmath5 level and by gan : v , cr and mn which have a deep semiconductor vbm @xcite . for gaas : v , this level anti - crossing results in a reduced @xmath117(@xmath0 ) exchange splitting compared to the cases where this crossing did not occur ( cr and beyond ) . for mn in iii - v s we see a reversal between hole - in - dbh case for gaas : mn with @xmath118_{cfr}$ ] @xmath119 configuration to the case of hole - in - cfr @xmath120_{cfr}$ ] @xmath121 for gan : mn . as the cfr is much more localized than dbh , this is reflected by a change in acceptor level depth . \2 . _ negative exchange splitting in the dbh manifold and low - spin configurations _ : for co , fe and mn in gaas the special position of the ga - vacancy level @xmath122(p ) _ between _ the exchange - split tm ion levels @xmath40(d ) and @xmath41(d ) ( fig . 11 ) , results in a hybridization - induced exchange splitting of the dbh states ( fig . 3 ) , opposite in direction to the splitting on the cfr levels ( fig . 3 ) . this results from the fact that @xmath58 is pushed down by @xmath41(@xmath5)-@xmath41(@xmath13 ) coupling more than the @xmath40(@xmath5)-@xmath40(@xmath13 ) coupling pushes @xmath57 up . such a negative exchange splitting was also observed in bulk mnte @xcite , cef@xmath21 @xcite and sr@xmath21femoo@xmath59 @xcite . as seen in table iii , the negative exchange splitting leads to the dbh orbital configuration ( @xmath123 ) and ( @xmath124 ) for mn and fe , respectively . this corresponds to a low - spin configurations for mn with a moment of 4 . co , on the other hand , has a configuration of [ @xmath125@xmath126@xmath127@xmath128 ( @xmath129)@xmath130 , so @xmath131=2@xmath19 . note that the moment @xmath131=4 for gaas : mn is a consequence of @xmath132 : the @xmath133 cfr levels give a spin of 5/2 , and the @xmath58-below-@xmath57 gives a configuration @xmath134 , so the total spin is 4/2 , and @xmath131=4 . had @xmath58 been above @xmath57 , we would have had the configuration @xmath135 with s=3 , and @xmath131=6 . as a result of the negative exchange splitting of the dbh states , one finds that the exchange splitting for the more delocalized @xmath3 levels is larger than that for the more localized @xmath56 levels . _ migration of the d - holes into the dbh acceptor states of gaas : cr , mn and co _ : the @xmath110 configuration of the trivalent tm ion corresponds to @xmath136 for cr and @xmath76 for mn in gaas and thus to 2 and 1 hole respectively , relative to @xmath137 . however , since for these impurities the @xmath81 and @xmath116 levels are deeper in energy than the @xmath4 levels ( fig . 11 ) , it is energetically favorable for the @xmath5 holes to `` float '' into the higher @xmath4 levels . as a result , we find that gaas : cr ( @xmath136 ) has a configuration @xmath138_{cfr}$]@xmath139 with two holes in the dbh , while for gaas : mn ( @xmath76 ) we find @xmath138_{cfr}$]@xmath140 with one hole , and gaas : fe ( @xmath6 ) has a closed - shell configuration of @xmath141_{cfr}$]@xmath142 . one expects co ( @xmath20 ) to have a configuration @xmath143_{cfr}$]@xmath144 , _ i.e. _ a ground state multiplet @xmath145 with a hole in @xmath56 . however , we find that the lowest energy configuration for co is @xmath146_{cfr}$]@xmath147 with a hole in the dbh level _ i.e. _ ground state multiplet @xmath148 . the migration of the holes from the tm - localized deep cfr levels into the dangling - bond manifold for cr , mn and co in gaas creates partially occupied @xmath13-@xmath5 hybridized states at the fermi level with significant delocalized as @xmath13 character . _ perturbation of the valence band maximum of the host _ : the valence band maximum of the pure host has @xmath0 symmetry . hence , they can interact with the states with the same symmetry on the dangling bonds as well as the tm ion . the consequent spin splitting of the vbm must depend on the relative separation of the levels involved as well as the coupling strength . our analysis for different tm - host combinations suggests that when the gap level has @xmath4 character the spin - splitting of the vbm is large ( @xmath17 0.4 ev for mn in gaas ) . changing the host semiconductor from gaas to gan , increases the energy separation between the dangling bond levels generated by the ga vacancy and the valence band levels . consequently the perturbation of the valence band levels resulting in the observed spin - splitting is smaller ( @xmath17 0.1 ev for mn in gan ) . this vbm spin - splitting of the impure system has been traditionally used to estimate the exchange interaction strength @xmath16 between the hole and the transition metal atom . we see that the spin splittings of the vbm of the impure system is grossly underestimates the spin - splitting of the ( dbh , cfr ) as seen in figs . 1,3 and 5 . we have analyzed the basic electronic structure of 3@xmath5 transition metal impurities in iii - v semiconductors . we find that the introduction of a 3@xmath5 impurity is accompanied by the introduction of a pair of states with @xmath0 symmetry in addition to nonbonding states with @xmath1 symmetry . not all 3@xmath5 impurities introduce holes . the basic symmetry and character of the hole state depends strongly on the semiconductor - impurity combination . we find that the hole has significant 3@xmath5 character . we have constructed a microscopic model which captures the basic aspects of the electronic structure of transition metal impurities in semiconductors . the elements of this model are the relative separation of the dangling bond and transition metal levels , the @xmath13-@xmath5 hybridization strength and the crystal - field and exchange splittings of the transition metal levels . we model a change in these interaction strengths by changing the semiconducting host , keeping the transition metal impurity fixed , - mn . we find that while the hole introduced by mn has significant 3@xmath5 character in gan , it is more delocalized in gaas . the symmetry ( @xmath1 vs. @xmath0 ) , the character ( dbh vs. cfr ) as well as the occupancy of the gap level , determine the magnetic ground state favored by the transition metal impurity . when the hole has dominantly host character , an exchange splitting is induced on the hole states which is opposite in direction to that on that the transition metal atom . the nature of the exchange coupling @xmath16 that exists between between the transition metal atom and the hole comes out automatically from such a microscopic model . the perturbation of the host valence band is not directly related to the coupling strength @xmath16 . when the hole has primarily dbh character , one finds the perturbation of the host valence band is larger . the basic picture that emerges from our first - principles calculations could be used to replace the more naive model - hamiltonian treatments which have assumed a host - like - hole picture , an unperturbed valence band and that the spin of the hole couples to the spin of the tm via a local exchange interaction .

galaxy mergers are likely to play a role in triggering active galactic nuclei ( agn ) , but the conditions under which this process occurs are poorly understood . in paper i , we constructed a sample of spatially offset x - ray agn that represent galaxy mergers hosting a single agn . in this paper , we use our offset agn sample to constrain the parameters that affect agn observability in galaxy mergers . we also construct dual agn samples with similar selection properties for comparison . we find that the offset agn fraction shows no evidence for a dependence on agn luminosity , while the dual agn fractions show stronger evidence for a positive dependence , suggesting that the merger events forming dual agn are more efficient at instigating accretion onto supermassive black holes than those forming offset agn . we also find that the offset and dual agn fractions both have a negative dependence on nuclear separation and are similar in value at small physical scales . this dependence may become stronger when restricted to high agn luminosities , though a larger sample is needed for confirmation . these results indicate that the probability of agn triggering increases at later merger stages . this study is the first to systematically probe down to nuclear separations of kpc ( kpc ) and is consistent with predictions from simulations that agn observability peaks in this regime . we also find that the offset agn are not preferentially obscured compared to the parent agn sample , suggesting that our selection may be targeting galaxy mergers with relatively dust - free nuclear regions .

1703.06143

accretion onto supermassive black holes ( smbhs ) , and the corresponding release of gravitational potential energy , power active galactic nuclei ( agn ) . this process requires that a significant amount of matter in the interstellar medium of a galaxy experience a loss of angular momentum sufficient to ultimately be captured by the smbh s accretion disk . numerical simulations suggest that major mergers of galaxies are an effective mechanism for removing angular momentum @xcite . observational evidence for this scenario includes bright quasi - stellar objects ( qsos ) that often show evidence of interactions or mergers at both obscured phases , such as dust - reddened qsos @xcite and ultra - luminous infrared galaxies ( ulirgs ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and in traditional qsos @xcite . this scenario is also consistent with the hierarchical paradigm of galaxy evolution , in which massive stellar bulges are capable of fueling smbh growth through mergers of gas rich galaxies @xcite and suggests that smbh growth may be linked with merger events in order to maintain observed correlations between smbhs and their host galaxies . for example , the masses of smbhs appear to be correlated with the stellar velocity dispersion @xcite and luminosities @xcite of the central stellar bulges , implying that the buildup of smbh mass is correlated with the buildup of stellar bulge mass @xcite . the structures of disk galaxies , on the other hand , are thought to be shaped by the collapse of gas via energy dissipation and smooth accretion from cooling gas within the surrounding dark matter halo , with a previous phase of mergers shaping the central stellar bulge and stellar halo @xcite . while subsequent major mergers will not necessarily destroy the disk if the gas supply of the progenitors is very high and the resulting bulge component is small @xcite , simulations have been successful at reproducing the observed properties of more massive bulges ( corresponding to substantial smbh growth ) through major mergers of spiral galaxies @xcite , implying that agn hosted by disk galaxies may be triggered via alternative routes . such mechanisms include instabilities internal to the host galaxy @xcite and minor mergers @xcite . however , the relative roles of internal instabilities , minor mergers and major mergers for growing smbhs is unclear . for example , the majority of local agn reside in late - type galaxies unlikely to have experienced a major merger @xcite , though their luminosities do not extend to the more powerful regimes seen in high - redshift quasars . these observables are explained in the model of @xcite where secular smbh growth dominates in the local universe but remains important only for low - luminosity agn triggering toward high - redshifts as the rate of gas - rich mergers increases . however , finding direct and consistent evidence in support of this picture has been difficult . numerous studies find that high luminosity agn show no preference for existing in galaxies with signs of merger activity when they are compared to a control sample of inactive galaxies ( @xcite , though see @xcite for a potential luminosity dependence ) . on the other hand , several studies find that the fraction of agn in mergers , out of a parent agn sample , increases with increasing luminosity @xcite . the above qualitative disagreements may be due to the variable conditions under which agn triggering may happen within a galaxy merger . for example , the stage of the merging galaxy system may be a particularly important parameter for the triggered accretion rate . the tidal torques induced by the merger are predicted to funnel gas and dust toward the nuclear region of the merging system , and as the merger evolves the smbhs will lose angular momentum due to dynamical friction , thereby migrating toward the nuclear region as well . therefore , at later merger stages , and thus smaller smbh separations , the supply of gas for accretion is greater , such that the accretion rate is likely to be higher @xcite , albeit with considerable uncertainty as to the timescales of agn duty cycles . while several studies have examined the dependence on merger stage using nuclear separations as a proxy @xcite , spatial resolution limits have precluded systematic analyses from observing the small nuclear separations when the accretion rate is predicted by simulations to peak @xcite . furthermore , simulations predict that the probability of triggering one versus two agn within a merger may be different and depend on properties of the host galaxies , such as the mass of their nuclear stellar cores @xcite . additionally , nuclear obscuration in mergers may hinder the establishment of a connection between mergers and agn at optical wavelengths . indeed , @xcite found that the fraction of galaxies with disturbed morphologies increases with the level of nuclear obscuration , suggesting that this may be a key phase in the evolution of agn in galaxies but which is hidden from most observations . to understand the conditions under which agn triggering is correlated with galaxy mergers , uniform merger samples with well - understood selection biases are necessary . while many galaxy merger candidates have been selected spectroscopically from velocity offset agn emission lines @xcite , the majority of them have been shown to host agn - driven outflows rather than dual smbhs based on follow - up observations @xcite . from imaging , several samples have been selected based on morphology , either visually @xcite or based on asymmetry @xcite . however , selection by morphology is not necessarily capable of quantifying the merger stage accurately if two nuclei are not visible . while samples based on galaxy pairs can measure separations , they are necessarily limited to earlier merger stages @xcite . however , resolved separations can be significantly reduced by spatially constraining the relative locations of two individual smbhs in a merging system @xcite . in this study , we exploit this concept by using a newly constructed sample of x - ray agn that are spatially offset from the nucleus of the host galaxy or a nearby companion galaxy from @xcite , hereafter referred to as paper i. the spatially offset agn sample can be used to quantify the merger stage based on separation , and was constructed using an astrometric registration procedure that detects offsets down to @xmath0 kpc . additionally , the sample also allows us to compare the merger scenarios for single agn formation against those of dual agn formation . this paper is organized as follows : in section [ sec : sample ] we described our sample , in section [ sec : analysis ] we analyze the effects of bolometric luminosity , nuclear separation , group environment and obscuration on the samples , in section [ sec : discussion ] we discuss the conditions that affect agn triggering in galaxy mergers and section [ sec : conclusions ] contains our conclusions . we assume the cosmological parameters of @xmath2 km s@xmath3 mpc@xmath3 , @xmath4 and @xmath5 throughout . in this section , we discuss the samples used in our analysis : an offset agn sample and two separate dual agn samples . the offset agn represent galaxy mergers hosting only a single agn that is off - nuclear ( section [ sec : offset_agn ] ) , and the dual agn represent galaxy mergers hosting two agn ( section [ sec : dual_agn ] ) . for our sample of galaxy mergers hosting only a single agn , we use the spatially offset agn from paper i. full details of this sample can be found in paper i , though here we summarize the main properties . the agn ( type 2 ) were originally selected from galaxies in the sdss seventh data release ( dr7 ) that are located in the agn regime of the baldwin - phillips - terlivich ( bpt ) diagram @xcite . from overlapping archival _ chandra _ coverage , we identified hard x - ray sources that are required to be within the sdss fiber and satisfy the x - ray criteria of an agn : unabsorbed hard x - ray luminosity of @xmath6 @xmath7 erg s@xmath3 , or a hardness ratio of @xmath8 @xmath9 where @xmath8 @xmath10 and @xmath11 and @xmath12 are the number of hard and soft x - ray counts , respectively . the overlapping sdss and _ chandra _ images were astrometrically registered with our pipeline described in paper i , establishing the _ complete _ parent agn sample ( 48 ) . spatially offset x - ray agn were then selected as those with significant spatial offsets ( @xmath13 in significance but @xmath14 kpc in projected physical separation ) from the nucleus of the host galaxy or a nearby companion galaxy , establishing the _ complete _ offset agn sample ( 18 ) . since the spatially offset agn detections require _ chandra _ imaging , the selection is biased toward agn that have been targeted by _ this bias will affect our analysis if the agn were targeted as potentially being in merging systems since it will not represent a random sampling of agn ( see section 7.2 of paper i ) . this bias affects 8 offset agn from the _ complete _ sample . removing these 8 agn leaves a sample of 40 parent agn and 10 spatially offset agn based on an unbiased sample of _ chandra _ observations . these sources make up the final offset agn sample , and they are used in all of our subsequent analyses . below we describe several unique characteristics of our offset agn sample and how we address them in our analyses : _ possible presence of dual agn : _ since our selection of offset agn in paper i imposed several conservative x - ray thresholds , we can not rule out the possibility that any of them are dual agn systems with the second agn either below the x - ray source detection threshold or below the x - ray agn thresholds ( see section 7.1 of paper i ) . one particular example is the source sdssj110851.04 + 065901.5 ( sdssj1108 + 0659 ) which was selected as a spatially offset agn in paper i but was also previously discussed in @xcite who interpreted the system to be consistent with a dual agn . while sdssj1108 + 0659 is excluded from the sample because it was targeted by a _ chandra _ program as a dual agn candidate , secondary agn that are currently undetected may exist in some of our other offset agn systems . for uniformity , however , we adopt the definitions used in paper i and note that our conclusions can only be said to apply to x - ray bright agn . we also acknowledge the possibility of two x - ray detected agn in one of our sources , sdssj111458.02 + 403611.41 ( sdssj1114 + 4036 ; see section 7.1 of paper i ) , though we note that using the secondary x - ray source ( sdssj1114 + 4036sw ) instead of the primary x - ray source ( sdssj1114 + 4036ne ) has a negligible effect on our results . _ sdss fiber size : _ as described in paper i , due to the @xmath15 sdss fiber radius , spatial offsets between the x - ray agn and galaxy centers may be contained entirely within the fiber or not ( see figure 3 of paper i ) . since the x - ray agn is always constrained to be within the fiber radius , this distinction is made between cases in which the x - ray agn is offset from a galactic core that is inside the fiber ( ` in - fiber ` offset agn ) versus a galactic core that is outside the fiber ( ` out - fiber ` offset agn ) . we emphasize that the distinction between the ` in - fiber ` and ` out - fiber ` subsamples is not a selection effect but merely an artifact of the sdss fiber size . however , a side effect of this artifact is that spectroscopic coverage is available for the offset galactic cores of the ` in - fiber ` subsample ( smaller separation pairs with unresolved secondary nuclei in sdss imaging ) but not for the ` out - fiber ` subsample ( larger separation pairs with resolved secondary nuclei in sdss imaging ) . in principle , this difference should not affect our results concerning the agn properties assuming the x - ray agn detection is associated with the optical agn detection from the fiber spectrum . still , given that the fiber coverage of the overall systems is generally larger for the ` in - fiber ` subsample , in each analysis subsection we consider the effects of removing the ` out - fiber ` offset agn . _ nature of the x - ray sources : _ since x - ray sources significantly in excess of @xmath6 @xmath16 erg s@xmath3 are known to be associated with accreting smbhs , the criterion of @xmath6 @xmath17 erg s@xmath3 is likely to rule out a non - agn contribution @xcite . sources that only pass the @xmath8 @xmath18 criterion , however , may have smaller values of @xmath6 and therefore more ambiguous physical origins . the x - ray luminosity function of off - nuclear x - ray sources does not extend far above @xmath19 erg s@xmath3 @xcite , with only a few brighter sources known ( e.g. * ? ? ? * ; * ? ? ? these objects , known as hyper luminous x - ray sources ( hlxs ) , are often associated with intermediate mass black holes ( imbhs ) . such sources in our sample with @xmath6 @xmath20 erg s@xmath3 are likely associated with either a lower luminosity agn or otherwise accretion onto imbhs with high hardness ratios @xcite , an event that is likely the result of a minor galaxy merger . at lower luminosities ( @xmath6 @xmath21 erg s@xmath3 ) , however , we can not rule out the possibility of stellar - mass systems in star - forming regions ( ultra - luminous x - ray sources ) that are passing through phases of unusually hard x - ray spectra that can mimic the higher hardness ratios typically seen in agn @xcite . finally , we noted in paper i that the ionizing nature of agn optically classified as low - ionization nuclear emission regions ( liners ) is ambiguous and may not originate from accretion onto nuclear massive black holes @xcite . therefore , in each analysis subsection we consider the effects of removing the subsample that does not pass the threshold of @xmath6 @xmath22 erg s@xmath3 or is optically classified as a liner . our goal in this subsection is to create a sample of dual agn from the literature for comparison to our offset agn . since no samples currently exist that satisfy both the optical and x - ray selection criteria from paper i , we have constructed two dual agn samples , one of which is optically - selected ( section [ sec : sdss_dual_agn ] ) and the other of which is x - ray - selected ( section [ sec : bat_dual_agn ] ) . similar to the offset agn parent sample , the parent sample of the optically - selected dual agn is derived from the sdss dr7 spectroscopic agn @xcite . since we require _ chandra _ detections to spatially isolate the agn relative positions , we have cross - matched the sdss spectroscopic agn with unique detections from the _ chandra _ source catalogue ( csc ; * ? ? ? * ) within @xmath15 to create the final parent sample of the optically - selected agn . from this parent sample , we use the previously identified dual agn systems found in the two studies that have selected dual agn starting from the sdss spectroscopic agn sample and for which @xmath23 imaging reveals spatially distinct x - ray agn @xcite . the two samples are described below : there are two systems from @xcite that the authors classify as dual agn : sdssj1108 + 0659 , also in our _ complete _ offset agn sample ( section [ sec : offset_agn ] ) , and sdssj114642.47 + 511029.6 ( sdssj1146 + 5110 ) . the x - ray agn in each pair are separated by @xmath13 and @xmath14 kpc , thereby satisfying the spatially offset criteria from paper i. however , in both systems one of the x - ray sources ( sdssj1108 + 0659nw and sdssj1146 + 5110sw ) passes the x - ray agn selection criteria from paper i , while the other x - ray source ( sdssj1108 + 0659se and sdssj1146 + 5110ne ) does not pass either of the @xmath6 or @xmath8 criteria . however , @xcite find sdssj1108 + 0659se likely to be an agn based on a one - dimensional analysis of the psf profiles and x - ray luminosities ( soft and hard ) that are several times the expected contribution from star - formation , and they find sdssj1146 + 5110ne likely to be an agn because the soft x - ray luminosity is more than an order of magnitude larger than the expected contribution from star - formation . furthermore , slit spectroscopy of these two systems from @xcite suggest that two agn may be present in each system . while sdssj1108 + 0659 is also in our _ complete _ offset agn sample , it is rejected from the ( unbiased ) offset agn sample and therefore only appears in the dual agn sample for our analyses . there is one system from @xcite that is classified as a dual agn : sdssj112659.54 + 294442.8 ( sdssj1126 + 2944 ) . as with the dual agn from @xcite , both of the x - ray agn in sdssj1126 + 2944 satisfy the spatially offset criteria from paper i ( @xmath13 and @xmath14 kpc ) . the brightest of the two x - ray agn ( sdssj1126 + 2944nw ) passes the @xmath6 criterion and the weaker source ( sdssj1126 + 2944se ) is consistent with the @xmath6 threshold when accounting for the uncertainty . these x - ray agn detections are also consistent with the orientation and separation of agn photoionized double [ oiii]@xmath245007 components from slit spectroscopy presented in @xcite . for completeness , we searched within the sdss agn - csc cross - matched sample for additional dual agn with separations of @xmath13 and @xmath14 kpc , and with velocity separations of less than 600 km s@xmath3 ( ( e.g . as in * ? ? ? * ) ) , finding none . finally , we have omitted all sources from the comparison sample that are not within the range of redshifts ( @xmath25 @xmath26 @xmath27 ) , bolometric luminosities , @xmath28 ( @xmath29 @xmath28 @xmath30 erg s@xmath3 ) and projected physical separations , @xmath31 ( @xmath32 @xmath31 @xmath33 kpc ) of the parent sample of paper i. this leaves a sample of 69 optically - detected parent agn and 6 optically - detected agn in 3 dual agn systems . we refer to this sample as the sdss dual agn . _ two caveats about the sdss dual agn sample : _ first , the sdss dual agn were originally selected for follow - up imaging ( @xmath34 ) as dual agn candidates because of explicit double - peaks in the narrow agn emission lines of the sdss spectra and therefore do not represent an unbiased sample of agn . second , the two dual agn systems from @xcite would only be selected as offset agn by the criteria from paper i , whereas we have adopted their dual agn interpretations ; in this sense , we can not claim the same x - ray properties for the dual agn as our offset agn sample . this choice was made to increase the sample size from one to three systems . @xmath35{wcs_sep_off.eps } & \hspace*{0.1 in } \includegraphics[width=3.4in]{wcs_sep_err_off.eps } \vspace*{-0.in } \end{array } $ ] the parent sample of the x - ray - detected dual agn consists of burst alert telescope ( bat ) detections from the 58 month survey @xcite that are crossmatched with agn @xcite . the x - ray - selected dual agn systems we use are the subset of this sample that appear in the @xcite sample of dual agn ( based on the presence of galaxy companions also hosting an agn ) and with separations of @xmath14 kpc . as with the optically - selected dual agn , the spatial centroids of the agn host galaxies are separated by @xmath13 . since the @xcite sample contains the full sample of bat agn in dual agn systems , we do not search for additional dual agn within the parent sample . furthermore , all of the bat agn are within the redshift range of the parent sample . this yields a parent sample of 246 bat - detected agn and 16 bat - detected agn in 8 dual agn systems ( ngc 6240 , mrk 739 , mrk 463 , iras 05589 + 2828 , eso 509-ig 066 , iras 03219 + 4031 , ngc 3227 and ngc 835 ) . we refer to this sample as the bat dual agn . we note that the agn classifications in the bat sample are not uniform and come from a variety of evidence including both optical and x - ray detections . in this section , we use our offset agn sample and dual agn samples to constrain the conditions that affect agn triggering in galaxy mergers . specifically , we examine the agn merger fractions ( section [ sec : frac ] ) , group environments ( section [ sec : environ_frac ] ) and level of nuclear obscuration ( section [ sec : loiii_lx ] ) . [ sec : analysis ] in this section , we first derive corrections for the known selection biases in the offset agn sample ( section [ sec : bias ] ) . then we investigate the agn merger fractions as a function of agn bolometric luminosity , @xmath28 ( section [ sec : lbol_frac ] ) and projected nuclear physical separation , @xmath31 ( section [ sec : phys_sep_frac ] ) . in each case , we do so for both the offset agn fractions ( @xmath36 ) and the dual agn fractions ( @xmath37 and @xmath38 ) . to reduce the statistical uncertainty , fractions are only shown in each bin if the number of parent agn is at least 2 ( which corresponds to @xmath39 confidence in the poisson count statistics but also allows for an adequate number of bins and dynamic range for our analysis ) . the binomial distribution , defined by the size of the parent sample and the success rate of offset agn occurrences within the parent sample , is used to compute the lower and upper quantiles defining the @xmath40 confidence interval around all fraction values . uncertainties associated with the functional parameterizations of fractions are the @xmath40 quantiles surrounding the median value of each parameter distribution obtained by adding simulated random uncertainties ( also drawn from the binomial distribution ) and refitting until the uncertainties converge . the sample of offset agn was uniformly defined by requiring that the angular offset between the agn and galaxy core or secondary agn , @xmath41 , be three or more times its standard uncertainty , @xmath42 : @xmath41@xmath43@xmath42 ( see paper i for details ) . therefore , from the parent agn sample , selection of those with real spatial offsets depends directly on only the two parameters @xmath41 and @xmath42 . figure [ fig : hist_bias ] shows the results of our selection process on the distributions of @xmath41 and @xmath42 for both the ` in - fiber ` and ` out - fiber ` subsamples , where the offset agn have a mean @xmath41 value greater than that of the parent sample , and a mean @xmath42 value less than that of the parent sample . we have used a two - sample kolmogorov - smirnov ( ks ) test to determine the null hypothesis probability that the parent and offset agn sample values come from the same distribution ( @xmath44 ) . for @xmath41 , the small value of @xmath44 @xmath45 strongly suggests that the offset agn are biased toward large values of @xmath41 and that we have likely missed offset agn with small values of @xmath41 . therefore , the selection has effectively introduced a minimum @xmath41 limit in the offset agn sample . for @xmath42 , on the other hand , the value of @xmath44 @xmath46 suggests that the selection is relatively insensitive to @xmath42 . to account for these direct selection effects , we have utilized monte - carlo simulations similar to those described in paper i. in short , the simulations produce offset nuclei with random projected physical separations ( @xmath47@xmath48@xmath47 @xmath14 kpc ) , redshifts ( @xmath49 @xmath50 @xmath51 ) and relative positional uncertainties ( @xmath52 @xmath53 @xmath54 ) that are drawn from uniform distributions . projecting @xmath48 onto the sky based on random orientations and phases , and then scaling to @xmath50 yields a simulated angular offset , @xmath55 . combined with @xmath53 , we selected offset nuclei using the same procedure as in paper i. we then calculated the recovered fraction of simulated offset nuclei as a function of @xmath55 and @xmath53 . values of @xmath36 have been corrected for biases introduced by large values of @xmath41 and small values of @xmath42 based on the results of these simulations . specifically , in each bin of @xmath28 or @xmath31 , we identified a simulated offset nucleus with values of @xmath55 and @xmath53 that most closely match @xmath41 and @xmath42 , respectively . then , we divided the observed agn merger fraction by the average recovery fraction of the matched simulated nuclei . the ranges of correction factors are quoted in sections [ sec : lbol_frac ] and [ sec : phys_sep_frac ] . the simulated recovery fractions should not be taken as estimates of the absolute recovery fractions because the true distributions of the parent sample parameters are unlikely to be uniform as in our simulations . therefore , they only provide estimates of the relative fractions as a function of @xmath55 and @xmath53 . since the corrections are relative , we have normalized them to unity ( i.e. no correction ) at the largest values of @xmath55 and @xmath53 . we also caution that the corrections based on these simulations do not account for any potential indirect biases which are instead discussed individually in each subsection as footnotes . we do not correct the values of @xmath37 and @xmath38 since the dual agn positional measurements and uncertainties are not uniformly measured ( section [ sec : dual_agn ] ) . numerical simulations of galaxy mergers have predicted that the dependence on mergers for agn triggering is positively correlated with the agn bolometric luminosity @xcite , and observational evidence of morphological disturbances in the host galaxies of high luminosity agn supports these predictions @xcite . however , other studies that have examined the morphological traits of agn and non - agn host galaxies find no statistical differences between samples at low and high agn luminosities @xcite . these null - results imply that agn triggering is relatively independent of galaxy mergers such that internal instabilities play a comparable if not stronger role in smbh growth . while these studies used different procedures and tests , their results are qualitatively in disagreement about the role that galaxy mergers play in triggering agn as a function of agn luminosity . therefore , we use our systematically selected sample of galaxy mergers to address the role that agn bolometric luminosity has on the agn merger fraction . @xmath56{lbol_oiii_off.eps } \\ \hspace*{-0.2 in } \includegraphics[width=0.5\textwidth]{lbol_dual_sdss_bat.eps } \vspace*{-0.1 in } \end{array } $ ] our estimates of @xmath28 for the parent sample of the offset agn and the parent sample of the sdss dual agn are calculated from the extinction - corrected [ ] @xmath57 luminosity ( @xmath58,corr.}}$ ] ; from @xcite based on balmer decrements ) since the original agn identifications of both samples were based on optical emission lines and are not affected by significant obscuration due to dust ( see section [ sec : loiii_lx ] ) . we use a bolometric correction of @xmath28 @xmath59@xmath58,corr.}}$ ] @xcite . since the companions of the bat dual agn were identified from non - uniform selection criteria , ultra - hard ( @xmath60 kev ) x - ray luminosities ( @xmath61 ) are available for only one agn in each system . therefore , we have chosen to derive @xmath28 from @xmath61 using the bolometric correction of @xmath28 @xmath62@xmath61 @xcite . while this choice means we can only use eight of the 16 agn in the bat systems , the use of ultra - hard x - rays mitigates the effect of nuclear obscuration @xcite . figure [ fig : frac_lbol_oiii ] ( top ) shows @xmath36 , calculated as the number of offset agn ( @xmath63 ) out of the number of parent agn ( @xmath64 ) , as a function of @xmath28 . figure [ fig : frac_lbol_oiii ] ( bottom ) shows @xmath37 and @xmath38 as a function of @xmath28 , where each is computed in the same manner as @xmath36 . to optimize the combination of signal and binning resolution , we have chosen bin sizes individually for each sample . due to the multiple orders of magnitude spanned by @xmath28 , the bins have been given logarithmically uniform spacing with sizes of 0.5 dex for the offset agn sample and 1 dex for both dual agn samples . after implementing the threshold of @xmath652 parent agn in each bin , the resulting @xmath28 bin ranges are @xmath66 erg s@xmath3 for @xmath36 , @xmath67 erg s@xmath3 for @xmath37 and @xmath68 erg s@xmath3 for @xmath38 . therefore , to examine both the offset and dual agn samples over the same @xmath28 values , we have chosen the plotting range of @xmath28 @xmath69 erg s@xmath3 . from @xmath28 @xmath70 erg s@xmath3 , @xmath36 is adequately fit by a power - law function ( @xmath71 ) with parameter values of @xmath72 @xmath73 and @xmath74 @xmath75 . this fit corresponds to a slope with a @xmath76 significance from zero . for comparison , the correction factors ( section [ sec : bias ] ) range from 1.14 to 1.21 , and the best - fit power - law parameters for the uncorrected fractions are @xmath72 @xmath77 and @xmath74 @xmath78 , corresponding to a slope with a @xmath79 significance from zero . to test the implications of using the ` out - fiber ` subsample ( section [ sec : offset_agn ] ) , we remove the ` out - fiber ` sources and find that doing so does not change the qualitative result that no significant change in @xmath36 is seen as a function of @xmath28 . we also find that removing sources with @xmath6 @xmath21 erg s@xmath3 or that are liners also has no effect on this qualitative result . from @xmath28 @xmath80 erg s@xmath3 , @xmath37 is adequately fit by a power - law function with parameter values of @xmath72 @xmath81 and @xmath74 @xmath82 . compared to @xmath36 , this fit corresponds to a steeper slope with a @xmath83 significance from zero . from @xmath28 @xmath84 erg s@xmath3 , @xmath38 is also adequately fit by a power - law function with parameter values of @xmath72 @xmath85 and @xmath74 @xmath86 . compared to @xmath37 , this fit corresponds to a shallower slope , though it still has a @xmath87 significance from zero . while the current data offers a null result for @xmath36 , a similar increase from low to high @xmath28 can not be ruled out due to the significant fit uncertainties . for comparison , in both panels of figure [ fig : frac_lbol_oiii ] we show the best - fit linear function to the agn merger fraction from @xcite . the parent sample used in @xcite consists of agn identified from x - ray , infrared and spectroscopic surveys , and the galaxy merger systems were taken from samples identified by visual classification . as a result , the number of agn in each system ( offset or dual agn ) is usually not possible to determine . in the top panel , we see that the function from @xcite is consistent with @xmath36 below @xmath28 @xmath88 erg s@xmath3 while over - predicting @xmath36 at higher @xmath28 by @xmath89 . this result provides tentative evidence that @xmath36 behaves differently from the @xcite function at high @xmath28 values , though the substantial uncertainties prohibit a firm conclusion . the @xmath36 @xmath28 values plotted also do not extend down to the luminosities of @xmath90 erg s@xmath3 , and thus the behavior of @xmath36 compared to the lower @xmath28 end of the @xcite sample is not known . the stronger positive correlations seen in the dual agn ( relative to the offset agn ) are in better qualitative agreement with the function from @xcite , though the bat - sample is generally over - predicted while the sdss sample is in agreement to within the @xmath91 uncertainties over nearly the full @xmath28 range plotted . numerical simulations have predicted that , in an evolving galaxy merger , the probability of observing an agn increases with decreasing separation of the two smbhs from the progenitor galaxies @xcite . observational evidence of rising agn merger fractions with decreasing nuclear separation supports these predictions @xcite . however , the galaxy mergers from those samples were identified from galaxy pairs , thereby limiting the nuclear separations to larger values ( several kpc or greater ) so that individual galaxies can be distinguished . simulations , on the other hand , predict that the agn observability continues to increase significantly below 1 kpc , and previous observational studies have not been able to probe the small separation regime where the dependence of the agn merger fraction on nuclear separation is predicted to peak . therefore , we use our sample of offset agn with resolved x - ray agn offsets from @xmath92 to @xmath1 kpc to examine the agn merger fraction from early to late merger stages . we have adopted the physical separations presented in @xcite and @xcite for the sdss dual agn and those presented in @xcite for the bat dual agn . @xmath93{physep_off.eps } \\ \hspace*{-0.2 in } \includegraphics[width=0.5\textwidth]{physep_dual_sdss_bat.eps } \vspace*{-0.1 in } \end{array } $ ] figure [ fig : frac_phys_sep ] ( top ) shows @xmath36 as a function of @xmath31 . since values of @xmath31 are not available for the parent agn sample , we have computed the minimum projected physical separation that could potentially be resolvable ( by @xmath94 ) based on the @xmath42 and @xmath26 values for each parent agn : @xmath95 . the values of @xmath36 in each @xmath31 bin are then the number of offset agn with @xmath31 within that bin ( @xmath63 ) out of the number of parent agn with @xmath95 less than or equal to the mean value of that bin ( @xmath96}}$ ] ) . note that a parent agn may be in multiple bins using this approach . figure [ fig : frac_phys_sep ] ( bottom ) shows @xmath37 and @xmath38 as a function of @xmath31 , where each is computed in the same manner as @xmath36 . as in section [ sec : lbol_frac ] , we have chosen bin sizes individually for each sample to optimize the combination of signal and binning resolution . the bins have been given linearly uniform spacing with sizes of 1.5 kpc for the offset agn sample , 3 kpc for the sdss dual agn sample , and 6 kpc for the bat dual agn sample . after implementing the threshold of @xmath652 parent agn in each bin , the resulting @xmath31 bin ranges are @xmath97 kpc for @xmath36 and @xmath98 kpc for both @xmath37 and @xmath38 . therefore , to examine both the offset and dual agn samples over the same @xmath31 values , we have chosen the plotting range of @xmath31 @xmath99 kpc . from @xmath31 @xmath99 kpc , @xmath36 is adequately fit by a power - law function with parameter values of @xmath72 @xmath100 and @xmath74 @xmath101 . this fit corresponds to a slope with @xmath102 significance from zero . the correction factors ( section [ sec : bias ] ) range from 0 for fractions in bins of large @xmath31 to 1.28 in the bin of smallest @xmath31 . for comparison , the best - fit power - law parameters for the uncorrected fractions are @xmath72 @xmath103 and @xmath74 @xmath104 , corresponding to a slope with @xmath105 significance from zero . as in section [ sec : lbol_frac ] , we test the implications of including the ` out - fiber ` subsample by removing those sources , finding that doing so has no qualitative effect on the significant rise in @xmath36 at small physical separations since the ` out - fiber ` offset agn have physical separations larger than the small - separation regime in which the rapid change in @xmath36 is seen . this test also confirms that the correlation is not due to the artifact introduced by the fiber size . likewise , removal of the sources with @xmath6 @xmath21 erg s@xmath3 or those that are liners has no qualitative effect on the significant rise in @xmath36 at small physical separations . from @xmath31 @xmath99 kpc , @xmath37 is adequately fit by a power - law function with parameter values of @xmath72 @xmath106 and @xmath74 @xmath107 . compared to @xmath36 , this fit corresponds to a shallower slope , but it is still has a @xmath108 significance from zero . from @xmath31 @xmath109 kpc , @xmath38 is also adequately fit by a power - law function with parameter values of @xmath72 @xmath110 and @xmath74 @xmath111 . compared to @xmath37 , this fit is steeper ( though still shallower than @xmath36 ) and has a @xmath112 significance from zero . we note that @xmath36 dramatically increases at small separations when compared to @xmath37 and @xmath38 ( as indicated by their respective magnitudes of @xmath113 ) . this difference may indicate that the value of @xmath113 in the @xmath36 trend is strongly driven by the data point at @xmath31@xmath114 kpc which is lacking in the sdss and bat dual agn samples . for comparison , in both panels of figure [ fig : frac_phys_sep ] we show a line representing the trend seen in the agn merger fraction as a function of projected pair separation from @xcite based on the _ wise _ color cut ( @xmath115 ) that has been empirically shown to select infrared bright agn with high reliability . in particular , the line shows the slope made by the change in the agn merger fraction in their pre - coalescence sample from the largest separation data point to the smallest separation data point plotted in their figure 2 . the @xcite sample utilizes galaxy mergers selected as galaxy pairs ( not distinguishing between the offset and dual agn scenarios ) that extends from a minimum separation of @xmath116 kpc ( based on the redshift range and the sdss fiber size ) out to @xmath31 @xmath117 kpc @xcite . above @xmath31@xmath118 kpc , @xmath36 , @xmath37 and @xmath38 are all consistent with the @xcite sample , while all three merger fractions would be under - predicted by a linear extrapolation of this slope at @xmath31@xmath119 kpc . observationally , galaxies that exist in dense environments , such as groups or clusters , generally have redder colors ( due to suppressed star - formation ) and more elliptical morphologies compared to galaxies in less dense environments @xcite . one route for dense environments to drive galaxies toward redder colors and elliptical morphologies is through interactions or mergers between galaxies . while the probability of a direct merger is low among satellite galaxies in clusters due to their high relative velocities , galaxy groups are the environments most likely to contain merging systems @xcite . therefore , we may expect to find a higher fraction of offset and dual agn in group environments . we use our sample of offset and dual agn to test this prediction by quantifying the density of their environments . we do so by first taking a catalogue of galaxy group members described in @xcite . the catalogue was constructed by applying host dark matter halo and satellite dark matter halo mass functions @xcite to galaxies from the sdss dr7 . based on the halo mass functions of each galaxy , the central most massive galaxy of a group and its satellites can be identified . thus , every galaxy in the catalogue has an assigned number of group members , @xmath120 , with masses above @xmath121 m@xmath122 ( below this mass limit the method is not sensitive ) which we take as a parameterization of the group density . we then matched the offset agn parent sample and the dual agn parent samples with the catalogue to assign them @xmath120 values . since the redshift range of the galaxy group catalogue is limited to @xmath123 , only a subset of the agn in our samples will have @xmath120 values . figure [ fig : hist_ngroup ] compares the @xmath120 distributions for the parent and offset agn . while the mean value of the offset agn ( 14.0 ) is larger than that of the parent agn ( 5.6 ) , the difference is not at a significant level based on a two - sample ks test ( @xmath44 @xmath124 ) . as in sections [ sec : lbol_frac ] and [ sec : phys_sep_frac ] , we test the implications of including the ` out - fiber ` subsample by removing those sources , finding that doing so does not change the qualitative result that no significant difference is seen between the two samples . likewise , removal of the sources with @xmath6 @xmath21 erg s@xmath3 or those that are liners has no qualitative effect on this result . too few of the sdss and bat agn are assigned @xmath120 values to statistically examine their environments . while the number statistics are small , this result suggests that offset agn tend to be in environments similar to the general type 2 agn population , or at least that the difference is small enough to be undetectable in our sample . this suggestion implies that the types of mergers leading to the spatially offset nature of an x - ray agn within the parent sample are not linked with over - dense environments compared to the parent sample . in fact , the environments of both the parent and offset agn have relatively small densities of group members compared to cluster environments ( @xmath120 @xmath125 ) . however , these densities only correspond to group members with masses above @xmath121 m@xmath122 , suggesting that mergers with lower - mass galaxies may play a role in producing the offset agn systems . the implications of this merger mass ratio effect are discussed in section [ sec : disc_single_dual ] . the nuclear regions of merging galaxy systems can potentially be heavily obscured as gas and dust will naturally settle toward the regions of largest gravitational potential . since the [ ] @xmath57 emission line originates far enough from the smbh to not be subject to nuclear obscuration , comparison with the x - ray luminosity , which does originate near the smbh accretion disk , can potentially reveal the presence of nuclear enshrouding material . therefore , we have tested for this effect in figure [ fig : plot_lx_loiii ] by plotting @xmath6 against @xmath58,corr.}}$ ] for the parent agn sample and offset agn sample . we have also shown in figure [ fig : plot_lx_loiii ] the mean value of a type 2 agn sample selected independently of mergers @xcite . the effect of our @xmath6 and hardness ratio ( @xmath8 ) selection criteria can be seen in figure [ fig : plot_lx_loiii ] . the @xmath6-selected subsample has @xmath6/@xmath58,corr.}}$ ] ratios similar to the merger - independent sample , whereas the @xmath8-selected subsample has systematically lower values of @xmath6 for a given value of @xmath58,corr.}}$ ] . since the @xmath6-selected subsample has an average column density that is larger by @xmath126 dex than that of the @xmath8-selected subsample , this difference between the two subsamples is consistent with underestimates of the column densities in the latter ( assuming that the distributions of intrinsic x - ray spectral slopes and intrinsic absorbing columns are the same ) . indeed , the relatively simple absorbed power - law fit to the chandra spectra ( paper i ) may be under - estimating the column densities in these sources due to low counts . @xmath127{loiii_lx_int_off_plot.eps } \end{array } $ ] while the mean @xmath6/@xmath58,corr.}}$ ] ratios of the offset agn are lower than those of the parent agn for the @xmath6-selected subsample ( @xmath128 versus @xmath129 ) and the @xmath8-selected subsample ( @xmath130 versus @xmath131 ) , they are in agreement to within the scatter . furthermore , the two - sample ks test results of @xmath44 @xmath132 and @xmath44 @xmath133 for the @xmath6-selected and @xmath8-selected subsamples , respectively , do not suggest a significant difference . therefore , for agn with a given value of @xmath6 and @xmath58,corr.}}$ ] , selection of x - ray sources that are spatially offset from the host galaxy core or a nearby galaxy core does not appear to introduce a bias toward obscuration . instead , this result indicates that our selection of offset agn , from the parent agn sample , coincides with a tendency to select relatively unobscured systems . the potential physical implications of this result are discussed in section [ sec : obscuration ] . several studies have placed estimates on the fraction of agn hosted by galaxies in mergers or merger - remnants @xcite . however , these studies have not estimated the fractions for the specific scenarios of offset agn or dual agn because the methods by which those studies selected galaxy mergers and cross - matched with agn does not uniformly allow for a distinction between systems in which one or both galaxies hosts an agn . the distinction between the two scenarios is a crucial step toward understanding the physics that govern accretion onto smbhs within galaxy mergers . since our selection method requires the agn to be spatially isolated within the merger , we can measure the number of agn and constrain the conditions of offset versus dual agn formation . in this section , we discuss the offset and dual agn scenarios in the context of agn luminosity ( section [ sec : disc_single_dual ] ) , merger stage ( section [ sec : disc_sep ] ) , the combined effect of agn luminosity and merger stage ( section [ sec : sep_lum ] ) , and finally nuclear obscuration ( section [ sec : obscuration ] ) . as mentioned in section [ sec : lbol_frac ] , some observational evidence suggests that high - luminosity agn are preferentially found in galaxies that are interacting , merging , or have merged in the past , while other studies find no evidence for agn hosts to have different morphologies from inactive galaxies . in section [ sec : lbol_frac ] and figure [ fig : frac_lbol_oiii ] , we used our agn merger samples ( selected independent of galaxy or merger morphology ) to address this discrepancy , producing the following results : based on their parameterizations ( section [ sec : lbol_frac ] ) , the offset and dual agn fractions behave differently as a function of @xmath28 . the offset agn fraction displays no significant evolution over the agn bolometric luminosity range probed ( @xmath76 significance ) as shown in figure [ fig : frac_lbol_oiii ] ( top ) . comparatively , the dual agn fractions show stronger evidence for a statistically significant increase at high agn bolometric luminosities ( @xmath87@xmath134@xmath83 significance ) as shown in figure [ fig : frac_lbol_oiii ] ( bottom ) . these results can be broadly interpreted as in agreement with claims of a positive correlation between the agn merger fraction and bolometric luminosity . specifically , observations of large - scale ( 10 - 100 kpc ) galaxy pairs have shown that the agn merger fraction is highest for major mergers @xcite , implying that major mergers are more efficient at removing angular momentum from gas and dust in their host galaxies . additionally , @xcite find that their bat - selected dual agn are preferentially found in major mergers . therefore , dual agn host systems may show a preference for major mergers so that enough fuel is available to power both agn . for example , of the three sdss dual agn systems , two are hosted by major mergers @xcite . since the sample from @xcite extends to high redshifts and was selected based on morphology , that sample may contain a relatively higher fraction of major mergers that allow for easier visual classification , possibly accounting for the agreement between that sample and the comparison sdss dual agn sample . by contrast , offset agn may show a preference to reside in minor mergers instead of major mergers . for example , in section [ sec : environ_frac ] we examined the distribution of the number of group members for both the parent and offset agn , finding no significant evidence for a difference between the two samples ( figure [ fig : hist_ngroup ] ) and that both samples reside in relatively low - density environments . however , since the procedure used for measuring group members is not sensitive to galaxies with masses @xmath135 m@xmath136 @xcite , we are likely missing many lower - mass group members . therefore , compared to galaxy mergers seen in dense environments of more massive galaxies , the offset agn are more likely to be undergoing mergers with galaxies of lower masses , corresponding to mass ratios that fall in the minor merger regime . however , we caution that the offset agn slope is consistent within @xmath91 with those of both dual agn samples ( figure [ fig : frac_lbol_oiii ] ) so that the true difference between offset and dual agn evolution with agn bolometric luminosity is poorly constrained . we also note that the effect of merger mass ratio on agn triggering is still poorly understood at small separations , particularly for offset agn , and may also depend on whether or not loss of angular momentum happens more efficiently in the major or minor galactic stellar core . currently , theoretical work has provided ambiguous results , with one model suggesting that the more luminous agn likely resides in the more massive stellar bulge @xcite , while a recent simulation of galaxy mergers has suggested that the accretion rate is higher for the agn in the less massive galaxy @xcite . in paper iii , follow - up imaging of our offset agn sample with @xmath137 will put constraints on these predictions by allowing estimates of the merger mass ratios and smbh accretion rates . as mentioned in section [ sec : phys_sep_frac ] , numerical work predicts that the probability of agn triggering becomes strongest at separations below @xmath138 kpc as the two smbhs dynamically evolve toward the region of greatest gravitational potential along with a significant amount of gas and dust for accretion . in section [ sec : phys_sep_frac ] and figure [ fig : frac_phys_sep ] , we used our agn merger samples ( with measured physical separations reaching below @xmath138 kpc ) to test this prediction , producing the following results : evidence for a negative correlation with nuclear separations below @xmath92 kpc is seen at @xmath39 significance for both the offset agn fractions ( @xmath102 ) and the dual agn fractions ( @xmath112@xmath134@xmath108 ) as shown in figure [ fig : frac_phys_sep ] . while the physical separations of the dual agn sample do not allow us to see this trend continue below @xmath139 kpc , we are able to do so for the offset agn sample due to our astrometric registration procedure . indeed , the slope magnitude for the offset agn fractions is larger ( by @xmath140 ) than those of the dual agn fractions , an indication that the merger fraction rises most strongly at the smallest separations . a similar , or larger , increase may be seen in the dual agn sample with a larger sample size or with data points at separations @xmath114 kpc . our finding that the agn merger fraction rises fastest and peaks ( down to our resolution limits ) below 1 kpc is consistent with numerical predictions . at large separations , the offset agn fractions and the dual agn fractions are consistent with the slope of the agn merger sample presented in @xcite . in fact , both the offset agn and sdss dual agn fractions are consistent with this slope over the full range shown ( @xmath141 kpc ) . at large separations , our results are also consistent with other studies @xcite . when examining the absolute values of the agn merger fractions , we see that the offset agn and sdss dual agn fractions ( below @xmath142 kpc ) of @xmath143 and @xmath144 , respectively , are in agreement when accounting for their @xmath91 uncertainties . that the offset agn and sdss dual agn fractions are consistent at small separations may be indicating that , among similarly selected samples , dual agn triggering becomes more likely at smaller separations and comparable to single agn triggering . this result is consistent with the expectation that there is an increased supply of gas for accretion at smaller nuclear separations . the generally larger values of the bat dual fraction may indicate a stronger overall dependence of agn triggering on nuclear separation in that sample , though this conclusion is tenuous due to the substantial uncertainties ( @xmath145 below @xmath142 kpc ) and @xmath91 overlap with the offset agn and sdss dual agn . since the frequency of agn observability in mergers peaks at small separations due to the increased availability of fuel for accretion , we may also expect that this regime coincides with enhanced accretion rates among agn . to investigate this prediction , we have examined the dependence of the offset agn fractions as a function of nuclear separation for two subsamples that are separated by bolometric luminosity : @xmath28 @xmath146 erg s@xmath3 and @xmath28 @xmath147 erg s@xmath3 ( figure [ fig : frac_phys_sep_lbollohi ] ) . due to the smaller numbers in each subsample compared to the full offset agn sample , we have used bin sizes of @xmath148 kpc . as shown in figure [ fig : frac_phys_sep_lbollohi ] , the fractions in the low- and high - luminosity bins are consistent within their uncertainties throughout the entire range of separations investigated . therefore , we see no statistically significant evidence that the occurrence of offset agn is dependent on luminosity at any given nuclear separation . however , we note that the greatest divergence between the two fractions ( by @xmath89 ) occurs at the smallest separation . while this result is tenuous due to the large uncertainties , it may hint that while an overall dependence of the offset agn fraction with bolometric luminosity is not seen ( section [ sec : lbol_frac ] ) , it does appear at small nuclear separations when the supply of gas for accretion is larger . this result is also qualitatively consistent with the results from @xcite in which the luminosities of dual agn increase at smaller separations . we caution that the observability of agn in these systems depends not only on luminosity but also on the timescale of activity which we can not measure . still , these combined results are overall consistent with the numerical predictions that the bolometric luminosities of agn in merging systems peak at merger stages corresponding to small separations @xcite . heavy obscuration of the nuclear regions of galaxies undergoing mergers is predicted by models of galaxy and quasar co - evolution in which mergers trigger enhanced levels of accretion onto agn but also pass through a stage of enhanced obscuration @xcite . therefore , the continuum emission of agn that are hosted by on - going galaxy mergers may preferentially exhibit signs of obscuration compared to agn in non - merging systems . extreme examples of agn obscuration are found in ulirgs that are rich in gas and dust @xcite , with most emission from the accretion disk obscured except for high energy photons such as hard x - rays . heavily obscured ( compton - thick ) agn , in which x - ray emission is severely obscured , may constitute a significant population of agn in galaxy mergers @xcite and therefore understanding the role that x - ray absorption plays in the link between galaxy mergers and agn is crucial . the effect of obscuration in optically - selected and x - ray detected dual or offset agn was noticed independently by @xcite and @xcite . their samples of dual and offset type 2 agn systems show systematically lower observed hard x - ray to [ ] @xmath57 luminosity ratios compared to the optically - selected type 2 agn sample from @xcite . since the [ ] @xmath57 emission line originates far enough from the smbh to not be subject to nuclear obscuration , this result suggests that dual and offset agn are suffering heavier nuclear obscuration than the general population of type 2 agn . in fact , @xcite have shown that even the absorption - corrected hard x - ray luminosities of their sample are still under - luminous compared to a general sample of type 2 agn that was cross - matched with the @xmath23 source catalogue @xcite . this has led to the suggestion that the low counts and/or intrinsically high absorbing columns result in systematically underestimated column densities . the analysis in section [ sec : loiii_lx ] showed that , while the offset agn sample has a lower mean hard x - ray to [ ] @xmath57 luminosity ratio than the parent agn , the difference is not at a significant level . therefore , no evidence of preferential nuclear obscuration is seen . this result is opposite that seen in @xcite and @xcite , and may be a result of the spatially offset selection introducing a bias toward face - on systems ( as numerically predicted in paper i ) with shallower absorbing columns . however , it may also be a result of the x - ray selection properties that target x - ray bright agn with intrinsically little nuclear obscuration . that the selection may preferentially target galaxy mergers with relatively dust - free nuclear regions means the sample is fundamentally different from the prototypical mergers seen in gas - rich systems such as ulirgs and may be probing a specific subclass of galaxy mergers . we have used our systematically constructed sample of spatially offset agn from paper i to constrain the parameters under which agn triggering is driven by galaxy mergers . due to the selection of galaxy mergers based on offset x - ray agn , our sample is not biased toward morphological disturbances or large projected physical separations , allowing us to investigate the agn merger fraction in major or minor mergers and at early or late merger stages . we have investigated the fractions of offset agn , and those of similarly constructed dual agn samples , out of their respective parent samples , as functions of agn bolometric luminosity and projected nuclear separation . additionally , we have examined their group environments and compared their x - ray to optical luminosity ratios to those of independent agn samples . our conclusions are as follows : 1 . the fraction of spatially offset agn shows no evidence for a dependence on agn bolometric luminosity , while the fractions of dual agn do show a positive dependence , increasing from @xmath149 at @xmath150 erg s@xmath3 to between @xmath151 and @xmath152 at @xmath153 erg s@xmath3 ( figure [ fig : frac_lbol_oiii ] ) . these results suggest that agn triggering is indeed linked to mergers but that this dependence may only become strong in the specific scenarios of high bolometric luminosities , dual agn activation and possibly major mergers . the offset agn group environments show no evidence for a difference from the parent agn sample ( figure [ fig : hist_ngroup ] ) , and both reside in environments with a low - density of massive galaxies . the lack of numerous massive companions may point toward a preference for minor mergers in the offset agn systems . the fractions of spatially offset agn and dual agn show evidence for a negative dependence on projected physical nuclear separation , increasing from @xmath149 at 19 kpc to between @xmath154 and @xmath155 at @xmath119 kpc ( figure [ fig : frac_phys_sep ] ) . the offset and dual agn fractions are similar at small separations , suggesting that the efficiency of dual agn triggering becomes similar to single agn triggering at late merger stages when significant material is available for accretion . we can only trace the dual agn sample down to @xmath156 kpc , while the resolution of our offset agn sample allows it to be traced down to @xmath1 kpc where we see the most significant increase . our sample of offset agn has allowed this analysis to be extended down to @xmath0 kpc for the first time . we see tentative evidence that the inverse dependence of the agn merger fractions on separation become strongest when restricted to a high agn luminosity subsample ( figure [ fig : frac_phys_sep_lbollohi ] ) . if real , this result would be consistent with numerical predictions that agn triggering probabilities increase with decreasing nuclear separations , and that this late merger stage also corresponds with the stage of highest agn luminosity . the hard x - ray to [ ] @xmath57 luminosity ratios of offset agn show no significant evidence for a difference from that of the parent agn ( figure [ fig : plot_lx_loiii ] ) , in contrast to the expectation from studies of many known merging galaxy systems . while this similarity may reflect a tendency to select face - on systems , it may also point toward a selection of intrinsically unobscured systems that are fundamentally different from merging systems with coincident nuclear obscuration and on - going star - formation . in paper iii of this series , we will present new and archival @xmath137 imaging for a subset of our offset agn sample to put constraints on the correlated evolution of smbhs and their host galaxies . in particular , we will determine the effect of merger mass ratio on smbh growth , and we will put constraints on the correlated triggering of star - formation and agn . + the authors would like to thank an anonymous referee for detailed and insightful comments that greatly improved the quality of the paper . support for this work was provided by nasa through chandra award number ar5 - 16010a issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of nasa under contract nas8 - 03060 . the scientific results reported in this article are based in part on observations made by the _ chandra x - ray observatory _ , and this research has made use of data obtained from the _ chandra data archive _ and the _ chandra source catalog _ , and software provided by the _ chandra x - ray center ( cxc ) _ in the application packages _ ciao _ , _ chips _ , and _ sherpa_.

we investigate gravitational collapses of charged black holes in string - inspired gravity models , including dilaton gravity and braneworld model , as well as gravity and the ghost limit . if we turn on gauge coupling , the causal structures and the responses of the brans - dicke field depend on the coupling between the charged matter and the brans - dicke field . for type iia inspired models , a cauchy horizon exists , while there is no cauchy horizon for type i or heterotic inspired models . for type iia inspired models , the no - hair theorem is satisfied asymptotically , while it is biased to the weak coupling limit for type i or heterotic inspired models . apart from string theory , we find that in the ghost limit , a gravitational collapse can induce inflation by itself and create one - way traversable wormholes without the need of other special initial conditions . yitp-14 - 47

1406.0976

to investigate the theory of everything which explains gravity and quantum theory in a consistent way , several candidates have been proposed @xcite with string theory being one of the most competitive ideas @xcite . string theory can explain not only the finite quantum interactions of gravitons but also the possible modifications of gravity and matter sectors @xcite . such modified gravity models or additional matter contents are useful to study various topics , for example inflationary and late time cosmology @xcite , black hole physics and applications to ads / cft @xcite or astrophysical applications @xcite etc . moreover , if string theory is the correct model of quantum gravity then such modifications should explain various traditional problems of gravity , e.g. , the singularity problem of black holes @xcite or cosmology @xcite . a typical modification via string theory is the existence of a dilaton field @xcite . this is in fact the same as the brans - dicke field @xcite which controls the interaction coupling strengths . the kinetic term and the potential term of the brans - dicke field can be changed by a conformal transformation from the string frame ( or the jordan frame ) to the einstein frame and is equivalent to a canonical scalar field model in the einstein frame @xcite . however , if the model couples to other interactions , then the matter sector is no longer canonical after the conformal transformation , an interesting example of which is the gauge field . in section [ sec : gen ] , we summarize various types of string effective actions and the different types of interactions they generate between the dilaton field and the gauge field . in this paper , we want to investigate responses of the brans - dicke field that depend on couplings between the brans - dicke field and the gauge field . naively speaking , they will form charged black holes and there are already a lot of papers which investigate stationary solutions @xcite . however , in general the dynamics will be complicated for realistic gravitational collapses . for such cases we expect complicated phenomena such as : * _ causal structure : _ in the general string frame ( jordan frame ) , the null energy condition can be violated and hence the dynamics of locally defined horizons @xcite can be complicated . in addition , the existence of the central singularity and the cauchy horizon curvature singularity @xcite should be affected by the dynamics of the brans - dicke field . * _ response of the brans - dicke field : _ already some solutions are known which violate the assumptions of the no - hair theorem @xcite . however , is it be possible to form scalar hairs from gravitational collapses ? for neutral black holes it was investigated by @xcite , but the responses due to the charge can cause quite different phenomena . this may be related to the destabilization of the moduli / dilaton field that should be controlled by a proper potential and may have observational implications @xcite . we will mainly focus these two topics by using numerical techniques applied to the double - null formalism . to the best knowledge of the authors , the double - null formalism is one of the most efficient techniques to investigate the interior of the black holes @xcite and we have priviously used this approach with good success @xcite . there are some previous works regarding string - inspired models using the double - null formalism @xcite and other methods @xcite as well as the authors previous works @xcite . in this paper we will investigate a wide parameter space , for example , we will cover not only directly string - inspired models , but also indirectly string - inspired models ( @xmath0 inspired model @xcite ) as well as natural generalizations of the parameters , e.g. , the ghost limit . in fact , our potential parameter space is huge and can give rise to a wide range of setups , for example : * _ model dependence _ : type i , type ii , heterotic , or braneworld inspired models , etc . * _ potential dependence _ : whether there is a potential that stabilizes the moduli or dilaton field or not @xcite . * _ asymptotic conditions _ : de sitter , minkowski , or anti de sitter @xcite . * _ symmetries _ : spherical , planar , hyperbolic , or other symmetries @xcite . * _ dimensions _ : 3 , 4 , or @xmath1 dimensions @xcite . although some of these topics were already investigated by the authors and others , still there are many interesting and various topics to be studied . in this paper , we will limit our focus to the basic model dependence with no dilaton potential , asymptotic flat , spherical symmetry and four dimensions , hence the paper s subtitle : `` _ _ i . causal structures and responses of the brans - dicke field . _ _ '' . to investigate more general features of mass inflation ( for a basic review of mass inflation , see appendix a and references therein ) , we need to investigate the cases with different potentials as well as various parameters . furthermore , it is interesting to change the asymptotic cosmological constant and different symmetries . the variation of dimensions and their semi - classical responses will be also interesting to study . these issues are beyond the scope of this paper , but will be investigated in future works . apart from the physics involved , this paper is by itself important by it s technical merit , since this is a quite generalized and complicated implementation of the double - null formalism . we are sure that there are many more variations to be made from this numerical technique and that this paper is an important step towards further useful or realistic string - inspired models . we hope that this paper can be a leading paper that will introduce upcoming papers . this paper is organized as follows ; in section [ sec : gen ] , we summarize various models inspired from string theory . in section [ sec : mod ] , we present our model in the double - null formalism . in section [ sec : dyn ] , we present and summarize two of our main results , specifically the issue of causal structures ( e.g. , dynamics of local horizons and existence of cauchy horizons ) and responses of the brans - dicke field ( e.g. , check the no - hair theorem and the direction of coupling bias ) . in section [ sec : dis ] , we summarize and discuss future issues . finally , in the appendices , we comment on basics of mass inflation ( appendix a ) , the consistency and convergence of our numerical scheme ( appendix b ) and present a catalog of simulations for various initial conditions ( appendix c ) . in this paper , we consider the following action ( @xmath2 ) : @xmath3,\end{aligned}\ ] ] where @xmath4 is the brans - dicke field , @xmath5 is the ricci scalar , @xmath6 is the brans - dicke coupling , @xmath7 is the kinetic term of the gauge field and @xmath8 is the coupling between the brans - dicke field and the gauge field . as we will see below , @xmath6 and @xmath8 in particular are model depent . of course , this model is limited up to order @xmath5 and hence does not correspond to real string theory , since the real theory should contain all possible interactions , including higher order curvature corrections . on the other hand , this model is sufficiently general to cover a lot of string inspired models and thus provide important intuitions concerning dynamical black hole physics . next , we illustrate some of the models and motivations which is covered by our action ( equation ( [ eq : bdscalar_or ] ) ) . [ [ dilaton - gravity ] ] dilaton gravity + + + + + + + + + + + + + + + every low energy effective action of string theory contains the following sector that includes the dilaton field @xcite : @xmath9 where @xmath10 is the space dimensions , @xmath11 is the length scale of string units , and @xmath12 is the dilaton field . this can be transformed by redefining @xmath12 , such as @xmath13 where @xmath14 is the @xmath15 dimensional gravitation constant . then , we recover the brans - dicke theory in the @xmath16 limit . in general , there will be higher order loop corrections to the low energy effective action . for example , in a model from heterotic string theory compactified on a @xmath17 orbifold @xcite , this can be well approximated by the correction to the brans - dicke coupling @xmath6 : @xmath18,\end{aligned}\ ] ] where @xmath19 is a positive constant of order one which is related to anomaly coefficients . therefore , the coupling should be field dependent and hence , quite large ranges of @xmath6 can be covered by string - inspired models . [ [ couplings - to - the - gauge - field ] ] couplings to the gauge field : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + when we further consider the coupling to the gauge field , the parameter @xmath8 may depend on the models @xcite . we illustrate some detailed examples ; type iia : : the effective action ( bosonic sector ) of the type iia is expanded by @xmath20 + ... , \end{aligned}\ ] ] where @xmath21 is the field strength tensor of the ns - ns two - form field @xmath22 , @xmath23 is the field strength tensor of the r - r one form field @xmath24 and @xmath25 , where @xmath26 is the r - r three - form field . therefore , after the dimensional reduction , it is reasonable to obtain the following effective action : @xmath27 + ... , \end{aligned}\ ] ] and after the redefinition of dilaton , we obtain the gravitational sector by @xmath28 and the two - form field term becomes @xmath29 therefore , type iia corresponds to the @xmath16 and @xmath30 case . type i : : the effective action of type i is expanded by @xmath31 + ... , \end{aligned}\ ] ] where @xmath32 is the mixed contribution of the r - r two - form @xmath33 and of the matrix valued one - form @xmath24 , @xmath23 is the field strength tensor of @xmath24 with the gauge symmetry by the @xmath34 group . therefore , after the dimensional reduction and the symmetry breaking to @xmath35 , we can reduce the action by @xmath36 + ... ; \end{aligned}\ ] ] and after the field redefinition , we obtain the same gravitational sector plus the two - form field term @xmath37 therefore , type i corresponds to the @xmath16 and @xmath38 case . heterotic : : the effective action of the heterotic theory is expanded by @xmath39 + ... , \end{aligned}\ ] ] where @xmath32 is the mixed contribution of the ns - ns two - form field @xmath22 and of the matrix valued one - form @xmath24 , @xmath23 is the field strength tensor of @xmath24 with the gauge symmetry by the @xmath34 or @xmath40 group . therefore , after the dimensional reduction and symmetry breaking to @xmath35 , we can reduce the action by @xmath41 + ... ; \end{aligned}\ ] ] and after the field redefinition , we obtain the same gravitational sector plus the two - form field term @xmath42 therefore , heterotic theory corresponds to the @xmath16 and @xmath43 case . [ [ braneworld - models ] ] braneworld models + + + + + + + + + + + + + + + + + in the randall - sundrum model @xcite , they introduced two branes , where two branes are connected by a warp factor . in the end , one side has a positive tension and the other side has a negative tension . according to garriga and tanaka @xcite , one can calculate the effective action on branes in the weak field limit as a brans - dicke - type theory and we obtain the @xmath6 parameter @xmath44 where @xmath45 is the distance between branes , @xmath46 is the length scale of the anti de sitter space and the sign @xmath47 denotes the sign of the tension . these braneworld inspired models can allow various range of @xmath6 near @xmath48 . [ [ fr - gravity ] ] @xmath0 gravity + + + + + + + + + + + + + + + + + + + + + + + + + + in general , string theory can introduce various higher curvature terms . if we mainly focus on the ricci scalar sectors , then we obtain so - called @xmath0 gravity . the action of @xmath0 gravity is @xmath49 we can modify this action to the scalar - tensor type model by introducing an auxiliary field @xmath50 . then , the gravity sector changes the form @xcite as follows @xmath51,\end{aligned}\ ] ] where we have a constraint @xmath52 . from this , by defining a new field @xmath53 , we can rewrite the action @xmath54,\end{aligned}\ ] ] where @xmath55 now we obtain the brans - dicke theory in the @xmath56 limit with a potential of the brans - dicke field , though we ignore the potential term in this paper . the prototype action of the brans - dicke theory with a @xmath35 gauge field becomes @xmath57,\end{aligned}\ ] ] where @xmath12 is a complex scalar field again ; from now this is not the dilaton field but the complex scalar field . ] with a gauge coupling @xmath58 , @xmath59 is a gauge field , and @xmath60 . the einstein equation becomes as follows : @xmath61 where the brans - dicke part of the energy - momentum tensors are @xmath62 and the matter part of the energy - momentum tensors are @xmath63 where @xmath64 here , we define @xmath65 and @xmath66 is the renormalized energy - momentum tensor to include hawking radiation , while in this paper , we will not consider this semi - classical term . the field equations are as follows : @xmath67 where @xmath68 we use the double - null coordinates @xmath69 assuming spherical symmetry , where @xmath70 is the retarded time , @xmath71 is the advanced time , @xmath72 , where @xmath73 and @xmath74 are angular coordinates . we introduce auxiliary variables , following the notation of @xcite : the metric function @xmath75 , the radial function @xmath76 , the brans - dicke field @xmath4 and a complex scalar field @xmath77 and define @xmath78 the einstein tensor is then given as follows : @xmath79 also , we can obtain the energy - momentum tensors for the brans - dicke field part and the scalar field part : @xmath80 @xmath81,\\ \label{eq : tsuv}t^{\mathrm{c}}_{uv } & = & \frac{{(a_{,v})}^{2}}{4\pi\alpha^{2}},\\ \label{eq : tsvv}t^{\mathrm{c}}_{vv } & = & \frac{1}{4\pi } z\overline{z},\\ \label{eq : tsthth}t^{\mathrm{c}}_{\theta\theta } & = & \frac{r^{2}}{4\pi\alpha^{2 } } \left [ ( w\overline{z}+z\overline{w } ) + iea(\overline{z}s - z\overline{s})+\frac{2{(a_{,v})}^{2}}{\alpha^{2 } } \right],\end{aligned}\ ] ] where we will define @xmath82 as the charge function . to implement the double - null formalism into our numerical scheme , it is convenient to represent all equations as first - order , mixed - derivative differential equations . note that @xmath83 the einstein equations for @xmath84 , @xmath85 , and the field equation for @xmath4 are then coupled as follows ( we define @xmath86 for any quantity @xmath87 ) : @xmath88 where @xmath89 after solving these coupled equations , we can write all equations : @xmath90 @xmath91 including the field equations @xmath92 where equations ( [ eq : solved ] ) and ( [ eq : s ] ) are evolution equations for @xmath84 , @xmath85 , @xmath93 and @xmath94 , which can be solved with the same integration scheme used in the authors previous papers @xcite . the gauge field @xmath95 can be evolved using the same numerical scheme via equation ( [ eq : a2 ] ) or we can evolve @xmath95 and @xmath96 via equations ( [ eq : a1 ] ) and ( [ eq : q1 ] ) . we have found the latter method to be more robust for this study and have used it for our integration via a standard 4th order runge - kutta scheme . consistency and convergence tests for our numerical implementation of the present set of equations are presented in appendix b. for our simulations , we need to specify initial conditions for the dynamic variables ( @xmath97 ) on the initial @xmath98 and @xmath99 null surfaces . for all simulations in this paper , we choose @xmath100 and computational domain to have size @xmath101 $ ] and @xmath102 $ ] . first off , we have the gauge freedom to choose @xmath76 along the initial null surfaces . here , we choose @xmath103 and @xmath104 such that the radial function for an in - going observer decreases and that for an out - going observer increases . in addition , we assume that the effective gravitational constant @xmath105 is asymptotically unity , hence we can set @xmath106 . [ [ in - going - null - direction ] ] in - going null direction + + + + + + + + + + + + + + + + + + + + + + + we use an in - going shell - shaped scalar field such that its interior is not affected by the shell . thus , we can simply choose @xmath107 , @xmath108 , @xmath109 and @xmath110 . then , the misner - sharpe mass function @xmath111 should vanish at @xmath112 . to satisfy this , it is convenient to choose @xmath113 , @xmath114 . we also choose the radial function at the initial point @xmath112 to be @xmath115 , thereby fully determining the radial function along the initial null segments . these initial conditions are in agreement with the constraint equations ( equation ( [ eq : e1 ] ) ) and thus completes the assignment of initial conditions along the in - going null segment . [ [ out - going - null - direction ] ] out - going null direction + + + + + + + + + + + + + + + + + + + + + + + + we can choose an arbitrary function for @xmath116 to induce a collapsing , in - going pulse . in this paper , we use @xmath117\end{aligned}\ ] ] for @xmath118 and otherwise @xmath119 . next , we can integrate constraint equation equation ( [ eq : e2 ] ) to determine @xmath120 on the @xmath121 surface , simultaneously with integrating equations ( [ eq : a1 ] ) and ( [ eq : q1 ] ) for determining @xmath122 and @xmath123 . we have now fully specified the five dynamic variables ( @xmath97 ) along the both in- and out - going null segments in a manner consistent with the constraint equations ( equations ( [ eq : e1 ] ) and ( [ eq : e2 ] ) ) . this finishes the basic assignment of initial conditions ( but cf . with paragraph below ) . [ [ free - parameters ] ] free parameters + + + + + + + + + + + + + + + for the scalar field pulse , we choose @xmath124 , @xmath125 and @xmath126 ( to include maximum charge ) , leaving just three free parameters @xmath127 , where @xmath6 is the brans - dicke coupling parameter , @xmath8 is the coupling between the matter sector and the brans - dicke field and @xmath58 is the gauge coupling . to summarize , we vary as follows : @xmath6 : : : @xmath128 ( @xmath0 limit ) , @xmath129 ( dilaton limit ) , @xmath130 ( braneworld limit ) , @xmath131 ( ghost limit ) , @xmath8 : : : @xmath128 ( type iia ) , @xmath132 ( type i ) , @xmath133 ( heterotic ) , @xmath58 : : : @xmath128 ( neutral black hole ) , @xmath134 ( charged black hole ) . causal structure of neutral black holes with @xmath135 . thin black curves are contours of a constant @xmath76 ; the difference between contours are @xmath133 . red curve is the apparent horizon @xmath136 , blue curve is the horizon @xmath137 , and thick black curve is the singularity @xmath138 . all subsequent causal structure plots has similar legend . ] causal structure of neutral black holes with @xmath139 . ] causal structure of neutral black holes with @xmath140 . ] classification of causal structures for neutral black holes . ] in this section , we report the possible causal structures and responses of the brans - dicke field for neutral and charged black holes of various string - inspired models obtained from numerical calculations by using the double - null formalism . in this paper , we wish to study charged black holes . but first , let us briefly summarize the case of neutral black holes ; figure [ fig : neutral_beta0 ] shows the causal structures , plotted are lines of constant r , apparent horizons and @xmath138 singularities of neutral black holes for @xmath135 , figure [ fig : neutral_beta05 ] shows neutral black holes for @xmath139 and figure [ fig : neutral_beta1 ] shows neutral black holes for @xmath140 , all figures are obtained by our numerical code . let us briefly look at the brans - dicke field equation in the neutral limit : @xmath141 where @xmath142 therefore , if @xmath143 , then since the term @xmath144 is positive , it pushes the brans - dicke field to the strong coupling limit ( @xmath145 ) and vice versa . if @xmath146 then it behaves oppositely . the @xmath43 case is unique and in this case , there is no back - reactions to the brans - dicke field and hence there is no dependence on @xmath6 ( the four figures in figure [ fig : neutral_beta1 ] are the same , bar numerical errors ) . causal structure for charged black holes with @xmath135 . ] as we observed in a previous paper @xcite , if @xmath143 , then during the gravitational collapse , the brans - dicke field is pushed to @xmath145 limit and after the collapse , the pushed field is repulsed and approaches the stationary limit . on the other hand , if @xmath146 , then brans - dicke is pushed to @xmath147 limit during gravitational collapses and approaches the stationary limit . because of this behaviour , there appears not only the apparent horizon @xmath136 , but also @xmath137 for @xmath143 . in addition , the apparent horizon can show strong oscillations for @xmath146 and perhaps it can create a cauchy horizon as a naked singularity . as @xmath148 decreases , the back - reaction decreases . this is clear , if we observe the location of @xmath137 horizons or the oscillations of the apparent horizon @xmath136 and compare with the @xmath30 case and the @xmath38 ; they tend to disappear , where these phenomena are the typical behaviors which are due to the brans - dicke field . we summarize the possible charge - less causal structures , inferred from our simulations , in figure [ fig : classification ] . for @xmath143 , the typical shape is ( a ) . the oscillations of the apparent horizon @xmath136 depends on the number of peaks of the initial collapsing field ( see ( b ) and figure [ fig : tvv_beta0 ] in appendix c , and compare with figure 5 of @xcite ) . this is one difference between @xcite and the current paper . the location of @xmath149 horizon appears and increases as the dynamics of the brans - dicke field increases , e.g. , as @xmath6 approaches @xmath48 or @xmath148 increases , as denoted in ( c ) . for @xmath146 , the typical shape is ( e ) , while the apparent horizon and the singularity can meet as was observed in @xcite and the causal structure can be ( f ) . for all cases , when the dynamics of the brans - dicke field is suppressed , the causal structure approaches ( d ) and follows that of einstein gravity . now we turn on the gauge coupling parameter @xmath150 . note that the case of @xmath126 is the case with maximal asymptotic charge ( for einstein gravity , this is demonstrated in @xcite ) , since the phase between the real part and the imaginary part become maximal . however , in general it does not mean that extreme ( or naked ) black holes can be formed ( this was also demonstrated by @xcite ) . the excessive charge will be repelled outside the horizon and the excessive charge is either repelled to future infinity or forms a stable charge cloud outside the horizon @xcite . hence in terms of gravitational collapses , @xmath126 allows the largest possible amount of charge . causal structure for charged black holes with @xmath139 . ] [ [ case - beta0 ] ] case @xmath135 : + + + + + + + + + + + + + + + + + + + + + + + + + + + figure [ fig : charge_beta0 ] shows causal structures of charged black holes with @xmath30 . the obvious feature is the existence of a cauchy horizon inside the outer apparent horizon . the singularity is always spacelike . however , in the limit @xmath151 , there will be nonzero distance ( in terms of @xmath70-coordinate ) between the outer apparent horizon and the singularity . this indicates the existence of the cauchy horizon . these global behaviors are the same as for the case of charged black holes in einstein gravity . in the left of figure [ fig : alpha_maintext ] ( as well as in figure [ fig : alpha_beta0 ] in appendix c ) , the metric function @xmath75 is plotted and there it can clearly be seen that inside the horizon , @xmath152 and which hence indicates mass inflation inside the charged black hole ( see appendix a ) . one important difference between this and the previous case , is the complicated horizon behaviors inside the apparent horizon . such non - trival behaviours are to be expected since we have both the dynamics of the matter field as well as the brans - dicke field inside the horizon and these cause the observed complicated dynamical behaviour of horizons . in einstein gravity , if we do not turn on the semi - classical effects , then there is no complicated horizon dynamics inside the horizon ; but , if we turn on the semi - classical effects , then the null energy condition is violated and hence the inside horizon structure becomes complicated @xcite . also , for modified gravity models such as @xmath0 gravity ( in the jordan frame ) , similar behaviors has been observed @xcite . comparison of @xmath75 for @xmath30 ( left ) and @xmath38 ( right ) , with the same @xmath153 and @xmath150 . the cases for this and other initial conditions are illustrated in appendix c. ] causal structure for charged black holes with @xmath140 . ] [ [ case - beta0.5 ] ] case @xmath139 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + figure [ fig : charge_beta05 ] shows causal structure for the case of @xmath38 . comparing this with the @xmath30 case , we see to two main differences . first , for @xmath143 , the cauchy horizon _ disappears_. in other words , there is a spacelike singularity and it approaches the apparent horizon in the @xmath151 limit . this was observed by @xcite ( for dilaton limit in the einstein frame ) and our results ( various @xmath6 in the jordan frame ) are consistent with it . this issue will be clarified when we study the response of brans - dicke field in the next section . looking at the @xmath75 function , there is the signature of mass inflation , see for example right of figure [ fig : alpha_maintext ] for @xmath150 and @xmath143 ( as well as in figure [ fig : alpha_beta05 ] in appendix c ) , since there is a region where @xmath75 approaches zero . however , this is not enough to generate a cauchy horizon as in the @xmath30 cases . secondly , for @xmath146 , there can be an @xmath149 horizon which will eventually form a _ cosmological horizon_. this was not observed in gravitational collapses of neutral matters . therefore , we can state that if @xmath146 , then a _ charged _ gravitational collapse can induce inflation and a baby universe . if we interpret the @xmath149 horizon as a throat of a ( one - way traversable ) wormhole , then we can make a one - way traversable wormhole using a charged gravitational collapse . of course , we know that if we consider the limit @xmath146 , then the brans - dicke field becomes a ghost and hence it can allow a wormhole or a baby universe @xcite . however , what we show is that such objects can be obtained just by _ gravitational collapse _ of a charged matter field , i.e. , in an extremely simple way while previously , complicated initial conditions were needed to obtain such a state @xcite . [ [ case - beta1 ] ] case @xmath140 : + + + + + + + + + + + + + + + + + + + + + + + + + + + the causal structure of this case , shown in figure [ fig : charge_beta1 ] , shares many qualitative properties with the @xmath38 case . in the neutral limit , there was no similar behaviors for @xmath146 limit , therefore , it is more clear that the existence of the inflating region really is comes from the effects of _ importantly , it is noted that for the case of @xmath16 ( dilaton gravity with heterotic theory ) , there exists a spacelike singularity . there were some discussions that extreme black holes can be regular @xcite . however , if the black hole is formed from a gravitational collapse , then we have to conclude that such a regular black hole state is impossible to obtain . [ [ summary - of - causal - structures ] ] summary of causal structures + + + + + + + + + + + + + + + + + + + + + + + + + + + + the intuitive nature of local @xmath136 horizons or @xmath137 horizons are the same as figure [ fig : classification ] . however , when we turn on the charge , the global structures can take on three different types . ( 1 ) for @xmath30 , there is the cauchy horizon ( ( g ) of figure [ fig : causal ] ) . ( 2 ) for @xmath143 and @xmath154 or @xmath133 , there is no cauchy horizon ( ( h ) of figure [ fig : causal ] ) . ( 3 ) for @xmath146 and @xmath38 or @xmath133 cases , the gravitational collapse can induce inflation ( ( i ) of figure [ fig : causal ] ) . if the final point of @xmath137 in the @xmath151 limit ( green dot in ( i ) ) does not meet the spacelike future infinity , then there will be a cauchy horizon in the @xmath151 limit . causal structures for charged black holes . ] plots of @xmath4 for @xmath135 . ] the brans - dicke field equation is @xmath141 where @xmath155 note that we can approximately write this as follows : @xmath156 the kinetic terms mainly contribute _ during _ the gravitational collapse ( i.e. in our simulations for @xmath157 ) , while the charge terms will contribute _ after _ the gravitational collapse ( i.e. in our simulations for @xmath158 ) and _ outside _ the apparent horizon . outside the black hole , the charge term will remain since the charge is a conserved quantity , while the kinetic term will disappear as the gravitational collapse ends . if we keep this in mind , then for the @xmath30 case , there will be essentially no effects after the gravitational collapse and the outside the horizon . in other words , the brans - dicke hair will decay as time goes on and approach a stationary limit , figure [ fig : bd_beta0 ] shows such behaviors . on the other hand , in the @xmath140 limit , there will be almost no back - reaction to the brans - dicke field during the gravitational collapse ; only after the gravitational collapse has finished will the brans - dicke field be affected . figure [ fig : bd_beta1 ] shows such behaviors very clearly . plots showing the brans - dicke field @xmath4 for @xmath139 . ] plots showing the brans - dicke field @xmath4 for @xmath140 . ] if @xmath159 , then one can see both effects simultaneously , such behaviour is shown in figure [ fig : bd_beta05 ] ; one can see the contrast between during ( @xmath160 ) and after ( @xmath161 ) the gravitational collapse , which is different from the cases of @xmath30 and @xmath43 . now let us think about the directions of the brans - dicke fields , whether it becomes greater or less than one . for @xmath143 , if the term @xmath144 is positive , it will push the brans - dicke field to the strong coupling limit ( @xmath145 ) ; if it is negative , then it will push to the weak coupling limit ( @xmath147 ) . to summarize : : : effects of kinetic terms : if @xmath162 and @xmath143 , then the kinetic terms always push the brans - dicke field to the strong coupling limit . : : effects of charge terms : if @xmath163 and @xmath143 , then the charge terms always push the brans - dicke field to the weak coupling limit . for the other cases ( e.g. , @xmath146 ) , opposite behaviors will be observed . this is consistent with the numerical results shown in figures [ fig : bd_beta0 ] , [ fig : bd_beta05 ] and [ fig : bd_beta1 ] . for @xmath164 and @xmath143 , during the gravitational collapse , @xmath145 appears ( shifts to blue color ) except @xmath43 . after the gravitational collapse , always we have @xmath147 ( shifts to red color ) except for @xmath30 . in particular we can see both phenomena at the same time for the case of @xmath139 . if @xmath146 , then all behaviors are opposite . we can summarize our results ( for @xmath143 ) as follows : @xmath165 : : : all effects push the brans - dicke field to the strong coupling limit . as time goes on , the brans - dicke hair will be formed in the strong coupling direction . @xmath30 : : : the brans - dicke field is pushed to the strong coupling limit , but only dynamically . as time goes on , the brans - dicke hair will disappear in the stationary limit . @xmath159 : : : the brans - dicke field is pushed to the strong coupling limit dynamically . on the other hand , as time goes on , the brans - dicke hair will be formed in the weak coupling direction . @xmath43 : : : the brans - dicke field does not affected by gravitational collapse . however , as time goes on , the brans - dicke hair will be formed in the weak coupling direction . @xmath166 : : : all effects push the brans - dicke field to the weak coupling limit . one final comment should be made regarding the no - hair theorem in the einstein frame . we know that for a quite general class of models , the no - hair theorem should be satisfied in the einstein frame , hence , there should be no scalar hair in the stationary limit . if this is the case in the einstein frame , then there should be no hair in the jordan frame , too @xcite . however , our model is consistent with these general discussions . after we change the einstein frame , the matter sector lagrangian has a non - trivial coupling between the matter field and the brans - dicke field : so to speak , @xmath167 , unless @xmath30 . this violates the assumptions of @xcite and it explains why a non - trivial scalar hair appears after the gravitational collapse has finished . furthermore , we can consistently observe that there is no scalar hair after the gravitational collapses for the @xmath30 case . in this paper , we have investigated charged black holes of string - inspired gravity models . two parameters are important : the brans - dicke parameter @xmath6 and the coupling between the brans - dicke field and the gauge field @xmath8 . this model covers lots of string - inspired models : the dilaton gravity @xmath16 with type iia ( @xmath30 ) , type i ( @xmath38 ) , heterotic ( @xmath43 ) , some braneworld inspired models @xmath143 and @xmath0 inspired models @xmath56 . in addition , we investigated the ghost limit @xmath146 . we used numerical techniques with double - null formalism to numerically solve the dynamic equations . we have focused on the causal structures and responses of the brans - dicke field , especially reactions via charges . for usual non - ghost limit @xmath143 , the direction of the response of the brans - dicke field , whether biased to strong coupling limit or weak coupling limit , is determined by @xmath8 . if @xmath43 , then there is no response of the brans - dicke field during the gravitational collapse ; if @xmath30 , then there is no response of the brans - dicke field via charges after a black hole is formed . this means that after the gravitational collapse , as long as @xmath163 , the brans - dicke field will be biased to the weak coupling limit . this is very crucial in order to understand the nature of charged black holes for @xmath163 . an important point is that in general such a black hole will have brans - dicke hair . another important point is that the brans - dicke field is biased to the weak coupling limit and this will _ screen _ the charge inside the horizon . therefore , this explains the absence of the cauchy horizon , while it exists for the @xmath30 case . however , we did not try to control the brans - dicke field using a potential and hence further points to be clarified , this issue will be studied in a future paper _ ii _ , which will focus on mass inflation and dependence on parameters and potentials . however , the current paper suggests that there are are many other parameters and issues which will be interesting to study . although not a string - inspired model per se , the @xmath146 and @xmath163 cases are also interesting in the sense that a charged matter collapse can induce inflation , while the outside still does not inflate . so , as long as the theory allows for the existence of ghosts , the formation of a ( at least one - way traversable ) wormhole does not require complicated initial conditions . of course , the existence of the ghost may be accompanied by complicated instabilities and this calls for further investigations . a detailed study of this issue is also postponed to a future paper . to see , not only outside , but also inside the black hole and to study the detailed relations between gravity and various field contents , probably the best way is to use the double - null formalism . the authors hope that this paper will be an important step to prepare for upcoming investigations of diverse and various interesting topics , including the interests of theoretical gravity , string theory , holography , astrophysical applications , as well as cosmology , by varying matter , dimensions , gravity model , and symmetry , etc . we would like to thank to bum - hoon lee and also give thanks for using computer facilities in center for quantum spacetime , sogang university . dy would like to thank to dong - il hwang for help at the beginning of this project . dy was supported by the jsps grant - in - aid for scientific research ( a ) no . 21244033 and leung center for cosmology and particle astrophysics ( lecospa ) of national taiwan university ( 103r4000 ) . jh was supported in part by the programs of the construction and operation for large - scale science data center at kisti ( k-14-l01-c06-s01 ) and the global hub for experiment data of basic science , nrf ( n-14-nm - ir06 ) and by the apctp topical research program . in this appendix , we briefly review mass inflation in einstein gravity . mass inflation is a typical phenomenon of a deep inside of a charged black hole . let us begin from a static charged black hole solution , so - called the reissner - nordstrom solution @xcite : @xmath168 where @xmath169 . this solution has two horizons : @xmath170 with asymptotic mass @xmath171 and charge @xmath172 ( if @xmath173 ) . we show that the inner horizon is unstable up to a small perturbation , in the sense that a local observer near the inner horizon who sees a small perturbation can experience an exponential growth of the energy density . for an in - going matter pulse , the energy - momentum tensor can be approximated by @xcite @xmath174 where @xmath175 is so - called the luminosity function that should be proportional to @xmath176 with a certain constant @xmath177 for a realistic matter @xcite . now , when there is an out - going observer who sees the in - going matter pulse , the observer measures the energy density @xmath178 , where @xmath179 is the null geodesic of the out - going observer . note that @xmath180 , where @xmath181 is the surface gravity of the inner horizon and @xmath182 is the affine parameter of the observer @xcite . therefore , the energy density becomes @xmath183 this is so - called mass inflation . first , this implies that the inner horizon of the reissner - nordstrom solution is unstable via a small perturbation @xcite . therefore , if we consider gravitational collapses , we can not trust the reissner - nordstrom solution for the internal causal structure and the internal structure is dramatically changed @xcite . this necessarily requires numerical simulations . secondly , as the local energy function exponentially increases , other curvature quantities also exponentially increases @xcite . in the end , the inner horizon becomes a physical curvature singularity with a non - zero area , where the singularity is null for classical cases , while space - like with semi - classical back - reactions @xcite . in terms of double - null coordinates , mass inflation is usually identified by the behavior of the metric function @xmath75 ( when we use the double - null metric @xmath184 ) ; @xmath75 exponentially decreases , i.e. , @xmath185 with a constant @xmath186 . this was repeatedly observed by numerical simulations @xcite . in the extremely small @xmath75 limit , all curvature quantities ( ricci scalar , kretchmann scalar , etc . ) behave like @xmath187 ( @xmath188 ) and hence curvatures exponentially diverge . all of these descriptions are on mass inflation in einstein gravity . on the other hand , if there is a complicated coupling with charge , matter , and gravity , then these expectations can be modified and this is one important motivation of this paper . in this appendix , we demonstrate that 1 ) our numerical solutions are converging when including the effects of nontrivial parameters and scalar fields and 2 ) that the converged numerical solution is indeed a solution which satisfies the constraint equations . the results presented here are very similar to those presented in @xcite which is not surprising , since the code in that paper and this one are essentially identical . for this reason we refer to that paper and references therein ( especially , @xcite ) , for further details of the numerical schemes and other details related to the design of the code . the initial conditions and computational domain for the tests in this appendix are similar to those used for making figures [ fig : charge_beta0 ] and [ fig : bd_beta0 ] for the @xmath189 , @xmath150 , @xmath135 case . this setup is clearly non - trivial and is thus a good test for the general convergence properties of our code . our computational domain for this convergence tests is , as for figure [ fig : causal ] , in the range @xmath101 $ ] and @xmath102 $ ] . next we demonstrate the convergence of the code by comparing a series of simulations with varying numerical resolution . for the tests in this appendix , we do a total of six simulations , with each simulation changing the base resolution . to limit the number of plots , we concentrate on displaying convergence results along the line @xmath190 , which is well inside the apparent horizon and passes through a region with very complex behaviors ( cf . figures [ fig : charge_beta0 ] and [ fig : bd_beta0 ] ) . thus , if we see convergence here , it is a fair indication that convergence requirements are satisfied throughout the whole of the computational domain . along this line ( @xmath190 ) , we calculate the relative convergence between two simulations ( one with a numerical resolution twice that of the other ) relative to a simulation with very high resolution : @xmath191 where @xmath192 denotes the dynamic variable @xmath193 at the @xmath194-th grid point of simulation with resolution @xmath195 and where @xmath196 denotes the dynamic variable of the same @xmath194-th point for a simulation with the highest numerical resolution done by us . obviously , this expression only makes sense for those @xmath194 points that coincide in all simulations . the first six plots in figure [ fig : convergence ] show the relative convergence , @xmath197 , for the dynamic variables @xmath75 , @xmath76 , @xmath4 , re @xmath45 , i m @xmath45 and @xmath95 , respectively . the lines in the figures are marked by their numerical resolution measured in terms of the most coarse resolution @xmath198 , the high resolution simulation used to calculate expression equation ( [ eq : xidef ] ) , has a numerical resolution of @xmath199 times the base resolution , i.e. , @xmath200 . from these figures , it is clear that the dynamic variables are converging for simulations of increasing resolution . furthermore , since we plot the _ relative _ convergence of the dynamic variables , we see that the relative change between the two highest resolution simulations show that the variables change on the order of @xmath201 or less , which we considered a quite acceptable . it is , however , not enough to demonstrate that the simulations are converging , they must also converge to a ` _ _ physical _ _ ' solution , i.e. , the residuals of the constraint equations ( equations ( [ eq : e1 ] ) and ( [ eq : e2 ] ) ) must converge to zero . to demonstrate this , we calculate the relative convergence of the constraint equation residuals , ( relative to the einstein - tensor ) in a similar way to equation ( [ eq : xidef ] ) : @xmath202 where @xmath203 denotes the residual of the constraint equation ( @xmath204 or @xmath205 ) , at the @xmath194-th point for simulation with resolution @xmath195 and where @xmath206 denotes the corresponding einstein - tensor component ( @xmath207 or @xmath208 respectively ) at the same point . the relative convergence of the residuals of the constraint equations are demonstrated in the bottom two plots of figure [ fig : convergence ] , where it is seen that they converge towards zero for higher resolution simulations . this indicates that not only are the numerical solutions converging for simulations of higher resolution , but that they are indeed converging towards a physical solution . finally , it is emphasized that the results presented in this appendix are not the only convergence tests that we have performed , they are merely representative of typical convergence behavior . for all results presented in this paper , we have performed a large number of simulations with varying resolutions to ensure that the results had converged to their physical solution . relative convergence of the dynamic variables and constraint equations along line of @xmath190 for simulation with parameters : @xmath189 , @xmath150 and @xmath135 . see text for details.,title="fig : " ] relative convergence of the dynamic variables and constraint equations along line of @xmath190 for simulation with parameters : @xmath189 , @xmath150 and @xmath135 . see text for details.,title="fig : " ] relative convergence of the dynamic variables and constraint equations along line of @xmath190 for simulation with parameters : @xmath189 , @xmath150 and @xmath135 . see text for details.,title="fig : " ] relative convergence of the dynamic variables and constraint equations along line of @xmath190 for simulation with parameters : @xmath189 , @xmath150 and @xmath135 . see text for details.,title="fig : " ] relative convergence of the dynamic variables and constraint equations along line of @xmath190 for simulation with parameters : @xmath189 , @xmath150 and @xmath135 . see text for details.,title="fig : " ] relative convergence of the dynamic variables and constraint equations along line of @xmath190 for simulation with parameters : @xmath189 , @xmath150 and @xmath135 . see text for details.,title="fig : " ] relative convergence of the dynamic variables and constraint equations along line of @xmath190 for simulation with parameters : @xmath189 , @xmath150 and @xmath135 . see text for details.,title="fig : " ] relative convergence of the dynamic variables and constraint equations along line of @xmath190 for simulation with parameters : @xmath189 , @xmath150 and @xmath135 . see text for details.,title="fig : " ] in this appendix , we summarize the metric function @xmath75 and energy - momentum tensor components @xmath209 and @xmath210 . figures [ fig : alpha_beta0 ] , [ fig : alpha_beta05 ] , and [ fig : alpha_beta1 ] are plots of @xmath75 for @xmath30 , @xmath132 , and @xmath133 , respectively . for typical situations of mass inflation , @xmath152 exponentially , and hence almost all curvature quantities ( ricci scalar , kretchmann scalar , etc . ) behave @xmath187 ( @xmath188 ) and hence curvatures exponentially diverge . figures [ fig : tuu_beta0 ] , [ fig : tuu_beta05 ] , and [ fig : tuu_beta1 ] are plots of @xmath209 for @xmath30 , @xmath132 , and @xmath133 , respectively . figures [ fig : tvv_beta0 ] , [ fig : tvv_beta05 ] , and [ fig : tvv_beta1 ] are plots of @xmath210 for @xmath30 , @xmath132 , and @xmath133 , respectively . note that around the apparent horizon @xmath211 , @xmath212 and hence , for an out - going observer , if @xmath213 , then the @xmath136 horizon changes the sign of @xmath214 from @xmath215 to @xmath216 , while @xmath217 , changes the sign of @xmath214 from @xmath216 to @xmath215 . also , around the apparent horizon @xmath137 , @xmath218 and hence , for an in - going observer , if @xmath219 , then the @xmath137 horizon changes the sign of @xmath220 from @xmath215 to @xmath216 , while @xmath221 , changes the sign of @xmath220 from @xmath216 to @xmath215 . for all figures , these rules are always satisfied even with complicated horizon dynamics . this is also a simple check of the consistency of our simulations . b. s. dewitt , phys . rev . * 160 * , 1113 ( 1967 ) ; + b. s. dewitt , phys . * 162 * , 1195 ( 1967 ) ; + b. s. dewitt , phys . rev . * 162 * , 1239 ( 1967 ) . c. kiefer , `` _ quantum gravity , _ '' oxford university press ( 2004 ) ; + c. rovelli , `` _ quantum gravity , _ '' cambridge university press ( 2004 ) ; + g. w. gibbons and s. w. hawking , `` _ euclidean quantum gravity , _ '' world scientific ( 1993 ) . m. b. green , j. h. schwarz and e. witten , `` _ superstring theory . vol . 1 : introduction , _ '' cambridge university press ( 1987 ) ; + m. b. green , j. h. schwarz and e. witten , `` _ superstring theory . 2 : loop amplitudes , anomalies and phenomenology , _ '' cambridge university press ( 1987 ) . s. kachru , r. kallosh , a. d. linde and s. p. trivedi , phys . d * 68 * , 046005 ( 2003 ) [ arxiv : hep - th/0301240 ] ; + j. p. conlon , fortsch . phys . * 55 * , 287 ( 2007 ) [ hep - th/0611039 ] . j. m. maldacena , adv . * 2 * , 231 ( 1998 ) [ int . j. theor . phys . * 38 * , 1113 ( 1999 ) ] [ arxiv : hep - th/9711200 ] . j. p. conlon and f. g. pedro , jhep * 1105 * , 079 ( 2011 ) [ arxiv:1010.2665 [ hep - th ] ] ; + d. hwang , f. g. pedro and d. yeom , jhep * 1309 * , 159 ( 2013 ) [ arxiv:1306.6687 [ hep - th ] ] . s. w. hawking and r. penrose , proc . lond . a * 314 * , 529 ( 1970 ) . a. borde , a. h. guth and a. vilenkin , phys . lett . * 90 * , 151301 ( 2003 ) [ arxiv : gr - qc/0110012 ] . m. gasperini , `` _ elements of string cosmology , _ '' cambridge university press ( 2007 ) . g. w. gibbons and k. -i . maeda , nucl . b * 298 * , 741 ( 1988 ) ; + d. garfinkle , g. t. horowitz and a. strominger , phys . d * 43 * , 3140 ( 1991 ) [ erratum - ibid . d * 45 * , 3888 ( 1992 ) ] . a. ashtekar and b. krishnan , living rev . * 7 * , 10 ( 2004 ) [ arxiv : gr - qc/0407042 ] ; + s. a. hayward , arxiv : gr - qc/9303006 ; + a. b. nielsen and d. yeom , int . j. mod . a * 24 * , 5261 ( 2009 ) [ arxiv:0804.4435 [ gr - qc ] ] . d. hwang and d. yeom , class . grav . * 27 * , 205002 ( 2010 ) [ arxiv:1002.4246 [ gr - qc ] ] . w. israel , phys . rev . * 164 * , 1776 ( 1967 ) ; + w. israel , commun . * 8 * , 245 ( 1968 ) . r. s. hamade and j. m. stewart , class . * 13 * , 497 ( 1996 ) [ arxiv : gr - qc/9506044 ] . r. parentani and t. piran , phys . lett * 73 * , 2805 ( 1994 ) [ arxiv : hep - th/9405007 ] ; + t. chiba and j. soda , prog . * 96 * , 567 ( 1996 ) [ arxiv : gr - qc/9603056 ] ; + s. ayal and t. piran , phys . d * 56 * , 4768 ( 1997 ) [ arxiv : gr - qc/9704027 ] ; + s. hod and t. piran , phys . lett * 81 * , 1554 ( 1998 ) [ arxiv : gr - qc/9803004 ] ; + s. hod and t. piran , gen . * 30 * , 1555 ( 1998 ) [ arxiv : gr - qc/9902008 ] ; + e. sorkin and t. piran , phys . d * 63 * , 084006 ( 2001 ) [ arxiv : gr - qc/0009095 ] ; + e. sorkin and t. piran , phys . d * 63 * , 124024 ( 2001 ) [ arxiv : gr - qc/0103090 ] ; + y. oren and t. piran , phys . d * 68 * , 044013 ( 2003 ) [ arxiv : gr - qc/0306078 ] ; + p. p. avelino , a. j. s. hamilton and c. a. r. herdeiro , phys . d * 79 * , 124045 ( 2009 ) [ arxiv:0904.2669 [ gr - qc ] ] . a. borkowska , m. rogatko and r. moderski , phys . d * 83 * , 084007 ( 2011 ) [ arxiv:1103.4808 [ hep - th ] ] ; + a. nakonieczna , m. rogatko and r. moderski , phys . d * 86 * , 044043 ( 2012 ) [ arxiv:1209.1203 [ hep - th ] ] ; + a. nakonieczna and m. rogatko , arxiv:1209.3614 [ hep - th ] . m. a. scheel , s. l. shapiro and s. a. teukolsky , phys . d * 51 * , 4208 ( 1995 ) [ arxiv : gr - qc/9411025 ] ; + m. a. scheel , s. l. shapiro and s. a. teukolsky , phys . d * 51 * , 4236 ( 1995 ) [ arxiv : gr - qc/9411026 ] . j. hansen , a. khokhlov and i. novikov , phys . d * 71 * , 064013 ( 2005 ) [ arxiv : gr - qc/0501015 ] ; + a. doroshkevich , j. hansen , i. novikov and a. shatskiy , int . d * 18 * , 11 ( 2009 ) [ arxiv:0812.0702 [ gr - qc ] ] ; + a. doroshkevich , j. hansen , d. novikov , i. novikov , d. h. park and a. shatskiy , phys . d * 81 * , 124011 ( 2010 ) [ arxiv:0908.1300 [ gr - qc ] ] . d. yeom , arxiv:0912.0068 [ gr - qc ] ; + d. hwang and d. yeom , class . * 28 * , 155003 ( 2011 ) [ arxiv:1010.3834 [ gr - qc ] ] ; + b. -h . lee and d. yeom , nuovo cim . c * 036 * , no . s01 , 79 ( 2013 ) [ arxiv:1111.0139 [ gr - qc ] ] ; + d. hwang , b. -h . lee and d. yeom , jcap * 1301 * , 005 ( 2013 ) [ arxiv:1210.6733 [ gr - qc ] ] . s. e. hong , d. hwang , e. d. stewart and d. yeom , class . * 27 * , 045014 ( 2010 ) [ arxiv:0808.1709 [ gr - qc ] ] ; + d. hwang and d. yeom , phys . d * 84 * , 064020 ( 2011 ) [ arxiv:1010.2585 [ gr - qc ] ] . d. hwang , b. -h . lee and d. yeom , jcap * 1112 * , 006 ( 2011 ) [ arxiv:1110.0928 [ gr - qc ] ] . j. hansen , b. -h . lee , c. park and d. yeom , class . * 30 * , 235022 ( 2013 ) [ arxiv:1307.0266 [ hep - th ] ] . d. hwang , h. -b . kim and d. yeom , class . * 29 * , 055003 ( 2012 ) [ arxiv:1105.1371 [ gr - qc ] ] . d. hwang , b. -h . lee , w. lee and d. yeom , jcap * 1207 * , 003 ( 2012 ) [ arxiv:1201.6109 [ gr - qc ] ] . s. foffa , m. maggiore and r. sturani , nucl . b * 552 * , 395 ( 1999 ) [ arxiv : hep - th/9903008 ] . l. randall and r. sundrum , phys . * 83 * , 3370 ( 1999 ) [ arxiv : hep - ph/9905221 ] . t. p. sotiriou and v. faraoni , rev . phys . * 82 * , 451 ( 2010 ) [ arxiv:0805.1726 [ gr - qc ] ] ; + s. nojiri and s. d. odintsov , int . j. geom . * 4 * , 115 ( 2007 ) [ arxiv : hep - th/0601213 ] . a. g. agnese and m. la camera , phys . d * 51 * , 2011 ( 1995 ) . j. hansen , d. hwang and d. yeom , jhep * 0911 * , 016 ( 2009 ) [ arxiv:0908.0283 [ gr - qc ] ] . t. p. sotiriou and v. faraoni , phys . lett . * 108 * , 081103 ( 2012 ) [ arxiv:1109.6324 [ gr - qc ] ] . h. reissner , ann . physik . , * 50 * , 106 ( 1916 ) ; + g. nordstrom , proc . kon . ned . , * 20 * , 1238 ( 1918 ) .

galaxy formation requires a process that continually heats gas and quenches star formation in order to reproduce the observed shape of the luminosity function of bright galaxies . to accomplish this , current models invoke heating from supernovae , and energy injection from active galactic nuclei however , observations of radio - loud active galactic nuclei suggest that their feedback is likely to not be as efficient as required , signaling the need for additional heating processes . we propose the self - annihilation of weakly interacting massive particles that constitute dark matter as a steady source of heating . in this paper , we explore the circumstances under which this process may provide the required energy input . to do so , dark matter annihilations are incorporated into a galaxy formation model within the millennium cosmological simulation . energy input from self - annihilation can compensate for * all * the required gas cooling and reproduce the observed galaxy luminosity function only for what appear to be extreme values of the relevant key parameters . the key parameters are : the slope of the inner density profile of dark matter haloes and the outer spike radius . the inner density profile needs to be steepened to slopes of or more and the outer spike radius needs to extend to a few tens of parsecs on galaxy scales and a kpc or so on cluster scales . if neutralinos or any thermal relic wimp with s - wave annihilation constitute dark matter , their self - annihilation is inevitable and could provide enough power to modulate galaxy formation . energy from self - annihilating neutralinos could be yet another piece of the feedback puzzle along with supernovae and active galactic nuclei . [ firstpage ] dark matter stars : evolution accretion early universe

0711.2302

a major challenge to our current understanding of structure and galaxy formation in the universe is the discrepancy between the theoretically predicted mass function of dark matter haloes and the observed shape of the luminosity function of galaxies ( kauffmann & white 1993 ; cole et al . 1994 ; somerville , primack & faber 2001 ; benson et al . 2003 ) . simply put , the number density of dark matter haloes falls off as a power - law at high masses , whereas the luminosity function of galaxies occupying such haloes terminates exponentially above a characteristic luminosity ( e.g. cole et al . 2001 ; huang et al . this implies that the supply of gas to a galaxy and the conversion of this gas into stars becomes preferentially inefficient in more massive systems ( white & rees 1978 ) . a key issue for galaxy formation theory is thus to illuminate the physical processes that heat and cool gas in massive galaxies as this cycle regulates the formation of new stars . in addition , one of the challenges for galaxy formation is to unravel the inter - play of baryons with the ubiquitous dark matter in galaxies . to this end , feedback processes operating in galaxies at both the low and high - mass ends of the halo mass function are required to explain the faint and bright - end slopes of the observed galaxy luminosity function . energy input from supernovae is thought to play a significant role in the regulation of star formation in low mass galaxies ( dekel & silk 1986 ; mac low & ferrara 1999 ) , however the supernova energy injected in high mass galaxies is too small to suppress gas cooling effectively . for such objects , to reconcile theory with observations a more energetic process that continually heats the gas and operates independently of star formation appears to be required ( croton et al . 2006 ; bower et al . 2006 ) . recent observations in nearby galaxies reveal a correlation between the masses of supermassive black holes and the velocity dispersion of the stellar component . this suggests that black holes might play a role in regulating star formation ( magorrian et al . 1998 ; tremaine et al . 2002 ; ferrarese & merritt 2002 ) . energy input from nuclear outflows driven by accreting black holes are currently favoured as the principal source of feedback driving the truncation of star formation in massive galaxies ( di matteo et al . 2005 ; croton et al . 2006 ; bower et al . 2006 ) . unlike supernovae feedback , the observed energy output from active galactic nuclei ( agn ) can far exceed that liberated from cooling gas in the hot halo . agn heating in two distinct forms has been proposed . episodic feedback from the agn via outflows generated during the merger process have been proposed by di matteo et al . ( 2005 ) ; sijacki & springel ( 2006 ) , the so - called ` quasar - mode ' . however , it is known that agn spend most of their lifetimes in a low accretion rate state , and therefore croton et al . ( 2006 ) and bower et al . ( 2006 ) argue for a more steady , so called ` radio - mode ' feedback that is long - lived . the details of both these processes are complex and the micro - physics is currently not well understood . here , we focus on an alternative heating mechanism , the energy steadily generated by the self - annihilation of dark matter particles in the inner regions of haloes . although the standard cosmological paradigm is predicated on the existence of non - baryonic dark matter ( dm ) particles , the precise nature of these particles and their interactions remain a puzzle . the neutralino is the current leading dark matter candidate . in this paper we explore if the energy supplied from the self - annihilation of dark matter in the centres of galaxies and cluster haloes could possibly play a significant role in the baryonic cooling / heating cycle ( ascasibar 2007 ; totani 2004 ; 2005 ; colafrancesco et al . detailed astrophysical implications of neutralino dark matter annihilations in galaxy clusters , with a specific application to the coma cluster have been calculated by colafrancesco , profumo & ullio ( 2006 ) . they performed a thorough analysis of the transport and diffusion properties of neutralino annihilation products , and investigated the resulting multi - frequency signals , from radio to gamma - ray frequencies . they also study other relevant astrophysical effects of neutralino annihilations , like the dm - induced sunyaev - zeldovich effect and the intracluster gas heating . annihilation luminosity is expected to be produced as a result of an enhancement in the density of the dark matter distribution in the inner - most regions of galaxies , due to the response of the dark matter to the adiabatic growth of a central black hole ( gondolo & silk 1999 ) and/or to the adiabatic compression suffered by the dark matter due to the collapse of baryons into the dark matter potential well ( blumenthal et al . 1986 ; ryden & gunn 1987 ; gnedin et al . 2004 ) . by including this steady heating source of annihilating neutralinos to suppress cooling flow gas in a sophisticated model of galaxy formation , we explore the effect of various key parameters in reproducing the bright - end of the observed galaxy luminosity function . the question we address is whether self - annihilation can provide an alternative mechanism to agn driven feedback within the context of the current paradigm . the outline of this paper is as follows . in section [ section2 ] we discuss the physics of neutralino annihilation ; a scenario for the coupling of this energy source to baryons in galactic nuclei is explored in section [ section3 ] . with this phenomenology developed , we incorporate this feedback scheme into a galaxy formation model . the details of the n - body simulation and galaxy formation model used in this implementation are discussed in section [ section4 ] . our results are presented in detail in section [ section5 ] and summarized in section 6 . we conclude with a discussion of the implications of our results and comparison of annihilation heating with other modes of feedback . unless otherwise stated we assume hubble constant parametrised as @xmath1 with @xmath2 . theoretical predictions of structure formation and evolution in a cold dark matter ( cdm ) cosmogony appear to be in very good agreement with current observations on most scales bar perhaps the smallest ( spergel et al . 2006 ; tegmark et al . 2004 ; seljak et al . 2005 ; cole et al . 2005 ) . in this well established paradigm , the bulk of the matter in the universe is comprised of cold , collisionless particles that seed the formation of structures from the gravitational amplification of their early fluctuations ( blumenthal et al . 1984 ; davis , efstathiou , frenk & white 1985 ) . weakly interacting massive particles ( wimps ) are the currently favoured dark matter candidates as their cross - sections are small enough that they act as essentially collisionless particles ( kolb & turner 1990 ) . the super - symmetric _ neutralino _ , arising in minimal super - symmetric extensions of the standard model of particle physics , is probably the most widely studied wimp candidate ( see review by jungman , kamionkowski & griest 1996 and references therein ) . in the minimal super - symmetric standard model ( mssm ) , the super - partners of the gauge bosons and the neutral higgs bosons , respectively called binos and higgsinos , mix into four majorana fermionic mass eigenstates , called neutralinos . in some regions of the super - symmetric parameter space , the lightest neutralino can naturally achieve a relic density that matches the observed cosmic dm abundance , thus making it a theoretically well motivated dm candidate . the neutralino mass can be anywhere between @xmath3 gev , which is the lower bound allowed by accelerator constraints ( eidelman et al . 2006 , see _ ibidem _ for the set of assumptions made on the underlying super - symmetric scenario ) , up to 100 tev in some extreme non - perturbative scenarios ( profumo 2005 ) , although masses larger than a few tev are commonly considered as `` unnatural '' , if one wants susy to solve the theoretical problems ( such as the hierarchy problem ) for which it was originally invented . an interesting alternative candidate arises in theories with universal extra dimensions ( ued ) , and corresponds to the first kaluza - klein state of the hypercharge gauge boson ( appelquist , cheng and dobrescu , 2002 ; servant and tait 2000 ) . the existence of a viable dark matter candidate in ued theories can be seen as a consequence of the conservation of momentum in a higher dimensional space . to generate chiral fermions at the zero mode , the extra dimensions must be modded out by an orbifold , leading to the conservation of the so - called kk - parity , such that all odd - level kk particles are charged under this symmetry , thus ensuring that the lightest ( first level ) kk state is stable . a lower bound on the mass of the lightest kaluza - klein particle ( lkp ) comes from electroweak measurements , and depending on the mass of the higgs , it can be as low as @xmath4 gev ( gogoladze and macesanu 2006 ) . although the annihilation cross section in these scenarios is fixed by the lkp mass , it is possible to achieve the correct relic density even for particles as heavy as several tev , provided that one includes the effect of co - annihilation with other kk particles ( burnell & kribs 2006 , kong & matchev 2006 , kakizaki et al . we note that the arguments presented here are valid more generally for any thermal relic wimp candidate with s - wave annihilation . regardless of its precise nature , it is possible to set cosmological constraints on the wimp mass @xmath5 under some simplifying , but rather general assumptions on its nature . the wimp mass is in fact constrained to be roughly @xmath6 and @xmath7 , by theoretical considerations of the thermal freeze - out in the early universe ( e.g. bertone , hooper and silk 2005 ) . the very same theoretical considerations suggest that the wimp annihilation rate in the local universe is far below the expansion rate of the universe . however , annihilations proceed in high density dm regions , such as the centres of dm haloes . we demonstrate in this work that this resultant annihilation luminosity could play an important role in galaxy formation by providing a strong source of feedback which prevents gas from cooling and forming stars in galaxies . wimp pair annihilation in high dm density regions will inevitably produce high energy neutrinos , positrons , anti - protons and gamma - rays . there are several on - going and future experiments focused on direct and indirect searches for the signature of neutralinos ( see reviews by bergstrom 2002 ; bertone , hooper & silk 2004 ) . in this paper we focus on gev mass - scale wimps , however for illustration purposes we also calculate the consequence of heating by mev dm . here we describe the key ingredients of the simple theoretical model constructed to estimate the self - annihilation luminosity of dm in the centres of galaxy and cluster scale haloes . in the concordance @xmath8cdm cosmology the thermally averaged annihilation rate of wimps , in absence of co - annihilations , is related to the dm relic density by @xmath9 . from this expression one can estimate the annihilation luminosity from neutralinos in a halo with a given density profile ( assuming they constitute all the dark matter ) : @xmath10 where the integral is performed over the halo volume @xmath11 , @xmath5 is the wimp mass , and @xmath12 is the dark matter density profile , commonly assumed to be spherically - symmetric , making it only dependent on the galactocentric radial coordinate @xmath13 . the density profile of a dark matter haloes that assemble in cosmological n - body simulations is well fit by the following parametrised functional form , referred to as the generalised navarro - frenk - white ( nfw ) profile ( navarro , frenk & white 1996 ) : @xmath14^{(\beta-\gamma)/\alpha } } ~ , \label{profile } \end{aligned}\ ] ] where @xmath15 , @xmath16 , and @xmath17 govern the slope of the profile on large and small scales respectively , with a transition in slope at the scale radius @xmath18 , defined as the ratio of the virial radius of the dm halo , @xmath19 , and the virial concentration parameter , @xmath20 : @xmath21 in this paper we only consider profiles for which the outer slope is @xmath22 , i.e. where @xmath23 . this leads to the following expression for the normalisation constant , @xmath24 , @xmath25 where @xmath26 is the virial mass of the system . the slope of the density profile in the inner regions of the halo @xmath17 is of interest for our work . a value of @xmath27 refers to the nfw profile , while @xmath28 corresponds to the so - called moore profile ( moore et al . note that eqn . ( [ profile ] ) is a fit to the output of numerical simulations which do not include either particle physics and hydrodynamic effects . such effects may alter the distribution of dm over time . the consequences of this possibility are discussed in the next section . in fact , it is the time evolution of the density profile that holds the key to possibly tapping the energy from the neutralino self - annihilation . we briefly note here that for a cluster scale dm halo , with typical virial mass 10@xmath29 , the luminosity generated from eqn . ( [ lum ] ) with central density profile @xmath30 is , at most , of order @xmath31 ergs@xmath32 . thus , even if this entire energy output coupled maximally to the cooling gas in the inner regions it would be insufficient to offset the cooling luminosity ( of order @xmath33 ergs@xmath32 ) in a typical cluster system ( totani 2004 ; 2005 ) . physical processes that further steepen the inner density profile slope are therefore required to make the annihilation viable as a feedback mechanism . fortunately , several astrophysical processes in the centres of dark matter haloes are expected to affect the density profile of dm in precisely the required fashion . density enhancements can arise in the inner regions of dark matter haloes from a number of different astrophysical processes . one such process is the adiabatic response of the dark matter to the infall and collapse of baryons ( blumenthal , flores & primack 1986 ; ryden & gunn 1987 ; gnedin et al . baryonic gas loses energy through radiative processes and falls into the centre of a dark matter halo to form stars . as a result of this redistribution of mass , the gravitational potential of the inner regions of the halo changes the dark matter responds to the subsequent deepening of the potential by altering its distribution and thereby enhancing its density . the increase in dark matter density from adiabatic compression can be calculated using adiabatic invariants . the cusp index @xmath17 in the nfw profile of eqn . ( 2 ) steepens to a value that depends on the density profile of the baryons . earlier calculations by blumenthal et al.(1986 ) over - predicted the steepening due to the assumptions of spherical symmetry and circular orbits . this was due , in part , to the fact that haloes in the hierarchical structure formation scenarios grow via multiple violent mergers and accretion along filaments , and particle orbits in the haloes are highly eccentric . gnedin et al . ( 2004 ) revisited this question using high - resolution cosmological simulations that included gas dynamics , radiative cooling , and star formation . they found that the dissipation of gas indeed increased the density of dark matter and steepened its radial profile in the inner parts of the halo when compared with haloes without cooling . comparisons with the earlier work of blumenthal et al.(1986 ) showed that the assumption of spherical symmetry induces a systematic over - prediction of the density enhancement in the inner 5% of the virial radius . gnedin et al . ( 2004 ) correct for this by providing a simple modification of the assumed adiabatic invariant which includes orbit - averaged particle positions . if the baryons have a radial density profile @xmath34 , then the spike ( i ) retains the same slope if @xmath35 ; ( ii ) or if @xmath36 , the contracted inner slope of the dm profile is steeper than its original value . for @xmath37 , the final inner dm density slope can be as steep as @xmath38 , respectively . in cluster sized systems , although baryons represent only a small fraction of the overall mass , they may be crucially important on scales comparable to the extent of the typical brightest , central cluster galaxies . there has been a lot of recent activity in the simulations community to understand the likely interactions between baryons and dark matter ( gnedin et al . 2004 ; nagai & kravtsov 2005 ; faltenbacher et al . 2005 ; gustafsson , fairbairn & sommer - larsen 2006 ) . for instance , in a recent simulation that included baryons , gustafsson , fairbairn and sommer - larsen ( 2006 ) claim that the central dm cusps steepen to @xmath39 , with an indication of the inner logarithmic slope converging on galaxy mass scales . gustaffson et al . ( 2006 ) claim that the difference in the extent of adiabatic contraction and subsequent response they find compared to gnedin et al . ( 2004 ) and other works originates in the differences in their stellar feedback prescription . so the extent of steepening due to adiabatic response is an unsettled issue at the present time due to the inherent uncertainity in the our understanding and implementation of star formation in simulations . on the observational side there have also been many recent attempts to disentangle the dark matter and baryonic components in clusters ( zappacosta et al . 2006 ; biviano & salucci 2006 ; mahdavi et al . 2007 ; sand et al . 2008 ) . extensive convergence studies have shown that modern highest resolution dissipationless simulations agree in their predictions : the average logarithmic slope of the density profile at @xmath40 is @xmath41 , with a substantial scatter of @xmath42 from object to object ( fukushige et al . 2004 ; tasitsiomi et al . 2004 ; navarro et al . 2004 ; reed et al . 2004 ; diemand et al . 2004 ; 2005 ) . at the same time , despite a significant decrease in the smallest reliably resolved scale , the logarithmic slope continues to get shallower with decreasing radius without reaching an asymptotic value . for the purposes of this work , for our default model we will assume a conservative value for the inner slope of the density profile of @xmath43 , which is the standard nfw value commonly used . on extremely small scales @xmath44 , i.e. at the very centre of the galaxy hosted by a dm halo , the gravitational potential is dominated not by dm ( as described by eqn . [ profile ] ) but by baryons , mainly comprising stars and frequently a supermassive black hole ( smbh ) . in fact , from the demography of nearby galaxies it appears that nearly every galaxy hosts a smbh , and their masses are well correlated with properties of the stellar component in the inner - most regions ( magorrian et al . 1998 ; tremaine et al . 2002 ; ferrarese & merritt 2002 ) . in our own galaxy , for instance , stars exhibit a cusp inside a few parsecs described by ( genzel et al . 2003 ) : @xmath45 this distribution is centred around the supermassive black hole at the galactic centre , whose mass is estimated to lie in the range @xmath46 ( ghez et al . 2005 ; genzel et al . 2003b ) . in fact , hooper , finkbeiner & dobler ( 2007 ) claim that the excess microwave emission from the region around the center of our galaxy detected by wmap ( wilkinson microwave anisotropy probe ) could be synchrotron emission from relativistic electrons and positrons generated in dark matter annihilations . hooper et al . ( 2007 ) find that the angular distribution of this `` wmap haze '' matches the prediction for dark matter annihilations with a cusped density profile , @xmath47 in the inner few kiloparsecs . comparing the intensity in different wmap frequency bands , they find that a wide range of possible wimp annihilation modes are consistent with the spectrum of the haze for a wimp with a mass in the 100 gev to multi - tev range . most interestingly , they find that to generate the observed intensity of the haze , the dark matter annihilation cross section is required to be approximately equal to the value needed for a thermal relic , @xmath48 @xmath49/s . more generally , the _ adiabatic _ growth of a massive object at the centre of a power - law distribution of dm with index @xmath17 induces a redistribution of the dm ( also referred to as a density ` spike ' ) , into a new power - law with a steepened index @xmath50 ( peebles 1972 , young 1980 , ipser & sikivie 1987 , quinlan et al . 1995 , gondolo & silk 1999 ) . @xmath51 where @xmath52 and the outer spike radius is approximated by @xmath53 where @xmath54 is the gravitational radius of influence of the black hole ( merritt & szell 2006 ) . the density @xmath55 is found by evaluating eqn . [ profile ] at the outer spike radius . for the milky way , @xmath56 . a physical cut - off to the otherwise diverging profiles of eqn . ( [ profile ] ) and eqn . ( [ spikeprofile ] ) at small @xmath13 is provided by the self - annihilation rate itself . we can define a limiting radius , @xmath57 , where the density reaches a maximal value given by @xmath58 where @xmath59 is the lifetime of the density spike ( see below ) . this limiting radius @xmath57 can then be found by equating and solving eqn . ( [ limdens ] ) and eqn . ( [ spikeprofile ] ) simultaneously . in reality , the inner cut - off radius can not be arbitrarily small . hence , we truncate the inner annihilation luminosity at @xmath60 ~ , \label{rcut}\end{aligned}\ ] ] where @xmath61 is the schwarzchild radius of the smbh , @xmath62 . this defines the inner spike radius , @xmath63 . in order to understand the time evolution of a dm density spike , we note that once the spike is established several astrophysical and particle physics effects act to disrupt it . the three principal effects that can damp a density spike are : ( i ) the self - annihilation process itself that depletes dm ( discussed above ) , ( ii ) the interaction between the dm and the surrounding baryons near the smbh , and ( iii ) spike disruption during galaxy ( or more specifically black hole ) mergers . dm and baryons interact gravitationally with each other , and stars typically have significantly larger kinetic energies than dm particles . gravitational encounters between these two populations will tend to drive them toward equipartition of energy , causing the dm to heat up , flattening the spike while retaining the shape of the density profile . as shown by bertone & merritt ( 2005a ; 2005b ) , the resultant spike evolution with time can be modelled as an exponential decay via @xmath64 and @xmath65 where @xmath66 is the time elapsed since the original spike formation @xmath59 measured in units of the heating time @xmath67 , @xmath68 we show results for the time evolution of a fiducial spike in at the centre of our galaxy in figure 1 and discuss the issue further in section 5.1 . mergers of black holes , and in particular major mergers ( i.e. smbh mass ratios greater than 0.3 ) , are efficient at flattening or destroying density spikes ( merritt et al . we note that the disruption is maximal during equal mass mergers , however these are extremely rare . after a merger the density spike will be re - established on a time - scale comparable to the relaxation time of stars ( merritt , harfst and bertone 2006 ) . if major mergers are too frequent then spikes will not be long - lived enough to enhance the annihilation luminosity . however , if they never happen the spike decay described above will result in inefficient heating . we encapsulate spike disruption in eqn . ( [ limdens ] ) by following the time since the last smbh major merger in each halo , @xmath59 , which approximates the spike lifetime . the major merger rate in the simulation and galaxy formation model is discussed further in section [ sec : spikes ] . using all the above computed quantities , we can finally estimate the time dependent annihilation luminosity at any time after the most recent smbh major merger as : @xmath69 @xmath70 in the limit that @xmath71 . in what follows , we evaluate eqn . ( 14 ) for dm haloes in the millennium run simulation . to explore the effects of the heating of cooling gas in the hot halo from neutralino annihilation we employ a model of galaxy formation ( croton et al . 2006 ) coupled to a high resolution n - body simulation , the so - called ` millennium run ' ( springel et al . below we give only a brief outline of these techniques ; the interested reader should refer to croton et al ( 2006 ) and references therein for further details . in the following sub - sections we describe the addition of neutralino heating to this model . the millennium run n - body simulation follows the dynamical evolution of approximately 10 billion dark matter particles in a periodic box of side - length @xmath72mpc with a mass resolution per particle of @xmath73 . the adopted cosmological parameter values are @xmath74 , @xmath75 , @xmath2 , and @xmath76 ( colless et al . 2001 , spergel et al . 2003 , seljak et al . 2005 ) . from the simulation outputs merger trees are constructed that describe in detail how structures grow as the universe evolves . these trees form the backbone onto which the galaxy formation model is coupled . inside each tree , virialised dark matter haloes at each redshift are assumed to attract ambient gas from the surrounding medium , from which galaxies form and evolve . the galaxy formation model effectively tracks a wide range of physics in each halo , including reionization of the inter - galactic medium at high redshift , including radiative cooling of hot gas and the formation of cooling flows , star formation in the cold disk and the resulting supernova feedback , black hole growth , metal enrichment of the inter - galactic and intra - cluster medium , and galaxy morphology shaped through mergers and merger induced starbursts . more specifically , once a dark matter halo has grown in mass to approximately @xmath77 infalling baryons no longer fall ` cold ' on to the central galaxy but instead shock heat to the virial temperature of the dark matter halo to form a quasi - static hot atmosphere ( croton et al . 2006 ) . as the density of hot gas increases the cooling time near the centre of the halo becomes short and a cooling flow of condensing gas forms . using simple thermodynamic and continuity arguments a cooling rate for this flow , @xmath78 , can be estimated under the assumption that the gas is approximately isothermal ( @xmath79 ) outside the very central regions at the virial temperature of the halo ( @xmath80 , where @xmath81 is the virial velocity of the dark halo ) : @xmath82 here @xmath19 is the virial radius of the halo , @xmath83 is the mass of gas in the hot phase , and @xmath84 is the cooling time of the gas at the cooling radius , which is defined by : @xmath85^{1/2}~ , \label{rcool}\ ] ] the cooling radius traditionally marks the radius out to which gas has had time to cool quasi - statically given the age of the system . in the above , @xmath86 is the mean particle mass , @xmath87 is the boltzmann constant , and @xmath88 is the cooling function , dependent on both the temperature @xmath89 and metallicity @xmath90 of the hot gas . despite their simplicity , ( 15 ) and ( 16 ) provide a good approximation to the rate at which gas is deposited at the centre in the similarity solution for a cooling flow ( bertschinger 1989 ; croton et al . 2006 ) . gas cooling in simulations appears to be much more efficient than observed in nature . for low mass haloes heating and expulsion of baryonic material by supernovae can prevent over - cooling , but on group and cluster scales , while the observed baryon fraction has the universal value ( balogh et al . 2001 ) , state of the art hydrodynamical simulations find a significantly larger fraction of cooled gas ( borgani et al . 2003 ; kravtsov , nagai & vikhlinin 2005 ) . in fact in clusters the problem is particularly severe ; observationally the x - ray luminosities in many clusters imply cooling rates in the inner regions of @xmath91 100 - 1000 @xmath92 ( e.g. fabian & nulsen 1977 ; edge 2001 ; fabian 2004 ) , whereas little or no associated star formation is detected . a decrease in x - ray gas temperature is often seen in the centres of clusters but x - ray spectroscopy detects very little cool gas below @xmath93 k at temperatures of @xmath94 ( peterson et al . 2001 ; 2003 ) . therefore it has been known for some time ( see fabian 1994 and references therein ) , the measured cooling times at the centre of observed clusters of galaxies are significantly shorter than the hubble time . this implies that gas condensation should be occurring from the conventional picture described above , however none is observed , nor are the secondary effects of cooling such as copious star formation in the central brightest cluster galaxy . such galaxies are mostly found to be ` red and dead ' . this behaviour may indicate the presence of a heat source that is supplying energy to the cooling gas to keep the central regions from condensing . this is the so - called classical ` cooling flow problem ' ( fabian 1994 ) . to arrest the expected runaway cooling in massive systems a heat source must be present . we test the idea of whether the heating from dark matter annihilation at the centres of massive systems can plausibly be a global solution to the cooling flow problem described above . another key aspect of the cooling flow problem is the fact that the cooling time of gas is short over a fairly large range of radii , particularly in groups and clusters , implying that the heating rate of any proposed mechanism ought to balance the cooling at all radii . below we describe our implementation of this dm annihilation powered heating into the galaxy formation model . the energy produced from the self - annihilation of neutralinos in the vicinity of the spike @xmath95 we argue will couple thermally with the baryons in galactic nuclei , thereby heating the gas . we investigate potential physical processes and the relevant time - scales that will likely enable the transfer of this energy to the baryons . we compute relevant time - scales for the centre of the milky way treating that as our fiducial case from which we scale to other masses . similar heurisitic estimates have been made by totani ( 2004 ; 2005 ) for galaxy clusters . * the dominant astrophysical process via which the self - annihilation energy interacts with baryons is unknown at the present time . * here for purposes of estimating the efficiency of potential astrophysical processes we use observational estimates of the density , pressure and temperature from the inner regions of our galaxy at roughly 200 pc from the galactic center ( bradford et al . 2005 ; morris & serabyn 1996 ) . the physical conditions present in this region are consistent with thermal , non - thermal and magnetic pressures that are several orders of magnitude higher than those present in the large - scale galactic disk . for the measured densities of @xmath96 , the pressure is @xmath97 , however turbulent pressures greatly exceed this value , approaching @xmath98 . for the inferred magnetic field strength of @xmath99 mg , the magnetic pressure is also large , of the order of @xmath100 . to estimate the time - scales for energy loss and therefore heating of the gas in the inner regions , we adopt the following scaling values for the density , pressure and magnetic field strength from the milky way : @xmath101 , @xmath102 , and @xmath103 mg . * note that the self - annihilation heating occurs for the @xmath104 k cooling gas from the hot halo and not to the @xmath105 k molecular gas at 200 pc . * using the above values we can address the question of how the annihilation luminosity is likely converted efficiently into thermal energy of the gas in the nucleus . we focus on a fiducial neutralino model where about @xmath106th of the annihilation energy @xmath107 goes into continuum gamma - rays [ channel a ] , @xmath108th goes into electrons and positrons [ channel b ] , @xmath109th goes to protons and anti - protons [ channel c ] , and the rest is imparted to neutrinos that are not relevant for heating . the spectral energy distribution of the products is such that @xmath110 peaks roughly at @xmath111 , @xmath111 and @xmath112 respectively . the average energy of secondary particles produced in the annihilation of a neutralino of mass 50 gev is of the order of @xmath113 . these annihilation products are representative of a wide class of super - symmetric scenarios , as determined with the darksusy code ( gondolo et al . 2005 ) , that are consistent with cosmological and accelerator constraints . examining channel b in detail , we assume for simplicity that all the electrons and positrons have roughly the same energy @xmath114 gev . since they are produced in an extremely high density environment , they will expand relativistically and form a bubble . as discussed in shu ( 1991 ) , the primary energy loss processes are examined below . the energy loss time - scale for these particles by coulomb collisions is given by : @xmath115 which is so short that all the electrons and positrons from the neutralino decay will very efficiently transfer energy to the gas cooling out of the hot halo . this is the principal physical process that will likely heat the gas and provide the feedback discussed in this work . on the other hand , the energy loss time from inverse compton scattering is long , @xmath116 and therefore inefficient . the time - scale for energy losses via brehmstrahlung is also short , @xmath117 in comparison using the same estimate for @xmath118 , the energy loss time - scale from synchrotron radiation is given by , @xmath119 we conclude from the above estimates that the conversion of the self - annihilation energy into thermal energy can occur efficiently for the gas in the inner regions of galactic nuclei as illustrated specifically for the milky way case . however , we do note here that our estimate of the magnetic field strength is on the high side . this energy input suppresses the gas cooling and quenches star formation . due to our lack of knowledge of neutralino properties , it is not possible to calculate the precise time - scale and physical process that thermally couples the by - products of the @xmath17-ray annihilation to baryons . for the purposes of this work , it is assumed that this coupling is efficient and all the available energy is transferred effectively to heating the gas in the inner regions of dm haloes . the estimate of time - scales for the milky way halo suggests that gas in the inner region on the scale of @xmath120 is likely directly involved in the heating process . we note here that an alternate dark matter driven gas heating mechanism exploiting inelastic scattering of x - dark matter particles with relic abundances comparable to neutralinos has been proposed by finkbeiner & weiner ( 2007 ) . they propose a wimp candidate with an excited state that maybe collisionally excited and de - excites by @xmath121 pair emission . the kinetic energy of these pairs they argue could heat intra - cluster gas and the gas in galaxies to compensate for cooling similar to what we explore here . the primary motivation for this model was to provide a possible interpretation for the 511 kev line observed by the integral satellite in the inner milky way consistent with the observed wmap haze and current constraints on the gamma - ray background . we model the self - annihilation luminosity of neutralinos as a steady input of energy that prevents gas cooling in preferentially massive galaxies / haloes . the assumptions and properties of simulated haloes that are needed to fully specify the dm density profile , bh mass , and the inner and outer spike radii are all taken as input by the galaxy formation model to calculate the annihilation luminosity for any system at any given time its evolution . the dm density profile for every halo in the millennium run at every time - step is approximated using the universal analytic function described by eqn . ( [ profile ] ) . note that , while in principle each profile can be directly measured from the distribution of bound dark matter particles , this is computationally prohibitive given the number of haloes in the millennium run ( up to 25 million at any given time - step ) , and the analytic formalism is accurate enough for our purposes here . to explore the dependence of annihilation heating on the inner slope of the halo density profile , @xmath17 , we take the inner slope as a free parameter in determining eqn . ( [ profile ] ) . in the next section we will vary @xmath17 and examine its effect on the evolution of the galaxy population . the remaining parameters in the density profile are either assumed fixed or taken directly from the simulation . for the default model , we assume @xmath122 and choose @xmath16 such that @xmath23 . this ensures that the outer slope of the halo density profile remains fixed at @xmath22 , which is known to be an accurate description of the results of numerical simulations ( navarro , frenk & white 1996 ) . we take the measured virial mass @xmath26 , and radius @xmath19 , required in eqn . ( [ rs ] ) and ( [ rho0 ] ) , directly from the simulation . to estimate the halo concentration parameter @xmath20 , we use the measured @xmath123 and @xmath124 and solve eqn . ( 5 ) in navarro et al . 1997 ( see croton , gao & white 2007 ) . this fully describes the density profile for each dark matter halo . once a value for the slope of the inner dark matter density profile , @xmath17 , has been assumed , the steepened spike index , @xmath50 ( eqn . [ spikegamma ] ) , can be calculated . this then fixes the spike density profile defined by eqn . ( [ spikeprofile ] ) , and therefore also the inner limiting density determined by the self - annihilation rate itself , eqn . ( [ limdens ] ) . the radius of this limiting density , and the inner spike radius , are calculated from eqn . ( [ spikeprofile ] ) and ( [ limdens ] ) . note that we limit the inner spike radius to always be equal to or greater than four times the schwarzchild radius ( eqn . [ rcut ] ) . however , this limit is rarely reached in practice . the outer spike radius is simpler to calculate for each halo in the galaxy formation model , since the model explicitly follows the growth of smbhs in each galaxy ( croton et al . this , along with the use of the virial velocity of the halo @xmath81 as a proxy for the inner velocity dispersion @xmath125 , allows the outer spike radius to be calculated using eqn . ( [ spikeouter ] ) . we consider a maximal heating model , assuming that all the energy available from the annihilation luminosity given by eqn . ( [ heating ] ) couples with the cooling gas in the hot halo . with this assumption , the cooling rate described by eqn . ( [ cooling ] ) is modified in the presence of dm annihilations : @xmath126 thus , if the heating rate from the annihilation flux is comparable to the energy released from gas cooling out of the hot x - ray halo , the cooling flow can be suppressed and this will starve the central galaxy from lack of new star forming material . under such circumstances galaxy growth will stall , altering the relationship between dark halo mass and galaxy luminosity in a way more compatible with observations . as the results in the following sections are considered , it is important to keep in mind that the galaxy formation model we use to obtain them is imperfect . its construction is largely based on observational phenomenology that , while well described in the mean , is often ill understood in detail . the largest uncertainty in our galaxy formation model relevant to our results is the cooling prescription described in section [ mr ] . the physics of cooling losses from hot plasma in a dynamically evolving multiphase medium is complicated . our prescription provides cooling rates that are a reasonable average approximation to that obtained from hydrodynamically simulated gas infall ( see yoshida et al . 2002 for a comparison ) . however , cooling in both the hydrodynamic simulations and our galaxy formation model is first calculated in the absence of any heating ( eqn . [ cooling ] ) . the model heating rate is then subtracted ( eqn . [ finalheating ] ) . but in reality , cooling and heating will occur simultaneously . energy injection from heating can potentially modify the properties of the gas ( notably temperature and density ) used to calculate later cooling . hence , under these circumstances subsequent cooling estimates will be different from those where the past heating history is neglected as in our model . due to these caveats , in absolute terms our heating rates must be treated with caution ; they are simply a relative measure that defines an upper limit on the energy required to stop gas from cooling . however , relative to the cooling rate they are expected to be accurate for the model . more detailed models can be constructed to take into account the past heating history when calculating the current cooling rate , but these estimates would be complicated and highly unconstrained . for example , the temperature and density gas profiles could change as a function of the energy injected , depth of potential , and redshift , for every galaxy in an evolving universe . these detailed and self - consistent responses of the baryons need to be taken into account . however , such improvements add considerable complexity that is beyond the scope of the present work . below , we present the results of implementing a simple model of annihilation heating . as discussed in sections 3.1 and 3.3 , the luminosity produced from dark matter annihilation in an unsteepened nfw - type halo falls short by several orders - of - magnitude when compared to the cooling losses from the x - ray emitting hot gas . for annihilations to be a plausible solution to the cooling flow problem one requires an enhancement of the dm density in the inner most regions of the galaxy , usually assumed to be produced by the presence of a super - massive black hole or in response to the adiabatic compression of baryons or perhaps both these processes . when present , density spikes are expected to evolve with time due to the annihilation itself . to illustrate this , in fig . [ fig : spike_evol ] we plot the time evolution of a dm density spike in the centre of a fiducial halo , modelled on the milky way . the x - axis is normalised to the radius of gravitational influence of the black hole , while the y - axis is normalised the density of the dm halo at this radius . the three different line styles denote the values of @xmath66 ( i.e. time in units of the heating time @xmath67 , eqn . [ heatingtime ] ) . the red line is for @xmath127 , green for @xmath128 , and blue for @xmath129 . the unmodified nfw profile ( i.e. @xmath130 ) is shown as a dotted curve . the central plateau in the inner regions is due to annihilations , and the maximum density drops with time . the evolution of the density spike is clearly a result of many complex interactions within the halo . the radius of gravitational influence of the black hole , while the y - axis is normalised to @xmath24 the density of the dm halo at @xmath131 . the 3 different line styles denote the values of @xmath66 ( i.e. time in units of the heating time @xmath67 ) . the solid lines are , from top to bottom , for @xmath127 ( red ) , @xmath128 ( green ) and @xmath129 ( blue ) . the `` unmodified '' nfw profile is shown as a dotted line . the central `` plateau '' for the curves is due to annihilations , and the maximum density is calculated for a toy model , to simply show the qualitative behaviour of the profile at small radii . the maximum density - the height of the plateau , varies with time . ] dark matter density spikes are fragile and transient a major merger can easily destroy a spike on a short time - scale . however , it is unclear whether the disruption is triggered by the merger of the dark matter haloes , central black holes or galaxies hosted in the haloes . the merger rates for these three populations of objects are not necessarily the same , and yet a typical approximation adopted is to the use extended press - schechter ( eps ) theory to obtain the merger rate . however eps only describes the properties of dm haloes . in reality it is probably the mergers of black holes that ought be most significant for the evolution of a density spike . using our cosmological model of galaxy formation we can check the frequency of mergers for haloes , galaxies , and the black holes that reside within them . in fig . [ fig : merger_rate ] we show for each object the mean number of major mergers at @xmath132 . we focus on this redshift range because it is primarily only at late times wherein heating is needed to prevent the cooling of gas in haloes . major mergers of smbhs are less common than major mergers of galaxies or dark matter haloes , with less than one occurring per system on average since @xmath133 at all host halo mass scales . this implies a typical spike survival time @xmath134 . [ fig : merger_rate ] also shows that the merger rates for all populations tends to increase with increasing mass although this flattens somewhat for the most massive haloes . we conclude from fig . [ fig : merger_rate ] that spikes can survive on gyr timescales despite the violent nature of the hierarchical growth and assembly in a @xmath8cdm universe . note , that even after a major merger spikes may later reform , typically on the stellar relaxation time - scale of the inner region of the galaxy . in this work , we further assume that the mass of central black holes grows significantly at each merger ( as the solution for adiabatic growth in valid only in the limit @xmath135 ( di matteo , springel & hernquist 2005 ) . we now investigate the enhanced annihilation luminosity due to the presence of density spikes described in section [ section4 ] . the annihilation heating model depends on a well defined set of parameters . we explore the dependences in detail . these dependencies are illustrated in the various panels of fig . [ fig : heating ] : the dark matter halo virial mass @xmath136 , the black hole mass @xmath137 , the mass of the dark matter particle @xmath138 , the average time between major mergers @xmath66 , the inner dm halo density profile slope @xmath17 , and the outer radius @xmath139 adopted for the dm density spike . we consider two halo mass ranges for each , galaxy scales masses ( @xmath140 , dashed lines ) and cluster masses ( @xmath141 , solid lines ) . our default model assumes common values for each of @xmath142 , @xmath143 , @xmath144 ( nfw ) , and @xmath145 ( the sphere of influence of the black hole ) . in each panel we vary one parameter keeping the rest fixed to clearly show the plausible range of heating rates that can be expected from the model ( the default values of each parameter are marked by vertical dotted lines ) . in addition , the top left panel shows the expected cooling rate assuming a hot gas fraction of 0.1 . the choice of 10% for the host gas fraction is motivated by the x - ray observations in galaxy groups and clusters ( laroque et al . 2006 ; vikhlinin et al . 2006 ; gonzalez , zaritsky & zabludoff 2007 ) . this is approximately the energy the dm heating needs to replenish . it is clear that even in the presence of density spikes it is usually difficult to produce enough flux to reheat all the cooling gas , and this is true for haloes of all masses ranging from galaxy scales to cluster scales . enhancement of between 1 - 2 orders - of - magnitude does occur when mergers are more frequent ( this works to offset the exponential decay of the spike amplitude shown in fig . [ fig : spike_evol ] ) or for lower mass dm particle candidates . however the heating luminosity is still significantly lower when compared with the cooling rate shown in the top left panel . the bottom two panels of fig . [ fig : heating ] show how sufficient heating can be produced , either through significant steepening of the inner dm halo density profile ( on which the spike sits ) , or a significantly larger outer spike radius or both . for galaxy - scale dm haloes annihilation heating can balance cooling whenever the inner density profile is steepened to values of @xmath146 , and for cluster - scale haloes when @xmath147 . similarly large values for the outer spike radius , @xmath148 on galaxy scales and @xmath149 on cluster scales , are required for annihilation heating to compensate for the gas cooling . although observationally the inner dm halo profile remains unconstrained on extremely small scales , it may be difficult to produce such steep slopes or large spike radii on the scales required for all galaxies and clusters , in order to effectively influence global properties . another challenge for our model is the requirement that heating and cooling balance needed at all radii as the cooling time for the gas is short for a range of radii specially on group and cluster scales . as developed here self - annihilation dumps all the energy at very small radii and here we assume due to lack of a more complete understanding at the present that all that energy is also thermalized efficiently at small radii . while this could lead to the generation of large entropy inversions in the x - ray gas that are typically not observed ( voit & donahue 2005 ) ; we argue that there are other efficient energy transport processes like thermal conduction available to deposit energy at larger radii ( ruszkowski & begelman 2002 ; ruszkowski , bruggen & begelman 2004 ) . it is useful to point out here that even for the alternative agn feedback that have been proposed the astrophysical process that couple radiation to baryons in the inner regions of galaxies , groups and clusters is poorly understood at present ( mccarthy et al . this is clearly a rich avenue for future work for all currently proposed feedback models . the suppression of cooling flow gas in massive haloes can have a dramatic effect on evolution of the galaxies that reside in them . this is due to the fact that across cosmic time , a growth of a galaxy is dictated by the availability and supply of star forming material . at late times this mostly comes in the form of gas condensing out of the hot halo . hence , any mechanism that suppresses gas cooling ultimately also prevents the galaxy from further star formation . our goal is to investigate under what circumstances the evolving density spikes explored in section [ sec : spikes ] and the resultant heating model described in section [ sec : heating ] can actually shut down star formation in massive galaxies when the full hierarchical evolution of galaxies is taken into account . in a set of three figures , fig . [ fig : lfs1][fig : lfs3 ] , we show the resultant local luminosity function predicted by our galaxy formation model with annihilation heating included . we plot the @xmath150 k - band galaxy luminosity function , for the models ( lines ) and observations for comparison ( symbols with error bars ) . the long - dashed line in each figure shows the conservative default model used in fig . [ fig : heating ] . the remaining lines in each figure illustrate the consequences of different parameter choices for the annihilation heating prescription , with each figure focused on a specific set of parameters that tune the annihilation heating model . as can been seen from all the luminosity function figures , the default model significantly overpredicts the abundance of the brightest galaxies . this is a consequence of inefficient annihilation flux heating for the default model , as seen in fig . [ fig : heating ] . essentially overcooling occurs in the centers of group and cluster systems , leading to excess star formation and overly bright and massive galaxies . note that this is not necessarily a failure of our underlying galaxy formation model - the cooling flow problem has a long history ( see review by fabian 2004 for details ) and failure of the default heating model is simply another manifestation of it ( in the absence of strong enough heating ) . the two additional lines in fig . [ fig : lfs1 ] , dotted - dashed and solid , show the galaxy luminosity function when the inner dm halo slope is steepened to @xmath151 and @xmath152 , respectively ( the default model has @xmath153 , the standard nfw profile ) . an inner slope of @xmath0 appears to be insufficient to produce enough heating , even with the presence of a central density spike . only inner halo slopes of @xmath154 or steeper are able to do this . in the context of current models such steep slopes do not arise naturally in dark matter haloes . a combination of steepening mechanisms needs to operate in a coordinated fashion to achieve these slopes . [ fig : lfs2 ] considers the combination of a steeper inner halo slope , here @xmath155 , and more extreme values for the inner and outer spike radii explored previously in fig . [ fig : heating ] . our choices for the spike radii are made to obtain the correct turnover in the galaxy luminosity function . the required values are , for the inner spike radius @xmath156 times smaller than the default value ( dashed - dotted line ) and for the outer spike radius @xmath157 times larger than the default value ( solid line ) . both parameter sets still over predict the abundance of very bright galaxies . this exercise is repeated in fig . [ fig : lfs3 ] with a more extreme inner slope of @xmath158 . the inner and outer spike radii choices that provide the best fit are now @xmath159 times smaller than the default inner spike radius ( dashed - dotted line ) and @xmath160 times larger than the default outer spike radius ( solid line ) . the brightest galaxies remain overabundant , but less so than for the previous values . in figs . 7 & 8 we plot the heating rate and cumulative heating rate respectively generated by dark matter self - annihilation as a function of radius for a galaxy scale halo and a cluster scale halo for various parameter choices . also marked as arrows for reference in these plots are 3 relevant physical scales : 4 @xmath61 ( inner - most arrow ) ; the outer spike radius ( middle arrow ) and the virial radius of the dm halo ( outer - most arrow ) . the plots emphasize that the equations predict that the heating flux comes from well inside the cooling radius . due to the simplicity of our assumed cooling model ( i.e. that the gas is isothermal ) it is not possible to accurately determine the cooling rate as a function of radius inside individual halos for a direct radial comparison of the heating cooling balance . in the context of our simple model as shown we can however calculate the global heating and cooling rates and compare . these plots also show that the heating from self - annihilation akin to other modes of feedback needs to propagate out to larger radii to be effective . the microphysics of these transport processes is poorly understood at this point . we conclude that for modest parameter choices we are unable to produce galaxies that match observed ones . of course , further combinations of these parameters are possible , however , it is unlikely that self - annihilation is the sole feedback process in galaxy formation . it is plausible that this process operates in addition to supernovae and agn feedback . if indeed neutralino self - annihilations contribute to the energetics of feedback in galaxies and clusters as proposed here , we can expect a range of observational signatures . if the model of annihilation heating developed here operates , we expect the existence of a range of inner dark matter density profile slopes in the centres of galaxies and galaxy clusters . the time evolution explored in this model suggests that the inner density slopes on the scale of tens of parsecs in galaxies are likely to be diverse . observationally , this is an extremely challenging length scale to probe and detect this diversity . studies of the velocity dispersion profiles of the stellar component in the vicinity of the black holes combined with strong lensing offer a potentially viable probe . mapping of rotation curves has also provided constraints on the density profile of the dark matter in the central regions of galaxies , ( see de blok , mcgaugh & rubin and references therein ) however on much larger scales , of the order of kpc , whereas the dm annihilation scenario leaves an imprint on much smaller scales . it does appear on that on kpc scales ( larger scales than relevant for dark matter self - annihilations ) there is compelling evidence for bimodality in the distribution of light ( baryons ) . the nuker group has studied this effect extensively using hubble space telescope data ( gebhardt et al . 1996 ; faber et al . 1997 ; lauer et al . 2002 ) . in a recent paper , combining several hst investigations on the central structure of early - type galaxies they find that the distribution of the logarithmic slopes of the central brightness profiles is bimodal ( lauer et al . 2007 ) . they claim that at the hst resolution limit , most galaxies are either power - law systems , which have steep cusps in surface brightness , or core systems , which have shallow cusps interior to a steeper envelope in the brightness distribution . there is a atrong correlation between the luminosity @xmath161 and inner profile slope , and it has been suggested that this correlation is likely due to core formation by binary bhs during mergers ( ferrarese et al . whether and how this observed bimodality in the surface brightness profiles of the baryonic component reflects the dark matter density profile on the smallest scales is unclear at the present . the physical scale on which dark matter annihilation manifests itself in the case of clusters is predicted to be of the order of @xmath91 kpc ( as shown in the bottom right hand panel of figure 3 ) . in this context , we predict that similarly in clusters there ought to be a diversity of density profile slopes on kpc scales . in clusters that have more complex dynamical histories , the dark matter spike is likely to have been disrupted progressively due to frequent mergers and these density spikes are also expected to have depleted from the annihilation process itself as a result . these growing clusters are systems in which the spike reassembly is most unlikely to occur rapidly . in the context of the self - annihilation feedback picture , these clusters are likely to have density profiles with a central plateau ( akin to the evolution shown in figure 1 ) . since clusters are the most recently assembled structures in the universe , we predict a range of inner density slopes in the central few kpc , some shallower than the predictions of dissipationless simulations and some steeper , depending on their dynamical history . dynamical history coupled with the modification produced due to the presence of the baryonic component ( stars or black holes ) is intricately coupled to the process of dm annihilation as we have shown , and the interplay of these process might dictate the slope of the dark matter density profile in the inner - most regions . observationally , the issue is once again challenging . strong lensing studies of the inner regions of clusters with radial and tangential arcs point to the possible existence of shallower density slopes and perhaps cores on scales of @xmath162 kpc . this is of the order of the scales on which we expect to see signatures of the annihilation process . since strong lensing constrains the total mass as a function of radius , disentangling the effect of the baryons to infer the density profile of the dark matter alone on these scales is difficult . the combination of gravitational lensing and dynamical data is uniquely capable of achieving this . sand et al . ( 2002 ; 2004 ) attempted this for a sample of strong lensing clusters . in more recent work , sand et al . ( 2008 ) study 2 clusters abell 383 and ms2137 - 23 combining strong lensing constraints with stellar velocity dispersion data for the brightest central cluster galaxy . they find that a shallower inner slope is preferred compared to predictions from simulations ( @xmath163 ) for abell 383 for a coarse lensing model . for ms2137 - 23 , no self - consistent model that incorporates strong lensing and the stellar velocity dispersion data was found to be a good fit . we also note that constraints derived from recent chandra observations also suggest shallow inner slopes ( @xmath164 ) on scales of @xmath165 ( vikhlinin et al . 2006 ; zappacosta et al . 2006 ; voigt & fabian 2006 ) . it is however a real challenge to extract robust constraints on the density profile slope at these small radii from observations . meanwhile , simulations that incorporate baryons are likely to improve in resolution in the near future and might offer a powerful test of our predictions . in this section we discuss the details of the agn heating paradigm in order to contrast with our dark matter annihilation model . we describe the mechanism and argue that there is a need for additional sources of feedback that dark matter self - annihilations may well provide . the proposed physical mechanisms for agn driven feedback are expected to occur in two modes : the ` quasar mode ' , where mergers trigger fuelling to central black holes activating an episodic , bright quasar phase accompanied by large - scale outflows ( di matteo et al . 2005 ; hopkins et al . reference ) , and the ` radio mode ' , which refers to steady feedback from low - level agn activity ( croton et al . 2006 ; bower et al . the need for these processes arise from the over - cooling problem that occurs on a range of mass scales , from galaxy groups to clusters ( somerville & primack 1999 ; cole et al . 2001 ; kauffamnn et al . 1999 ; benson et al . 2003 ) . agn feedback in the quasar mode likely occurs during the epochs of efficient cold gas feeding to the central black holes via a thin accretion disk , at high accretion rates ranging between @xmath166 . this phase is short - lived and the thermal coupling of agn energy is fairly weak ( at @xmath167 5% level ) . in this mode the agn - driven wind removes residual gas at the end of the merger , leading to suppression of subsequent star formation and self - regulated bh growth that reproduces the observed @xmath168 relation ( springel et al . 2005 ) . however , for a typical massive galaxy in the local universe a quasar event was long ago in its history . such galaxies instead host bhs that are accreting at much lower rates ; in fact most spend much of their lifetime in these radiatively inefficient states . radio activity is associated with these low accretion rate states and radio jets are seen in many massive galaxies . the coupling of jet energy with host gas can be very efficient and models of effervescent heating with a combination of sound waves , weak shocks and bubbles can heat a large fraction of the gas in clusters ( ruszkowski et al . 2004 ; churazov et al . 2001 ; bruggen et al . 2005 ) and produce features that match x - ray observations ( fabian et al . 2005 ) . a detailed exploration of the theoretical consequences of the steady radio - mode feedback from low luminosity agn has been presented in croton et al . this was done using the same semi - analytic model used in this paper , but with dm annihilation heating replaced with radio - mode heating . the authors of croton et al . showed that for a set of energetically and observationally plausible parameters such a model could simultaneously explain : ( i ) the low observed mass drop - out rate in cooling flows ; ( ii ) the exponential cut - off at the bright end of the galaxy luminosity function ; and ( iii ) the fact that the most massive galaxies tend to be bulge - dominated systems in clusters and found to contain systematically older stars than lower mass galaxies . in a recent paper best et al . ( 2007 ) study a sample of radio - loud agn in nearby groups and clusters from the sloan digital sky survey ( sdss ) . using observational estimates of the mechanical output of radio jets , they estimate the time - averaged heating rate associated with recurrent radio source activity for all group and cluster galaxies . they find that within the cooling radius the radio - mode heating associated with galaxy groups and low mass clusters is sufficient to offset the cooling flow from the extended hot halo . in the most massive brightest cluster galaxy systems , however , radio mode heating alone is not enough . they conclude that other processes acting in massive clusters must also be contributing to the suppression of cooling flow gas . importantly for this work , although agn appear to be making an observable contribution to the evolution of gas dynamics in dark matter haloes , alone they only comprise part of the full physical picture . the presence of smbhs at the centres of massive systems that drive agn winds also provide enhancement in the annihilation rate of dm that , under the right circumstances , can produce sufficient heating flux to arrest cooling gas . composite dm annihilation and agn heating models will be a natural extension of our work and there is need for all feedback mechanisms to be better understood . one of the current challenges for all models including the self - annihilation proposed here is developing a better understanding of the micro - physics of how to thermally couple to the gas at all radii where the heating and cooling are required to balance for feedback to be effective . using the millennium run n - body simulation coupled with a sophisticated model of galaxy formation that includes the heating of cooling flow gas through neutralino annihilation , we have shown that : * density spikes that support the annihilation flux at the levels required to offset cooling flows are stable enough over long enough time - scales to maintain a reasonably constant heating source ( fig . [ fig : merger_rate ] ) . * models that appear to be extreme at the present time ( given our current understanding of dm density profiles ) are required to produce enough heating flux to offset the predicted cooling rates . to obtain the required heating rates we either need to steepen the inner dm density slope to values @xmath169 or increase the outer spike radius . for galaxy sized haloes the outer spike radius is required to be of the order of @xmath148 , and for cluster haloes @xmath170 ( fig . [ fig : heating ] and [ fig : lfs1 ] ) . * the efficiency of heating in the dm annihilation model scales with halo mass ( and circular velocity ) , therefore this mechanism does provide preferential suppression of star formation in more massive haloes as required to explain current observations of the luminosity function ( fig . [ fig : heating ] ) . in this treatment , we have assumed that the mass of central black holes grows significantly at each merger ( as the solution for adiabatic growth in valid only in the limit @xmath171 . we note here that an alternative source of annihilations may be provided by mini - spikes around inter - mediate mass black holes as suggested by bertone , zentner & silk ( 2005 ) . it is clear that feedback and energy injected into the inter - stellar medium of galaxies and the intra - cluster medium is a complex process , and that a combination of astrophysical processes , including the one explored here , are likely at play . one of the key uncertainties in the model explored in this paper arises from the fact that we lack an understanding of the physics through which the annihilation flux is expected to couple with the cooling hot halo gas . while we have discussed some possibilities , like coulomb collisions , brehmstrahlung and synchrotron radiation . colafrancesco et al . ( 2007 ) have explored these in more detail for the case of the heating of gas in the coma cluster due to dm annihilations . experimental confirmation of super - symmetery from the lhc at cern might throw new light on the viability and likely couplings for the neutralino . additionally while following the cumulative heating history is very challenging to do it is needed to really understand the detailed energetics of the gas . there is incontrovertible evidence for the presence of copious amounts of dm on all scales in the universe , so dm annihilation is an inescapable phenomenon . however , how much energy is released in the process and how efficiently it couples to the baryonic component are unclear . pn acknowledges the use of the yale its high performance computing facility . dc acknowledges support from nsf grant ast00 - 71048 . the millenium simulation was carried out by the virgo supercomputing consortium at the computing center of the max - plank society in garching . the simulation data is publicly available at `` http://www.mpa-garching.mpg.de/millenium '' . the authors thank david spergel , joel primack , david merritt , douglas finkbeiner and sergio colafrancesco for useful comments on the manuscript . ferrarese , l. , cote , p. , blakeslee , j. , mei , s. , merritt , d. & west , m. , 2006 , in `` black holes : from stars to galaxies - across the range of masses '' , proceedings iau symposium no . 238 , eds . v. karas & g. matt . , astro - ph/0612139

we consider the statistical experiment of functional linear regression ( flr ) . furthermore , we introduce a white noise model where one observes an it process , which contains the covariance operator of the corresponding flr model in its construction . we prove asymptotic equivalence of flr and this white noise model in lecam s sense under known design distribution . moreover , we show equivalence of flr and an empirical version of the white noise model for finite sample sizes . as an application , we derive sharp minimax constants in the flr model which are still valid in the case of unknown design distribution . .

1211.4742

we consider the statistical problem of functional linear regression ( flr ) . in its standard version , one observes the data @xmath0 where @xmath1 are i.i.d . random variables taking their values in @xmath2)$ ] , that is , the set consisting of all continuous functions on the interval @xmath3 $ ] , and @xmath4 with @xmath5 where @xmath6 denotes the @xmath7)$]-inner product throughout this work . the i.i.d . error variables @xmath8 are assumed to be centered and normally distributed with the variance @xmath9 . moreover , all @xmath10 are independent . the goal is to estimate the regression function @xmath11)$ ] . in general , we allow for such a structure of the function class @xmath12 which does not determine @xmath13 up to finitely many real - valued parameters . thus we consider a nonparametric estimation problem . moreover we assume that @xmath14 and @xmath15 \leq c_{x,0 } \exp(-c_{x,1 } x^{c_{x,2}})$ ] for all @xmath16 and some finite constants @xmath17 where @xmath18 , @xmath19 denotes the @xmath20)$]-norm of some element of that space . thus the tails of the design distribution are restricted . such conditions are usual in nonparametric regression problems . the flr model has obtained considerable attention in the statistical community during the last years , which is reflected in the large amount of literature on this topic . various of estimation procedures have been proposed to make the regression function @xmath13 empirically accessible ( see , e.g. , @xcite ) . the minimax convergence rates in flr are investigated , for example , in @xcite . in @xcite , adaptive estimation in flr is considered . generalizations of flr are discussed in @xcite . a central limit theorem for flr is derived in @xcite . in @xcite , practical applications of flr in the field of medical statistics are described ; the authors consider two real data sets on primary biliary cirrhosis and systolic blood pressure . for a comprehensive introduction to the field of functional data analysis in general , see @xcite . in order to compare two statistical models , it is useful to prove asymptotic equivalence between those models . for the basic concept and a detailed description of this strong asymptotic property , we refer to @xcite and @xcite . also , a review on this topic is given in the following section . as an important feature , if two models @xmath21 and @xmath22 are asymptotically equivalent , then @xmath21 adopts optimal convergence rates and sharp asymptotic constants with respect to _ any _ bounded loss function from model @xmath22 and vice versa . thus , the theory of asymptotic equivalence does not only capture special loss functions such as the mean integrated squared error ( mise ) or the pointwise mean squared error ( mse ) but includes various types of semi - metrics between the estimator and the target function @xmath13 and also addresses the estimation of characteristics of @xmath13 , such as its support or its mode . furthermore , superefficiency phenomena also coincide in both models when considering subclasses @xmath23 of the target parameter space @xmath12 . in particular , research has focussed on proofs of asymptotic equivalence of experiments where @xmath24 i.i.d . data are observed , whose distribution depends on some parameter @xmath25 , and experiments where @xmath13 occurs in the drift of an empirically accessible it process . for instance , nussbaum @xcite considers an asymptotically equivalent white noise model for density estimation , while brown and low @xcite introduce such a model for nonparametric regression . in recent related literature on regression problems , carter @xcite studies the case of unknown error variance , and reiss @xcite extends asymptotic equivalence to the multivariate setting . returning to model ( [ eq:1.1 ] ) , we suppose that the nuisance parameters @xmath26 and @xmath27 , that is , the distribution of the @xmath28 , are known . that allows us to exclude those quantities from the parameter space of the experiment and to fully concentrate on the estimation of @xmath13 . this condition is also imposed in most papers dealing with asymptotic equivalence for nonparametric regression experiments . the work of @xcite represents an exception where the corresponding white noise model becomes more difficult and , apparently , less useful to derive adoptable asymptotic properties . with respect to asymptotic equivalence , we restrict our consideration to the case of known @xmath27 . however , in section [ s : new ] , we will show that the sharp minimax asymptotics with respect to the mise are extendable to the case of unknown design distribution . the main purpose of the current work is to prove asymptotic equivalence of model ( [ eq:1.1 ] ) and a statistical inverse problem in the white noise setting . that latter model is described by the observation of an it process @xmath29 , @xmath30 $ ] , @xmath31 , driven by the stochastic differential equation @xmath32(t)\,dt + n^{-1/2 } \sigma \,dw(t ) , \ ] ] where @xmath33 denotes a standard wiener process on the interval @xmath3 $ ] , and @xmath34 denotes a linear operator mapping from the hilbert space @xmath7)$ ] to itself . these models are also widely studied in mathematical statistics ( see , e.g. , @xcite and @xcite ) . they have their applications in the field of signal deblurring and econometrics . we will concentrate on a specific version of model ( [ eq:1.1.1 ] ) where @xmath34 is equal to the unique positive symmetric square root @xmath35 of the covariance operator @xmath36 , that is , @xmath37 and @xmath38 for any @xmath39)$ ] . thus , the observation @xmath29 , @xmath30 $ ] , is defined by @xmath31 and @xmath40(t)\,dt + n^{-1/2 } \sigma \,dw(t ) .\ ] ] in @xcite , the authors remark on the similarity of models ( [ eq:1.1 ] ) and ( [ eq:1.1.2 ] ) . in the current paper , we will rigorously establish asymptotic equivalence between those models . as an interesting feature , additional observation of the data @xmath41 would be redundant in model ( [ eq:1.1.2 ] ) . all information about the design points is recorded by @xmath36 in ( [ eq:1.1.2 ] ) . therefore , all what is observed in the corresponding white noise experiment is the process @xmath29 , @xmath42 $ ] . after the general introduction to the property of asymptotic equivalence as used in the current paper in section [ s1.0 ] , we will first prove ( nonasymptotic ) equivalence of model ( [ eq:1.1 ] ) and an empirical version of model ( [ eq:1.1.2 ] ) where @xmath36 is replaced by a noisy counterpart in section [ s2 ] . in section [ s3 ] , we prove asymptotic equivalence of ( [ eq:1.1 ] ) and ( [ eq:1.1.2 ] ) under some additional technical conditions . in section [ s : new ] , we show that the sharp lower bound which follows from the results of the previous section can be attained by specific estimators in the realistic case of unknown design distribution . a discussion of the findings and their conclusions are provided in section [ s5 ] . to recall the definition of asymptotic equivalence , we consider two ( sequences of ) statistical experiments @xmath43 , @xmath44 , with a joint parameter space @xmath12 , which may depend on @xmath24 . the lecam distance between @xmath45 and @xmath46 is defined by @xmath47 where @xmath48 is the total variation distance , and @xmath49 denotes the collection of so - called transitions ( see @xcite and @xcite for their exact definition ) . the statistical experiments @xmath45 and @xmath46 are called asymptotically equivalent if @xmath50 converges to zero as @xmath51 , while they are called equivalent if @xmath52 for all @xmath24 . in the framwork of our note , we will not use that general definition of ( asymptotic ) equivalence but our proofs lean on following sufficient conditions for these properties : we consider the following sufficient condition for asymptotic equivalence of @xmath45 and @xmath53 : we define the sets @xmath54 , @xmath44 , @xmath55 , which contains all real - valued integrable random variables @xmath56 on the domain @xmath57 satisfying @xmath58 a.s . thus any kind of bounded loss functions are captured by the classes @xmath54 so that the expectation @xmath59 with respect to the distribution @xmath60 describes an arbitrary bounded and normalized statistical risk for estimating the parameter @xmath13 under the observation scheme @xmath61 . now we define two sequences @xmath62 , @xmath63 , of @xmath64-measurable mappings from @xmath65 to @xmath66 . as an essential condition , the @xmath67 must not depend on @xmath13 . hence , @xmath67 may be interpreted as a transformation of the data from an observation contained in the space @xmath57 to an observation which lies in @xmath68 . thus a statistician who intends to construct an estimation procedure for @xmath13 may always apply this transformation @xmath67 to an observation @xmath69 . then we obtain asymptotic equivalence of @xmath70 and @xmath46 when we can show the existence of such transformation sequences @xmath62 , @xmath63 , so that @xmath71 as @xmath72 for all @xmath63 . accordingly , we have equivalence if the left - hand side in ( [ eq : aseq ] ) equals @xmath73 for any @xmath24 . intuitively speaking , after transforming the data drawn from model @xmath45 according to @xmath74 , the distance between any bounded statistical risk in model @xmath46 on one hand and for the transformed data from model @xmath45 becomes small for large @xmath24 or is equal to zero for any @xmath24 , respectively . the same condition must also hold true when exchanging @xmath75 and @xmath76 . in the specific framework of our note , we assume , in addition , that all transformations @xmath67 must not depend on the nuisance parameter @xmath26 . that compensates the unrealistic condition of known @xmath26 . in particular , @xmath26 is not used to transform the data or to construct decision procedures or estimators . therefore , our results also addresses the case of unknown @xmath26 . nevertheless , @xmath26 must be viewed as uninteresting for the statistician , that is , it must not explicitly occur in the loss functions @xmath77 . thus , the problem of estimating @xmath26 is not covered by our approach . assume that the experiment @xmath46 describes the observation of @xmath78 for @xmath79 in experiment @xmath45 where @xmath80 is a sufficient statistic for @xmath13 in experiment @xmath45 . then @xmath45 and @xmath46 are statistically equivalent ( i.e. , their lecam distance vanishes ) and , hence , asymptotically equivalent . that assertion holds true whenever the experiments are polish spaces . this criterion is satisfied as all probability spaces considered in the current work are @xmath81 , @xmath2)$ ] , @xmath7)$ ] and some set products of those classes ( see lemma 3.2 in @xcite ) . if some experiments @xmath45 and @xmath46 on one hand , and @xmath46 and @xmath82 on the other hand , are ( asymptotically ) equivalent , then @xmath45 and @xmath82 are ( asymptotically ) equivalent , too . also , ( asymptotic ) equivalence of @xmath45 and @xmath46 is a symmetric relation between the experiments . assume that some experiments @xmath45 and @xmath46 may be decomposed into two independent experiments @xmath83 and @xmath84 , @xmath63 , respectively . moreover , we suppose that the experiments @xmath85 and @xmath86 on one hand and the experiments @xmath87 and @xmath88 on the other hand are ( asymptotically ) equivalent . then , the combined experiments @xmath45 and @xmath46 are ( asymptotically ) equivalent as well . now , @xmath21 denotes the underlying experiment of the flr model ( [ eq:1.1 ] ) ; it is defined by @xmath89)^{(n ) } \times\mathbb{r}^{(n)}$ ] , @xmath90 denotes the borel @xmath91-algebra when considering the uniform metric on the functional components and the euclidean on the real - valued components of @xmath92 . the corresponding probability measures @xmath93 are well defined by the assumptions of the model ( [ eq:1.1 ] ) . the parameter space @xmath94)$ ] will be specified later . still , the observations @xmath0 may be viewed as random variables having their domain on some basic probability space @xmath95 . we define the linear covariance operator @xmath96 ) \to l_2([0,1])$ ] by @xmath97 ) .\ ] ] writing @xmath98 , we realize that @xmath36 is a hilbert schmidt integral operator where @xmath99 by the cauchy schwarz inequality and the tail condition imposed on the distribution of @xmath100 . hence @xmath36 is a continuous and compact operator . we have @xmath101 for all @xmath102 $ ] so that the operator @xmath36 is self - adjoint . furthermore , it is positive ; that is , by fubini s theorem we have @xmath103 of the separable hilbert space @xmath7)$ ] which consists of eigenfunctions of @xmath36 . the corresponding eigenvalues are denoted by @xmath104 . the sequence @xmath105 converges to zero and may be viewed as monotonously decreasing without loss of generality . those results are also used , for example , in @xcite . furthermore , for @xmath36 as for any compact self - adjoint positive operator from @xmath7)$ ] to itself , there exists a unique compact self - adjoint positive operator @xmath106 from @xmath7)$ ] to itself such that @xmath107 ; then @xmath106 is called the square root of @xmath36 . we have @xmath108 for any @xmath109 . we may define an empirically accessible version @xmath110 of @xmath36 by replacing the expectation by the average ; more precisely , we have @xmath111 ) .\ ] ] thus , @xmath110 may be viewed as the operator @xmath36 when @xmath27 equals the uniform distribution on the discrete set @xmath112 . therefore , all properties derived for @xmath36 in the previous paragraph can be taken over to @xmath110 . in particular , @xmath113 , integer @xmath109 , denotes the orthonormal basis of the eigenfunctions of @xmath110 with the eigenvalues @xmath114 . now we consider the conditional probability density @xmath115 of the data @xmath116 given the design functional observations @xmath41 in model ( [ eq:1.1 ] ) . this density shall be understood with respect to the @xmath24-dimensional lebesgue measure . we derive that @xmath117 with the vectors @xmath118 and @xmath119 . moreover , denotes the euclidean norm . expanding @xmath120)$ ] in the orthonormal basis @xmath121 gives us that @xmath122 we impose the following condition on the distribution @xmath27 : @xmath123 & = & 0 , \qquad \mbox{for any deterministic linear subspace } \nonumber \\[-8pt ] \\[-8pt ] & & \hspace*{32pt}l \subseteq l_2([0,1])\mbox { with } \dim l < \nonumber\end{aligned}\ ] ] intuitively , this assumption provides that the probability mass of the @xmath28 fills the whole of @xmath7)$ ] . somehow , ( [ eq : cond_x ] ) is the functional data analog for continuity of a distribution of some real - valued random variables . it is satisfied when we take an appropriate gaussian process for @xmath124 , for instance . condition ( [ eq : cond_x ] ) yields that the linear space generated by @xmath41 is @xmath24-dimensional almost surely . otherwise , at least one of the @xmath28 must be included in the linear hull of the other design variables . according to ( [ eq : cond_x ] ) that occurs with probability zero when employing the conditional probability measure given the data @xmath125 . finally , applying the expectation , we obtain the desired result for the unconditional distribution . we realize that the range of @xmath110 is included in the linear hull of @xmath41 . by definition , @xmath113 is contained in that @xmath24-dimensional space whenever @xmath126 . as the @xmath113 form an orthonormal basis at most @xmath24 of the eigenvalues @xmath114 are nonvanishing . furthermore , the linear independence of the @xmath41 implies that the functions @xmath127 , @xmath128 , are linearly independent , too , so that the range of @xmath110 is equal to the linear hull of @xmath129 . clearly , the range of @xmath110 also coincides with the linear hull of all @xmath113 with @xmath126 , from what follows @xmath130 for all @xmath131 and @xmath132 . also , we have @xmath126 for @xmath131 and @xmath133 for @xmath134 . hence , ( [ eq : s2.n.2 ] ) leads to the representation @xmath135 for all @xmath131 . equation ( [ eq : s2.n.3 ] ) is equivalent to the system of linear equations @xmath136 with the vector @xmath137 and the matrix @xmath138 with the components @xmath139 , @xmath140 . then the conditional density @xmath141 as in ( [ eq : s2.n.1 ] ) may be written as @xmath142 we consider that the @xmath143th component of the matrix @xmath144 is equal to @xmath145 thus @xmath144 is a diagonal matrix containing @xmath146 as its @xmath147th component . we denote the diagonal matrix having @xmath148 as its @xmath147th component by @xmath149 . obviously , @xmath149 is invertible , and we define @xmath150 . we have @xmath151 where @xmath152 denotes the identity matrix . also , this yields that @xmath153 and that @xmath154 is an orthogonal matrix . thus , @xmath155 for any vector @xmath156 . equality ( [ eq : s2.n.4 ] ) provides that @xmath157 referring to the notation of ( [ eq : aseq ] ) , we consider the expectation @xmath158 where @xmath159 . we derive that @xmath160 \\[-8pt ] & = & e { \int\cdots\int}r_{1,n,\theta}(x_1,\ldots , x_n;\mathbf{a}\mathbf{z } ) ( 2\pi)^{-n/2 } \nonumber\\ & & \hphantom{e { \int\cdots\int } } { } \times\sigma^{-n } \exp \bigl ( - \|\mathbf{z } - \mathbf{d } \mathbf{f}\| ^2 / ( 2\sigma^2 ) \bigr ) \ , dz_1 \cdots dz_n \nonumber\\ & = & e r_{1,n,\theta}(\mathbf{x},\mathbf{a}\mathbf{z } ) , \nonumber\end{aligned}\ ] ] where @xmath161 denotes a vector consisting of independent normally distributed random variables where @xmath162 has the mean @xmath163 and the variance @xmath9 , conditionally on the @xmath91-algebra generated by @xmath164 . therefore , the @xmath162 may be represented as @xmath165 where @xmath166 are i.i.d . @xmath167-distributed random variables . the @xmath8 are independent of the @xmath26-algebra generated by @xmath164 . we have applied the integral transformation @xmath168 where @xmath169 due to the orthogonality of @xmath154 . note that the sign of the eigenfunctions @xmath170 may still be chosen ; we can arrange that @xmath171 . now we define the statistical experiment @xmath22 with the same parameter space @xmath12 as @xmath21 , @xmath172 and @xmath173 as the probability measure generated by the random variable @xmath174 with @xmath175 as in ( [ eq : s2.n.6 ] ) . in the notation of section [ s1.0 ] , paragraph ( i ) , we use the mapping @xmath176 defined by @xmath177 , @xmath178)^{(n)}$ ] , @xmath179 , as the data transformation from @xmath22 to @xmath21 . by definition , the matrix @xmath154 does not depend on the parameter @xmath13 but only on the data @xmath41 and the known orthonormal basis @xmath180 . also , it does not depend on @xmath26 as requested in the previous section . we have already derived that @xmath154 is an orthogonal matrix so that @xmath181 is a bijective mapping from the set @xmath182)^{(n ) } \times\mathbb{r}^n$ ] to itself . hence , its reverse mapping @xmath183 may be used as the data transformation @xmath184 . then , according to ( [ eq : aseq ] ) , we have proved the following lemma . [ l:1 ] under condition ( [ eq : cond_x ] ) , the statistical experiments @xmath21 and @xmath22 are equivalent . the random variables @xmath8 , integer @xmath185 , as occurring in ( [ eq : nn.1 ] ) , may be represented by @xmath186 where @xmath187 denotes a standard wiener process on @xmath3 $ ] which is independent of @xmath164 . we deduce that the @xmath188 are an independent sequence of @xmath167-distributed random variables . moreover , they are independent of @xmath41 although @xmath113 depends on these design variables . that can be shown via the conditional characteristic function of @xmath189 given @xmath129 ; that is , @xmath190 & = & \exp \biggl(- \frac12 \sigma^2 \biggl\|\sum_{j=1}^\infty s_j \hat{\varphi}_j \biggr\| _ 2 ^ 2 \biggr ) \\ & = & \exp \biggl(- \frac12 \sigma^2 \sum_{j=1}^\infty s_j^2 \biggr)\end{aligned}\ ] ] for all real - valued sequences @xmath191 with @xmath192 . applying the expectation to the above equality , the unconditional characteristic function of @xmath193 turns out to coincide with the conditional one . we have @xmath194 for all @xmath131 where @xmath195 , @xmath30 $ ] , denotes an it process satisfying @xmath196(t)\,dt + \sigma\ , dw(t ) , \ ] ] and @xmath197 . the differential @xmath198 shall be understood in the it sense . now we define the statistical experiment @xmath199 with a completely functional observation structure . we fix that @xmath200)^{(n+1)}$ ] with the corresponding borel @xmath26-algebra @xmath201 . the probability measure @xmath202 is defined via the observation of @xmath164 as in @xmath22 and the it process @xmath195 , @xmath30 $ ] , as defined in ( [ eq : s2.n.7 ] ) . the definition ( [ eq : s2.n.6.1 ] ) of @xmath203 can be extended to @xmath134 straightforwardly . as @xmath204 , we obtain that @xmath205 moreover , @xmath195 is uniquely determined by the @xmath203 for all integers @xmath109 and vice versa . that can be seen as follows : @xmath206}(s)\,dz(s ) = \sum_{j=1}^\infty\langle 1_{[0,t ] } , \hat{\varphi}_j \rangle z_j\ ] ] for all @xmath30 $ ] where the infinite sum must be understood as an @xmath207-limit . that seems to cause some troubles as we only observe one element of the probability space . however , convergence in probability implies almost sure convergence of a subsequence so that @xmath195 is fully accessible by the observation of all @xmath203 . on the other hand , by a similar argument , all @xmath203 are accessible ( in practice , that means approximable arbitrarily precisely ) by a trajectory of the process @xmath208 . hence the data set @xmath209 is independent of the @xmath210 , conditionally on the @xmath26-algebra generated by @xmath164 . furthermore , the distribution of the @xmath203 , @xmath211 , does not depend on the target parameter @xmath13 so that @xmath203 , for @xmath134 , does not contain any information about @xmath13 . we conclude that @xmath212 is a sufficient statistic for the observation scheme in the experiment @xmath199 . we can utilize result ( ii ) from section [ s1.0 ] in order to prove equivalence of the experiments @xmath22 and @xmath213 . considering paragraph ( iii ) from section [ s1.0 ] , we may establish equivalence of the experiments @xmath21 and @xmath199 . this result is presented in the following theorem . [ t:1 ] under condition ( [ eq : cond_x ] ) , the flr statistical experiment @xmath21 is equivalent to the model @xmath199 where one observes @xmath164 and the it process @xmath195 , @xmath214 $ ] , as defined in ( [ eq : s2.n.7 ] ) . in the previous section , we have derived a statistically equivalent white noise model for the flr problem . however , the it process @xmath208 in ( [ eq : s2.n.6.1 ] ) contains the noisy operator @xmath110 in its construction . in the current section , we will replace it by the covariance operator @xmath36 . for that purpose , we split the original experiment @xmath21 into two independent parts @xmath215 and @xmath216 where @xmath215 is based on the observation of the data @xmath217 , @xmath218 , and @xmath219 consists of the residual data @xmath217 , @xmath220 . the selection of the integer parameter @xmath221 is deferred . the strategy of splitting the sample in the current context leans on @xcite . applying theorem [ t:1 ] to each of the experiments @xmath222 , @xmath63 , we obtain equivalence of @xmath223 and the experiments @xmath224 for @xmath63 . therein , @xmath225 is defined by the observation of @xmath226 and the it process @xmath227 , @xmath214 $ ] , specified by @xmath228 and @xmath229(t)\,dt + \sigma\ , dw_1(t ) , \ ] ] and accordingly @xmath230 is defined by the observation of @xmath231 and the it process @xmath232 , @xmath233 $ ] , specified by @xmath234 and @xmath235(t)\,dt + \sigma\ , dw_2(t ) .\ ] ] furthermore , @xmath236 , @xmath63 , denotes the empirical covariance operator constructed by the data @xmath237 and @xmath238 , respectively . also note that @xmath239 and @xmath240 are two independent standard wiener processes . using criterion ( iv ) in section [ s1.0 ] , the experiment @xmath241 , which combines the independent experiments @xmath225 and @xmath230 , we deduce that @xmath241 and @xmath21 are equivalent . from the experiment @xmath225 we construct an estimator @xmath242 for @xmath13 . we define that @xmath243(t)\,dz_1(t ) \varphi_k , \ ] ] where @xmath34 is an integer - valued smoothing parameter still to be selected . condition ( [ eq : cond_x ] ) guarantees that all @xmath244 are positive since , otherwise , @xmath245 would yield that @xmath246 a.s . so that @xmath124 would lie in the linear hull of @xmath247 . thus the estimator @xmath242 is well defined . we introduce the data transformation @xmath248 where @xmath249(t)\,dt\\ & & \hspace*{62pt } \qquad \quad { } + ( n - m)^{1/2 } \int_0^\cdot[\gamma ^{1/2 } \hat{\theta}_1](t)\,dt \biggr ) .\end{aligned}\ ] ] the transformation is fully accessible by the data drawn from the experiment @xmath225 and the assumed knowledge of the distribution of @xmath164 . we set @xmath250 where @xmath251 ) \times c_0([0,1 ] ) \times c_0^{n - m}([0,1 ] ) \times c_0([0,1])$ ] and @xmath252 is the corresponding borel @xmath26-algebra . the data structure of @xmath241 is represented by @xmath253 when inserting the data set as an argument of the mapping @xmath254 . note that @xmath255 may be inserted in the definition of the estimator @xmath242 . the integral occurring in the definition of @xmath242 is not defined for all continuous functions @xmath256 but for almost all trajectories of @xmath257 . for the other negligible trajectories the integral may conventionally be put equal to zero to make the mapping @xmath254 well defined on the whole of its domain . we define the experiment @xmath258 where ones observes the data @xmath259 where the data @xmath260 are obtained under experiment @xmath241 . the experiment @xmath261 is defined on the probability space @xmath262 . considering the definition of @xmath254 , we realize that the shift contained in the forth component is still available in the experiment @xmath258 as the other components are kept . therefore , @xmath254 is an invertible transformation so that the experiments @xmath241 and @xmath258 are equivalent . in the experiment @xmath258 , the component @xmath263 is still an it process conditionally on @xmath264 . now we introduce the experiment @xmath265 with @xmath266 where one observes the data @xmath267 and the it process @xmath268 , @xmath30 $ ] with @xmath269 and @xmath270(t)\,dt + \sigma\ , dw_2(t ) .\ ] ] in the notation of section [ s1.0 ] , we consider that @xmath271 \\[-8pt ] & & \qquad \leq 2 e \biggl\{1 - \exp \biggl(- \frac1{2\sigma^2 } \|\delta_{5,6}\| _ 2 ^ 2 \biggr ) \biggr\}^{1/2}\nonumber \\ & & \qquad \leq 2 \biggl\{1 - \exp \biggl ( - \frac1{2\sigma^2 } e \|\delta_{5,6}\|_2 ^ 2 \biggr ) \biggr\}^{1/2 } , \nonumber\end{aligned}\ ] ] where @xmath272 therein , we have used girsanov s theorem , @xmath273 for @xmath274 as @xmath275 , the bretagnolle huber inequality and jensen s inequality in the last step . now we study the expectation occurring in ( [ eq:4.1.t ] ) by parseval s identity with respect to the basis @xmath276 and the orthogonal expansion of @xmath277 with respect to @xmath278 where @xmath276 denotes the eigenfunctions of @xmath279 and @xmath280 the corresponding eigenvalues . @xmath281 where we have used the cauchy schwarz inequality for sums and the elementary inequality @xmath282 for all @xmath283 . therein @xmath284 is still to be selected . also , the independence of @xmath279 and @xmath242 has been utilized . then , we apply the cauchy schwarz inequality with respect to the discrete random variable @xmath285 satisfying @xmath286 = |\langle\varphi_j , \hat{\varphi}_{k,2}\rangle|^2 $ ] for all integers @xmath287 and some fixed integer @xmath185 , conditionally on @xmath238 . we conclude that @xmath288 we consider that @xmath289 \nonumber\\ & \leq & \mathrm{const.}\cdot(n - m)^{-1 } \bigl(c_j^2 \lambda_j + c_{x,0 } \exp(-c_{x,1 } c_j^{c_{x,2 } } /2 ) \bigr ) \nonumber\end{aligned}\ ] ] for @xmath24 sufficiently large where the sequence @xmath290 remains to be determined . in order to obtain those results , we impose the following : [ condx ] we assume that condition ( [ eq : cond_x ] ) holds true ; @xmath291 for all integer @xmath109 and some @xmath292 ; @xmath293 ; @xmath294 \leq c_{x,0 } \exp(-c_{x,1 } x^{c_{x,2}})$ ] for all @xmath16 and some finite constants @xmath17 . condition [ condx ] imposes a polynomial lower bound on the sequence of the eigenvalues of @xmath36 . this assumption is very common in flr ( see , e.g. , @xcite ) . when condition [ condx ] is fixed the underlying inverse problem can be viewed as a moderately ill - posed problem unlike severely ill - posed problems where exponential decay of the eigenvalues occurs . condition [ condx ] also corresponds to the deconvolution setting with ordinary smooth error densities in the related field of density estimation based on contaminated data . as an example for a stochastic process which satisfies condition [ condx ] , we mention the random variables @xmath295 where the @xmath296 , integer @xmath185 , form an arbitrary orthonormal basis of @xmath7)$ ] ; the @xmath297 are i.i.d . real - valued centered random variables with a continuous distribution which is concentrated on some compact interval , and @xmath298 . we stipulate that @xmath299 . easy calculations yield that the coefficients @xmath300 and @xmath296 are the eigenvalues and the eigenvectors of the corresponding covariance operator @xmath36 , respectively . stipulating that the sequence @xmath301 is bounded above ( as satisfied , e.g. , by the fourier polynomials ) , we can show that condition [ condx ] is fulfilled . in particular , the random variable @xmath302 is continuously distributed for any @xmath303)\setminus \{0\}$ ] since @xmath304 for at least one integer @xmath185 so that the distribution of @xmath302 is just the convolution of an absolutely continuous distribution and some other distribution ; hence the distribution of @xmath305 has a lebesgue density so that condition ( [ eq : cond_x ] ) can be verified . all other assumptions contained in condition [ condx ] can easily be checked . another even more important example for design distributions are the gaussian processes @xmath306 , @xmath42 $ ] , where @xmath187 denotes a standard wiener process and @xmath26 is a sufficiently smooth function which is bounded from above and below by positive constants . these processes satisfy condition [ condx ] as well where @xmath307 . the decay condition can be verified via the famous reflection principle of wiener processes . returning to the investigation of an upper bound on @xmath308 , the following inequality is evident : @xmath309 \\[-8pt ] & & { } \times \sum _ { j=1}^\infty j^{-\gamma } \bigl(c_j^2 \lambda_j + c_{x,0 } \exp ( -c_{x,1 } c_j^{c_{x,2 } } ) \bigr)^{1/2 } . \nonumber\end{aligned}\ ] ] we deduce that @xmath310(t)\,dz_1(t ) \biggr|^2 \nonumber\\ & & \qquad = 1_{\{j ' > k\ } } |\langle\varphi_{j ' } , \theta\rangle |^2 \nonumber\\ & & \qquad \quad { } + 1_{\{j'\leq k\}}\nonumber\\ & & \hphantom { { } + } \qquad \quad { } \times e \biggl|\lambda_{j'}^{-1 } \langle \hat{\gamma}_1 \theta,\varphi_{j'}\rangle- \langle\theta,\varphi_{j ' } \rangle+ \sigma m^{-1/2 } \lambda_{j'}^{-1 } \int[\hat{\gamma}_1^{1/2 } \varphi_{j'}](t)\,dw_1(t ) \biggr|^2 \nonumber\\ & & \qquad = 1_{\{j ' > k\ } } |\langle\varphi_{j ' } , \theta\rangle |^2 \nonumber \\[-8pt ] \\[-8pt ] & & \qquad \quad { } + 1_{\{j'\leq k\ } } \lambda_{j'}^{-2 } \{e | \langle\hat{\gamma}_1 \theta,\varphi_{j'}\rangle- \langle\gamma \theta,\varphi_{j ' } \rangle |^2 + \sigma^2 m^{-1 } e \|\hat{\gamma } _ 1^{1/2 } \varphi_{j ' } \|_2 ^ 2 \}\nonumber \\ & & \qquad \leq 1_{\{j ' > k\ } } |\langle\varphi_{j ' } , \theta\rangle & & \qquad \quad { } + 1_{\{j'\leq k\ } } \lambda_{j'}^{-2 } \{m^{-1 } \|\theta\|_2 ^ 2 e \| x_1\|_2 ^ 2 |\langle x_1,\varphi_{j'}\rangle|^2 + \sigma^2 m^{-1 } \lambda _ { j ' } \ } \nonumber\\ & & \qquad \leq 1_{\{j ' > k\ } } |\langle\varphi _ { j ' } , \theta\rangle |^2 \nonumber\\ & & \qquad \quad { } + 1_{\ { j'\leq k\}}\nonumber\\ & & \hphantom { { } + } \qquad \quad { } \times \lambda_{j'}^{-2 } \bigl \{m^{-1 } \|\theta\|_2 ^ 2 \bigl(c_{j'}^2 \lambda_{j ' } + c_{x,0 } \exp(-c_{x,1 } c_{j'}^{c_{x,2}}/2 ) \bigr ) + \sigma^2 m^{-1 } \lambda_{j ' } \bigr\ } . \nonumber\end{aligned}\ ] ] for further investigation of the asymptotic quality of the estimator @xmath242 , some conditions on @xmath27 and @xmath12 are required . they are stated such that combined with condition [ condx]all previously imposed assumptions concerning those characteristics are included . [ condt ] we assume that @xmath311 for all @xmath55 and some constants @xmath312 and @xmath313 , which are uniform with respect to @xmath55 . condition [ condt ] says that the @xmath55 are uniformly well approximable with respect to the orthonormal basis consisting of the eigenfunctions of @xmath36 . the parameter @xmath314 describes the degree of this approximability . if the @xmath315 were some fourier polynomials , then condition [ condt ] could be interpreted as sobolev constraints on the set of the target functions . we apply the parameter selection @xmath316 , and we fix that @xmath317 with some constants @xmath318 sufficiently large and that @xmath319 . also , we choose @xmath320 . inserting that result into ( [ eq:4.3.t ] ) , we deduce by conditions [ condx ] and [ condt ] that @xmath321 for some @xmath322 , due to the inequality @xmath323 and the suitable selection of @xmath324 . revisiting inequality ( [ eq:4.1.t ] ) , we have finally proved by section [ s1.0 ] , paragraph ( i ) that the experiments @xmath258 and @xmath265 are asymptotically equivalent . in the experiment @xmath265 , the observation of @xmath325 allows us to construct an estimator @xmath326 for @xmath13 as well . it is given by @xmath327(t)\,ds_2(t ) \varphi_k , \ ] ] where the parameter @xmath34 can be adopted from the estimator @xmath242 . we specify the transformation @xmath328 with @xmath329(t)\,dt + m^{1/2 } \int_0^\cdot[\gamma^{1/2 } \hat{\theta}_2](t)\,dt,\mathbf{x}_2,s_2 \biggr ) .\end{aligned}\ ] ] again the shift of the second component is accessible by the other components which are maintained under the mapping so that @xmath330 is invertible . therefore , we define the experiment @xmath331 by the observation of @xmath332 with @xmath333 as under the experiment @xmath265 . hence , we put @xmath334 and obtain that @xmath331 is equivalent to @xmath265 . we define the experiment @xmath335 by the observation of @xmath336 on the probability space @xmath337 where @xmath338 denotes the it process with @xmath339 and @xmath340(t)\,dt + \sigma\ , dw_1(t ) .\ ] ] we can show that @xmath335 is asymptotically equivalent to @xmath331 analogously to the proof of the asymptotic equivalence of @xmath258 and @xmath341 . the only remarkable modification concerns the application of the estimator @xmath326 instead of @xmath242 . however , even for that term we establish an upper bound at the same rate as for estimator @xmath242 since the asymptotic order of @xmath221 and @xmath342 coincide . taking a closer look at the data drawn from @xmath335 , we realize that the random variables @xmath343 are independent . that occurs as we have replaced the empirical covariance operators by the true deterministic one . furthermore the data sets @xmath344 do not carry any information on @xmath13 so that @xmath345 represent a sufficient statistic for the whole empirical information obtained under @xmath335 . by section [ s1.0 ] , paragraph ( ii ) , we conclude that @xmath335 is equivalent to the experiment @xmath346 in which only the observations @xmath345 are available . thus we put @xmath347 ) \times c_0([0,1])$ ] and @xmath348 equal to the corresponding borel @xmath26-algebra . we define the transformation @xmath349 with @xmath350 by @xmath351 with the matrix @xmath352 we easily verify that @xmath154 is invertible so that the experiment @xmath353 which is defined by the observation of @xmath354 is equivalent to the experiment @xmath346 . we consider the characteristic function of the @xmath7)\times l_2([0,1])$]-valued random variable @xmath355 , @xmath356(t)\,dt \biggr\rangle \biggr ) \\ & & \quad \qquad { } \times\exp \biggl(-\frac1{2n } \sigma^2 { \int\!\!\int}t_1(x_1 ) \min\{x_1,x_2\ } t_1(x_2)\,dx_1 \,dx_2 \biggr ) \\ & & \quad \qquad { } \times\exp \biggl(-\frac1{2 } \biggl[\frac1 m + \frac1{n - m } \biggr ] \sigma ^2 { \int\!\!\int}t_2(x_1 ) \min\{x_1,x_2\ } t_2(x_2)\,dx_1 \,dx_2 \biggr)\end{aligned}\ ] ] for any @xmath357)$ ] so that @xmath358 and @xmath359 are two it processes satisfying @xmath360 and @xmath361(t)\,dt + n^{-1/2 } \sigma\ , dw_3(t ) , \\ dt_2(t ) & = & \sigma \biggl(\frac1 m + \frac1{n - m } \biggr)^{1/2}\,dw_4(t ) , \end{aligned}\ ] ] where @xmath362 and @xmath363 are two independent wiener processes . thus @xmath358 and @xmath359 are independent , and @xmath359 is totally uninformative with respect to the target function @xmath13 . applying section [ s1.0 ] , paragraph ( ii ) again , we have established equivalence of @xmath353 and the experiment @xmath364 , which is equipped with @xmath365)$ ] and the corresponding borel @xmath26-algebra @xmath366 , and characterized by the observation of the process @xmath358 , which coincides with the process @xmath367 as defined in ( [ eq:1.1.2 ] ) . summarizing we have shown asymptotic equivalence of the experiments @xmath21 and @xmath368 . that provides our final main result , which will be given as a theorem below . [ t:2 ] under the conditions [ condx ] and [ condt ] , the flr experiment @xmath21 with known design distribution and independent @xmath369-distributed regression errors is asymptotically equivalent to the white noise experiment @xmath368 where only the it process @xmath367 as in ( [ eq:1.1.2 ] ) is observed . we can combine our results with theorem 1 in @xcite , which is due to @xcite , in order to derive a sharp minimax result with respect to the mise for the flr problem under known covariance operator . it follows from there that this sharp minimax risk corresponds to the sequence @xmath370 where @xmath324 is the unique solution of the equation @xmath371 under the conditions of theorem [ t:2 ] . more concretely , there exists an estimator @xmath372 of @xmath13 in the flr model , which satisfies @xmath373 thus , any other estimator in the underlying model satisfies the above equation when @xmath374 is replaced by @xmath375 . we have established sharp asymptotic constants . critically , we mention that the loss function @xmath376 is apparently not bounded . still , asymptotic equivalence yields coincidence of sharp minimaxity for the loss function @xmath377 for some @xmath378 sufficiently slowly . we can show that , in the white noise inverse problem , the sharp constant result is extendable to the truncated loss function . using theorem [ t:2 ] , we have a sharp lower bound for the flr model even for the truncated loss function . however , the design distribution @xmath27 is assumed to be known and occurs in the definition of the minimax estimator . on the other hand , the lower bound as derived from theorem [ t:2 ] in the previous section provides a lower bound for the flr model in the case of unknown @xmath27 as well since nonknowledge of @xmath27 can not improve this lower bound . thus if we succeed in showing that some estimator achieves these asymptotic properties under the assumption of unknown @xmath27 , then sharp asymptotic minimaxity is extended to this more realistic condition . assuming that all conditions of theorem [ t:2 ] except the knowledge of @xmath27 hold true , we propose the estimator @xmath379 for @xmath13 where @xmath380 for some @xmath381 . the weights @xmath382 remain to be specified . using the techniques of the papers of @xcite and @xcite , the mise of @xmath383 is equal to @xmath384 we stipulate that for @xmath385 all weights @xmath386 shall be put equal to zero . for all other @xmath387 we have @xmath388 for @xmath24 sufficiently large so that @xmath389 \\[-2pt ] & \leq & \lambda_k^{-1 } + \lambda_k^{-1 } \biggl(\frac1{\log k + 1 } + ( 2+\log k)^2 o(n^{2\rho-1 } ) \biggr ) , \end{aligned}\ ] ] where we have used markov s inequality and that @xmath390 is bounded from above by the expected squared hilbert schmidt norm of @xmath391 and , hence , by @xmath392 ( see , e.g. , @xcite ) . that requires the following assumption : @xmath393 ( see also @xcite ) . we conclude that the second term in equation ( [ eq : sh2 ] ) has the same asymptotic order as @xmath394 under the above restriction with respect to the selection of the weights . focusing on the first term in ( [ eq : sh2 ] ) , we deduce by the cauchy schwarz inequality that @xmath395 & & \hphantom { \biggl\ { \biggl ( } { } + \mathrm{const.}\cdot \biggl(\sum_k e |\langle\hat { \varphi}_k - \varphi_k,\theta\rangle |^2 \biggr)^{1/2 } \biggr\}^2 .\end{aligned}\ ] ] we consider that @xmath396 & \leq & \mathrm{const.}\cdot c_\theta\sum_{k , j } j^{-2\beta } e & = & \mathrm{const.}\cdot\sum_j j^{-2\beta } e |\langle\hat { \varphi}_j - \varphi_j , \varphi_j\rangle |^2 \nonumber\\[-2pt ] & & { } + \mathrm { const.}\cdot\sum_j j^{-2\beta } \sum_{k\neq j } e |\langle\hat { \varphi}_k - \varphi_k , \varphi_j\rangle |^2 \\[-2pt ] & \leq & \mathrm{const.}\cdot\sum_j j^{-2\beta } e\|\hat{\varphi}_j - \varphi_j \|_2 ^ 2 \nonumber\\[-2pt ] & & { } + \mathrm{const.}\cdot\sum_j j^{-2\beta } \sum _ { k\neq j } e |\langle\hat{\varphi}_k , \hat{\varphi}_j - \varphi _ j\rangle |^2 \nonumber\\[-2pt ] \nonumber & \leq & \mathrm{const.}\cdot\sum_j j^{-2\beta } e\|\hat{\varphi}_j - \varphi_j \|_2 ^ 2 \nonumber\end{aligned}\ ] ] by exploiting the orthonormality of the @xmath113 and the @xmath296 as well as condition [ condt ] and parseval s identity . bhatia , davis and mcintosh @xcite provide that the squared @xmath7)$]-distance between @xmath113 and @xmath296 is bounded from above by the squared hilbert schmidt norm of @xmath397 multiplied by @xmath398 via condition ( [ eq : as01 ] ) . thus we have @xmath399 do not depend on @xmath13 whenever @xmath400 returning to the consideration of the first term in ( [ eq : sh2 ] ) , we conclude that its asymptotic order reduces to that of @xmath401 then this term is bounded from above by @xmath402 & & \hphantom{\sum_k |w_k - 1 |^2 |\langle\varphi_k,\theta\rangle|^2 + \mathrm{const.}\cdot\sum_{k=1}^{\lfloor n^{\rho/\alpha}/\log n\rfloor } } { } + o ( n^{-2\beta\rho/\alpha } ( \log n)^{2\beta } ) \\[-2pt ] & & \qquad \leq o ( n^{-2\beta\rho/\alpha } ( \log n)^{2\beta } ) + \mathrm{const.}\cdot c_\theta / n + \sum_k |w_k - 1 |^2 condition [ condt ] and again the results of @xcite . the term @xmath403 is asymptotically negligible [ i.e. , bounded by @xmath404 whenever @xmath405 as we have already imposed the condition ( [ eq : sh4 ] ) . it follows that the mise of ( [ eq : sh1 ] ) may be reduced to its asymptotically efficient terms , that is , @xmath406 \\[-9pt ] & & { } + o ( n^{-1 } \log^{\alpha+1}n ) . \nonumber\end{aligned}\ ] ] the right - hand side of ( [ eq : sh5 ] ) , however , corresponds to the mise of an oracle estimator which uses the true versions of the eigenvalues and eigenfunctions of @xmath36 instead of the empirical ones . also , it follows from @xcite and @xcite that the estimator @xmath372 as in ( [ eq : sh1 ] ) attains the sharp asymptotic minimax risk when the weights are chosen as @xmath407 when writing @xmath408 with an appropriate deterministic parameter @xmath324 . more precisely , we consider @xmath409 which we define by the unique zero of the function @xmath410 with @xmath411 for @xmath412 where @xmath413 and @xmath414 are continuous montonically decreasing and increasing , respectively . the selection @xmath415 leads to asymptotic sharp optimality ( see , e.g. , @xcite ) . clearly , we have @xmath416 . otherwise , not even the convergence rates are optimal as the required balance between the bias and the variance term is violated . by condition ( [ eq : sh4 ] ) our additional assumption saying that that @xmath417 for @xmath418 is verified under this optimal selection of the weights when stipulating that @xmath419 as we have assumed @xmath420 . still , the suggested selector is an oracle choice as it requires knowledge of the true eigenvalues @xmath421 . that motivates us to consider a data - driven selector @xmath422 of @xmath324 . first we split the sample @xmath0 into two independent data sets @xmath423 , @xmath424 . the first data set @xmath425 consists of @xmath221 pairs @xmath426 where @xmath427 , and @xmath428 contains all the other observations . we employ @xmath425 to estimate the function @xmath13 while the second data set ( training data ) is used to provide an selector of @xmath324 . concretely , we fix @xmath429 as the unique zero of @xmath430 where @xmath431 therein , @xmath432 indicates that the estimator is based on the second data set . then we define our selector of @xmath422 as @xmath433 . this truncation takes into account the a priori knowledge about the true @xmath409 so that @xmath434 almost surely for @xmath24 sufficiently large . thus determining @xmath422 does not require knowledge of @xmath27 . now let us consider the mise of the estimator @xmath435 where the index indicates the incorporated choice of the parameter @xmath324 . by ( [ eq : sh5 ] ) , we derive that @xmath436 where the terms contained in @xmath437 do not depend on @xmath438 . as the asymptotic order of @xmath221 and @xmath24 coincides , the estimator based on @xmath221 data attains the same asymptotic rates and constants as the estimator which uses even @xmath24 data , so our above calculations remain valid . therefore , the estimator @xmath439 attains sharp minimax rates and constants whenever @xmath440 uniformly with respect to @xmath55 . the first term in ( [ eq : nnew1 ] ) is bounded from above by @xmath441 . the second term has the upper bound @xmath442\ ] ] for some sequence @xmath443 tending to infinity sufficiently slowly . we deduce by markov s inequality that ( [ eq : nnew1 ] ) is satisfied if @xmath444 for some fixed integer @xmath445 . the assertion @xmath446 , for some positive - valued sequence @xmath447 with @xmath448 , implies that @xmath449 or @xmath450 we have already imposed that @xmath451 so that @xmath452 for all @xmath453 . that , however , yields @xmath454 where @xmath455 denotes the hilbert schmidt norm of an operator . therein we have used the findings of @xcite again and the monotonicity of the functions @xmath456 , @xmath413 , @xmath414 as well as the definitions of @xmath409 and @xmath429 . we deduce by markov s inequality that @xmath457 \\ & = & s_n^{2\nu } + \mathrm{const.}\cdot n^{2\rho\mu } s_n^{-2\mu } \gamma _ n^{2\mu } e \|\hat{\gamma } ' - { \gamma } \|_{\mathrm{hs}}^{2\mu}\end{aligned}\ ] ] for any integer @xmath458 . as all moments of @xmath100 are finite by condition [ condt ] we derive that @xmath459 where we recall that @xmath460 is based on the training data set , thus on @xmath461 observations . as @xmath462 we conclude by suitable choice of @xmath463 that @xmath464^{2\nu } ) , \ ] ] where @xmath465 denotes some sequence tending to zero at an algebraic rate . choosing @xmath445 sufficiently large , we can finally verify ( [ eq : nnew1 ] ) yielding the following proposition which summarizes the investigation carried out in this section . [ p:1 ] we consider the flr model in the setting of theorem [ t:2 ] except the condition that @xmath27 is known . in addition , suppose that ( [ eq : as01 ] ) , ( [ eq : sh4 ] ) and @xmath466 . then , estimator ( [ eq : sh1 ] ) with the weight selector ( [ eq : weight ] ) , which does not use @xmath27 in its definition , attains the sharp minimax rate and constant with respect to the mean integrated squared error ; viewed uniformly over the function class @xmath12 which is defined via condition [ condt ] . hence , under some additional conditions on the model , we have established sharp minimaxity in the case where @xmath27 is unknown . only an arbitrary number between @xmath467 and @xmath468 is supposed to be known . we have proved equivalence of the flr model and a white noise model involving an empirical covariance operator in theorem [ t:1 ] . we mention that @xmath26 and @xmath27 can be treated as real nuisance parameters in section [ s2 ] ; more precisely , knowledge of those quantities is not needed to apply the data transformations . in contrast , for the asymptotic approximation in section [ s3 ] , @xmath27 must be known . nevertheless , section [ s : new ] shows that , with respect to the mise , the sharp asymptotic minimax risk can be taken over to the case of unknown design distribution . furthermore , under specific parametric assumption on @xmath27 , the condition of known @xmath27 can obviously be justified . cai and hall @xcite explicitly mention gaussian processes as examples for the random design functions @xmath28 . for instance , assuming that @xmath28 can be represented as @xmath469 with independent standard wiener processes @xmath470 as already suggested in the previous section , we realize that the function @xmath471 is precisely reconstrucable based on only one observation @xmath124 . then as @xmath471 is known the distribution @xmath27 is known as well . therefore , under this shape of @xmath27 , the assumption of known @xmath27 is not unrealistic at all . this phenomenon is typical for the functional data approach and does not occur in multivariate linear regression with finite - dimensional covariates . from that point of view , the assumption of known design distribution causes less trouble in flr compared to more standard regression problems . still this does not address the completely nonparametric case for @xmath27 and @xmath13 . as an interesting restriction , we have assumed that @xmath472 in condition [ condt ] . therefore , the quality of the approximation of the target curve @xmath13 in the orthonormal basis consisting of the eigenvalues of the covariance operator of the design variables must be sufficiently high . if this basis consisted of fourier polynomials then that assumption could be interpreted as a smoothness condition on @xmath13 . that corresponds to the theorems in @xcite and @xcite where hlder conditions are imposed , which correspond to @xmath473 , in order to prove asymptotic equivalence of the white noise model on one hand and density estimation and nonparametric regression on the other hand . otherwise , counterexamles can be constructed ( see @xcite ) . to our best knowledge our work represents the first proof of white noise equivalence in a statistical inverse problem . it seems reasonable that the essential condition is extended to @xmath474 in this setting as the selection @xmath475 describes the setting of direct estimation ( noninverse problems ) . still , the question of whether our results are extendable to some @xmath476 remains open . in section [ s : new ] we have studied the case of unknown @xmath27 ; however , the regularity parameter @xmath314 is still assumed to be known . therefore , another interesting problem , which can not be addressed within the framework of this paper , is whether this sharp risk can be achieved by an adaptive estimator , which does not use @xmath314 and @xmath477 in its construction . approaches to adaptivity in flr are studied in @xcite ; however , that report seems to focus on optimal rates rather than optimal constants . also , combining theorem [ t:2 ] and the results of brown and low @xcite , we conclude that , under reasonable conditions , the flr model is also equivalent to the standard nonparametric regression problem , under which the data @xmath478(x_j ) + \sigma \varepsilon_j , \qquad j=1,\ldots , n , \ ] ] are observed where the @xmath8 are i.i.d . and @xmath167-distributed , and the homogeneous fixed design setting @xmath479 , @xmath131 , is applied . the author is grateful to markus reiss for a discussion on this paper and to the reviewers for their inspiring comments .

asymmetries in heavy quark pair and dijet production in electron - proton collisions allow studies of gluon tmds in close analogy to studies of quark tmds in semi - inclusive dis . here we present expressions for azimuthal asymmetries for both unpolarized and transversely polarized proton cases and consider the maximal asymmetries allowed . the latter are found to be rather sizeable , except in certain kinematic limits which are pointed out . in addition , we consider the small- limit and expectations from a mclerran - venugopalan model for unpolarized and linearly polarized gluons and from a perturbative , large transverse momentum calculation for the t - odd gluon tmds . comparison to related observables at rhic and lhc is expected to provide valuable information about the process dependence of the gluon tmds . in particular this will offer the possibility of a sign change test of the gluon sivers tmd and two other t - odd gluon tmds . this provides additional motivation for studies of azimuthal asymmetries in heavy quark pair and dijet production at a future electron - ion collider .

1605.07934

heavy quark pair and dijet production in electron - proton collisions offer direct probes of the gluons inside the proton . they allow to map out detailed information about the momentum distribution of the gluons , including the transverse momentum dependence @xcite . this is of particular interest in the case of transversely polarized protons , which are known to exhibit large left - right asymmetries . such asymmetries have not yet been studied in heavy quark pair or dijet production in electron - proton collisions , but form a prominent part of the physics case for an electron - ion collider ( eic ) proposed in the u.s . especially the study of the gluon sivers effect is of interest here . the sivers effect @xcite refers to the transverse momentum distribution of unpolarized quarks and gluons inside a transversely polarized proton , where the transverse momentum forms a @xmath1 distribution around the transverse spin direction . for transversely polarized protons there is a strong analogy between quark and gluon distributions . for both there is the sivers effect already mentioned , but there is also a parallel between the helicity flip distributions of transversely polarized quarks and of linearly polarized gluons inside transversely polarized protons , for which there are two each . even though the latter distributions have different chirality and t behavior , their transverse momentum structures are such that they lead to the same azimuthal asymmetries in certain processes . this parallel we will discuss explicitly for the quark asymmetries in semi - inclusive deep inelastic scattering ( sidis ) process ( @xmath2 ) and the gluon asymmetries in open heavy quark pair production ( @xmath3 ) . the transverse momentum dependent distribution functions ( tmds ) of gluons inside transversely polarized protons are also of interest because of their process dependence and their associated small @xmath0 behavior . tmds are inherently process dependent , expressed through their gauge link dependence . for the processes under consideration here , the gauge links are exclusively future pointing , and can be related directly to the gluon tmds appearing at lhc in higgs or heavy scalar quarkonium production ( see section [ sec : links ] ) . this in principle allows to cross check the results obtained at lhc and eic for unpolarized protons and perhaps for transversely polarized protons too in case a polarized fixed target experiment , called after@lhc @xcite , will be performed in the future at lhc . the gluon tmds appearing in these processes correspond to the weiszcker - williams ( ww ) distributions at small @xmath0 @xcite . recently it was pointed out in ref . @xcite that unlike the dipole gluon tmds ( which have one future and one past pointing gauge link ) , the ww gluon tmds for a transversely polarized proton are suppressed with respect to the unpolarized gluon tmd by a factor of @xmath0 . this implies that the transverse spin asymmetries in heavy quark pair and dijet production in electron - proton collisions will become suppressed in the small-@xmath0 limit . a test of this at eic would be very interesting , as the eic can probe both small and large @xmath0 regions . at rhic heavy quarkonia in the forward region are not accessible , but other processes can be studied that in the small-@xmath0 limit probe the dipole distributions that are not suppressed with respect to the unpolarized gluon tmd towards small @xmath0 . such processes may also be studied very precisely at large @xmath0 at after@lhc @xcite . for a further discussion we refer to ref . @xcite . obtaining a consistent picture among the various processes at small and large @xmath0 is important for testing our understanding of transverse spin effects and the tmd formalism in general . in this paper we will restrict to the ww - type functions that appear in the heavy quark pair and dijet production at an eic . the outline of the paper is as follows . in section [ sec : def ] we provide the definition of gluon tmds in terms of qcd operators and discuss some phenomenological models which are relevant in the small-@xmath0 region . details of the calculations of the cross section for heavy quark pair production in dis are discussed in section [ sec : hq ] , together with the expressions for the azimuthal asymmetries and their upper bounds . in section [ sec : links ] we comment on the process dependence of the gluon tmds involved . our results for dijet production are given in section [ sec : dijet ] . summary and conclusions are presented in section [ sec : concl ] . transverse momentum dependent distribution functions ( tmds ) of gluons inside a proton with four - momentum @xmath4 and spin vector @xmath5 are defined through a matrix element of a correlator of the gluon field strengths @xmath6 and @xmath7 , evaluated at fixed light - front ( lf ) time @xmath8 , where @xmath9 is a lightlike vector conjugate to @xmath4 . explicitly , the correlator is given by @xcite @xmath10 } f^{\nu\sigma}(\xi)\,u^\prime_{[\xi , { 0}]}\,\big ] \,|p , s \rangle\,\big\rfloor_{\text{lf}}\,,\end{aligned}\ ] ] where the gluon momentum @xmath11 is decomposed as @xmath12 , while @xmath13}$ ] and @xmath14}$ ] are process dependent gauge links that ensure gauge invariance . the longitudinal and transverse component of the proton spin are defined through the sudakov decomposition @xmath15 with @xmath16 and @xmath17 being the proton mass . for an unpolarized ( @xmath18 ) and transversely polarized ( @xmath19 ) proton , the following parametrizations for the correlator in terms of gluon tmds will be employed : @xmath20 where we have introduced the symmetric and antisymmetric transverse projectors @xmath21 and @xmath22 , with @xmath23 , respectively . the symmetric part of the correlator @xmath24 , i.e. @xmath25 , is parametrized by three gluon tmds that are all t - odd . the naming scheme used here for the gluon tmds is based on the analogy with the quarks @xcite , but it should be emphasized that the quark and gluon tmds with the same name are not necessarily linked by evolution and can have quite different properties under symmetry operations . the @xmath26 functions for quarks are chiral - odd and do not mix with the chiral - even @xmath26 functions of the gluons . below we will make use of the combination @xmath27 which , despite its name , is not related to the well - known transversity distribution @xmath28 for quarks that has no gluonic analogue . although both @xmath28 and @xmath29 denote helicity flip distributions , the quark distribution is chiral - odd , t - even , and survives after transverse momentum integration , while the gluon distribution is chiral - even , t - odd , and vanishes upon integration over transverse momentum . both distributions require another helicity flip somewhere else in the process , which will require consideration of different types of processes . but as we will show , the angular dependences in such distinct processes can be the same , which in our view justifies the naming scheme . in order to estimate the size of the asymmetries in the small-@xmath0 limit , we discuss here some aspects of the small-@xmath0 behavior of the various gluon tmds , starting with the linearly polarized gluon distribution . it has been shown that both the dipole and ww type linearly polarized gluon distributions can be studied in the small-@xmath0 formalism as they both have the same @xmath30 enhancement in the small @xmath0-region as the unpolarized gluon distributions . it is thus possible to compute them using a saturation model . for example , in the mclerran - venugopalan ( mv ) model @xcite , the ww type linearly polarized gluon distribution is given by @xcite @xmath31 where the saturation scale is defined as @xmath32 and @xmath33 is the transverse area of the nucleon / nucleus . in the mv model , there is a relation @xmath34 , where @xmath35 is the number of nucleons inside a nucleus , which is 1 in our case . although the mv model is argued to work better for a large nucleus , its application to a nucleon target turns out to be quite successful phenomenologically @xcite . for numerical estimations a regularized mv model expression is required for the small-@xmath0 ww type gluon tmds @xcite , @xmath36 } \left \ { 1-e^ { - \frac{r_\perp^2 q_{s0}^2}{4}{\rm ln } \left [ \frac{1}{r_\perp^2 \lambda^2_{qcd } } + e \right ] } \right \}\,,\end{aligned}\ ] ] where @xmath37 is a starting saturation scale to be taken from fits to experimental data . the ratio between linearly polarized and unpolarized gluon distributions in the mv model is then given by @xmath38}\ , \left \ { 1-e^ { - \frac{r_\perp^2 q_{s0}^2}{4}{\rm ln } \left [ \frac{1}{r^2_\perp \lambda^2_{qcd } } + e \right ] } \right \}}{\int_0^\infty \frac{{\mathrm{d}}r_\perp}{r_\perp}\ , j_0 \left ( | \bm p_{{\scriptscriptstyle t}}| r_\perp \right ) \ , \left \ { 1-e^ { - \frac{r_\perp^2 q_{s0}^2}{4}{\rm ln } \left [ \frac{1}{r_\perp^2 \lambda^2_{qcd } } + e \right ] } \right \}}\,.\ ] ] numerically we find that @xmath37 should be larger than about 1.1 times @xmath39 , in order for the ratio to stay below 1 for all @xmath40 values ( this is contrary to the claim in @xcite ) . alternatively one could consider replacing @xmath41 in the numerator by @xmath42 , and take @xmath43 to be sufficiently larger than 1 . in fig . [ fig : bound - mv ] we show the ratio for the case @xmath44 gev@xmath45 at @xmath46 and @xmath47 gev . here the value @xmath48 gev@xmath45 is based on the gbw model fits to hera data @xcite . if one uses instead the value @xmath49 gev@xmath45 in combination with @xmath50 gev , as in @xcite , the resulting function would violate the positivity bound in a restricted region of intermediate transverse momentum . as can be seen from the figure , unlike the dipole case for which the ratio is saturated for all @xmath40 @xcite , the ratio in this ww case is not saturated for @xmath40 values below 10 gev , but still sizeable except for very small @xmath40 values . sizeable asymmetries may thus also be expected at eic at small but not too small @xmath0 . now we turn to discuss the behavior of the ww type t - odd gluon tmds in the small @xmath51 limit . at large @xmath40 , three t - odd ww type gluon tmds inside a transversely polarized target can be perturbatively computed @xcite following the standard collinear twist-3 approach @xcite . as mentioned in the introduction , these calculations have suggested that the ww gluon tmds are suppressed by a factor of @xmath0 as compared to the corresponding dipole type ones . however , in ref . @xcite , only the asymptotic small-@xmath0 behavior of the ww gluon tmds is discussed in the context of the collinear twist-3 approach without giving the full expressions . below we present the complete results for these gluon distributions , which are relevant at moderately small @xmath0 . one notices that such sub - asymptotic behavior can not be studied in the mv model , as in general , only the leading logarithm contribution is taken into account in saturation models . for the ww type gluon sivers function , one has @xmath52 \right .\ + \frac{1+\xi^2}{2\xi(1-\xi)_+ } n(x , z ) \nonumber \\ & & - \!\!\!\!\!\ ! \left .\ \left .\ \frac{1+(1-\xi)^2}{2\xi(1-\xi)_+ } n(z , z - x ) -\frac{\xi^2+(1-\xi)^2}{2\xi(1-\xi)_+}n(x , x - z ) \right \ } \right \}\ , , \label{fsivers}\end{aligned}\ ] ] where @xmath53 and @xmath54 . here @xmath55 is the collinear twist-3 quark - gluon correlation function , commonly known as the qiu - sterman function @xcite , while @xmath56 is a tri - gluon correlation function in which three gluons are in a symmetric color singlet state @xcite . the tri - gluon correlation function has the symmetry properties : @xmath57 , which below will be used to simplify the expressions . terms proportional to @xmath58 $ ] are ignored , as they are suppressed in the limit @xmath59 . the large-@xmath60 expression of the gluon tmd @xmath61 takes a similar form , @xmath62 where again the term proportional to @xmath63 is ignored . finally , the gluon tmd @xmath64 is given by @xmath65 where the tri - gluon correlation function @xmath66 is defined as @xmath67/z$ ] . with these results , we can study the small-@xmath0 asymptotic behavior of the ww t - odd gluon tmds . first of all , as observed in @xcite , there is a complete cancelation between the soft gluon pole contributions and the hard gluon pole contributions in the leading power of @xmath68 for the gluon distributions @xmath69 and @xmath61 in the small-@xmath0 limit . the leading contribution to the gluon tmd @xmath64 also vanishes by invoking the burkardt sum rule @xcite , @xmath70 which is stable under qcd evolution @xcite . however , beyond the leading logarithm @xmath71 approximation , there is no exact cancellation among the various contributions . in the small-@xmath0 limit , the subleading contributions to the gluon tmds are given by @xmath72 we emphasize that the calculation presented here is based on the large-@xmath40 dglap - type formalism . unlike the t - odd dipole type gluon tmds @xcite , these t - odd ww type gluon tmds can not be treated in the small-@xmath0 formalism due to the lack of the leading logarithmic contribution , at least not in a conventional way @xcite . nevertheless , we will use these compact results as guidance for modelling the ww gluon tmds at small @xmath0 for all @xmath40 . for example , if we assume the burkardt sum rule were satisfied at each value of @xmath73 , i.e. @xmath74 , one obtains the simple result @xmath75 if , on the other hand , we assume that the tri - gluon correlation were zero at some low initial scale and is purely dynamically generated by the qiu - sterman function @xmath76 , one would have ( note that @xmath77 ) @xmath78 finally , if for some reason it turns out that @xmath79 , one obtains : @xmath80 these models will be used below to obtain simple estimates for some ratio of asymmetries . all three models differ from the leading logarithmic result for the t - odd dipole functions @xmath81 . the t - odd gluon tmds have to satisfy the following positivity bounds @xcite @xmath82 in none of the three models we presented , one can satisfy the bounds for all three functions at the same time . this is in contrast to the dipole model for which all bounds can be saturated simultaneously . it should be emphasized that for small @xmath0 the maximal values are not expected for the ww - type functions that are probed at eic in heavy quark pair or dijet production . as mentioned , the ww t - odd gluon tmds are suppressed by a factor of @xmath0 with respect to the ww unpolarized gluon tmd . the ww t - even gluon tmd @xmath83 is not suppressed by a factor of @xmath0 , but by a logarithm of @xmath84 in the region @xmath85 @xcite , whereas for @xmath86 it was found to saturate the bound . this is compatible with the sizeable but not maximal effects we obtained in the mv model , fig . [ fig : bound - mv ] . this is why below we will show results for the mv model , rather than for @xmath83 that saturates the positivity bound for all @xmath40 . the calculation of the cross section for the process @xmath87 where the proton is polarized with polarization vector @xmath5 and the other particles are unpolarized , proceeds along the same lines of ref . @xcite , to which we refer for details . the heavy quark - antiquark pair in the final state is almost back - to - back in the plane perpendicular to the direction of the incoming proton and the photon exchanged in the reaction , that we take as @xmath88-axis . therefore one has @xmath89 , where @xmath90 and @xmath91 , with @xmath92 ( @xmath93 ) being the transverse momenta of the heavy quark and antiquark , satisfying the relation @xmath94 . this is often referred to as the back - to - back correlation limit . in a reference frame in which azimuthal angles are measured w.r.t . the lepton plane ( @xmath95 ) , denoting by @xmath96 , @xmath97 and @xmath98 the azimuthal angles of the three - vectors @xmath99 , @xmath100 and @xmath101 , respectively , the cross section can be written as @xmath102 where @xmath103 are the rapidities of the quarks , @xmath104 is the inelasticity variable , and @xmath105 , with @xmath106 , is the bjorken-@xmath0 variable . at leading order in perturbative qcd , we find @xmath107 \delta(1-z_1-z_2 ) \ , , \label{eq : csu}\end{aligned}\ ] ] and @xmath108 + \cos(\phi_s-\phi_{{\scriptscriptstyle t } } ) \bigg [ b_0^t \sin 2 \phi_{{\scriptscriptstyle t}}\nonumber \\ & \quad + \ , b_1^t \sin(2\phi_{{\scriptscriptstyle t}}-\phi_\perp ) + b_2^t \sin 2 ( \phi_{{\scriptscriptstyle t}}-\phi_\perp ) + b_3^t \sin ( 2\phi_{{\scriptscriptstyle t}}-3\phi_\perp ) + b_4^t\sin(2\phi_{{\scriptscriptstyle t}}- 4\phi_\perp ) \bigg ] \nonumber\\ & \quad + \bigg [ b_0^{\prime\ , t } \sin(\phi_s+\phi_{{\scriptscriptstyle t } } ) + b_1^{\prime \,t } \sin(\phi_s + \phi_t-\phi_\perp ) + b_2^{\prime\,t } \sin(\phi_s+\phi_{{\scriptscriptstyle t}}-2\phi_\perp ) \nonumber \\ & \qquad \quad + b_3^{\prime\ , t } \sin(\phi_s+\phi_{{\scriptscriptstyle t}}-3\phi_\perp ) + b_4^{\prime\ , t } \sin(\phi_s+\phi_{{\scriptscriptstyle t}}-4 \phi_\perp ) \bigg ] \bigg \}\,\delta(1-z_1-z_2)\ , , \label{eq : cst}\end{aligned}\ ] ] where @xmath109 , with @xmath110 . the normalization factor @xmath111 is given by @xmath112 with @xmath113 , @xmath114 , where @xmath115 is the ( anti)quark mass . the explicit expressions of the terms @xmath116 in eqs . ( [ eq : csu ] ) and ( [ eq : cst ] ) , with @xmath117 , are given by @xmath118 where @xmath119 . we note that the three ratios @xmath120 are all equal to each other , i.e. to the ratio of the gluon sivers function @xmath121 and the unpolarized gluon distribution @xmath122 . the hard scattering functions @xmath123 describe the interaction of an unpolarized gluon with a photon which can be in different polarization states @xcite , @xmath124\,{\cal a}_{u+l}^{\gamma^*g\to q { \overline q } } -y^2 \,{\cal a}_{l}^{\gamma^*g \to q { \overline q}}\,,\nonumber \\ { \cal a}_{1}^{e g\to e q { \overline q } } & = ( 2-y ) \sqrt{1-y}\,{\cal a}_{i}^{\gamma^*g \to q { \overline q}}\,,\nonumber \\ { \cal a}_{2}^{e g\to e q { \overline q } } & = 2 ( 1-y)\ , { \cal a}_{t}^{\gamma^*g \to q { \overline q}}\,,\nonumber \\ \label{eq : agstar}\end{aligned}\ ] ] where the subscripts @xmath125 , @xmath126 , @xmath127 , @xmath19 refer to the specific polarization of the photon @xcite . introducing the notation @xmath128 , @xmath129 , @xmath130 , we have @xmath131\frac{q^2}{m_\perp^2 } \,\right . \nonumber \\ & \qquad\ \left . -\ , z(1-z)[1 - 2 z(1-z)]\frac{q^4}{m_\perp^4 } \right \}\ , , \label{eq : aqqb2}\\ { \cal a}_{l}^{\gamma^ * g\to q { \overline q } } & = 8\ , \frac{z^2(1-z)^2}{d^3 } \left ( 1-\frac{m_q^2}{m_\perp^2}\right ) \frac{q^2}{m_\perp^2}\ , , \\ { \cal a}_{i}^{\gamma^ * g\to q { \overline q } } & = 4 \,\sqrt{1-\frac{m_q^2}{m_\perp^2 } } \ , \frac{z(1-z)(1 - 2 z)}{d^3}\,\frac{q}{m_\perp } \bigg [ 1- z ( 1-z ) \frac{q^2}{m_\perp^2 } -2 \frac{m_q^2}{m_\perp^2 } \bigg ] \ , , \label{eq : aqqbphil}\\ { \cal a}_{t}^{\gamma^ * g\to q { \overline q } } & = 4\,\frac{z(1-z)}{d^3}\,\bigg ( 1-\frac{m_q^2}{m_\perp^2 } \bigg ) \bigg [ z ( 1-z ) \frac{q^2}{m_\perp^2 } + \frac{m_q^2}{m_\perp^2 } \bigg ] \ , , \label{eq : aq}\end{aligned}\ ] ] with @xmath132 analogously , the @xmath133 , @xmath134 and @xmath135 , with @xmath136 , describe the scattering of a linearly polarized gluon with a photon . they can be written as @xmath137 where @xmath138\,{\cal b}_{u+l}^{\gamma^*g\to q { \overline q } } -y^2 \,{\cal b}_{l}^{\gamma^*g \to q { \overline q}}\,,\nonumber \\ { \cal b}_{j}^{e g\to e q { \overline q } } & = ( 2-y ) \sqrt{1-y}\,{\cal b}_{j i}^{\gamma^*g \to q { \overline q}}\qquad ( j = 1,3 ) \,,\nonumber \\ { \cal b}_{k}^{e g\to e q { \overline q } } & = 2 ( 1-y)\ , { \cal b}_{k t}^{\gamma^*g \to q { \overline q}}\qquad ( k = 0,4)\ , , \label{eq : agstar}\end{aligned}\ ] ] with @xmath139\,\frac{q^2}{m_\perp^2 } - 2\ , \frac{m_q^2}{m_\perp^2 } \bigg \ } \bigg ( 1-\frac{m_q^2}{m_\perp^2 } \bigg ) \,,\nonumber \\ { \cal b}_{l}^{\gamma^*g\to q { \overline q } } & = - 4\,\frac{z^2(1-z)^2}{d^3}\ , \bigg(1-\frac{m_q^2}{m_\perp^2 } \bigg ) \frac{q^2}{m_\perp^2}\,,\nonumber \\ { \cal b}_{1 i}^{\gamma^*g \to q { \overline q } } & = 2 \,\frac{z(1-z)(1 - 2 z)}{d^3}\ , \bigg [ z ( 1-z ) \frac{q^2 } { m_\perp^2 } + \frac{m_q^2}{m_\perp^2 } \bigg ] \sqrt{1-\frac{m_q^2}{m_\perp^2}}\frac{q}{m_\perp } \,,\nonumber \\ { \cal b}_{3 i}^{\gamma^*g \to q { \overline q } } & = -2 \,\frac{z(1-z)(1 - 2 z)}{d^3}\,\bigg [ 1 -\frac{m_q^2}{m_\perp^2 } \bigg ] ^{3/2}\frac{q}{m_\perp}\,,\nonumber \\ { \cal b}_{0 t}^{\gamma^*g \to q { \overline q } } & = -\frac{z(1-z)}{d^3 } \bigg [ z ( 1-z)\frac{q^2 } { m_\perp^2 } + \frac{m_q^2}{m_\perp^2 } \bigg]^2\,,\nonumber \\ { \cal b}_{4 t}^{\gamma^*g \to q { \overline q } } & = \frac{z(1-z)}{d^3}\,\bigg ( 1 - \frac{m_q^2}{m_\perp^2 } \bigg ) ^2~.\end{aligned}\ ] ] the expressions involving @xmath140 have appeared before in ref . @xcite , but here a few redefinitions are made in order to display the similarity between the contributions of the various @xmath26 functions . in particular , the following observable , @xmath141 has been defined , which could provide direct access to @xmath140 . using the earlier mv model expressions for the ratio between linearly polarized and unpolarized tmds , in fig . [ figrp ] we show our estimates for @xmath142 at @xmath143 , @xmath144 , @xmath145= 6 gev and @xmath146 gev@xmath45 , for both charm and bottom production . for this and the subsequent numerical studies we take the charm mass @xmath147 gev and the bottom mass @xmath148 gev . we observe sizeable asymmetries in this mv model . of course , the center of mass energy has to be sufficiently large for these curves to indeed be in the small-@xmath0 range . for an eic the considered @xmath149 ranges from about 20 to 150 gev @xcite , therefore @xmath150 for @xmath151 requires @xmath152 . for @xmath153 gev one should thus not consider the @xmath154 gev@xmath45 curve . we include the large @xmath155 curves for a high energy eic or for an analysis of hera data which were taken at @xmath156 gev . for larger @xmath104 and @xmath145 values the asymmetries decrease . for @xmath157 @xmath142 vanishes identically . in the mv model , as a function of @xmath158 , calculated at @xmath159 gev , @xmath143 , @xmath144 and at different values of @xmath155 , for charm ( left panel ) and bottom ( right panel ) production in the process @xmath160.,title="fig:",scaledwidth=48.0% ] in the mv model , as a function of @xmath158 , calculated at @xmath159 gev , @xmath143 , @xmath144 and at different values of @xmath155 , for charm ( left panel ) and bottom ( right panel ) production in the process @xmath160.,title="fig:",scaledwidth=48.0% ] in the mv model , as a function of @xmath158 , calculated at @xmath161 gev , @xmath143 , @xmath162 , and different values of @xmath155 , for charm ( left panel ) and bottom ( right panel ) production in the process @xmath160.,title="fig:",scaledwidth=48.0% ] in the mv model , as a function of @xmath158 , calculated at @xmath161 gev , @xmath143 , @xmath162 , and different values of @xmath155 , for charm ( left panel ) and bottom ( right panel ) production in the process @xmath160.,title="fig:",scaledwidth=48.0% ] analogously , in fig . [ figr ] we provide our model estimates for the other asymmetry defined in ref . @xcite , @xmath163 which is calculated at @xmath143 , @xmath162 , @xmath145= 10 gev and @xmath146 gev@xmath45 . at large @xmath164 values the asymmetry is close to the maximal values allowed by the positivity bound . for larger @xmath104 and @xmath145 values the asymmetries decrease again , but the asymmetry does not vanish at @xmath157 . if we introduce the combination @xmath29 of eq . ( [ eq : h1 ] ) , it turns out that the cross section in the transversely polarized proton case , when integrated over @xmath98 , has only three independent azimuthal modulations : @xmath165 , @xmath166 , and @xmath167 . each one of them is related to a different t - odd gluon tmd . we observe that these angular structures and the tmds they probe , are closely analogous to the case of quark asymmetries in sidis ( @xmath168 ) , where the role of @xmath169 is played by @xmath170 , cf . the same applies to the @xmath171 asymmetry considered before . in order to single out separate @xmath96 dependent terms , we define the following azimuthal moments @xmath172}{\int { \mathrm{d}}\phi_{{\scriptscriptstyle t}}\,{\mathrm{d}}\phi_\perp\ , \left [ { \mathrm{d}}\sigma(\phi_s,\,\phi_{{\scriptscriptstyle t}},\,\phi_\perp ) + { \mathrm{d}}\sigma(\phi_s + \pi,\,\phi_{{\scriptscriptstyle t}},\,\phi_\perp)\right ] } \ , , \label{eq : mom}\end{aligned}\ ] ] where the denominator is given by @xmath173 and , since @xmath174 , we obtain @xmath175{\cal a}^{\gamma^*g \to q { \overline q } } _ { u+l } -y^2 { \cal a}_l^{\gamma^*g \to q { \overline q } } } \,\frac{\vert \bm q_{{\scriptscriptstyle t}}\vert}{m_p}\ , \frac{h_{1}^{g}(x,\bm q_{{\scriptscriptstyle t}}^2)}{f_1^g(x,\bm q_{{\scriptscriptstyle t}}^2)}\ , , \label{eq : a2}\\ a_n^{\sin(\phi_s-3\phi_{{\scriptscriptstyle t } } ) } & = - \frac{\vert \bm q_{{\scriptscriptstyle t}}\vert ^3}{m_p^3}\ , \frac{b_0^t}{2 a_0^u } = - \frac{2(1-y)\ , { \cal b}_{0t}^{\gamma^*g \to q { \overline q } } } { [ 1+(1-y)^2]{\cal a}^{\gamma^*g \to q { \overline q } } _ { u+l } -y^2 { \cal a}_l^{\gamma^*g \to q { \overline q } } } \ , \frac{\vert \bm q_{{\scriptscriptstyle t}}\vert^3}{2 m_p^3}\ , \frac{h_{1t}^{\perp \,g}(x,\bm q_{{\scriptscriptstyle t}}^2)}{f_1^g(x,\bm q_{{\scriptscriptstyle t}}^2)}\label{eq : a3}\,~.\end{aligned}\ ] ] note that the asymmetries in eqs . ( [ eq : a2 ] ) and ( [ eq : a3 ] ) vanish in the limit @xmath176 when the virtual photon is longitudinally polarized , which for @xmath177 corresponds to the limit @xmath59 . based on the absence of the leading logarithmic term for the ww t - odd gluon tmds , it is expected that also eq . ( [ eq : a1 ] ) vanishes in this limit , but it does not do so automatically through a kinematic suppression term as for the other two asymmetries . we note that a measurement of the ratio @xmath178 would probe directly the relative magnitude of @xmath179 and @xmath29 without dependences on any of the other kinematic variables in the process . in the small-@xmath0 models discussed in the previous section , one would obtain for this ratio : 0 in model i , @xmath180 in model ii , and @xmath181 in model iii . these cases are all distinguishable from the dipole case , which would give 1 . note that this ratio need not be bounded between @xmath182 and @xmath183 . as mentioned before , the ww t - odd gluon tmds are not expected to satisfy the positivity bounds , but the latter can be used to determine the maximum values of the above defined asymmetries , and thereby exclude less promising kinematic regions . it can be easily seen that the sivers asymmetry is bound to 1 , while the asymmetries in eqs . ( [ eq : a2 ] ) and ( [ eq : a3 ] ) have the same upper bound , which we denote by @xmath184 . the maximal values of @xmath185 for the latter are shown in fig . [ fig : anw_y ] for charm and bottom production as a function of @xmath104 , at @xmath186 , @xmath187 , @xmath188 and @xmath189 gev@xmath45 . these maximal values are attained at some specific values of @xmath190 , which in this case corresponds to the lowest @xmath190 value considered . the charm and bottom production upper bounds on @xmath185 as a function @xmath190 are shown in fig . [ fig : anw ] , again at @xmath186 , @xmath144 and @xmath187 , @xmath188 and @xmath189 gev@xmath45 . the upper bounds of these asymmetries are equal to the @xmath191 bounds of the weighted cross section @xmath192 , which are presented in fig . 2 of ref . @xcite for @xmath193 . alternatively , one can define the azimuthal angles w.r.t . @xmath98 instead of @xmath194 , which can then be integrated over . in this case , only the terms @xmath195 , @xmath196 and @xmath197 will contribute to the cross section in eq . ( [ eq : cst ] ) . after introducing again the combination in eq . ( [ eq : h1 ] ) , one finds , in analogy to eqs . ( [ eq : a1])-([eq : a3 ] ) , @xmath198{\cal{b}}_{u+l}^{\gamma^*g \to q { \overline q}}-y^2{\cal b}_l ^{\gamma^*g \to q { \overline q } } } { [ 1+(1-y)^2]{\cal a}^{\gamma^*g \to q { \overline q } } _ { u+l } -y^2 { \cal a}_l^{\gamma^*g \to q { \overline q } } } \ , \frac{\vert \bm q_{{\scriptscriptstyle t}}\vert}{m_p}\ , \frac{h_{1}^{g}(x,\bm q_{{\scriptscriptstyle t}}^2)}{f_1^g(x,\bm q_{{\scriptscriptstyle t}}^2)}\ , , \label{eq : a2b}\\ a_n^{\sin(\phi_s^\perp-3\phi_{{\scriptscriptstyle t}}^\perp ) } & = \frac{\vert \bm q_{{\scriptscriptstyle t}}\vert ^3}{m_p^3}\ , \frac{b_2^t}{2 a_0^u } = \frac{[1+(1-y)^2 ] { \cal{b}}_{u+l}^{\gamma^*g \to q { \overline q}}-y^2{\cal b}_l ^{\gamma^*g \to q { \overline q } } } { [ 1+(1-y)^2]{\cal a}^{\gamma^*g \to q { \overline q } } _ { u+l } -y^2 { \cal a}_l^{\gamma^*g \to q { \overline q } } } \ , \frac{\vert \bm q_{{\scriptscriptstyle t}}\vert^3}{2 m_p^3}\ , \frac{h_{1t}^{\perp \,g}(x,\bm q_{{\scriptscriptstyle t}}^2)}{f_1^g(x,\bm q_{{\scriptscriptstyle t}}^2)}\,~. \label{eq : a3b}\end{aligned}\ ] ] as before , the asymmetries in eqs . ( [ eq : a2b])-([eq : a3b ] ) have the same upper bound @xmath199 , presented in figs [ fig : anw_y_perp ] and [ fig : anwp ] for different values of the kinematic variables @xmath145 , @xmath104 , @xmath73 , @xmath155 . this bound is equal to @xmath200 , the upper bound of @xmath201 discussed in fig . 1 of ref . @xcite for @xmath193 . from fig . [ fig : anw_y_perp ] one sees that in this case the asymmetries do not need to vanish when @xmath202 . this means that for small-@xmath0 studies the asymmetries @xmath199 may be of more interest than @xmath203 . due to the zero crossing that occurs in the maximal value of @xmath204 , for charm production the asymmetry in the region of @xmath155 between 5 and 10 gev@xmath45 is expected to be small . for bottom production this would apply to @xmath155 closer to 100 gev@xmath45 . although these conclusions are based on the maximal asymmetries and therefore robust , the exact location of the zero crossing in @xmath104 may be affected by higher order corrections and thus may be a different function of @xmath155 than obtained here . on the asymmetries @xmath205 and @xmath206 for charm ( left panel ) and bottom ( right panel ) production in process @xmath160 . estimates are given as a function of @xmath104 , at different values of @xmath207 gev ( chosen in such a way that the asymmetries are maximal ) and @xmath155 , with @xmath143.,title="fig:",scaledwidth=48.0% ] on the asymmetries @xmath205 and @xmath206 for charm ( left panel ) and bottom ( right panel ) production in process @xmath160 . estimates are given as a function of @xmath104 , at different values of @xmath207 gev ( chosen in such a way that the asymmetries are maximal ) and @xmath155 , with @xmath143.,title="fig:",scaledwidth=48.0% ] , but for the upper bounds @xmath208 of the asymmetries @xmath209 and @xmath210.,title="fig:",scaledwidth=48.0% ] , but for the upper bounds @xmath208 of the asymmetries @xmath209 and @xmath210.,title="fig:",scaledwidth=48.0% ] in this section we discuss the gauge link structure of the gluon tmds in the process @xmath211 and its consequences . the subprocess @xmath212 probes a gluon correlator with two future pointing wilson lines , commonly referred to as @xmath213 links , at small @xmath0 corresponding to the ww type functions . in contrast , the process @xmath214 in the kinematic regime where gluons in the polarized proton dominate , effectively selects the subprocess @xmath215 , which probes a gluon correlator with a @xmath213 and @xmath216 link ( future and past pointing ) , at small @xmath0 corresponding to the dipole type functions . these two processes thus probe distinct and entirely independent t - odd gluon tmds . in this way processes in @xmath217 collisions and @xmath218 collisions can be completely complementary . however , it is important to also note that one can study processes in @xmath218 collisions that are fully related to the ones in @xmath217 collisions . for example , the process @xmath219 in the back - to - back correlation limit @xcite , probes t - odd gluon tmds with two past - pointing wilson lines , which are related by an overall sign change to the ones with two future - pointing links . as a result , we can make the new tmd - formalism prediction that the gluon sivers function probed in @xmath211 is of opposite sign to the one probed in @xmath220 : @xmath221}(x , p_{{\scriptscriptstyle t}}^2 ) = - f_{1t}^{\perp\ , g \ , [ p^\uparrow\ , p\to \gamma \ , \gamma \ , x ] } ( x , p_{{\scriptscriptstyle t}}^2),\ ] ] and the same relation applies to the other two t - odd gluon tmds , @xmath29 ( or @xmath222 ) and @xmath223 . here one can consider instead of a @xmath224 pair any other color singlet state in @xmath225-dominated kinematics , such as @xmath226 or @xmath227 pairs @xcite . this is the gluonic analogue of the famous sign change relation between the quark sivers tmd probed in sidis and drell - yan @xcite . in our view this provides additional motivation to study gluon sivers effects at rhic or after@lhc and compare it to eic studies in the future , irrespective of any theoretical expectations about ww - type gluon tmds at ( very ) small @xmath0 . for t - even functions there is no sign change , such that the process @xmath228 @xcite at lhc ( or instead of the higgs boson a ( pseudo-)scalar heavy quarkonium state @xcite ) , probes the same @xmath83 function as the process @xmath229 at eic : @xmath230}(x , p_{{\scriptscriptstyle t}}^2 ) = h_{1}^{\perp\ , g \ , [ p\ , p\to h \ , x ] } ( x , p_{{\scriptscriptstyle t}}^2).\ ] ] the l.h.s . involves the gluon tmds with two @xmath216 links and the r.h.s . with two @xmath213 links , which for t - even functions are equal . on the other hand , a process like higgs+jet production probes a more complicated link structure @xcite . the same applies to heavy quark - antiquark pair production in proton - proton collisions or proton - nucleus collisions @xcite , although that process is expected to be tmd factorizing at small @xmath0 only . in conclusion , the comparison between tmd observables that can be studied at lhc and at eic can be related or complementary depending on the process . the angular structure of the cross section for the process @xmath231 is the same as in eqs . ( [ eq : cs])-([eq : cst ] ) , with @xmath232 the terms @xmath116 at lo receive contributions from two subprocesses , @xmath233 and @xmath234 . they can be written in the form @xmath235 where @xmath236 , with @xmath237 being the number of colors . the hard scattering functions @xmath238 are the same as the ones calculated in ref . @xcite for unpolarized scattering . they read @xmath239\,{\cal a}_{u+l}^{\gamma^ * q\to q g } \ , -y^2 \,{\cal a}_{l}^{\gamma^ * q\to q g}\ , , \nonumber \\ { \cal a}_1^{e q\to q g } & = ( 2-y ) \sqrt{1-y}\,{\cal a}_{i}^{\gamma^ * q\to q g}\,,\nonumber \\ { \cal a}_2^{e q\to e q g } & = 2 ( 1-y)\ , { \cal a}_{t}^{\gamma^ * q\to q g}\,,\label{eq : aqstar}\end{aligned}\ ] ] where @xmath240 \,\frac{q^2}{\bm k_\perp^2 } + \left [ z^2 ( 1-z)^2\right ] \left [ 1+(1-z)^2\right]\ , \frac{q^4}{\bm k_\perp^4}\right \}\ , , \\ { \cal a}_{l}^{\gamma^ * q\to q g } & = 4 \,\frac{z^2(1-z)^3}{d_0 ^ 2}\,\frac{q^2}{\bm k_\perp^2 } \,,\\ { \cal a}_i^{\gamma^ * q\to q g } & = - 4 \,\frac{z^2(1-z)^2}{d_0 ^ 2}\,\left [ 1 + ( 1-z)^2 \,\frac{q^2}{\bm k_\perp^2 } \right ] \,\frac{q}{\vert \bm k_\perp \vert}\ , , \\ { \cal a}_{t}^{\gamma^ * q\to q g } & = \,2 \frac{z^2(1-z)^3}{d_0 ^ 2}\ , \frac{q^2}{\bm k_\perp^2}\ , , \label{eq : a2ej}\end{aligned}\ ] ] with @xmath241 only the process @xmath234 contributes to the terms @xmath133 , @xmath134 and @xmath242 , which can be written as @xmath243 the explicit expressions for @xmath244 and @xmath245 can be obtained from the corresponding ones for the process @xmath246 , presented in the previous section , by taking the limit @xmath247 . for the process under study it is possible to define azimuthal moments similar to those introduced in the previous section for heavy quark pair production . the denominators of the various asymmetries for dijet production will be given by @xmath248 in a kinematic region where @xmath0 is small enough , such that the quark contributions can be neglected , the maximal values of the asymmetries for dijet production will be the same as the corresponding ones for heavy quark pair production in the limit @xmath249 . we then recover the expressions of @xcite for the linearly polarized gluon asymmetries . in @xcite the effects of the ww @xmath83 in a cgc - formalism calculation including bk evolution with and without running coupling effects were studied for dijet production at eic . the results for @xmath250 were found to be large for @xmath251 gev , @xmath157 , @xmath143 and @xmath252 , reaching values up to 0.4 at larger transverse momentum of the dijet pair , close to the bound of 0.5 . although the selected kinematics do not correspond to small @xmath0 values unless @xmath253 is rather small for dijet production considerations ( @xmath254 gev in order for @xmath255 ) , the theoretically predicted logarithmic suppression @xcite for @xmath85 may thus only become apparent for much smaller @xmath0 values when @xmath256 is much larger than typically considered for eic . our mv model estimates for @xmath257 and @xmath258 in the dijet case are very similar to the charm production case for @xmath259 gev@xmath45 . for smaller @xmath155 , which corresponds to smaller @xmath0 values , the asymmetries are unfortunately significantly smaller than the @xmath260 gev@xmath45 curves for @xmath261 . for this reason we do not include separate plots for those asymmetries . note that upon inclusion of the quark contributions to the denominator , the asymmetries will decrease further in magnitude . for the t - odd gluon tmd asymmetries , upper bounds ( without quark contributions ) are shown in figs . [ fig : anw - y - dijet ] and [ fig : anw - dijet ] . again they are similar to the @xmath261 case for large @xmath155 and substantially smaller for small @xmath155 . we emphasize that at small @xmath0 these bounds are not expected to be satisfied . the simple models presented could be used to scale the asymmetries to more reasonable values in the small-@xmath0 case . we have selected @xmath262 gev based on the multijet analysis of small @xmath263 hera data in @xcite . we have studied the azimuthal asymmetries in heavy quark pair and dijet production in dis process , which provide direct access to the ww - type gluon tmds . we found that in these processes the gluon tmds arise in the same way as the quark tmds do in semi - inclusive dis . the measurements of these azimuthal asymmetries at an eic would allow a sign change test of the t - odd gluon tmds by comparing to corresponding observables at rhic and after@xmath264lhc . furthermore , it was shown that the asymmetries maximally allowed by the positivity bounds are rather sizeable , except in some kinematic limits . although ww - type gluon tmds suffer from suppression in the small-@xmath0 region , the effects from linearly polarized gluons are still expected to be sizeable in heavy quark pair production at an eic , but less so in dijet production because of the requirement that the transverse momentum of each jet must be large . for the sivers gluon tmd the bounds are always maximal and provide little guidance , especially in the small-@xmath0 region where the ww - type function is suppressed by a factor of @xmath0 with respect to the unpolarized gluon tmd . for the other two t - odd gluon tmds that arise with @xmath166 and @xmath167 modulations , their ratio can be exploited to test small-@xmath0 expectations . we have provided some simple model expectations for that ratio . it will also be interesting to compare these quantities to the dipole type t - odd gluon tmds that arise in other processes , such as virtual photon plus jet production in polarized proton - proton collisions at rhic . this allows to experimentally test the theoretical expectation that the ww and dipole type t - odd gluon tmds have different small @xmath0 asymptotic behavior . hence , the studies of these asymmetries could form a prominent part of both the spin physics program and the small-@xmath0 physics program at a future eic . this research is partially supported by the european research council ( erc ) under the fp7 `` ideas '' programme ( grant agreement no . 320389 , qwork ) and the european union s horizon 2020 research and innovation programme ( grant agreement no . 647981 , 3dspin ) . d. boer , s. j. brodsky , p. j. mulders and c. pisano , phys . lett . * 106 * , 132001 ( 2011 ) . c. pisano , d. boer , s. j. brodsky , m. g. a. buffing and p. j. mulders , jhep * 1310 * , 024 ( 2013 ) . d. boer _ et al . _ , arxiv:1108.1713 [ nucl - th ] . a. accardi _ et al . _ , arxiv:1212.1701 [ nucl - ex ] . d. w. sivers , phys . d * 41 * , 83 ( 1990 ) . s. j. brodsky , f. fleuret , c. hadjidakis and j. p. lansberg , phys . rept . * 522 * , 239 ( 2013 ) . j. p. lansberg _ et al . _ , epj web conf . * 85 * , 02038 ( 2015 ) . f. dominguez , c. marquet , b. w. xiao and f. yuan , phys . d * 83 * , 105005 ( 2011 ) . f. dominguez , b. w. xiao and f. yuan , phys . lett . * 106 * , 022301 ( 2011 ) . d. boer , m. g. echevarria , p. mulders and j. zhou , phys . lett . * 116 * , 122001 ( 2016 ) . a. schfer and j. zhou , phys . d * 88 * , 014008 ( 2013 ) . k. kanazawa , y. koike , a. metz and d. pitonyak , adv . high energy phys . * 2015 * , 257934 ( 2015 ) . j. p. ma and c. wang , phys . d * 93 * , 014025 ( 2016 ) . d. boer , c. lorc , c. pisano and j. zhou , adv . high energy phys . * 2015 * , 371396 ( 2015 ) . j. mulders and j. rodrigues , phys . d * 63 * , 094021 ( 2001 ) . s. meissner , a. metz and k. goeke , phys . d * 76 * , 034002 ( 2007 ) . l. d. mclerran and r. venugopalan , phys . d * 49 * , 2233 ( 1994 ) . a. metz and j. zhou , phys . d * 84 * , 051503 ( 2011 ) . f. dominguez , j. w. qiu , b. w. xiao and f. yuan , phys . d * 85 * , 045003 ( 2012 ) . j. l. albacete , n. armesto , j. g. milhano , p. quiroga - arias and c. a. salgado , eur . j. c * 71 * , 1705 ( 2011 ) . t. lappi and h. mantysaari , phys . d * 88 * , 114020 ( 2013 ) . j. l. albacete , n. armesto , a. kovner , c. a. salgado and u. a. wiedemann , phys . lett . * 92 * , 082001 ( 2004 ) . r. enberg , k. j. golec - biernat and s. munier , phys . d * 72 * , 074021 ( 2005 ) . k. j. golec - biernat and m. wsthoff , phys . d * 59 * , 014017 ( 1998 ) . x. ji , j. w. qiu , w. vogelsang and f. yuan , phys . lett . * 97 * , 082002 ( 2006 ) . j. zhou , f. yuan and z. t. liang , phys . d * 78 * , 114008 ( 2008 ) . j. w. qiu and g. sterman , phys . * 67 * , 2264 ( 1991 ) . x. d. ji , phys . b * 289 * , 137 ( 1992 ) . h. beppu , y. koike , k. tanaka and s. yoshida , phys . d * 82 * , 054005 ( 2010 ) . a. schfer and j. zhou , arxiv:1308.4961 [ hep - ph ] . m. burkardt , phys . d * 69 * , 057501 ( 2004 ) . c. lorc , jhep * 1508 * , 045 ( 2015 ) . j. zhou , phys . d * 92 * , 074016 ( 2015 ) . j. zhou , phys . d * 89 * , 074050 ( 2014 ) . j. bartels , b. i. ermolaev and m. g. ryskin , z. phys . c * 70 * , 273 ( 1996 ) . t. brodkorb and e. mirkes , z. phys . c * 66 * , 141 ( 1995 ) . d. boer and p. j. mulders , phys . d * 57 * , 5780 ( 1998 ) . j. w. qiu , m. schlegel and w. vogelsang , phys . lett . * 107 * , 062001 ( 2011 ) . w. j. den dunnen , j. p. lansberg , c. pisano and m. schlegel , phys . lett . * 112 * , 212001 ( 2014 ) . lansberg and h. s. shao , nucl . b * 900 * , 273 ( 2015 ) . j. c. collins , phys . b * 536 * , 43 ( 2002 ) . p. sun , b. w. xiao and f. yuan , phys . d * 84 * , 094005 ( 2011 ) . d. boer , w. j. den dunnen , c. pisano , m. schlegel and w. vogelsang , phys . lett . * 108 * , 032002 ( 2012 ) . d. boer and c. pisano , phys . d * 86 * , 094007 ( 2012 ) . d. boer and c. pisano , phys . d * 91 * , 074024 ( 2015 ) . e. akcakaya , a. schfer and j. zhou , phys . d * 87 * , 054010 ( 2013 ) . a. dumitru , t. lappi and v. skokov , phys . lett . * 115 * , 252301 ( 2015 ) . s. chekanov _ et al . _ [ zeus collaboration ] , nucl . b * 786 * , 152 ( 2007 ) .

we study the semi - discrete directed random polymer model introduced by oconnell and yor . we obtain a representation for the moment generating function of the polymer partition function in terms of a determinantal measure . this measure is an extension of the probability measure of the eigenvalues for the gaussian unitary ensemble ( gue ) in random matrix theory . to establish the relation , we introduce another determinantal measure on larger degrees of freedom and consider its few properties , from which the representation above follows immediately

1506.05548

in this paper we consider a directed random polymer model in random media in two ( one discrete and one continuous ) dimension introduced by oconnell and yor @xcite . for @xmath0 independent one - dimensional standard brownian motions @xmath1 and the parameter @xmath2 representing the inverse temperature , the polymer partition function is defined by @xmath3 here @xmath4 for @xmath5 and @xmath6 represents the energy of the polymer . in the last fifteen years much progress has been made on this oconnell - yor polymer model , by which we can access some explicit information about @xmath7 and the polymer free energy @xmath8 @xcite . the first breakthrough was made in the zero temperature @xmath9 case . in this limit , @xmath10 becomes @xmath11 where @xmath12 is the ground state energy . for @xmath13 , the following relation was established @xcite : @xmath14^n}\prod_{j=1}^ndx_j \cdot p_{\text{gue}}(x_1,\cdots , x_n;t ) , \label{t0result } \\ & p_{\text{gue}}(x_1,\cdots , x_n;t)= \prod_{j=1}^n\frac{e^{-x_j^2/2t}}{j!t^{j-1}\sqrt{2\pi t}}\cdot \prod_{1\le j < k\le n}(x_k - x_j)^2 , \label{t0result0}\end{aligned}\ ] ] where @xmath15 is the probability density function of the eigenvalues in the gaussian unitary ensemble ( gue ) in random matrix theory @xcite . this type of connection of the ground state energy of a directed polymer in random media with random matrix theory was first obtained for a directed random polymer model on a discrete space @xmath16 @xcite by using the robinson - schensted - knuth(rsk ) correspondence . . can be regarded as its continuous analogue . note that is written in the form of a product of the vandermonde determinant @xmath17 . this feature implies that the @xmath18-point correlation function is described by an @xmath19 determinant , i.e. the eigenvalues of the gue are a typical example of the determinantal point processes @xcite . in addition based on this fact and explicit expression of the correlation kernel , we can study the asymptotic behavior of @xmath13 in the limit @xmath20 . in @xcite , it has been shown that under a proper scaling , the limiting distribution of @xmath13 becomes the gue tracy - widom distribution @xcite . in this paper , we provide a representation for a moment generating function of the polymer partition function which holds for arbitrary @xmath21 : @xmath22 = \int_{\r^n}\prod_{j=1}^ndx_j \ , f_f(x_j - u ) \cdot w(x_1,\cdots , x_n;t ) , \label{result } \\ w(x_1,\cdots , x_n;t ) = \prod_{j=1}^n\frac{1}{j!}\cdot \prod_{1\le j < k\le n } ( x_k - x_j)\cdot \det\left(\psi_{k-1}(x_j;t)\right)_{j , k=1}^n , \label{w1}\end{gathered}\ ] ] where @xmath23 is the fermi distribution function and @xmath24 for more details see definition [ defdgue ] and theorem [ thmmain ] below . this is a simple generalization of to the case of finite temperature . we easily find that it recovers in the zero - temperature limit ( @xmath25 ) . note that the function @xmath26 is also written as a product of two determinants and thus retains the determinantal structure in . in most cases , to find a finite temperature generalization of results for zero - temperature case is highly nontrivial and in fact often impossible . but for the oconnell - yor polymer model and a few related models , rich mathematical structures have been discovered for finite temperature and the studies on this topic entered a new stage @xcite . in @xcite , oconnell found a connection to the quantum toda lattice , and based on the developments in its studies and the geometric rsk correspondence , it was revealed that the law of the free energy @xmath27 is expressed as @xmath28}dx_1\int_{\r^{n-1}}\prod_{j=2}^{n } dx_{j}\cdot m(x_1,\cdots , x_n;t ) . \label{whit0}\end{aligned}\ ] ] here the probability measure @xmath29 , which is called the _ whittaker measure _ , is defined by the density function @xmath30 in terms of the whittaker function @xmath31 ( for the definition , see @xcite ) and the sklyanin measure @xmath32 ( see below ) as follows , @xmath33 where @xmath34 represents @xmath35 . in contrast to , the density function is not known to be expressed as a product of determinants and the process associated with does not seem to be determinantal . nevertheless some determinantal formulas for the oconnell - yor polymer have been found : first in @xcite , oconnell showed a determinantal representation for the moment generating function ( lhs of ) in terms of the sklyanin measure . ( see below . ) next in @xcite , borodin and corwin obtained a fredholm determinant representation for the same moment generating function ( see below ) . a direct proof of the equivalence between the two determinantal expressions was given in @xcite . in @xcite , by considering its continuous limit , the authors also obtained an explicit representation of the free energy distribution for the directed random polymer in two continuous dimension described by stochastic heat equation ( she ) @xcite . the distribution in this limit , which describes the universal crossover between the kardar - parisi - zhang ( kpz ) and the edwards - wilkinson universality class , was first obtained in @xcite and can be interpreted also as the height distribution for the kpz equation @xcite . furthermore in @xcite , they consider not only the oconnell - yor model but a class of stochastic processes having the similar fredholm determinant expressions , the macdonald processes , the probability measures on a sequence of partitions which are written in terms of the macdonald symmetric functions and include the whittaker measure defined by as a limiting case . the purpose of this paper is to investigate further the mechanism of appearance of such determinantal structures and is the central formula in our study . although @xmath36 defined by is not a probability measure but a signed measure except when @xmath25 , a remarkable feature of this measure is that it is _ determinantal _ for arbitrary @xmath37 contrary to the whittaker measure . this determinantal structure allows us to use the conventional techniques developed in the random matrix theory and thus from the relation we readily get a fredholm determinant representation with a kernel using biorthogonal functions which is regarded as a generalization of the kernel with the hermite polynomials for the gue . in , the parameter @xmath37 , which originally represents the inverse temperature in the polymer model appears in the fermi distribution function @xmath38 with the chemical potential @xmath39 as well as @xmath40 in rhs . this fact with the determinantal structure suggests that the rhs might have something to do with the free fermions at a finite temperature . related to this , a curious relation of the height of the kpz equation with fermions has been discussed in @xcite . for establishing the relation , we introduce a measure on a larger space @xmath41 . by integrating the measure in two different ways , we get its two marginal weights . in one formula appears a determinant which solves the @xmath0 dimensional diffusion equation with some condition ( see , , and ) and the other one with a symmetrization is exactly the rhs of . the relation follows immediately from the equivalence of these two expressions . our approach is similar to the one by warren @xcite for getting the relation . actually in the zero - temperature limit @xmath25 , we see that the integration of the measure is written in terms of the probability measure introduced in @xcite , which describes the positions of the reflected brownian particles on the gelfand - tsetlin cone . note that the macdonald processes ( especially the whittaker process in our case ) @xcite are also another generalizations of @xcite . although the whittaker process has rich integrable properties , they do not inherit the determinantal structure of @xcite . on the other hand , our measure is described without using the whittaker functions and keeps the determinantal structure . furthermore combining with the fact that the quantity can be rewritten as the fredholm determinant found in @xcite ( corollary [ corfred ] and proposition [ propfred2 ] below ) , our approach can be considered as another proof of the equivalence between and in @xcite . one feature of our proof is to bring to light the larger determinantal structure behind the two relations . this paper is organized as follows . in the next section , after stating the definition of a determinantal measure , we give our main result , theorem [ thmmain ] and its proof . the proof consists of two major steps : we first introduce in lemma [ lemoc ] a determinantal representation for the moment generating function which is a deformed version of the representation in @xcite . next we introduce another determinantal measure on larger space @xmath41 and then we find two relations about its integrations which play a key role in deriving our main result . in sec . 3 we show that this approach can be considered as an extension of the one in @xcite . in sec.4 . , we consider the fredholm determinant formula with biorthogonal kernel obtained by applying conventional random matrix techniques to our main result . the scaling limit to the kpz equation is discussed in sec.5 . we check that our kernel goes to the one obtained in the studies of the kpz equation . a concluding remark is given in the last section . in this section , we introduce a measure @xmath36 , state our main result and give its proof . [ defdgue ] let @xmath42 be @xmath43 for @xmath44 , a function @xmath26 is defined by @xmath45 * remark . * we find that @xmath26 is a real function on @xmath46 since by definition @xmath40 is real for any @xmath47 and @xmath48 . but in general , the positivity of this measure is not guaranteed . for example @xmath49 shows a damped oscillation and can take a negative value for some @xmath50 . thus at least for the case @xmath51 , @xmath52 can be negative . we discuss the zero - temperature limit @xmath53 of @xmath26 . noting @xmath54 , we see @xmath55 where we used the integral representations of the @xmath56th order hermite polynomial @xmath57 ( see e.g. section 6.1 in @xcite ) , @xmath58 note that @xmath59 is a monic polynomial ( i.e. the coefficient of the highest degree is 1 ) and @xmath60 thus we find @xmath61 where @xmath15 is defined by . the function @xmath26 can be regarded as a deformation of which keeps its determinantal structure . in this paper , we provide a determinantal representation for the moment generating function of the polymer partition function in terms of the function . [ thmmain ] @xmath62 where @xmath23 is the fermi distribution function . by , and the simple facts @xmath63 we find that the zero temperature limit of becomes . because of the determinantal structure of @xmath26 , we can get the fredholm determinant representation for the moment generating function by using the techniques in random matrix theory . recently another fredholm determinant representation has been given based on properties of macdonald difference operators @xcite . the equivalence between them will be shown in sec . [ fdf ] . here we provide a proof of theorem [ thmmain ] . our starting point is the representation for the moment generating function given in @xcite : @xmath64 where @xmath65 and @xmath32 is the sklyanin measure defined by @xmath66 this relation was obtained by using the properties of the whittaker functions @xcite and the whittaker measure . [ lemoc ] @xmath67 where @xmath68 is defined below and @xmath69 with @xmath65 . we will discuss an interpretation of in the next section . in this definition , we have arranged @xmath70s in the reversed order so as to relate , the zero - temperature limit of , to the stochastic processes defined later in . * noting the relation @xmath71 we rewrite rhs of as @xmath72 where in the last equality , we used the andrief identity ( also known as the cauchy - binet identity ) @xcite : for the functions @xmath73 , @xmath74 such that all integrations below are well - defined , we have @xmath75 we notice that the factor @xmath76 in can be written as @xmath77 where we used the reflection formula for the gamma function and the relation . from and , we arrive at the desired expression . from ( [ lemma3 ] ) , we see that for the derivation of our main result ( [ main ] ) , it is sufficient to prove the relation @xmath78 where @xmath68 is defined below and @xmath79 is given in definition [ defdgue ] . note that this is a relation for the integrated values on @xmath80 . to establish this we introduce a measure on the larger space @xmath41 . [ deftr ] let @xmath81 be an array @xmath82 where @xmath83 and @xmath84 . we define a measure @xmath85 by @xmath86 here @xmath87 , @xmath88 is given by @xmath89 with @xmath90 and @xmath91 is defined by using the fermi and bose distribution functions , @xmath92 and @xmath93 respectively as follows . @xmath94 * remark . * the reason why both the bose and fermi distributions appear in our approach is not clear . the interrelations between them ( see - below ) will play an important role in the following discussions . as in fig [ vtgt ] . we usually represent the array @xmath95 graphically in the triangular shape . although no ordering is imposed on @xmath95 , in the zero - temperature limit , @xmath96 has the support on the ordered arrays as in fig . [ vtgt ] ( a ) ( see ) . [ vtgt ] ( b ) represents the other ordered array called the gelfand - tsetlin pattern ( see ) . ( 350,90 ) ( 90 , 80)(a ) ( 105,60)@xmath97 ( 155,60)@xmath98 ( 205,60)@xmath99 ( 145,45 ) ( 195,45 ) ( 130,30)@xmath100 ( 180,30)@xmath101 ( 170,15 ) ( 155,0)@xmath102 ( 250,80)(b ) ( 265,60)@xmath97 ( 315,60)@xmath98 ( 365,60)@xmath99 ( 275,50 ) ( 305,45 ) ( 330,50 ) ( 355,45 ) ( 290,30)@xmath100 ( 340,30)@xmath101 ( 300,20 ) ( 330,15 ) ( 315,0)@xmath102 as discussed later we will find that the moment generating function of the oconnell - yor polymer model is expressed as the integration of this measure @xmath96 over @xmath41 . we have other choices for the definition of @xmath96 which give the same integration value . one example is @xmath103 this comes form the following consideration . let @xmath104 be a function which is symmetric under permutations of @xmath105 for each @xmath106 . then we see that @xmath96 and @xmath107 have the same integration value : @xmath108 it can be shown as follows . from the symmetry of @xmath104 , lhs of the equation above becomes @xmath109 here @xmath110 is defined by @xmath111 where @xmath112 is the permutation of @xmath113 and @xmath114 denotes @xmath115 with @xmath116 . we easily find the equivalence @xmath117 . note that @xmath118 here in the last equality , @xmath119 is defined by using @xmath120 and @xmath121 as @xmath122 , where we regard @xmath120 as an element of @xmath112 with @xmath123 . further in the last equality we used @xmath124 . substituting into and using the definition of the determinant , we have @xmath117 . the function @xmath107 has a similar determinantal structure to the schur process @xcite . the schur process is a probability measure on the sequence of partitions @xmath125 where @xmath126 , described as products of the skew schur functions @xmath127 . for the ascending case ( see definition 2.7 in @xcite ) , the probability measure is expressed as @xmath128 where @xmath129 are positive variables . we note that @xmath130 is expressed as a @xmath131th order determinant and @xmath132 as a @xmath0th order determinant by the jacobi - trudi identity @xcite , @xmath133 where @xmath134 is a complete symmetric polynomial with degree @xmath131 and @xmath135 is the length of the partition @xmath34 . thus and have a common structure of @xmath0 products of determinants with increasing size times an @xmath0th order determinant . in the following we provide the relations about two marginals of @xmath96 , from which immediately follows . for this purpose , we give two formulas for @xmath68 and @xmath136 . first we define a multiple convolution @xmath137 for a functions @xmath138 on @xmath139 and an integral operator @xmath140 with the kernel @xmath141 as @xmath142 using this definition , the formulas are written as follows : [ lem9 ] we regard all integrations below as the cauchy principal values . for @xmath143 , @xmath144 with @xmath145 and @xmath146 , we have @xmath147 where @xmath148 is an @xmath18th order polynomial with the coefficient of the highest degree being @xmath149 . a proof of this lemma will be given in appendix [ b ] . the polynomial @xmath148 in is defined inductively by - . but in our later discussion we will not use its explicit form . from and , we readily obtain for @xmath150 , @xmath151 where we define @xmath152 . using and , we obtain the following relations . [ mar ] let the measures @xmath153 and @xmath154 be @xmath155 then we have @xmath156 here @xmath157 is defined by and @xmath158 where @xmath159 is defined below . we easily see that can be obtained from these relations and : integrating the both hand sides of them over the remaining degrees of freedom ( @xmath160 for and @xmath161 for ) , we get two different expression about the integrated value of @xmath96 @xmath162 where @xmath163 . rhs of the second relation is further rewritten as @xmath164 and the symmetrized @xmath165 in this equation is nothing but @xmath166 since @xmath167 here in the second equality we used the fact that @xmath168 is a @xmath169th order polynomial with the coefficient of the highest degree being @xmath170 and @xmath171 . * proof of theorem [ mar ] . * first we derive . by the definition of , lhs of becomes @xmath172 here @xmath173 is defined below . applying to this equation we obtain . next we derive . we see that the factor @xmath174 in @xmath175 can be decomposed to @xmath176 and from the integration of the factor for each @xmath131 is represented as @xmath177 where @xmath148 is given in . . follows immediately from this relation . the purpose of this section is to have a better understanding of the two quantities , @xmath26 and @xmath178 , which arose as partially integrated quantities of @xmath96 in theorem [ mar ] ( for @xmath179 a symmetrization is also necessary , see ( [ align ] ) ) . we will first consider the evolution equations of these two quantities . next we will see that the zero - temperature limit of the equation for @xmath26 is nothing but the evolution equation for the brownian particles with reflection interaction while @xmath26 satisfies the one for the gue dyson s brownian motion @xcite regardless of the value of @xmath37 . furthermore we will find that our idea using @xmath96 in an enlarged space @xmath41 ( theorem [ mar ] ) is similar to the argument in @xcite although we need a modification of @xcite about the ordering in an enlarged space . let us first summarize the properties of @xmath180 @xmath181 all of which are easily confirmed by simple observations : @xmath182 where @xmath40 in and @xmath173 in are defined by and below . . is equivalent to while is obtained from the relation @xmath183 for @xmath184 . this relation is easily given by with @xmath185 . we see that due to and the multilinearity of a determinant , @xmath178 satisfies the diffusion equation . @xmath186 in addition , by , it satisfies the condition @xmath187 for @xmath188 . though this condition is unusual , we will see that it is regarded as a finite temperature generalization of the neumann boundary conditions at @xmath189 in the zero temperature limit ( see ) . on the other hand , from with the harmonicity of the vandermonde determinant in , we see that @xmath26 satisfies the kolmogorov forward equation of the gue dyson s brownian motion @xcite , which is a dynamical generalization of the gue , @xmath190 the time evolution equation for the gue dyson s brownian motion can be transformed to the imaginary - time schrdinger equation with free - fermionic hamiltonian ( e.g. see chapter 11 in @xcite ) . on the other hand note that the density function of the whittaker measure does not solve such a simple free - fermionic time evolution equation . let us consider the zero temperature limit of the equations with and . note that for @xmath191 , @xmath192 where @xmath173 is defined below and @xmath193 and @xmath194 are the step functions defined by @xmath195 in addition we have @xmath196 where @xmath197 is defined for @xmath198 and @xmath199 as @xmath200 here we summarize a few properties of the function which are the zero temperature limit of for @xmath180 . @xmath201 where in , @xmath40 is defined by and @xmath202 is the @xmath131th order hermite polynomial @xcite . the second equality in has appeared as . note that corresponds to the zero - temperature limit of , since rhs of goes to @xmath203 in the zero - temperature limit and thus the integral operator with the kernel @xmath204 is equivalent to differentiation in the zero temperature limit when its action is restricted to @xmath205 . let @xmath206 be the zero - temperature limit of @xmath178 defined on @xmath80 . from , we find @xmath207 the function @xmath208 appeared as a solution to the schrdinger equation for the derivative nonlinear schrdinger type model @xcite . as discussed in @xcite , using and with basic properties of a determinant , we find that for @xmath209 , @xmath208 satisfies the diffusion equation , @xmath210 with the boundary condition @xmath211 the probabilistic interpretation of @xmath208 has been given in @xcite . let @xmath212 be the stochastic processes with @xmath0-components described by @xmath213 where @xmath214 satisfying @xmath215 represent initial positions , @xmath216 denotes the standard brownian motion and @xmath217 is twice the semimartingale local time at zero of @xmath218 for @xmath219 while @xmath220 . the system describes the @xmath0-brownian particles system with one - sided reflection interaction , i.e. the @xmath221th particle is reflected from the @xmath222th particle for @xmath223 . in @xcite , warren found that the transition density of this system from @xmath224 to @xmath70 , @xmath225 is written as @xmath226 . such kind of determinantal transition density was first obtained for the totally asymmetric simple exclusion process ( tasep ) in @xcite . furthermore , based on the determinantal structures , various techniques for discussing the space - time joint distributions for the particle positions or current have been developed for tasep @xcite and the reflected brownian particle system @xcite . on the other hand , we have seen in that the zero temperature limit of @xmath26 is the gue density @xmath15 . note that @xmath15 also satisfies since it holds for arbitrary @xmath37 i.e : @xmath227 from , , and , we find that the zero - temperature limit of is @xmath228^{n}}\prod_{\ell=1}^ndx_{\ell}\cdot\mathcal{g}(x_1,\cdots , x_n;t ) = \int_{(-\infty , u]^{n}}\prod_{j=1}^ndx_{j}\cdot p_{\text{gue}}(x_1,\cdots , x_n;t ) . \label{wresult}\end{aligned}\ ] ] in @xcite warren showed that this relation , which connects the two different processes , is obtained in the following way . first one introduces a process on the @xmath229-dimensional gelfand - tsetlin cone whose two marginals describe the above two processes . the gelfand - tsetlin cone @xmath230 is defined as @xmath231 for the graphical representation of an element of @xmath232 , see fig . [ vtgt ] ( b ) . next we introduce a following stochastic process on @xmath233 . let @xmath234 with @xmath235 be a process defined by @xmath236 where @xmath237 are the @xmath229 independent brownian motions starting at the origin , @xmath238 represent the initial positions and the process @xmath239 and @xmath240 are twice the semimartingale local time at zero of @xmath241 and @xmath242 respectively . . describes the interacting particle systems where each @xmath243 is a brownian motion reflected from @xmath244 to a negative direction and from @xmath245 to a positive direction . in @xcite , borodin and ferrari also introduced similar processes on the discrete gelfand - tsetlin cone where the probability measure at a particular time is described by the schur process @xcite . the pdf of the system at time @xmath246 can be given explicitly : for the case of @xmath247 , it is expressed as @xmath248 where @xmath95 is defined above and @xmath249 represents the indicator function on gt@xmath250 . the pdfs of the two marginals , @xmath251 and @xmath161 for @xmath252 was obtained as follows : ( proposition 6 and 8 in @xcite)[propw ] @xmath253 where @xmath208 , @xmath254 , @xmath193 and @xmath255 are defined by , , and respectively . * remark . * note that @xmath256 in can be replaced by an arbitrary function on @xmath80such that it corresponds to @xmath256 in the region @xmath257 . for later discussion on a generalization of finite temperature , we chose it as @xmath256 on the whole @xmath80 . we see that the relation is obtained from this theorem . by decomposing the integral on @xmath95 in two different ways , we clearly have @xmath228^{n(n-1)/2}}d\underline{x}_n\mathcal{q}_{\text{gt}}(\underline{x}_n;t ) & = \int_{(-\infty , u]^n}\prod_{j=1}^{n}dx^{(j)}_1 \int_{\r^{n(n-1)/2}}da_1\mathcal{q}_{\rm gt}(\underline{x}_n;t)\notag\\ & = \int_{(-\infty , u]^n}\prod_{j=1}^{n}dx^{(n)}_{j } \int_{\r^{n(n-1)/2}}da_2\mathcal{q}_{\rm gt}(\underline{x}_n;t ) . \label{gt1}\end{aligned}\ ] ] applying and to this equation , we get @xmath258^n}\prod_{j=1}^{n}dx^{(j)}_1 \g(x^{(1)}_1,\cdots , x^{(n)}_1;t)\prod_{j=1}^{n-1}1_{>0}(x^{(j+1)}_1-x^{(j)}_1)\\ = \int_{(-\infty , u]^n}\prod_{j=1}^{n}dx^{(n)}_{j } n!p_{\rm{gue}}(x^{(n)}_1,\cdots , x^{(n)}_n;t)\prod_{j=1}^{n-1}1_{>0}(x^{(n)}_j - x^{(n)}_{j+1 } ) . \label{gt3}\end{gathered}\ ] ] due to the symmetry of @xmath15 under the permutations of @xmath259 , we readily see that rhs of this equation is equal to rhs of . also we find that lhs of becomes @xmath260^n}\prod_{j=1}^{n}dx^{(j)}_1\cdot \g(x^{(1)}_1,\cdots , x^{(n)}_1;t)\prod_{j=1}^{n-1}1_{>0}(x^{(j+1)}_1-x^{(j)}_1)\notag\\ & = \int_{(-\infty , u]^n}\prod_{j=1}^{n}dx^{(j)}_1\cdot \g(x^{(1)}_1,\cdots , x^{(n)}_1;t)\prod_{j=1}^{n-1}\left(1_{>0}(x^{(j+1)}_1-x^{(j)}_1)+1_{>0}(x^{(j)}_1-x^{(j+1)}_1)\right)\notag\\ & = \int_{(-\infty , u]^n}\prod_{j=1}^{n}dx^{(j)}_1\cdot \g(x^{(1)}_1,\cdots , x^{(n)}_1;t),\end{aligned}\ ] ] where in the first equality we used for @xmath261 and @xmath262^n$ ] @xmath228}dx^{(k)}_1 \g(x^{(1)}_1,\cdots , x^{(n)}_1;t)\prod_{j=1}^{k-2}1_{>0}(x^{(j+1)}_1-x_1^{(j)})\cdot 1_{>0}(x^{(k-1)}_1-x^{(k)}_1)=0 . \label{gt2}\end{aligned}\ ] ] note that @xmath256 is defined on @xmath80 and is finite even outside the region @xmath257 . ( see remark . of proposition [ propw ] . . is obtained from the following observation : putting the last factor @xmath263 in the @xmath264th row of the determinant @xmath256 in then applying , we get the determinant which has the same two rows . thus is obtained from proposition [ propw ] . this is similar to the situation of and theorem [ mar ] . this naive observation gives us the impression that the pdf @xmath252 is the zero - temperature limit of the weight @xmath96 . however in fact this is not the case . . from and one has @xmath266 from and , it is further rewritten as @xmath267 where @xmath268 is the indicator function on an ordered set @xmath269 defined by @xmath270 for the graphical representation of an element of , see fig . [ vtgt ] ( a ) . comparing with , we see that they have the same form but their supports ( @xmath233 and @xmath271 ) are different . we further notice that @xmath271 with an additional order @xmath272 corresponds to gt@xmath273 . hence our approach using @xmath274 can be regarded as a modification of warren s arguments on @xmath233 to the ones on the partially ordered spece @xmath271 . let us focus on two marginals @xmath275 and @xmath276 for @xmath274 . by taking the zero - temperature limit of theorem [ mar ] , we have the following analogue of proposition [ propw ] : [ mar0 ] @xmath277 where for the definition of @xmath153 and @xmath154 , see , @xmath278 is given by and @xmath279 * proof . * it is obtained by taking the zero - temperature limit @xmath9 in theorem [ mar ] . as discussed in , @xmath15 can be interpreted as the symmetric version of @xmath280 : @xmath281 therefore by the similar discussion in , we see that the relation is obtained also from proposition [ mar0 ] . the fact that both proposition [ propw ] and [ mar0 ] lead to implies the relation @xmath282^{n(n+1)/2}}d\underline{x}_n \mathcal{q}_{\rm gt}(\underline{x}_n;t ) . \label{p11}\end{aligned}\ ] ] this equivalence of their integration values is generalized in the following way . [ vgt ] let @xmath283 be the function defined above . then we have @xmath284^{n(n+1)/2}}d\underline{x}_n f_{\rm{sym}}(\underline{x}_n)\mathcal{q}_{\rm gt}(\underline{x}_n;t ) \label{p11}\end{aligned}\ ] ] an essential step of the proof of this proposition is represented as the following [ vgt2 ] @xmath285 the proof of this lemma will be given in appendix [ pvgt ] . using this lemma we readily derive proposition [ vgt ] . * proof of proposition [ vgt ] . * substituting the definition of @xmath274 into , we see that the lhs of is rewritten as @xmath286 where in the second equality we use lemma [ vgt2 ] . the function @xmath26 has a notable determinantal structure that it is described by a product of two determinants . this allows us to apply the results of random matrix theory and determinantal point processes developed in @xcite and to get the fredholm determinant representation . to see this we provide a lemma let @xmath287 be @xmath288 where the contour encloses the origin anticlockwise with radius smaller than @xmath37 . we find @xmath289 and @xmath40 satisfy the biorthonormal relation : [ lembior ] for @xmath290 , we have @xmath291 * proof . * substituting the definitions and into lhs of , one has @xmath292 as the integrand in this equation is analytic on @xmath293 with respect to @xmath294 , we can shift the integration path as @xmath295 . then using @xmath296 we find @xmath297 the residue calculus shows that the function @xmath289 is a @xmath169th order polynomial in @xmath50 and the coefficient of the highest order is @xmath170 . as the vandermonde determinant in is expressed as @xmath298 @xmath26 is rewritten as a product of two determinants @xmath299 from lemma [ lembior ] and , we obtain a fredholm determinant representation for the moment generating function . throughout this paper , we follow @xcite for the notation on fredholm determinants . [ propfred1 ] @xmath300 where @xmath301 is an arbitrary function such that the left hand side is well - defined and in the right hand side @xmath302 represents a fredholm determinant defined by @xmath303 here @xmath304 and @xmath305 is written in terms of the biorthogonal functions @xmath306 and @xmath307 as @xmath308 * proof . * we readily obtain this representation by applying the techniques in @xcite with lemma [ lembior ] to lhs of . for reference , here is an outline of the proof . first , using the andrief ( cauchy - binet ) identity , we have @xmath309 where @xmath310 is defined as @xmath311 in the first equality of , we used with and in the last one we used lemma [ lembior ] . we further rewrite @xmath312 as @xmath313 by using @xmath314 applying the identity for fredholm determinants , @xmath315 and noting @xmath316 we arrive at our desired expression . combining this proposition with theorem [ thmmain ] , we readily obtain [ corfred ] @xmath317 where the right hand side is the fredholm determinant with the kernel @xmath318 , @xmath319 , and @xmath320 is defined in . as in , we see @xmath321 which is due to another representation of the @xmath56th order hermite polynomial @xmath57 ( see e.g. section 6.1 in @xcite ) , @xmath322 where the contour encloses the origin anticlockwise . from and , we find @xmath323 here rhs appears as a correlation kernel of the eigenvalues in the gue random matrices @xcite . thus @xmath320 is a simple biorthogonal deformation of the kernel with hermite polynomials which appears in the eigenvalue correlations of @xmath324 gue random matrices . using this fredholm determinant expression , we can understand a few asymptotic properties of the partition function by applying saddle point analyses to the kernel as will be discussed in sec . [ kpzsl ] . in @xcite , oconnell first introduced the probability measure on @xmath80 which is called the whittaker measure @xmath29 whose density function @xmath30 is defined in terms of the whittaker function @xmath31 ( see @xcite ) , @xmath325 where throughout this paper we denote @xmath326 and @xmath327 is defined by . then he showed the following relation about the distribution of the free energy @xmath328 ( see theorem 3.1 and corollary 4.1 in @xcite ) , @xmath28}dx_1\int_{\r^{n-1}}\prod_{j=2}^{n } dx_{j}\cdot m(x_1,\cdots , x_n;t ) . \label{whit}\end{aligned}\ ] ] the density function @xmath329 is also a finite temperature extension of @xmath330 . actually it has been known that @xmath30 converges to @xmath331 in the zero - temperature limit . ( see sec.6 in @xcite ) . in contrast to @xmath79 , however , this extension does not inherit the determinantal structure which @xmath15 has and thus we can not apply the techniques in random matrix theory which is useful especially for asymptotic analyses of the gue . this fact necessitated the developments of new methods @xcite . by using the techniques of the macdonald difference operators @xcite and the duality @xcite , one can get a fredholm determinant expression for the moment generating function of the partition function , which allows us to access the asymptotic properties . ( @xcite ) @xmath332 where @xmath333 denotes the contour enclosing only the origin positively with radius @xmath334 and the kernel @xmath335 is written as @xmath336 here @xmath337 satisfies the condition @xmath338 . we can show the equivalence between the two expressions and . [ propfred2 ] @xmath339 where @xmath340 and @xmath341 and @xmath335 are defined and respectively . * proof . * substituting the definitions and into , we have @xmath342 for the definition of @xmath333 , see below . here we changed @xmath343 in and shift the path of @xmath294 by @xmath337 which is larger than the radius of @xmath344 . we notice that although the last expression in consists of two terms proportional to @xmath345 and @xmath346 , the integration of the term proportional to @xmath345 with respect to @xmath344 vanishes . thus we see @xmath347 where we set @xmath348 here we use the relation for fredholm determinants , @xmath349 , where the kernel @xmath350 on rhs reads @xmath351 using the relation @xmath352 we perform the integration over @xmath50 in as @xmath353}. \label{xint}\end{aligned}\ ] ] note that because of the conditions @xmath354 and @xmath338 ( see below and respectively ) , is applicable to the above equation . thus from and , we have @xmath355 } \frac{w^ne^{w^2t/2-wu}}{v'^ne^{v'^2t/2-v'u}}\frac{1}{w - v}\frac{\gamma(1+v'/\beta)^n } { \gamma(1+w/\beta)^n}\notag\\ & = l(v , v';t)\end{aligned}\ ] ] in this section , we discuss a scaling limit of the oconnell - yor polymer model . when both @xmath0 and @xmath246 are large with its ratio @xmath356 fixed , it has been known that the polymer free energy @xmath27 defined below is proportional to @xmath0 on average and the fluctuation around the average is of order @xmath357 @xcite . furthermore recently it has been shown in @xcite that the limiting distribution of the free energy fluctuation under the @xmath357 scaling is the gue tracy - widom distribution @xcite . this type of the limit theorem has been obtained also for other models related to the oconnell - yor model @xcite . these results reflect the strong universality known as the kpz universality class . although we expect that the same result on the tracy - widom asymptotics can be obtained from our representation , we consider another scaling limit where the partition function goes to the solution to the stochastic heat equation ( she ) ( or equivalently , the free energy goes to the solution to the kardar - parisi - zhang ( kpz ) equation ) . this scaling limit to the kpz equation has also been known to be universal although in a weaker sense compared with the kpz universality stated above @xcite . the height distribution of the kpz equation has been obtained for a droplet initial data in @xcite . since then , explicit forms of the height distribution have been given for the kpz equation and related models for a few initial data @xcite . in particular for the oconnell - yor model , the limiting distribution of the polymer free energy has been obtained by applying the saddle point method to the kernel @xcite . in this section , we confirm that a similar saddle point analysis can be applicable to our biorthogonal kernel . since our kernel has a simple form , we find that the nontrivial part of this problem reduces only to the asymptotic analyses of the functions @xmath40 and @xmath358 . before discussing the saddle point analysis , let us briefly review the scaling limit to the kpz equation . hereafter we will write out explicitly the dependence on @xmath37 of the polymer partition function as @xmath359 . let @xmath360 . by it s formula , we easily find that it satisfies the stochastic differential equations @xmath361 where we set @xmath362 and interpret the second term of this equation as it type . now let us take the diffusion scaling for : we set @xmath363 and at the same time we scale @xmath37 as @xmath364 then take the large @xmath365 limit . the scaling exponent @xmath366 in is known to be universal : it characterize the disorder regime referred to as the intermediate disorder regime @xcite , which lies between weak and strong disorder regimes in directed polymer models in random media in @xmath367 dimension . this @xmath368 scaling can be explained in the following heuristic way . let @xmath369 be @xmath0 independent one dimensional standard brownian motions . for @xmath370 , we have @xmath371 where @xmath372 represents the expectation value with respect to the brownian motions . now we consider its large @xmath365 limit under the same scaling as i.e. @xmath373 and @xmath374 noting that @xmath375 under , we see @xmath376 this suggests in a heuristic sense , @xmath377 here @xmath378 with @xmath379 and @xmath380 is the space - time white noise with mean @xmath381 and @xmath337-function covariance , @xmath382 thus considering , we choose the scaling of @xmath37 . under the scaling and , the following limiting property is established . @xmath383 here @xmath384 is the solution to the she with the @xmath337-function initial condition , @xmath385 where @xmath378 is the space - time white noise with mean @xmath381 and @xmath337-function covariance . the she is known to be well - defined if we interpret the multiplicative noise term as it - type @xcite . using this equation , the solution to the kpz equation can be defined via @xmath386 which is called the cole - hopf solution to the kpz equation . recently a new regularization for the kpz equation was developed in @xcite ( see also @xcite ) . according to @xcite , a rigorous estimate about the convergence to the she has been obtained for the oconnell - yor model @xcite based on the results in @xcite . this type of convergence has been discussed also in interacting particle processes @xcite . for reference we offer a sketch of the derivation of . for this purpose , we provide the following lemma , for @xmath387 defined above , one has @xmath388 where @xmath389 denotes the poissonian density and @xmath390 . @xmath391 denotes the region of the integration @xmath392 and the it integrals on rhs , referred to as the multiple it integrals @xcite , are performed in time order ( i.e. the order of @xmath393 ) . * proof . * by the definition of @xmath7 , we have @xmath394 with @xmath395=0 and the integrand of rhs is expressed as @xmath396 here we use the relation on a one - dimensional standard brownian motion @xmath397 : one has for @xmath398 and @xmath143 , @xmath399 where the it integrals on rhs , referred to us the multiple it integrals , are performed in time order ( i.e. the order of @xmath393 ) @xcite . using this , we get @xmath400 substituting this into , and performing the integration on @xmath401 , we have @xmath402 where we set @xmath403 . now we introduce the new variables @xmath404 by the relation @xmath405 then one has @xmath406 leading to @xmath407 further from , we have @xmath408 where we set @xmath409 . substituting these and into and noting the summations @xmath410 can be summarized as the simple form @xmath411 , we obtain . note that under the scaling , the poissonian density @xmath412 goes to the gaussian density @xmath413 , i.e. @xmath414 furthermore by theorems 4.3 and 4.5 in @xcite , for a function @xmath415 that converges to @xmath416 under the scaling @xmath417 and @xmath418 , we have @xmath419 where @xmath420 is the space - time white noise with the @xmath337-covariances . thus from , and , we have under the scaling , @xmath421 where @xmath422 . since we easily find that rhs of this equation is the solution of the she with @xmath337-function initial data , we obtain . in @xcite , borodin and corwin discussed the asymptotics of the fredholm determinant under the scaling limit to the kpz equation , especially the limiting property of the kernel based on the saddle point method . here we check that a similar saddle point method is applicable to our biorthogonal kernel . the scaling limit we consider is discussed above , but here we adopt its rephrased version stated in @xcite , @xmath423 where @xmath424 is @xmath425 which is more suitable for our purpose . to see the equivalence between and , we rewrite the relation as @xmath426 where we scale @xmath427 as @xmath428 furthermore focusing on the scaling property of the partition function @xmath429 , we find @xmath430 in distribution . noticing under the scaling @xmath431 where @xmath424 is defined in , we find that is equivalent to . for the moment generating function , implies @xmath432 where on lhs , @xmath39 is set to be @xmath433 with @xmath434 , and in the last equality in we used . the notions of the kpz universality class tell us that the fluctuation of the height @xmath435 and the position @xmath436 are scaled as @xmath437 and @xmath438 respectively for large @xmath439 . considering them , we set @xmath440 where @xmath441 . the first term @xmath442 represents the macroscopic growth with a constant velocity . the height fluctuation is expressed as @xmath443 and the term @xmath444 reflects the fact that the she with delta - function initial data in corresponds to the parabolic growth in the kpz equation @xcite . thus substituting @xmath445 , @xmath446 into , we arrive at the modified scaling @xmath447 hence is rewritten as @xmath448 with the scaling . this is the scaling limit of the moment generating function from the oconnell - yor polymer to the kpz equation . it has been known that rhs of this equation can be represented as the fredholm determinant @xcite , @xmath449 where the kernel @xmath450 is expressed as @xmath451 note that @xmath452 does not appear in rhs of this equation . this kernel first appeared in the studies of the kpz equation for the narrow wedge initial condition @xcite . from the relation we readily get the distribution of the scaled height @xmath443 given in . by combining the formula for the oconnell - yor polymer and the limiting relation from the oconnell - yor polymer to the kpz equation , we can obtain by showing @xmath453 under . this was indeed already discussed in @xcite by using the kernel . here we show that the kernel appears rather easily from the scaling limit of our biorthogonal kernel . using the saddle point method , we get the following : [ asyker ] @xmath454 here the kernel is expressed in terms of @xmath455 and @xmath456 defined by and respectively as @xmath457 and we set @xmath39 to be and @xmath458 since the factor @xmath459 in does not contribute to the fredholm determinant , we get ( though for a complete proof one has to prove the convergence of the fredholm determinant itself , not only the kernel ) . note that has a similar structure to the kernel in the gue random matrices . when we discuss certain large @xmath0 limits in the gue such as the bulk and the edge scaling limit , the nontrivial step reduces to the scaling limit of the hermite polynomial in . the same thing happens in our case : the only nontrivial step for getting is the asymptotics of the functions @xmath40 and @xmath358 . based on the saddle point method , we obtain the following results of which the proof is given in appendix [ a ] . @xmath460 where we set @xmath70 as and @xmath131 and @xmath246 as @xmath461 the constant @xmath462 is represented as @xmath463 in terms of @xmath464 defined by , and in appendix [ a ] . on the other hand , when we take the same limit for the other representation , we can apply the saddle point analysis also to the kernel and can get the limiting kernel . but since it does not correspond to the kernel directly , we need an additional step to show the equivalence between the fredholm determinant with the limiting kernel and that with ( see sec . 5.4.3 in @xcite ) . * proof of proposition [ asyker ] . * combining the estimate with the simple fact @xmath465 under and , we immediately obtain the result . for the oconnell - yor directed random polymer model , we have established the representation ( [ main ] ) of the moment generating function for the partition function in terms of a determinantal function which is regarded as a one - parameter deformation of the eigenvalue density function of the gue random matrices . there are some special mathematical structures behind the oconnell - yor model which play a crucial role in deriving the relation . the first one has been the determinantal representation which is essentially the one with the sklyanin measure in @xcite . next we have introduced another determinantal measure in enlarged degrees of freedom . our main theorem has been readily obtained from a simple fact about two marginals of this measure ( theorem [ mar ] ) . we can regard our approach as a generalization of the one in @xcite which retains its determinantal structures . to see this we needed to reinterpret the dynamics in the gelfand - tsetlin cone introduced in @xcite using the weight supported on the partially ordered space @xmath271 . our approach is a natural generalization of @xcite from this viewpoint . it would be an interesting future problem to find a clear relation with the macdonald process @xcite , which is another generalization of @xcite . applying familiar techniques in random matrix theory to the main result , we have readily obtained the fredholm determinant representation of the moment generating function whose kernel is expressed as the biorthogonal functions both of which are simple deformations of the hermite polynomials . the asymptotics of the kernel under the scaling limit to the kpz equation can be estimated easily by the saddle point analysis . first we give a proof of . for this purpose , it is sufficient to show the case of @xmath466 , @xmath467 furthermore setting @xmath468 , @xmath469 , one sees that is rewritten as @xmath470 where @xmath471 and we take the branch cut of @xmath472 to be the positive real axis . hence here we prove . we set the contour @xmath473 as depicted in fig . [ contour ] with @xmath474 . from the cauchy integral theorem , we find @xmath475 dividing the contour @xmath473 into @xmath476 as in fig . [ contour ] , we find that by simple calculations , @xmath477 where note that the factors @xmath478s come from the cut locus of @xmath479 . from , , we get @xmath480 which leads to . the contour @xmath473 on @xmath293 , where @xmath481 . it consists of the paths @xmath482 . ] next we give a proof of . for this purpose we first show the following relation . let @xmath483 , @xmath484 , be @xmath485 then we have @xmath486 where @xmath487 is a @xmath131th order polynomial of @xmath50 where the coefficient of the highest degree is @xmath488 . this relation can be derived by considering the integration of @xmath489 with respect to @xmath294 along the contour @xmath473 in fig . [ contour ] with @xmath490 and @xmath491 . note that @xmath492 where rhs corresponds to the residue of @xmath493 at @xmath494 . as in the previous case , one easily gets @xmath495 substituting into , we find @xmath496 thus we obtain @xmath497 which leads to . here we show . we find that is rewritten as @xmath498 where @xmath148 is defined below and the functions @xmath499 and @xmath500 on @xmath501 are defined by @xmath502 and @xmath503 . we prove ( [ hformula ] ) by using ( [ iformula ] ) and by mathematical induction : suppose that holds for @xmath504 . then we get @xmath505 where @xmath506 is the coefficient of @xmath507 in @xmath508 and in the last equality we used the assumption for the mathematical induction and . considering , we arrive at . to show lemma [ vgt2 ] , we will use the following identity . for @xmath509 satisfying @xmath510 and @xmath511 , we have @xmath512 where @xmath513 is the permutation of @xmath514 . for the proof of , it is sufficient to show for @xmath515 , @xmath516 where , as in , we assume the condition @xmath517 . this can easily be obtained by noting that @xmath518 where in the first equality we used the fact that the factor @xmath519 can be omitted in this equation thanks to the factor @xmath520 with the condition @xmath521 and the last equality follows from the fact that the term with @xmath522 cancels the term with @xmath523 where @xmath523 is defined in terms of @xmath522 as @xmath524 and @xmath525 with @xmath526 for @xmath527 . using , we have for @xmath517 @xmath528 where for the first and the third equality , we used with @xmath529 and @xmath530 respectively . performing the procedure in repeatedly , we arrive at . now we give a proof of the lemma by the mathematical induction . the case @xmath51 in is trivial . suppose that it holds for @xmath531 . then noticing @xmath532 we see that lhs of is written as @xmath533 where in the second equality we used the assumption for @xmath531 . note that in the rightmost side of , the condition @xmath534 holds for the support of @xmath535 . thus we can apply to the rightmost side . we see that it becomes @xmath536 which completes the proof of lemma [ vgt2 ] . in this appendix , we give a proof of based on the saddle point method in a similar way to sec . 5.4.3 in @xcite . here we deal with the case of general @xmath452 while the case of @xmath537 was considered in @xcite . we focus mainly on the limit about @xmath40 in since the case @xmath358 can also be estimated in a parallel way . changing the variable as @xmath538 , becomes @xmath539 where @xmath540 substituting and into , we arrange the first three terms in ascending order of powers of @xmath0 as @xmath541 using the stirling formula @xmath542 for the last term in , we have @xmath543 thus from and , @xmath544 can be expressed as @xmath545 here @xmath546 , which does not depend on @xmath547 is written as @xmath548 we note that @xmath549 above has a double saddle point @xmath550 such that @xmath551 . we expand @xmath552 around @xmath553 . noting @xmath554 , @xmath555 , @xmath556 , we get @xmath557 where @xmath558 and @xmath559 are @xmath560 thus from and under the scaling @xmath561 , we have @xmath562 where @xmath563 are defined in and and @xmath564 is @xmath565 further changing the variable @xmath566 , we obtain @xmath567 hence from and , we get the limiting form of @xmath568 @xmath569 which is nothing but . as with , we rewrite @xmath358 by the change of variable @xmath570 , @xmath571 where @xmath572 is given in . applying the same techniques as above to this equation , we get the result for @xmath358 . the work of t. i. and t. s. is supported by kakenhi ( 25800215 ) and kakenhi ( 25103004 , 15k05203 , 14510499 ) respectively .

the properties of the velocity field in the local volume ( km s ) have been difficult to constrain due to a lack of a consistent set of galaxy distances . the sparse observations available to date suggest a remarkably quiet flow , with little deviation from a pure hubble law . however , velocity field models based on the distribution of galaxies in the 1.2 jy iras redshift survey , predict a quadrupolar flow pattern locally with strong infall at the poles of the local supergalactic plane . we probe this velocity field and begin to establish a consistent set of galactic distances . we have obtained images of nearby galaxies in the and band from the w.m . keck observatory and in f814w and f555w filters from the _ hubble space telescope_. where these galaxies are well resolved into stars we can use the tip of the red giant branch ( trgb ) as a distance indicator . using a maximum likelihood analysis to quantitatively measure the magnitude of the trgb we determine precise distances to seven nearby galaxies : leo i , sextans b , ngc 1313 , ngc 3109 , ugc 03755 , ugc 06456 , and ugc 07577 .

astro-ph0204192

although tremendous progress has been made in recent years in mapping the large - scale gravitational field ( out to scales of @xmath4 km s@xmath1 ) , relatively little data are available to study the velocity field of galaxies in the local neighborhood ( @xmath5 km s@xmath1 ) . the major impediment to constraining the local flow is the lack of a consistent set of distances to nearby galaxies . locally , the deviations from a pure hubble law have been predicted to be large and measurable , as much as 1 magnitude in distance modulus . if gravitational instability has generated all the large scale structure in the universe , then it must also be responsible for the extremely anisotropic local galaxy distribution ( see fig . [ figure : skydist ] ) @xcite . under this assumption , it is possible to predict the velocity field from this distribution . the catalogs from such full sky galaxy redshift surveys as the 1.2 jy iras survey @xcite allow us to estimate the large scale gravity field out to a redshift @xmath6 km s@xmath1 . in the limit of linear perturbations at late times , one expects the peculiar velocity at a point in space to be proportional to the gravity field at that point . [ figure : skydist ] demonstrates that according to linear theory , the local velocity field is expected to show substantial deviations from uniformity , with quadrupole distortions dominating . outflow is directed toward virgo and fornax on opposite sides of the sky , while substantial infall occurs from the poles of the local distribution . such a flow is expected within any gravitational model and would act to depopulate galaxies from the local poles . the bulk of the local galaxy distribution does reside on the positive and zero contours of figure [ figure : skydist ] ( the local supercluster plane ) , while the regions of negative peculiar velocity are deficient but not entirely devoid of galaxies . it is possible that linear theory does not apply on this scale , but in that case one would expect a more chaotic flow of similar amplitude . observations of local deviations from hubble flow do not really support the expectations of fig . [ figure : skydist ] , especially in the southern direction ( see the review by @xcite ) . however , the data are sparse because large spiral galaxies suitable for tully - fisher or cepheid period luminosity analysis are very rare locally , and the sky coverage , particularly in the expected infalling regions , is poor . we have thus set out to provide new distances to a complementary sample of nearby galaxies that were chosen to lie near the maximum predicted infall . according to stellar evolution theory , the tip of the first ascent red giant branch ( trgb ) marks the onset of the core helium flash in low mass stars . observationally , this phenomenon causes a distinct and abrupt termination of the bright end of the red giant branch luminosity function . this discontinuity translates directly into an excellent distance indicator : the bolometric luminosity of the trgb for low mass stars is predicted to vary by only @xmath7 mag for ages ranging from 2 up to 15 gyr @xcite and for metallicities encompassing the entire range represented by galactic globular clusters ( @xmath8 < -0.7 $ ] ) @xcite . the discontinuity is also found empirically to be stable at the @xmath7 mag level in the @xmath2 band for the same set of stellar properties . the trgb method has a strong empirical , as well as physical , basis . distances obtained using the trgb agree with the cepheid period luminosity relation ( where direct comparisons have been made ) at a @xmath95% level , e.g. @xcite an exception is for m33 : @xcite . furthermore , since all visible galaxies must have had a first generation of stars , the trgb method is applicable to populations in all morphological types , including spiral , elliptical , and even irregular galaxies , provided their metallicity is sufficiently low . the trgb method also requires far less telescope time for observations than cepheids since repeat visits are unnecessary , a considerable advantage . a good review of the trgb , its applications and theoretical underpinnings , is given in @xcite . we have used the tip of the red giant branch to obtain distances to a sample of local galaxies ( mostly dwarfs ) in order to derive their peculiar velocities . in 2 we discuss the observations we have made of several galaxies using both the w.m . keck observatory and the _ hubble space telescope _ ( _ hst _ ) . in 3 we present the methods for measuring the trgb that we have employed and include detailed information on the analysis and error propogation . finally , in 4 we present the color magnitude diagrams , luminosity functions , and derived distances for several of the galaxies studied . we discuss the derived peculiar velocities and their implications in paper ii @xcite . to select targets for observation , we assembled an all - sky sample of 502 galaxies likely to be in the local volume ( cz @xmath10 km s@xmath1 ) . this sample consists of all objects identified as galaxies ( type g ) with heliocentric radial velocities less than + 900 km s@xmath1 within the nasa / ipac extragalactic database ( ned ) . we then transformed all velocities to the local group frame @xcite and selected those having radial velocities less than + 550 km s@xmath1 . their distribution on the sky ( in galactic coordinates ) is shown in fig . [ figure : skydist ] . undoubtedly , many ( @xmath11 ) of these galaxies are not actually within the local neighborhood but are cluster members whose large peculiar velocities arising from virial motions bring them into our sample . this is clearly the case toward the virgo cluster ( @xmath12 , @xmath13 ) . we are most interested in galaxies located in the regions of strong infall predicted by the flow models of @xcite and @xcite . we have observed a subset of these galaxies with the keck telescope in hawaii and/or the _ hubble space telescope_. we observed target galaxies on the nights of 1997 december 22 and 23 under photometric conditions using the low resolution imaging spectrometer ( lris ) @xcite . at that time lris operated at the cassegrain focus of the 10-m keck ii telescope at the w.m . keck obseratory located on mauna kea , hawaii ( it now resides on keck i ) . the instrument includes a backside - illuminated tek @xmath14 ccd detetector with an imaging scale of 0.215 arcseconds / pixel and a field of view of @xmath15 arcminutes . we read out the ccd in the two - amplifier mode . each galaxy was observed through both johnson @xmath3 and cousins @xmath2 broadband filters . for the brightest , best resolved galaxies we took one exposure in each filter , of typically 300 sec duration . for the fainter , less well resolved galaxies we took between three and eight 250500 second exposures in @xmath2 and one to two 300800 second exposures in @xmath3 . see table [ table : obs ] for details . over both nights four landolt standard star fields were observed @xcite in order to calibrate the photometry . the images were reduced according to standard procedures using software written in idl . the ccd frames were subtracted for bias level ( determined from the overscan region appropriate to that amplifier ) and dark current . the right side of each frame was scaled to match the left using the gain ( 1.97 @xmath16/count for the left amplifier , 2.1 @xmath16/count for the right ) and the respective bias levels . in @xmath2 sky flats were constructed using all of a single night s image frames themselves , while in @xmath3 dome flats were used to remove pixel - to - pixel variations . the resulting image frames were found to be uniform in sensitivity across their widths to better than 1% . frames for individual galaxies were registered and combined through a median filter to remove cosmic rays and increase the signal - to - noise ratio . these combined frames were used for finding stars . due to geometric and optical distortions , the point spread function ( psf ) varies as a function of position on the lris ccd chip . it was therefore necessary to compute an empirical psf for each image from a hand - selected list of @xmath17 bright , unsaturated stars that had no stellar neighbors within @xmath18fwhm . the psf for each data image was modeled using the daophot package within iraf . stars were identified and instrumental magnitudes were computed using the fortran versions of daophot ii / allframe @xcite . these programs , modified for fits i / o , were obtained from k. gebhardt at the university of california , santa cruz . several of the fainter targets were not well resolved into stars , and thus were not usable for determining trgb distances . thus , we did not perform full photometry on these objects . the instrumental magnitudes @xmath19 and magnitude errors for each image in a filter were combined using a robust mean ( a weighted mean iteratively refined by a factor of @xmath20 for each weight , @xmath21 ) and converted to standard magnitudes @xmath22 with the equation @xmath23 where @xmath24 is the exposure time , ac is the aperture correction , x is the airmass for each observation , and c@xmath25 , c@xmath26 , c@xmath27 , and c@xmath28 are constants . the aperture correction , ac , was computed by comparing the magnitudes of bright isolated stars in each object found through an aperture of 40 pixels ( 8.6 arcsec ) in diameter to the magnitude found via psf fitting . values of the aperture corrections range from @xmath29 to @xmath30 mag , with standard errors of @xmath31 to @xmath32 mag . the broadband filters on lris are very close to the standard system , so the color terms , c@xmath27 and c@xmath28 , were found to be negligable . the extinction coefficients ( in magnitudes / airmass ) given by c@xmath26 are 0.12 at 550 nm and 0.07 at 800 nm @xcite . the magnitude zero point , c@xmath25 , was computed by using the above information and aperture photometry of both landolt standard stars and the additional stars in landolt fields recently calibrated to the standard system @xcite . the interactive iraf task _ fitparams _ in the * noao.digiphot.photcal * package was used to compute a non - linear least squares fit of the observed magnitudes to the calibrated magnitudes of these standards . the resulting zero points have an rms error of 0.02 mag in both @xmath2 and @xmath3 . we also obtained images of thirteen galaxies with the wide field planetary camera 2 ( wfpc2 ) aboard the _ hubble space telescope _ between august 13 , 1999 and may 9 , 2000 ( program 8199 , pi m. davis ) . galaxies were observed in two exposures from 1200 - 1300s each in the f814w filter ; three objects were also observed with two 1200 - 1300s exposures in the f555w filter . table [ table : obs ] details the wfpc2 observations . these thirteen galaxies were chosen specifically because they are located in regions of the sky where strong inflow is predicted by the iras data , especially around @xmath33 and @xmath34 . the calibrated and data quality images were retrieved from the space telescope science institute ( stsci ) where preliminary ccd processing , including bias- and dark - subtraction , flat fielding , and a shutter shading correction , are automatically applied using calibration data closest to the time of the observation . details of these initial reductions are given in @xcite and @xcite . for photometric measurements we used the psf - fitting package hstphot developed by a. dolphin specifically for analyzing hst / wfpc2 data . hstphot has been used recently in a number of stellar population studies @xcite . it has been optimized for the undersampled conditions present in wfpc2 data and works very well in crowded fields . hstphot is written in c and runs non - interactively from the command line . in our reductions , hstphot was an order of magnitude faster than allframe , and identified @xmath35 more stars . the algorithms used by hstphot most resemble those of dophot @xcite ; tests of hstphot photometry including a comparison with an independent dophot analysis are discussed in @xcite . our own limited tests of hstphot reliability as measured against allframe photometry showed no gross systematics , even at faint magnitudes . hstphot determines stellar magnitudes using a library of tiny tim psfs @xcite that have been built for all the commonly used wfpc2 filters . the psfs are modified internally for geometric distortions using the equations of @xcite , and corrected for the 34@xmath36 row error @xcite . the final stellar magnitudes are calibrated using the charge - transfer efficiency and zero - point corrections derived in @xcite . the data were analyzed using hstphot according to the following recipe : ( 1 ) mask bad pixels and columns using the data - quality image ; ( 2 ) reject cosmic rays ; ( 3 ) flag and mask hot pixels ; ( 4 ) generate a sky image ; ( 5 ) find stars and make photometric measurements ; and ( 6 ) determine and apply aperture corrections . the output from the _ multiphot _ procedure in hstphot was trimmed according to the following criteria to generate the final , calibrated star list : @xmath37 , @xmath38 , @xmath39 , and @xmath40 . we augmented our sample of hst data by searching the wfpc2 archive for suitable nearby galaxies with observations in the f814w and f555w filters . the blue compact dwarf galaxy ugc 06456 was found to have a well defined trgb as discussed in more detail in these data were reduced in the same way as described above . early estimates of the apparent magnitude of the trgb used a simple `` eyeball '' estimate of its location in a color - magnitude diagram ( cmd ) , e.g. @xcite . in the 1990s more quantitative methodologies were introduced @xcite . in our analysis we have used two distinct methods to quantitatively estimate the trgb magnitude from the shape of the luminosity function of a population of stars . after the exhaustion of hydrogen fuel in the core of a star there is a rapid increase in the rate at which its luminosity will change with time . because this evolution is so quick , stars that appear above the main sequence turnoff point in the cmd must have had the same main sequence lifetime , and hence the same initial mass , to within a few percent . therefore , the luminosity function of a star cluster ( or single population of stars ) past the main sequence is primarily a reflection of the rate of luminosity evolution for those stars . see the review by @xcite for details in interpreting the luminosity functions of populations of evolved stars . low - mass , post - main - sequence stars develop an inert isothermal core supported by electron degeneracy pressure . the luminosity of these stars is almost entirely dependent on the mass of that core , because all of their energy generation occurs in a thin hydrogen burning shell surrounding the core . the temperature of the shell is set by the amount of potential energy converted to heat by the gravitational contraction of the core . once on the red giant branch ( the vertical track in the cmd ) the stars rate of luminosity evolution is primarily dependent on the rate of increase of the mass of the core . due to the simple dependence of luminosity on core mass , red giant branch luminosity functions are expected to be very simple and depend only very weakly on parameters such as age , metallicity , or mass of the envelope of the star . indeed theoretical and empirical studies find the bright end of red giant branch luminosity functions to have a simple power - law behavior ; i.e. , they are straight lines in the magnitude - log(counts ) plane @xcite . models all predict essentially the same power - law slope and are in excellent agreement with observations . the luminosity function of the red giant branch terminates at the trgb , the core mass at which the helium flash occurs , and the stars then adjust their structures rapidly and move to a new location in the cmd ( the horizontal branch of helium core burning stars ) . the magnitude of the trgb is the location in the luminosity function where the red giant branch power law truncates . brightward of that break are other populations of stars ( asymptotic giant branch and red supergiant stars mostly ) but they are rare and have a distinctly different luminosity function behavior . from our data we have determined that if the red giant branch luminosity function power - law is given as @xmath41 where @xmath42 is the apparent magnitude , then @xmath43 . this fixed slope can also be found in numerous other studies of red giant branch luminosity functions ( * ? ? ? * for example ) though that fact is rarely explicitly stated . the first trgb estimation method we have applied simply uses the fact that there will be a discontinuity in the luminosity function at the location of the trgb . the first derivative of the luminosity function should therefore show a large peak at the discontinuity . early applications often used histograms to approximate the luminosity function and a sobel edge detection filter to estimate the first derivative @xcite , but this had the major drawback that the estimates could be strongly dependent on the histogram binning size . later * hereafter referred to as smf96 ) updated this method by employing a gaussian - smoothed luminosity function , @xmath44 , and a continuous edge detection function @xmath45 where @xmath46 ( as defined by smf96 ) is the mean photometric error within a bin of @xmath47 mag about magnitude @xmath42 . we have used the method of smf96 , with some minor modifications , to estimate the trgb magnitude of our objects . due to the natural power law form of the luminosity function we are most concerned with fractional changes . thus , we use a logarithmic edge detection function , @xmath48 another modification to the smf96 method is the definition of @xmath46 used . for each galaxy observed we determine a polynomial fit to our data for the photometric flux errors as a function of flux . the function is then transformed to magnitude error as a function of magnitude . thus , we define @xmath46 as simply the photometric error given by the fit at magnitude @xmath42 . the region of the luminosity function brighter than the trgb is sparsely populated ( mostly by agb stars ) and thus very noisy ; the large first derivative noise spikes can make it difficult to robustly identify the trgb automatically . we thus weight the edge detection response by the poisson noise in the smoothed luminosity function . even that fails to quiet some noise spikes and we are thus forced to limit this analysis to the region of the luminosity function that we know by inspection contains the trgb . see the figures in 4.3 for results of the edge detection ( e.g. fig . [ figure : leoi_ccut_lf ] ) binning the data is not necessary , and we have employed a maximum likelihood analysis similar in spirit to the luminosity function analysis of ( * ? ? ? * see also references therein ) . define the luminosity distribution function of stars in the galaxy under investigation to be @xmath49 where @xmath42 is the apparent magnitude and * a * is the list of parameters defining the function . the probability @xmath50 that star @xmath51 has an observed magnitude @xmath52 within a range @xmath53 is then @xmath54 where @xmath55 and @xmath56 are lower and upper bounds for the validity of the parameterized distribution function . the likelihood @xmath57 that the observed distribution of stellar magnitudes is described by this function is simply the product of the individual probablities , since each star is drawn randomly from this distribution . it is most convenient to consider @xmath58 @xmath59 where the sum extends over all stars @xmath51 with @xmath60 stars in total . for this analysis we use a simple , broken power - law model to describe the luminosity distribution function , but we also take into account the smoothing induced by measurement errors for the apparent luminosities . we use the form @xmath61 and define @xmath62 as @xmath63 we fix @xmath64 but allow as free parameters the trgb magnitude @xmath65 , the slope of the luminosity function brighter than the trgb @xmath66 , and the strength of the discontinuity at the trgb , @xmath67 . the smoothing function @xmath68 is the same error function mentioned in 3.1.1 used to estimate @xmath46 . it is a function of the apparent magnitude but not of the parameters of the fit . our procedure is then to compute the likelihood function on a grid in the parameters @xmath65 , @xmath66 , and @xmath67 and to determine the point of maximum likelihood . the method used is self normalizing ; the likelihood contours are quite peaked , with @xmath69 error widths in @xmath65 usually less than @xmath70 mag ( e.g. fig [ figure : sextansb_halo_ccut_like ] ) . to test for covariance we marginalize the likelihood by computing @xmath71 covariance with the variables , @xmath66 , and @xmath67 is found to be negligible , and the peaks in most of the marginalized likelihoods occur at the same magnitude as the peaks in the full likelihoods . in the cases reported in 4.3 there are no differences , and the largest difference in all cases is only @xmath72 mag . we fit gaussians to the marginalized likelihoods and find that the @xmath73 errors in @xmath65 are typically only a few hundredths of a magnitude . the best fitting @xmath74 curves for each galaxy are displayed in the figures of 4.3 . note that these fits are quite acceptable in each case . smoothing by the photometric errors converts the discontinuity in @xmath62 at @xmath75 into a continuous distribution @xmath74 . since the smoothing becomes larger for fainter magnitudes , the model automatically degrades the apparent slope of the trgb edge as a function of distance . the likelihood analysis works best if the red giant branch power law is observed over a range of one magnitude or more . in cases in which we failed to detect the trgb , we seldom observed a power law slope for the red giant branch , with crowding and incompleteness effects leading to a rolloff of detected stars that did not fit the trgb form . in these cases our model for @xmath62 is inadequate and the likelihood analysis is meaningless . in any event , it is necessary to set the limits of the magnitude range for which the likelihood analysis is performed ; it is critical to ensure that the faint limit avoids regions where incompleteness is non - negligible . the bright limit is less critical , but it obviously must be set brighter than the trgb . we find that setting it at @xmath76 or more magnitudes brighter than the trgb gives good results . we have used a different approach for estimating random error in the trgb magnitude from most previous authors . errors from the method of smf96 are generally obtained from the sobel edge detection response . the uncertainty of @xmath77 is taken as either the @xmath78 in a gaussian fit to the peak response or the fwhm of the peak , e.g. @xcite . instead we use bootstrap resampling of the data ( * ? ? ? * and references therein ) to simulate the act of making the same observation multiple times . this allows for the most realistic estimate of the standard deviation in the mean . the magnitudes of all the stars brighter than a given limiting magnitide @xmath79 in an entire population have some luminosity function , @xmath80 . we observe only a random fraction of the whole stellar population and so determine an observed luminosity function , @xmath81 . from this luminosity function we determine @xmath77 via either edge detection or maximum likelihood analysis ; let us simply refer to the result of either process as @xmath82 . we then randomly resample , with replacement , all of the observed magnitudes ( i.e. , any given star may appear more than once or not at all ) . each stellar magnitude , @xmath83 , carries with it some magnitude error , @xmath84 . thus , during the resampling @xmath83 is perturbed by its errors such that the resampled magnitude , @xmath85 , will be randomly drawn from a gaussian distribution about the original magnitude with @xmath84 as its standard deviation . for each resampling a bootstrap luminosity function , @xmath86 , is then determined , and @xmath87 is measured via @xmath88 . we perform the bootstrap resampling 500 times for each galaxy and take the standard deviation of the distribution of @xmath87 as the uncertainty in the measurement of @xmath77 . we find that in most cases the estimated uncertainty is a few hundredths to one tenth of one magnitude . details for each galaxy are given in with a robust measurement of the @xmath2 magnitude of the trgb , @xmath77 , we may now calculate the distance modulus . we start with the following equation : @xmath89 where @xmath90 is the bolometric magnitude of the trgb and @xmath91 is the bolometric correction to the @xmath2 magnitude of the trgb . the quantity @xmath90 was found by ( * ? * hereafter referred to as da90 ) empirically to be a function of the metallicity [ fe / h ] : @xmath92 - 3.81 .\ ] ] they also found that @xmath91 is a function of @xmath93 color , @xmath94 da90 determined an empirical relation between @xmath93 color and metallicity , but @xcite found it necessary to alter the relation slightly for more distant ( and hence fainter ) systems , obtaining @xmath95 } = -12.64 + 12.6(v - i)_{-3.5 } - 3.3(v - i)_{-3.5}^{2 } , \ ] ] where @xmath96 is the dereddened red giant branch @xmath93 color measured at an absolute @xmath2 magnitude of @xmath97 . the absolute magnitude of the trgb is given by @xmath98 to determine @xmath99 we iteratively calculate the distance modulus and the metallicity until they converge . the amount of galactic extinction ( and hence reddening ) is calculated for each galaxy using the @xmath2 and @xmath3 extinctions given in the nasa / ipac extragalactic database ( ned ) , which are based upon the dust maps of @xcite . to measure the @xmath100 color of the trgb , we calculate a histogram of the colors of stars within @xmath101 mag . we then fit a gaussian to that histogram ( binned at 0.01 mag ) and take its peak as a first guess for @xmath102 . to determine the uncertainty in this value we carry out a bootstrap resampling procedure similar to that used in determination of the uncertainty in @xmath77 . stellar magnitudes and colors are randomly resampled ( with replacement ) and perturbed by their internal random errors . the histogram calculation and gaussian fitting are then carried out for each bootstrap resampling ( we used 800 bootstrap samples ) . we fit a gaussian to the final distribution of @xmath102 values and use its peak and standard deviation as the best estimate of @xmath102 and its uncertainty . the same procedure is used to determine @xmath96 . these values are given for each galaxy in table [ table : dist ] . the calibration relations were derived over the metallicity range of the galactic globular clusters ( @xmath103 } \leq -0.7 $ ] ) and are expected to work well over ages spanning 2 - 15 gyr . the zero point was calibrated to the same level as that of the rr lyrae distance scale ( used to estimate the distances of the galactic globular clusters ) . @xcite states generally that for this range @xmath104 mag . more specifically , using the above calibration relations and metallicity range one finds @xmath105 mag with an rms uncertainty of @xmath106 mag and a systematic error of @xmath107 mag . @xcite treated the trgb as a secondary distance indicator and calibrated a zero point from galaxies with cepheid distances that yielded @xmath108 mag . also , @xcite determined a new trgb calibration based on photometry and a distance estimate from a detached eclipsing binary in the galactic globular cluster @xmath109 centauri . they found @xmath110 mag . for this paper we use the calibration of @xcite , but in paper ii we will use the calibration of @xcite for consistency with other recent trgb measurements . the true distance modulus for each galaxy is computed as @xmath111 we consider total errors in distance to be of two forms for these data . there are internal errors that are specific to each galaxy and external systematic errors that are the same for all galaxies . the internal uncertainties in our distances are determined as the sum in quadrature of random and internal systematic errors . the random errors in the distance moduli are taken as the quadrature sum of the random error in measuring @xmath77 ( estimated via bootstrap resampling ) and @xmath112 as such @xmath113 uncertainties in @xmath112 arise from the errors in measuring the @xmath114 color at @xmath77 and at @xmath115 as described in the previous section . these errors are propagated through the calibration equations ( 11 - 14 ) : @xmath116 @xmath117 } , \ ] ] and @xmath118 } = |12.6 - 6.6 ( v - i)_{-3.5}| \sigma_{(v - i)_{-3.5},{\rm bootstrap } } .\ ] ] internal systematic errors include the uncertainties in the aperture corrections , the normalization to the standard photometric system ( _ i.e. _ the photometric zero points ) , and the foreground extinction and reddening . thus , we have @xmath119 @xmath120 and @xmath121 the systematic magnitude errors ( @xmath122 and @xmath123 ) are the quadrature sums of the aperture and zero point corrections in each bandpass . the uncertainties in extinction ( @xmath124 ) and reddening ( @xmath125 ) are taken to be 10% of their values ( as per @xcite ) . the values of @xmath126 range from @xmath127 mag ( sextans b and ugc 06456 using the maximum likelihood method ) to @xmath128 mag ( leo i and ugc 03755 using the edge detection method ) . the values of @xmath129 range from @xmath127 mag to @xmath130 mag . for galaxies without observations in @xmath3 we use the full range of color / metallicity under which the calibrations were determined . this yields an rms uncertainty of @xmath106 mag . the total internal uncertainties in the true distance moduli are thus @xmath131 ranging from @xmath130 mag ( sextans b and ugc 06456 using the maximum likelihood method ) to @xmath132 mag ( leo i and ugc 03755 using the edge detection method ) . the principle identified external systematic errors in the determination of the distance moduli arise from the uncertainty in the rr lyrae distance scale for globular clusters ( @xmath133 mag ; see @xcite ) , and the uncertainty in the trgb calibration relations ( @xmath134 mag ) : @xmath135 @xcite give the dispersion in equation ( 12 ) as @xmath136 mag and the dispersion in equation ( 11 ) as @xmath137 mag . thus , the total systematic error common to all our trgb distance measurements is @xmath107 mag . in order to gain confidence in our photometry and trgb detection methodology we observed a few local galaxies with keck that have been studied quite extensively by previous authors , including leo i , ngc 3109 , and sextans b. we find that our derived distance moduli for these galaxies are in excellent agreement with those of previous authors . in our hst dataset we were able to successfully detect the trgb for three galaxies and possibly a fourth . the definite detections were ugc 07577 , ugc 06456 , and ngc 1313 , and the possible detection was ugc 03755 . there were several galaxies in our combined keck/_hst _ dataset , such as ngc 2366 and ngc 2683 , for which we were unable to successfully determine distances . based upon distance estimates by previous authors using various methods we believe that these galaxies are sufficiently far away that their red giant branch stars were not resolved in our data . a summary of trgb distances and uncertainties is presented in table [ table : dist ] . color - magnitude diagrams ( cmds ) , luminosity functions , and other details for these galaxies are now presented . the dwarf spheroidal galaxy leo i is thought to be one of the most distant satellites of the milky way . our observations with keck were able to fully resolve leo i into stars down to @xmath138 mag ( see fig . [ figure : leoi_image ] ) . the color - magnitude diagram for the entire field reveals a very simple stellar population of red giants ( fig . [ figure : leoi_cmd ] ) . there appear to be almost no blue stars that may contaminate the luminosity function . the red giant branch is extremely well defined with an unambiguous tip at @xmath139 mag . plotted in fig . [ figure : leoi_cmd ] is an red giant branch locus for leo i. we determine this locus by using the value of @xmath140 to interpolate between the two nearest of the four galactic globular cluster loci determined by da90 : m15 , m2 , ngc 1851 , and 47 tuc . the few stars above the tip are likely asymptotic giant branch ( agb ) stars . also visible is the red clump of helium burning stars at @xmath141 mag and @xmath142 mag . [ figure : leoi_ccut_lf ] shows the logarithmic luminosity function for all stars with @xmath143 mag . the upper red giant branch power law is quite obvious and has a clear break . both the maximum likelihood model and the modified sobel edge detector agree on the location of the break . for the maximum likelihood analysis we determine @xmath144 mag as shown in fig [ figure : leoi_ccut_like ] . bootstrap resampling gives an uncertainty of @xmath145 mag , with a distribution shown in fig . [ figure : leoi_ccut_lf ] . the edge detector also finds @xmath144 mag with a bootstrap uncertainty of @xmath134 mag . these uncertainties are among the largest of the galaxies reported in this paper , despite leo i being the closest by far . the large uncertainty is due to the small number of stars at the trgb . with so few stars at the trgb it is easier for it to move by a few hundredths of a magnitude with each bootstrap . @xcite studied the effect of small population sizes and estimated that if more than @xmath11 stars can be imaged in the first full magnitude interval then any systematic effect in the distance modulus would be well less than @xmath70 mag . there are just over 100 stars in the first magnitude of the luminosity function of leo i. thus , using the method described in 4.1 we find @xmath146 mag and [ fe / h ] = @xmath147 dex . with a line of sight extinction of @xmath148 mag we find @xmath149(internal)@xmath107(systematic ) mag for the maximum likelihood case . this is in excellent agreement with a previously published distance modulus for leo i of @xmath150 mag @xcite . another estimate of the trgb for leo i was made by @xcite , who found @xmath151 mag ( yielding @xmath152 mag for the assumed @xmath153 mag ) . the difference between their measurement of @xmath77 and that presented here is @xmath154 mag , again in good agreement . sextans b is a dwarf irregular galaxy located at the edge of the local group . it was well resolved in our keck observations . [ figure : sextansb_cmd ] shows the cmd for the @xmath155 resolved stars with good photometry in the lris field . the stellar population appears fairly simple ; the red giant branch is quite prominent and red giants clearly dominate the light of the galaxy . there is little contamination from blue stars . the stars above the red giant branch are likely agb stars that do not obscure the conspicuous trgb . in order to minimize these contaminants and those arising from crowding and photometric errors we examined the cmd of only the stars in an arbitrary halo region . [ figure : sextansb_halo_reg ] shows the positions of the 9500 stars in this halo region and their cmd . the trgb was already obvious in fig . [ figure : sextansb_cmd ] but it can now be seen more clearly . additionally , we make a color cut in the halo star list . [ figure : sextansb_halo_ccut_lf ] shows the resulting luminosity function of halo stars with @xmath156 mag . again , the power - law distribution with index @xmath64 is evident and has an unambiguous break at @xmath157 mag in the maximum likelihood analysis as shown in fig . [ figure : sextansb_halo_ccut_like ] ( also @xmath157 mag in the edge - detector ) . bootstrap resampling of stars in the color - cut halo region gives an uncertainty of @xmath127 mag , shown in panels 3 and 4 of fig . [ figure : sextansb_halo_ccut_lf ] . the trgb color is estimated to be @xmath158 mag and the metallicity is [ fe / h ] = @xmath159 dex . the @xcite dust maps give the foreground extinction as @xmath160 mag . we assume that internal extinction is negligible as we are focusing on the halo region that is presumably mostly dust - free . thus for the maximum likelihood method we determine a population ii distance modulus of @xmath161(internal)@xmath107(systematic ) mag corresponding to a heliocentric metric distance of @xmath162 mpc . this result is in superb agreement with a previously published study of sextans b by @xcite using both the trgb ( @xmath163 ) and cepheids ( @xmath164 mag ) . ngc 3109 is another dwarf irregular galaxy , classified sm iv , at the edge of the local group . it has a flattened disk region apparently viewed nearly edge - on . our keck observations have brilliantly resolved the galaxy into stars . in fig . [ figure : ngc3109_cmd ] we present the cmd for the entire lris field . this cmd is considerably more complex than those of sextans b or leo i. a red giant branch is evident but there is large contamination from blue main sequence and asymptotic giant branch stars . there are also likely some red supergiants in the mix to further complicate matters . as we did with sextans b , we selected stars from a halo region to reduce contamination around the trgb . the pixel map and cmd for these stars is shown in fig . [ figure : ngc3109_halo_reg ] . the red giant branch is now much more apparent as is its tip . the luminosity function of halo stars with @xmath165 mag is shown in fig . [ figure : ngc3109_halo_ccut_lf ] . the two trgb detection methods agree quite well and we find @xmath166 mag for the maximum likelihood analysis , as shown in fig . [ figure : ngc3109_halo_ccut_like ] , and @xmath167 mag for the logarithmic edge detector . the best estimate of the trgb color is @xmath168 mag , with [ fe / h ] = @xmath169 dex . for the maximum likelihood method and using an extinction value of @xmath170 mag we find the distance modulus @xmath171(internal)@xmath107(systematic ) mag , which corresponds to a distance from the sun of @xmath172 mpc . this is within the errors of a previously published work on ngc 3109 @xcite which found @xmath173 mag using the tip of the red giant branch . the dwarf irregular galaxy ugc 07577 was one of our _ hst _ target galaxies . based on our experience with the keck data we decided to minimize _ hst _ observation time by only obtaining images in the fw814 filter . this was done for galaxies not expected to have considerable contamination from blue main sequence or asymptotic giant branch populations . even a large agb population is not likely to affect the trgb detection severely as it would surely have a different power - law slope in @xmath2 band from the red giant branch , so the break in the red giant branch power - law at the trgb would remain . the galaxy was well resolved into stars ( see fig . [ figure : ugc07577_image ] ) . the galaxy ( transformed into standard cousins @xmath2 ) was broken up into core and halo regions as was done for the keck objects . the luminosity functions of the entire field and of the halo are shown in fig . [ figure : ugc07577_lf ] , and fig . [ figure : ugc07577_halo_lf ] respectively . all show the same familiar red giant branch power - law slope , @xmath64 . the likelihood function for the entire field is maximized at @xmath174 mag and the edge detector gives @xmath175 mag . for the halo region the likelihood function is maximized at @xmath176 mag as shown in fig . [ figure : ugc07577_halo_like ] and this is the value we adopt . we assume that the mean metallicity of the ugc 07577 red giants is within the calibrated range of the galactic globular clusters . thus we adopt an absolute magnitude of @xmath177 mag . using the maximum likelihood value and an extinction of @xmath178 mag we derive a distance modulus for ugc 07577 of @xmath179(internal)@xmath145(systematic ) mag , which is a heliocentric distance of @xmath180 mpc . ugc 07577 has been little studied previously , but one recent estimate by @xcite gives the distance as 4.8 mpc . this is extremely discrepant , however they were estimating the distance using the magnitudes of the brightest blue super giant stars . several authors have commented on the errors associated with this method , e.g. @xcite . while there have been differences between their estimates the errors are generally uncomfortably large ( @xmath181 mag for the brightest blue stars ) . ngc 1313 is classified as a sb(s)d barred spiral and was described by @xcite as a transition object from the normal barred spirals to the irregular magellanic type barred spirals . it has considerable ongoing star formation and an active nucleus . certainly the mixed stellar populations and the internal extinction of the disk are cause for concern in making a clear distance measurement based on the trgb . for these reasons we obtained _ hst _ observations in the halo region of this galaxy 3.67 arcmin from its center ( see fig . [ figure : ngc1313_dss_overlay ] ) . in this region we expect the dominant stars to be population ii red giants . we obtained images only in the f814w bandpass for ngc 1313 as we expected little contamination from other stellar types in this region of the galaxy . in fig . [ figure : ngc1313_halo_lf ] we present the @xmath2 band luminosity function for the entire wfpc2 field of view . the tell - tale power - law signature ( @xmath64 ) of the red giant branch is visible from @xmath182 mag down to the photometric limits at @xmath183 mag . while there are many stars brighter than @xmath182 mag they would appear to cause little doubt that the red giant branch terminates there . the maximum likelihood analysis gives the magnitude of the trgb for ngc 1313 as @xmath184 mag , as shown in fig . [ figure : ngc1313_halo_like ] , and the modified sobel edge detector gives @xmath185 mag . since we do not have color information for the ngc 1313 halo we assume that its metallicity is within the calibrated range and thus use @xmath177 mag . using the maximum likelihood value and the @xcite extinction of @xmath186 mag we obtain a population ii distance modulus for ngc 1313 of @xmath187(internal)@xmath107(systematic ) mag corresponding to a heliocentric distance of @xmath188 mpc . using the magnitudes of brightest resolved stars , @xcite estimated the distance of ngc 1313 at 4.5 mpc . @xcite additionally gives the distance to ngc 1313 as 3.7 mpc . in addition to our own _ data we searched through the _ hst _ archive at stsci for wfpc2 observations of galaxies in our all sky sample in at least f814w . while we found several good candidates , most of the observations we found were for disk regions within spiral galaxies . we have found at least one excellent dataset to add to our study . the observations ( program 6276 , pi j. westphal ) were taken as part of a project to study peculiar and interacting galaxies . from the archive data we used three 1400s exposures in f555w and three 1400s exposures in f814w of the blue compact dwarf galaxy ( bcd ) ugc 06456 . we reduced these data as described in 2.2 for our own _ hst _ observations . the observations resolved the galaxy brilliantly into stars revealing a bright and knotted population of blue stars at its core and a smooth background of red stars . [ figure : ugc06456_cmd ] shows our color - magnitude diagram for the entire wfpc2 field . evident in the cmd is a strong population of blue main sequence stars as well as a prominent red giant branch . there are also a small number of agb stars visible above the red giant branch . we were able to remove the contamination from the blue stars by excluding the stars from the central region as shown in fig . [ figure : ugc06456_halo_reg ] . furthermore we exclude stars with colors outside the range @xmath189 mag . [ figure : ugc06456_halo_ccut_lf ] shows the resulting luminosity function . the trgb is extremely obvious and the maximum likelihood analysis finds the trgb at @xmath190 mag , as shown in fig . [ figure : ugc06456_halo_ccut_like ] , and the color of the trgb is @xmath191 mag . we thus take the metallicity of the red giant branch to be [ fe / h ] = @xmath192 . using an extinction of @xmath193 mag we find the distance modulus for ugc 06456 to be @xmath194(internal)@xmath107(systematic ) mag and hence the heliocentric distance to be @xmath195 mpc . this is in good agreement with the trgb analysis carried out by previous authors using these same data @xcite who found @xmath196 and @xmath197 mag respectively . ugc 03755 is a typical dwarf irregular galaxy that has been little studied by previous authors . it was of particular interest to us because it lies in the region of the sky with the strongest predicted infall ( see fig . [ figure : skydist ] ) . [ figure : ugc03755_image ] shows one of our f814w exposures of ugc 03755 . the galaxy appears well resolved into stars . however , inspection of the cmd ( fig . [ figure : ugc03755_cmd ] ) shows that we did not resolve the galaxy very deeply , especially in the f555w bandpass . there is only the slightest hint of a red giant branch , and the f555w photometry limits cut into it almost immediately . making region or color cuts does not actually help a detection in this case . however , upon inspection of the luminosity function for ugc 03755 ( fig . [ figure : ugc03755_lf ] ) we can see the red giant branch power - law signature for @xmath198 mag brighter than the magnitude limits in @xmath2 band . there is no strong discontinuity in the luminosity function as seen with the previous galaxies but there is a definite break in the power - law . we interpret this break as the tip of the red giant branch at @xmath199 mag , as measured weakly by the maximum likelihood algorithm and shown in fig . [ figure : ugc03755_like ] . because the red giant branch is not well defined we chose to treat ugc 03755 as a case in which we did not have color data and hence we assume that its metallicity is within the calibrated range and adopt @xmath177 mag . we use the @xcite line of sight extinction of @xmath200 mag , which yields a distance modulus of @xmath201(internal)@xmath107(systematic ) mag . this corresponds to a heliocentric distance of @xmath202 mpc . we take this as a lower limit on the distance to this galaxy , for if it were closer we should have seen a definitive red giant branch . likewise we take this distance as the upper limit for our ability to measure distances in our _ hst _ dataset . since all of our exposure times were the same we would not resolve the red giant branch population of any galaxy that is farther than this . @xcite estimated @xmath203 mag for this galaxy , based upon the magnitudes of the brightest resolved stars . in another paper @xcite they estimated the calibration errors in their method to be @xmath204 mag . there is no strong agreement , but again we believe that ugc 03755 can not be any closer than @xmath205 mpc . with success comes failure ; we were not able to determine trgb distances to many of our target galaxies . based on our successful measurements and on work done by previous authors , we do not think that our failures were caused by contaminating stellar populations . rather we think that in most cases the data were simply not deep enough to resolve out red giant branch stars . hence , the galaxies are too far away for our data to reveal the trgb . we estimate that in our keck data we would not have been able to detect any red giant branch ( or tip thereof ) fainter than @xmath206 mag . we take the limiting magnitude as at least @xmath76 magnitudes brighter than the turn over in the luminosity function . this means that we could not have successfully measured a trgb distance for any galaxy with a distance greater than @xmath207 mpc . this was unfortunate as many of our keck targets are expected members of the m81 group . one example is ngc 2366 . this dwarf irregular galaxy has been estimated to lie at a distance @xmath208 mpc , which is consistent with other m81 group members ( * ? ? ? * and references therein ) . our cmd for this galaxy ( fig . [ figure : ngc2366_cmd ] ) shows no recognizable structure , and the luminosity function turns over at about @xmath209 mag never having any red giant branch power - law slope . clearly our data for this object did not go deep enough to resolve a red giant branch population . the situation is much the same for our _ hst _ data . the typical detection limit in the _ hst _ data is @xmath210 mag , corresponding to a linear distance limit of @xmath211 mpc . an example of a galaxy we failed to measure a trgb distance for is ngc 2683 . this is a spiral galaxy seen nearly edge - on . we avoided the disk region and observed only in the halo region of the galaxy to minimize any population i contamination . a recent survey of galaxies by @xcite that used surface brightness fluctuations as a distance indicator gave a distance for this galaxy of @xmath212 mpc . this galaxy is then clearly outside the reach of our data . there is little wonder that we see no hint of a red giant branch power - law in the luminosity function ( fig . [ figure : ngc2683_lf ] ) . we have successfully used the trgb as a distance indicator for seven local galaxies observed with _ hst _ and keck . using knowledge of the luminosity function behavior of the red giant branch , we conducted a maximum likelihood procedure to estimate the @xmath2 band magnitude of the trgb . the uncertainty in the estimates of @xmath77 and @xmath213 were calculated via bootstrap resampling which makes no assumption about the underlying distributions and allows one to model the act of making repeated measurements of the same quantity . bootstrap resampling also allows us to better estimate the error in @xmath77 when the number of red giant branch stars at that magnitude is small . we find that compared to sobel edge detection the maximum likelihood analysis is equally accurate but more precise , and we find no systematic difference in the derived values of @xmath77 between the two methods . indeed we find the weighted mean absolute difference in @xmath77 to be @xmath214 mag ( the simple average absolute difference in @xmath77 is @xmath215 mag ) . the average ratio of uncertainties is @xmath216 . therefore , on average , the maximum likelihood method is 38% more precise than the sobel edge detection . it should also be noted that other authors ( e.g. @xcite ) have found sobel edge detection to be a biased estimator of the trgb . the bias depends on the amount of smoothing in the luminosity function at the location of the trgb . for our data this bias was less than 0.04 mag which is less than random errors and therefore we would not be sensitive to it in a comparison to the maximum likelihood . it is a positive feature of the trgb method that it does not depend very sensitively on the statistical technique used to define the tip . the trgb method will be most useful in further broadening our knowledge of distances to local galaxies and allow us to study the local flow in greater detail . bryan mndez acknowledges support from a ford foundation predoctoral fellowship for minorities . some of the data presented herein were obtained at the w.m . keck observatory , which is operated as a scientific partnership among the california institute of technology , the university of california and the national aeronautics and space administration . the observatory was made possible by the generous financial support of the w.m . keck foundation . support for proposal # 8199 was provided by nasa through a grant from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 . this research has made use of the 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 . km s@xmath1 ( in the frame of the local group ) derived from the iras galaxy distribution ( see @xcite , for details ) . contours are in units of 50 km s@xmath1 with the dash dot lines representing infall peculiar velocity and solid lines representing outflow peculiar velocity . the amplitude of the flow is almost linear in @xmath217 , where @xmath218 is the linear bias in the iras galaxy distribution relative to the mass distribution . the plot assumes @xmath219 . galaxies with @xmath0 km s@xmath1 are plotted.,height=696 ] mag ( top panel ) . overplotted is the best - fit model luminosity function determined via maximum likelihood analysis . the second panel plots the weighted logarithmic edge detection response ; the dot dash line indicates the trgb location . the third and fourth panels plot the distribution of the bootstrap results for edge detection and maximum likelihood methods , respectively.,height=768 ] mag ( top panel ) . overplotted is the best - fit model luminosity function determined via maximum likelihood analysis . the second and third panels plot @xmath73 , @xmath220 , and @xmath69 contours of the likelihood function parameters @xmath67 and @xmath66 , respectively . the fourth panel plots the marginalized likelihood . a gaussian fit with @xmath221 mag is overplotted as a dashed line.,height=768 ] mag ( top panel ) . overplotted is the best - fit model luminosity function determined via maximum likelihood analysis . the second panel plots the weighted logarithmic edge detection response with a dot dash line indicating the trgb . the third and fourth panels plot the distribution of the bootstrap results for edge detection and maximum likelihood methods , respectively.,height=768 ] mag ( top panel ) . overplotted is the best fit model luminosity function determined via a maximum likelihood analysis . the second and third panels plot @xmath73 , @xmath220 , and @xmath69 contours of the likelihood function parameters @xmath67 and @xmath66 . the fourth panel plots the marginalized likelihood . a gaussian fit with @xmath222 mag is overplotted as a dashed line.,height=768 ] . the second and third panels plot @xmath73 , @xmath220 , and @xmath69 contours of the likelihood function parameters @xmath67 and @xmath66 . the fourth panel plots the marginalized likelihood . a gaussian fit with @xmath223 mag is overplotted as a dashed line.,height=768 ] . the second and third panels plot @xmath73 , @xmath220 , and @xmath69 contours of the likelihood function parameters @xmath67 and @xmath66 . the fourth panel plots the marginalized likelihood . a gaussian fit with @xmath222 mag is overplotted as a dashed line.,height=768 ] . the second and third panels plot @xmath73 , @xmath220 , and @xmath69 contours of the likelihood function parameters @xmath67 and @xmath66 . the fourth panel plots the marginalized likelihood . a gaussian fit with @xmath224 mag is overplotted as a dashed line.,height=768 ] . the second and third panels plot @xmath73 , @xmath220 , and @xmath69 contours of the likelihood function parameters @xmath67 and @xmath66 . the fourth panel plots the marginalized likelihood . a gaussian fit with @xmath221 mag is overplotted as a dashed line . , height=768 ] . the second and third panels plot @xmath73 , @xmath220 , and @xmath69 contours of the likelihood function parameters @xmath67 and @xmath66 . the fourth panel plots the marginalized likelihood . a gaussian fit with @xmath224 mag is overplotted as a dashed line.,height=768 ] lrrcllrrrl leo i & 225.982 & 49.110 & 23/12/97 & keck / lris & i / v & 2/1 & 300/300 & 0.85 - 0.94/0.92 & success + sextans b & 233.200 & 43.784 & 22/12/97 & keck / lris & i / v & 1/1 & 300/300 & 0.75/0.78 & success + holmberg i & 140.729 & 38.662 & 23/12/97 & keck / lris & i / v & 1/1 & 300/300 & 0.97/1.08 & poor image + holmberg ii & 144.283 & 32.691 & 22/12/97 & keck / lris & i / v & 4/1 & 300/600 & 0.82 - 0.93/0.86 & shallow + holmberg ix & 141.977 & 41.052 & 22/12/97 & keck / lris & i / v & 1/1 & 300/300 & 0.75/0.88 & shallow + ic 342 & 138.173 & 10.580 & 22/12/97 & keck / lris & i / v & 4/1 & 300/600 & 0.77 - 0.90/1.02 & crowded + ic 2574 & 140.210 & 43.605 & 23/12/97 & keck / lris & i / v & 3/1 & 250 - 300/300 & 0.91 - 1.02/1.02 & shallow + ngc 1560 & 138.367 & 16.022 & 22/12/97 & keck / lris & i / v & 4/1 & 300/600 & 1.10 - 1.29/1.40 & shallow + ngc 2366 & 146.419 & 28.535 & 22/12/97 & keck / lris & i / v & 4/1 & 300/600 & 0.77 - 1.11/0.94 & shallow + ngc 2903 & 208.711 & 44.540 & 22/12/97 & keck / lris & i / v & 3/1 & 300/400 & 0.74 - 0.78/0.86 & shallow + ngc 2976 & 143.914 & 40.902 & 22/12/97 & keck / lris & i / v & 3/1 & 300/400 & 0.87 - 0.94/0.84 & shallow + ngc 3109 & 262.101 & 23.070 & 22/12/97 & keck / lris & i / v & 1/1 & 300/300 & 0.88/0.90 & success + ugc 02684 & 166.322 & -32.747 & 22/12/97 & keck / lris & i / v & 10/2 & 300 - 400/600 - 800 & 0.61 - 0.89/0.72 - 0.96 & unresolved + ugc 02716 & 166.777 & -31.825 & 22/12/97 & keck / lris & i / v & 7/2 & 300 - 400/600 - 800 & @xmath225 & unresolved + ugc 03974 & 203.100 & 18.541 & 22/12/97 & keck / lris & i / v & 8/2 & 400 - 500/800 - 900 & @xmath225 & unresolved + ugc 04115 & 207.003 & 20.897 & 22/12/97 & keck / lris & i / v & 8/2 & 300 - 400/600 - 800 & 0.64 - 0.83/0.76 & unresolved + ugc 05272 & 195.416 & 50.563 & 23/12/97 & keck / lris & i / v & 4/1 & 300/600 & 0.77 - 0.80/0.78 & unresolved + ugc 05340 & 199.887 & 51.614 & 23/12/97 & keck / lris & i / v & 4/1 & 300/600 & 0.74 - 0.92/0.94 & unresolved + ic 342 & 138.173 & 10.580 & 14/08/99 & _ hst_/wfpc2 & f814w / f555w & 2/2 & 1300/1300 & @xmath225 & crowded + iras 12483 - 1311 & 302.747 & 49.415 & 22/01/00 & _ hst_/wfpc2 & f814w & 2 & 1200 & @xmath225 & bad z + ngc 1313 & 283.472 & -44.654 & 14/01/00 & _ hst_/wfpc2 & f814w & 2 & 1300 & @xmath225 & success + ngc 2683 & 190.457 & 38.761 & 15/02/00 & _ hst_/wfpc2 & f814w & 2 & 1300 & @xmath225 & shallow + ngc 2903 & 208.711 & 44.540 & 02/05/00 & _ hst_/wfpc2 & f814w & 2 & 1200 & @xmath225 & shallow + ngc 6503 & 100.574 & 30.640 & 08/11/99 & _ hst_/wfpc2 & f814w & 2 & 1300 & @xmath225 & shallow + ugc 03476 & 180.766 & 10.513 & 13/02/00 & _ hst_/wfpc2 & f814w / f555w & 2/2 & 1300/1300 & @xmath225 & shallow + ugc 03698 & 172.971 & 21.622 & 11/02/00 & _ hst_/wfpc2 & f814w & 2 & 1300 & @xmath225 & shallow + ugc 03755 & 206.013 & 9.715 & 26/03/00 & _ hst_/wfpc2 & f814w / f555w & 2/2 & 1200/1200 & @xmath225 & marginal + ugc 03860 & 177.806 & 23.934 & 02/07/00 & _ hst_/wfpc2 & f814w & 2 & 1300 & @xmath225 & shallow + ugc 03974 & 203.100 & 18.541 & 22/01/00 & _ hst_/wfpc2 & f814w & 2 & 1200 & @xmath225 & shallow + ugc 04115 & 207.003 & 20.897 & 24/09/99 & _ hst_/wfpc2 & f814w & 2 & 1200 & @xmath225 & shallow + ugc 07577 & 137.751 & 72.945 & 09/05/00 & _ hst_/wfpc2 & f814w & 2 & 1300 & @xmath225 & success + ugc 06456@xmath226 & 127.836 & 37.327 & 10/07/95 & _ hst_/wfpc2 & f814w / f555w & 3/3 & 1400/1400 & @xmath225 & success + [ table : obs ] llllllllll leo i & @xmath227 & @xmath228 & @xmath229 & @xmath147 & @xmath230 & @xmath231 & @xmath232 & @xmath233 & @xmath234 ngc 3109 & @xmath235 & @xmath236 & @xmath162 & @xmath169 & @xmath237 & @xmath238 & @xmath239 & @xmath240 & @xmath172 sextans b & @xmath241 & @xmath242 & @xmath243 & @xmath159 & @xmath244 & @xmath245 & @xmath246 & @xmath247 & @xmath162 ugc 07577 & @xmath248 & @xmath225 & @xmath225 & @xmath225 & @xmath249 & @xmath232 & @xmath250 & @xmath251 & @xmath180 ngc 1313 & @xmath252 & @xmath225 & @xmath225 & @xmath225 & @xmath249 & @xmath253 & @xmath254 & @xmath255 & @xmath188 ugc 06456 & @xmath256 & @xmath236 & @xmath257 & @xmath192 & @xmath258 & @xmath231 & @xmath232 & @xmath259 & @xmath195 ugc 03755 & @xmath260 & @xmath225 & @xmath225 & @xmath225 & @xmath249 & @xmath261 & @xmath262 & @xmath263 & @xmath202 leo i & @xmath264 & @xmath228 & @xmath229 & @xmath265 & @xmath266 & @xmath231 & @xmath232 & @xmath267 & @xmath268 ngc 3109 & @xmath269 & @xmath236 & @xmath270 & @xmath271 & @xmath237 & @xmath238 & @xmath239 & @xmath272 & @xmath273 sextans b & @xmath274 & @xmath242 & @xmath243 & @xmath275 & @xmath244 & @xmath245 & @xmath246 & @xmath276 & @xmath277 ugc 07577 & @xmath278 & @xmath225 & @xmath225 & @xmath225 & @xmath249 & @xmath232 & @xmath250 & @xmath279 & @xmath280 ngc 1313 & @xmath281 & @xmath225 & @xmath225 & @xmath225 & @xmath249 & @xmath253 & @xmath254 & @xmath282 & @xmath283 ugc 06456 & @xmath284 & @xmath285 & @xmath286 & @xmath287 & @xmath258 & @xmath231 & @xmath232 & @xmath288 & @xmath289 ugc 03755 & @xmath290 & @xmath225 & @xmath225 & @xmath225 & @xmath249 & @xmath261 & @xmath262 & @xmath291 & @xmath292 ngc 2366 & @xmath293 & @xmath225 & @xmath225 & @xmath225 & @xmath294 & @xmath231 & @xmath232 & @xmath295 & @xmath296 ugc 03755 & @xmath297 & @xmath225 & @xmath225 & @xmath225 & @xmath294 & @xmath261 & @xmath262 & @xmath298 & @xmath299 ngc 2683 & @xmath300 & @xmath225 & @xmath225 & @xmath225 & @xmath294 & @xmath301 & @xmath246 & @xmath302 & @xmath303 ngc 2366 & @xmath304 23.00 & @xmath225 & @xmath225 & @xmath225 & @xmath294 & @xmath231 & @xmath232 & @xmath304 26.94 & @xmath304 2.44 ugc 03755 & @xmath304 24.50 & @xmath225 & @xmath225 & @xmath225 & @xmath294 & @xmath261 & @xmath262 & @xmath304 28.34 & @xmath304 4.65 ngc 2683 & @xmath304 25.00 & @xmath225 & @xmath225 & @xmath225 & @xmath294 & @xmath301 & @xmath246 & @xmath304 28.95 & @xmath304 6.15 [ table : dist ]

axisymmetric magnetohydrodynamic ( mhd ) simulations have been applied to investigate the interrelation of a central stellar magnetosphere and stellar wind with a surrounding magnetized disk outflow and how the overall formation of a large scale jet is affected . the initial magnetic field distribution applied is a superposition of two components - a stellar dipole and a surrounding disk magnetic field , in both either parallel or anti - parallel alignment . correspondingly , the mass outflow is launched as stellar wind plus a disk wind . our simulations evolve from an initial state in hydrostatic equilibrium and an initially force - free magnetic field configuration . due to initial differential rotation and induction of a strong toroidal magnetic field the stellar dipolar field inflates and is disrupted on large scale . stellar and disk wind may evolve in a pair of collimated outflows . the existence of a reasonably strong disk wind component is essential for collimation . a disk jet as known from previous numerical studies will become de - collimated by the stellar wind . in some simulations we observe the generation of strong flares triggering a sudden change in the outflow mass loss rate by a factor of two and also a re - distribution in the radial profile of momentum flux and jet velocity across the jet . we discuss the hypothesis that these flares may trigger internal shocks in the asymptotic jets which are observed as knots .

0810.4154

astrophysical jets are launched by magnetohydrodynamic ( mhd ) processes in the close vicinity of the central object an accretion disk surrounding a protostar or a compact object @xcite . however , the principal mechanism which actually launches the outflow - the transition from accretion to ejection - for a certain disk at a certain time is still not completely understood . during the recent decade , numerical simulations of mhd jet formation became more and more feasible and substantially helped to improve our understanding of how jets emerge . in general , these simulations may be distinguished in those taking into account the time evolution of the disk structure ( see e.g. @xcite ) and others considering the disk surface as a fixed - in - time boundary condition for the disk wind or jet ( see e.g. @xcite ) . the first approach allows to study the launching process directly in particular the mechanism lifting matter from the disk plane into the outflow . however , this approach is computationaly very expensive and yet limited by spatial and time resolution . in order to study the jet formation process - the acceleration and collimation of a disk / stellar wind - it is essential to follow the dynamical evolution of the system for many ( several thousands ) of rotational periods and on a sufficiently large grid with appropriate resolution . for such a goal , the second approach is better suited . since it is computationally less expensive it also allows to perform series of simulations for parameter studies . it is clear that the prescription of mass flow rate and magnetic flux profile constrains the result of the simulation more than a consistent simulation of the jet - disk evolution which could in fact provide the mass loss rate from the disk into the jet . on the other hand , the current status of mhd disk modeling has its own limitations . in particular the magnetic field structure in the disk is a rather open question unless radiative mhd , global simulations of dynamo - active disk models provide fully self - consistent results . the aim of this paper was to investigate the interaction and interrelation between a stellar wind and a disk wind and represents a unique approach in that field . for this first step it is of advantage to govern the simulation by well understood boundary conditions . further , for our goal it is essential to study the long term evolution of the outflow at considerable distances from the star . if we would include the disk evolution in the simulation it would be hard to reach the appropriate time scales . in fact all disk simulations so far stop at earlier time scales . future work should include the disk evolution for the jet launching . one may further distinguish between the different initial setup for these simulations - some of them consider a pure stellar dipole ( see e.g. @xcite ) , others a pure disk field ( see e.g. @xcite ) . the case of a superposed stellar and disk magnetic field is yet rarely treated in simulations , however , the first such model configuration was discussed already by @xcite . one example is @xcite who superposed a central dipole with an aligned vertical disk field . in this paper , we study the long - term evolution of a two - component mhd outflow consisting of a stellar wind launched from a stellar magnetosphere and a surrounding disk wind . it essential to follow the time evolution for very long term in order to be able to take into account the evolution also of the outer regions of the disk magnetosphere as much as possible . the magnetic field of protostellar jets ( and probably also those of micro - quasars ) most probably consists of two components - a central stellar , probably dipolar field plus a field component provided by the surrounding accretion disk ( either generated by a disk dynamo or advected from interstellar space or both ) . in the following we qualitatively expose the main aspects of the interaction between a central stellar field and the disk magnetic field and how that may affect the overall jet formation . _ enhanced magnetic flux_. the stellar field adds magnetic flux to the system . assuming a polar field strength @xmath0 and a stellar radius @xmath1 , the dipolar field scales with @xmath2 this has to be compared to the disk poloidal magnetic field provided either by a disk dynamo or by advection of ambient interstellar field . the latter can be estimated by equipartition arguments , and is limited to @xmath3 the stellar magnetic field will not remain closed , but will inflate due to shear between the disk surface and the star ( e.g. @xcite ) . the additional poynting flux that threads the disk may support the mhd jet launching and may provide an additional energy reservoir for the conversion of magnetic energy to jet kinetic energy , thus , implying a greater asymptotic jet speed ( michel scaling , see @xcite ) . _ additional magnetic pressure_. the central magnetic field provides additional magnetic pressure implicating possible de - collimation of the overall outflow . the stellar magnetic field may drive a strong stellar wind which will remove stellar angular momentum . this stellar outflow will interact with the surrounding disk wind . the observed protostellar jets may consist of two components a stellar wind and the a wind , both with a strength depending on intrinsic ( yet unknown ) parameters . _ angular momentum exchange_. in the scenario of magnetic disk locking " , the stellar field threading the disk will re - arrange the global angular momentum budget . angular momentum of the star is transfered by the dipolar field and deposited at the inner disk . thus , matter orbiting in this region will be accelerated to slightly super - keplerian rotation . this has two interesting implications . ( i ) due to the super - keplerian speed this disk material could be more easily expelled into the corona by magneto - centrifugal launching @xcite and forms a disk wind . ( ii ) excess angular momentum in that disk area will slow down accretion unless removed by some further process . the disk outflow launched from this very inner part of the disk can be an efficient way to do this . this scenario is similar to the x - wind models @xcite . _ non - axisymmetric effects_. in addition to the simple picture of an axisymmetric configuration , an inclined stellar magnetic dipole will add non - axisymmetric effects . a moderate non - axisymmetric perturbation may result in warping of the inner disk , and , thus , a precession of the outflow launched from this area . for extreme cases of inclination , jet formation may be completely prevented . a rotating inclined dipole further implies a time - variation of the magnetic field strength at the inner disk radius . this may lead to a time - variation in the accretion rate and , also , the mass outflow flow rate . numerical simulations of the warping process @xcite indicate that the warp could evolve into a steady state precessing rigidly . disks may be warped by the magnetic torque that arises from the a slight misalignment between the disk and star s rotation axis @xcite . -axis ) we distinguish jet inflow boundary conditions ( shaded areas of ghost cells outside active grid ) from the star ( @xmath4 ) and the disk region from @xmath5 ) . the mass flux from star or disk is prescribed by the profile for the inflow density and inflow velocity ( dashed arrows ) . the latter is typically of about @xmath6 of the local keplerian speed . the stellar boundary is in rigid rotation , the disk boundary in keplerian rotation . the disk inner radius is at the co - rotation radius . in the gap area between disk and star ( @xmath7 ) a minimum `` floor '' mass flux is defined in order to keep the simulation going . thin contours indicate initial poloidal field lines for one of the simulation runs . [ fig_model ] , width=264 ] the simulation of outflows from star - disk magnetospheres is a difficult numerical task due to strong gradients in magnetic field strength , density / pressure , and also the velocity field . still , the first simulations of a stellar dipole surrounded by a disk have been presented in seminal papers as early as 1984 by @xcite . probably due to the success of the disk wind models @xcite the formation of outflows from stellar magnetospheres became somewhat unattended before the topic was re - discovered in the early 90ies with several model suggestions @xcite . numerical , stationary - state mhd solutions of the star - disk jet formation problem were presented subsequently @xcite . time - dependent simulations of dipolar magnetospheres started again later @xcite . however , the time scale of these simulations including the disk evolution was short - a few inner disk rotations . goodson et al . @xcite in a series of papers presented simulations runs of dipolar star - disk magnetospheres for up to 150 ( inner ) disk rotations , in particular considering the reconnection / flaring behavior of the magnetic field close to the inner disk radius . as another result , a highly collimated , narrow jet emerged along the rotation axis for which the authors also presented forbidden emission line maps . simulations by @xcite followed a similar setup , however , based on an increased grid scale of @xmath8 and run for times scales of 150 days ( stellar case ) . magnetic flares emerging from the inner disk location lead to episodic mass ejections . these flares were both triggered and triggering a variation of the disk mass accretion rate and a corresponding change in the inner disk radius . reconnection is strongly interrelated to magnetic diffusivity ( resp . electric resistivity ) , however , such model assumptions were not particularly specified in these papers . @xcite investigated the setup of an ambient vertical field aligned with a central dipolar field . their scenario is somewhat different to what we consider : the surrounding magnetic field is rather weak and vertical and , thus , not an active outflow - generating disk field as in our case . the authors identified three classes of field geometries - as in our setup depending on the direction of the ambient field . as a main result , in all three cases the collimation of the central outflow was observed , similar to earlier suggestions by @xcite . in comparison , @xcite did not treat the disk structure in their simulations , allowing them to follow the evolution of the the stellar magnetosphere for more than 2000 rotation periods of the inner disk . these simulations showed for the first time that the axial jet feature observed in some of the simulations before @xcite is in fact an intermittent feature and disappears on the long term . the initial axial jet appears essentially during relaxation process of the hydrostatic initial condition towards a hydrodynamic steady state . @xcite were first to introduce a physical stellar wind boundary condition in addition to the disk wind . the emerging two component outflow from star and disk remained un - collimated . this was understood as caused by the fact that no net electric current was driven along the outflow ( since the initial dipolar field distribution ) . since the stellar dipolar field becomes very weak along the disk , a classical magneto - centrifugal wind launching becomes inefficient . this is another reason for the lack of collimation observed in these simulations . simulations of the disk - star interaction @xcite focus on the dipolar accretion process and potential angular momentum exchange between star and disk . no evidence was found for persistent outflows and only episodic ejections from the inner disk area occurred . time - dependent simulations of winds from an initially spherically symmetric stellar magnetic field were performed by @xcite , evolving exactly into the classical stationary weber - davis solution . @xcite derive the angular momentum loss of pure stellar winds by numerical simulations . the setup is similar to the stationary - state models calculated by @xcite and @xcite . the outflow angular momentum loss derived from these simulations is comparable to what can be derived from the stationary models . while the simulations from dipolar magnetospheres failed to show collimated outflows , mhd simulations of disk winds did actually proof the self - collimating characteristics of mhd winds . after seminal papers by @xcite and @xcite this approach was further developed taking into account the time - dependent change of disk field inclination @xcite , turbulent magnetic diffusivity in the jet @xcite , a variation in the disk wind magnetic field and density profile @xcite , or non - axisymmetric effects @xcite , and also the disk dynamical evolution @xcite . recent simulations by @xcite investigated the `` topological stability '' of two - component ( star - disk ) self - similar solutions derived from stationary mhd . in difference to our approach their simulations do not start from an initial hydro-_static _ state , but from an initial dynamical steady state solution of the mhd equations , with analytical extrapolations in the case of radially self - similar solutions ( which are singular on the axis ) . @xcite presented two - component outflow simulations including the treatment of the disk evolution . however , although the wind dynamics indeed consists of two components , the magnetic field distribution basically resembled a monotonous field profile ( i.e. no stellar dipole , no stellar outflow involved ) . we perform axisymmetric mhd simulations of jet formation for a set of different magnetic field geometries and mass fluxes . the general model setup follows @xcite , @xcite and @xcite , however , with important modifications . the original zeus-3d ideal mhd code @xcite extended for physical magnetic resistivity ( see description and tests in @xcite ) is used . for the purpose of this paper the magnetic diffusivity was set to such a low level that it does not affect the overall collimation of the outflow . however , resistivity / diffusivity is essential for our simulations , as it allows for magnetospheric reconnection phenomena . the set of mhd equations considered is the following , @xmath9 @xmath10 + \nabla ( p+p_a ) + \rho\nabla\phi - \frac{\vec{j } \times \vec{b}}{c } = 0,\ ] ] @xmath11 @xmath12 + p ( \nabla \cdot\vec{v } ) - { \frac{4\pi}{c^2}}\eta \vec{j}^2= 0,\ ] ] with the usual notation for the variables ( see @xcite ) . the magnetic diffusivity is space and time dependent and is denoted by @xmath13 . we apply a polytropic equation of state for the gas with the polytropic index @xmath14 . as @xcite we have added turbulent alfvnic pressure in order to allow to keep the corona `` cool '' . in fact , we do not solve the energy equation [ eq_energ ] and apply the internal energy of the gas reduced to @xmath15 . two major reasons to do this are both computational speed and numerical stability . some of our long - term simulations did last already more than two months of cpu time and it would be impossible to reach the desired evolutionary steps if the energy equation would have been solved . following @xcite this approach also allows to combine gas pressure forces and gravity under the same derivative in the code . thus , instead of subtracting gradients , we apply the gradient of the difference , which results in perfect stability of the initial state . note , however , that recent work by clarke and collaborators seems to indicate that relaxation of the polytropy assumption may affect the dynamical evolution in certain domains of the jet , in particular regions with shocks or contact discontinuities @xcite . the magnetic diffusivity can be considered as turbulent and , thus , be related to the alfvnic turbulent pressure @xmath16 if we assume that it is primarily due to the turbulent alfvnic waves which are responsible for the turbulent alfvnic pressure applied in the simulations . in our previous work @xcite we have derived a toy parameterization relating both effects by parameterizing the _ turbulent magnetic diffusivity _ similar to the shakura - sunyaev parameterization of turbulent viscosity , @xmath17 , where @xmath18 and @xmath19 and @xmath20 are the characteristic dynamical length scale and velocity of the system , respectively . with that we obtain @xmath21 and with @xmath22 and @xmath23 , it follows that @xmath24 , or , normalized , @xmath25 . for the chosen polytropic index this implies a magnetic diffusivity @xmath26 if @xmath19 is constant ( see also the discussion in @xcite ) . we again stress the point that in the present paper the magnetic diffusivity is on a such a low level ( much below the critical level found in @xcite ) that it does not affect the dynamics , but allows for reconnection . for the numerical grid , we use the `` scaled grid '' option by zeus with the element size decreasing inwards by a factor of 0.99 . the size of the cylindrical grid is @xmath27 elements resulting in a physical grid size for all simulations is @xmath28 corresponding to @xmath29au for @xmath30 . thus , the disk gap ( @xmath31 ) is resolved with 11 grid elements . time is measured in rotation periods ( keplerian orbits ) at the inner disk radius . in summary , compared to our previous studies @xcite , the main new feature included in the present approach is that the initial magnetic field distribution consists of two components - a stellar dipolar field and a disk field . in difference to recent studies by @xcite we included both outflow components and magnetic field components in the simulation box , in particular treating the star - gap - disk boundary . along the @xmath32-boundary we distinguish between the star extending from @xmath33 to @xmath34 , the gap region extending from @xmath35 to @xmath36 , and the disk region from @xmath37 to @xmath38 , see fig.[fig_model ] . a keplerian disk is taken as a ( fixed in time and space ) boundary condition for the mass inflow from the disk surface into the corona and the magnetic flux . the stellar surface is approximated by a rigidly rotating `` disk '' in cylindrical coordinates . the stellar rotation is chosen such that the inner disk radius is located at the co - rotation radius . the initial magnetic field is purely poloidal . magnetic field lines are anchored in the disk and the rotating star and are in co - rotation with their respective foot point . the poloidal magnetic field profile along the @xmath32-boundary remains fixed in time and is , hence , determined by the choice of the initial magnetic field distribution . the disk region governs the mass inflow from the disk surface into corona ( denoted as `` disk wind '' ) . in addition to that we prescribe a stellar wind with different mass load . the hydrodynamic boundary conditions are `` inflow '' along the @xmath32-axis for the disk and stellar region and either `` inflow '' ( very light mass flow ) or `` reflecting '' for the gap region , `` reflecting '' along the symmetry axis and `` outflow '' along the outer boundaries . matter is `` injected '' from the disk and the star into the corona parallel to the poloidal magnetic field lines with very low velocity @xmath39 and with a density @xmath40 . the proportionality constants are typically @xmath41 and @xmath42 for both , stellar and disk wind ( but usually different for both components ) . along the gap we impose a floor value for the density @xmath43 and a similarly low velocity . for the injection velocity , the assumption is that the initial disk wind speed is in the range of the sound speed in the disk , @xmath44 . these mass loss rates from star and disk , respectively , are our other main parameters besides the respective magnetic flux ( see tab . [ tab_all ] ) . @xmath45 ( from top to bottom ; runs a2 , a3 , a4a ) . [ fig_ini ] , title="fig:",width=245 ] resp . @xmath45 ( from top to bottom ; runs a2 , a3 , a4a ) . [ fig_ini ] , title="fig:",width=245 ] resp . @xmath45 ( from top to bottom ; runs a2 , a3 , a4a ) . [ fig_ini ] , title="fig:",width=245 ] as initial state we prescribe a force - free magnetic field along with a gas distribution in hydrostatic equilibrium . both is essential in order to avoid artificial relaxation processes caused by a non - equilibrium initial condition . the initial density distribution is @xmath46 . as initial magnetic field distribution we superpose a dipolar stellar field and a disk potential field . for the disk field component we apply the model of @xcite and @xcite . we prescribe the magnetic field distribution as derivative of the magnetic flux distribution @xmath47 ( i.e. the @xmath48-component of the vector potential ) . for the superposed magnetic flux from star and disk we have @xmath49 with the stellar and disk magnetic flux @xmath50 and @xmath51 . the poloidal magnetic field follows from the derivatives @xmath52 and @xmath53 , properly calulated in the staggered mesh in order to obtain a numerically divergence - free and force - free initial field structure . the dimensionless disk thickness @xmath54 with @xmath55 for @xmath56 is introduced in order to avoid kinks in the field distribution . several combinations of both field components were investigated , parameterized by combinations of @xmath51 and @xmath50 . figure [ fig_ini ] shows three examples for the initial field configuration for different strength and alignment of the field components . similar field geometries have been discussed already by @xcite . @xcite did consider similar configurations , however , superimposing a stellar dipole with a weak vertical disk field . in order to allow for a clear comparison between all our runs , the respective magnetic field component _ profiles _ and wind density component _ profiles _ are identical . for simulation run a4a . [ fig_quasi ] , width=302 ] we now discuss a number of mhd jet formation simulations covering a wide parameter range ( see tab . [ tab_all ] ) concerning both the magnitude of the disk and stellar wind mass load and magnetic flux , respectively . the results presented here are preliminary in the sense that not all simulation runs could be performed over time scales sufficiently long enough for the mhd flow _ as a whole _ to reach the grid boundaries or to establish a stationary state . this is due to the comparatively large physical grid size in combination with steep gradients in the disk wind parameters which in general allow only for a weak outflow from large disk radii . in particular the steep decline of the stellar field in combination with a reasonable disk mass loss rate is numerically problematic ( see below ) . therefore , for a true comparison between different runs it is essential to also take into account the dynamical state of the outflow . in order to check for resolution issues , we have also run a set of simulations with twice the resolution , without seeing significant differences ( runs a4b , a5 , a6 ) . thus , since we are interested in many parameter runs evolving for a long time , we mainly concentrate on the lower resolution simulations . we describe first the general evolution of jet formation in our simulations ( see also @xcite ) and later discuss specific features . figure [ fig_evol1 ] shows how the field structure evolves in time for the simulation run a4a . in this case the stellar dipolar field and the disk field are aligned . thus , the x - point is initially along the rotational axis . we have stopped this simulation after 3600 keplerian rotations at the inner disk radius . this corresponds to 6 rotations at the outer disk radius and makes it clear that the outer parts of the disk wind have not yet completely evolved into a quasi - stationary state . this is a generic problem of all disk - jet simulations published so far as soon as large disk radii are considered . the evolution during the very first time steps from the initial shows that above the emerging outflow the initial steady state corona is still present . the x - point which was initially located at @xmath57 along the rotational axis , moves upwards to @xmath58 ( for @xmath59 ) and @xmath60 ( for @xmath61 ) and is later swept out of the computational domain for @xmath62 ( see top of fig . [ fig_evol1 ] and discussion end of sect.4.2 ) . since the initial condition is still kept in steady state , artificial dynamical re - configuration is prevented . this demonstrates that the numerical resolution is sufficient for our problem also in the out parts of the grid . the next dominant feature is observed still at early stages of about ten rotations . the initial dipolar field breaks up due to the magnetic pressure of the toroidal magnetic field induced by differential rotation between star and disk . the magnetic pressure gradient drives this outflow . furthermore , an intermediate axial jet is launched due to the re - arrangement of the initially hydrostatic corona to a new dynamical equilibrium state . in fact , as the magnetic field along the axis is squeezed by lateral dynamical pressure , the material is accelerated along the rotational axis . after the dipole is broken up , there is few direct magnetic connection between star and disk anymore and the differential rotation induced toroidal field decreases . the outflows from star and disk continue to grow gaining higher kinetic energy and momentum . mhd self - induction of toroidal magnetic field leads to collimation and magnetic acceleration . compared to pure disk winds @xcite the outflow is clearly less collimated as being de - collimated by the central stellar wind . at intermediate time scales quasi - stationary states may emerge . this is demonstrated for example in simulation a4a in fig . [ fig_quasi ] by showing the time evolution of the density at point @xmath63 . two plateaus are clearly seen at different density levels indicating two quasi - stationary states of the flow evolution . one is from @xmath64 till @xmath65 , the other from @xmath66 when the outflow is slowly re - adjusting from the flaring events ( see below ) . note , however , that at this time the outer disk has rotated only about a fifth of an orbit . thus , the field and flow above the outer disk will further evolve in time and again disturb the enclosed structure in quasi - steady state . we observe that over even longer time scale such quasi - stationary states may be reached ( and be disturbed ) again and probably again and again . we believe that this feature is due to the still ongoing evolution of the outer disk wind . the final states of some representative simulation runs are shown in fig . [ fig_final1 ] and [ fig_final2 ] . we will discuss them in sect.[sect_final ] below . as seen already from the initial field configuration , the simulations of the anti - aligned field configuration ( as e.g. a4a ) reveal a change of sign in the magnetic flux distribution from star and disk ( negative magnetic flux is indicated by dashed contours in fig . [ fig_reversal ] ) . the location of flux reversal is accompanied with a concentration of contour lines in our figures which may be confused with the existence of a shock . this is however an artifact of the choice of contour levels for the magnetic flux . in fact , it makes sense to follow the same field lines emerging from the stellar surface also back into the disk surface . since the strong gradient of the stellar dipolar field , magnetic flux levels aroung @xmath67 are concentrated . figure [ fig_reversal ] shows that the radial profiles of density , velocity and poloidal field accross the flux reversal . note that the location of magnetic field reversal ( the neutral field line ) is at a smaller radius @xmath68 compared to the magnetic flux reversal at about @xmath69 . within the magnetic field reversal magnetic flux is accumulated . beyond the field reversal the magnetic flux decreases and then becomes negative . our version of zeus code has physical magnetic diffusivity implemented ( see @xcite for explanation and tests ) . this allows to consider reconnection processes . we observe reconnection flares along some of the outflows ( see fig . [ fig_evol2 ] ) . these flares are similar to coronal mass ejection . they rapidly evolve and propagate along the neutral field line . once formed , reconnection islands ( or rather `` tori '' in our axisymmetric setup ) propagate across the jet magnetosphere within a few rotation times and leave the computational domain . the flares typically expand and reconnect within 70 time units equivalent to 70 orbital periods of the inner disk ( and 70 stellar rotations ) . we may also naively measure the flare propagation speed by the proper motion of the field lines observed in the simulation . this gives an average `` flare propagation speed '' of about unity , as they travel 80 spatial units in 70 time units - the keplerian speed at their foot - point . the flares propagate along the neutral field line , and leave the physical grid at a radius corresponding to about 7 - 10 au from the axis . -boundary versus time . note the change of mass flux of about @xmath70 during the flare events . the high mass flux for @xmath71 indicates sweeping off of the initial hydrostatic corona . [ fig_mflux_i ] , title="fig:",width=302 ] -boundary versus time . note the change of mass flux of about @xmath70 during the flare events . the high mass flux for @xmath71 indicates sweeping off of the initial hydrostatic corona . [ fig_mflux_i ] , title="fig:",width=302 ] in the following we mainly concentrate on simulation a4a . figure [ fig_small1 ] is a magnification of the inner area of fig.[fig_evol2 ] and demonstrates that the flares are actually from a y - point above the disk , located at ( @xmath72 ) note that the stellar magnetosphere remains closed also for foot - points along the disk with @xmath73 ( at @xmath74 ) , respectively @xmath75 ( @xmath74 ) , beyond the co - rotation radius . in this case the flare eruptions are launched close to the inner disk area , but _ not _ at the inner disk radius . @xcite also observed flares or reconnection events in their simulations of a dipolar magnetosphere connected top a surrounding accretion disk . since the evolution of the disk structure was treated in their simulation , they were able to observe time - dependent accretion along the equatorial plane . the frequent ejection events of au - sized knots along the rotational axis were correlated with a time - variation of the accretion rate . this is different in our case , where the reconnection / flares seem to be triggered by the evolution of the outer disk wind . as mentioned , during the run - time of our simulations ( up to 3600 inner disk rotations ) the corona ( or disk wind respectively ) above the outer disk has not yet reached a steady dynamical state . thus , in our case , as the outer disk outflow evolves , the cross - jet pressure equilibrium is changed and forces the inner magnetic field configuration to adjust accordingly . since goodson et al did not specify the distribution and magnitude of the magnetic diffusivity in their simulations , it is difficult to compare the results in detail . another difference is the duration of the simulation . while our physical grid size is only marginally larger , the time scale of our simulations is substantially longer ( 3600 inner disk rotations corresponding to about 36,000 days compared to 130 days ) . this is particularly important when considering the dynamical evolution above the outer part of the disk and the long - term behavior . the flare events observed in our simulations are accompanied by a temporal change in outflow mass flux , velocity , or momentum , respectively . figure [ fig_mflux_i ] shows the mass loss rate in axial direction integrated across the upper r - boundary versus time . during the first flare we see a 10%-variation in the mass flux followed by a sudden decrease of mass flux by a factor of two during flare two . figures [ fig_mflux_mv ] and [ fig_mflux_v ] show the profiles of jet momentum and poloidal velocity across the upper r - boundary . we see that during flaring these profile indicate a re - arrangement of momentum and velocity profile across the jet . before the flare , the radial jet momentum profile is broad ( see fig . [ fig_mflux_mv ] upper left for @xmath76 ) . in fact , the profile ( fig . [ fig_mflux_mv ] , @xmath76 ) remains very similar for several 100 rotations before until the flare event starts . after the flare has passed the grid , the jet momentum profile is concentrated within a cylindrical sheet of radius @xmath77 and thickness @xmath78 ( fig . [ fig_mflux_mv ] lower right at @xmath79 ) . similar to the situation before the flare , the jet momentum profile for time steps after the flare ( @xmath80 ) look almost identical ( see bottom sub - figures for @xmath81 and @xmath82 ) . this behavior is somewhat mirrored in the poloidal velocity profile . what is interesting for shock formation in the asymptotic jet , is a temporal change in jet velocity at certain jet radii ( see below ) . the maximum velocities in the outflow reach about two times the keplerian speed at the inner disk radius and vary by a factor of two . along the neutral field line the outflow velocity is fast - magnetosonic ( see figs . [ fig_small1 ] and [ fig_final2 ] where the fast surface is indicated ) . as an estimate for the reconnection time scale we may apply the sweet - parker approach with @xmath83 with the dynamical ( alfvn ) time scale @xmath84 and the diffusive time scale @xmath85 . with our model of turbulent magnetic diffusivity @xmath86 we find @xmath87 . for @xmath88 , and @xmath89 at the reconnection area with @xmath90 ( for simulation a16 ) that @xmath91 . this is similar to the duration of the reconnection flare we observe between @xmath92 and @xmath93 . note an interesting feature in both of the general model setups ( aligned and anti - aligned field geometry ) concerning the reconnection geometry . figures [ fig_small1 ] and [ fig_small2 ] show the inner structure of the star - disk magnetosphere . these simulations ( a04a , a16 ) were launched from differently aligned field geometries ( aligned / anti - aligned disk - stellar field ) . in the anti - aligned case the initial x - point is along the axis . however , as the magnetic field evolves , the dipole expands and sweeps off the initial field along the axis . a new x - point evolves by disruption of some of the closed dipolar field loops . the new x - point is located above the disk close to the inner disk radius ( see above ) . this situation is not very different from the the aligned case for which the initial x - point , located along the equatorial plane , moves up to a certain ( small ) height above the disk . in summary , our simulations show that no magnetic x - point remains along the equatorial plane as e.g. assumed in the shu et al . x - wind model . instead we find from both initial configurations an x - point above the disk ( sometimes also called y - point , @xcite ) , located at time - averaged distances ( @xmath94 ) or ( @xmath95 ) , respectively . the generation of knots in protostellar jets is a long - standing puzzle . in particular , it is unclear whether the shocks / knots are triggered by the internal engine or due to interaction with the ambient medium . a strong argument for the first possibility is the existence of some perfectly symmetric jets as hh212 @xcite , however , the majority of jet sources looks asymmetric . internal shocks along the jet flow may be caused e.g. by a time - dependent velocity variation of the material injected into the jet ( e.g. @xcite ) . even if we have observational support for the idea of the jet knots being triggered by the central engine , we do not know _ how _ the central engine does that . the time scale derived from typical knot separations @xmath96 and velocities @xmath97 correspond to @xmath98 which is clearly different from the keplerian time scale close to the inner disk edge from where the jets are launched . considering the ejection of large - scale flares and the follow - up re - arrangement of outflow density and velocity distribution in our simulations , one is tempted to hypothesize that the origin of knots is triggered by such flaring events . in our simulations ( a4a , a16 ) the time scale of flare generation is about 500 - 1000 rotational periods of the inner disk , corresponding to about 30 - 60 years ( assuming an inner disk rotation period of , say two times the co - rotation period of a typical ttauri star of about 10 days ) . the variation in the hydrodynamic parameters lasts for about 30 - 40 inner disk rotation times ( respectively about @xmath99 days for a 10day stellar rotation period ) . the further evolution and generation of further flaring events is of course not known as it is beyond our computation time . the essential point , however , is that we detect a _ long _ time scale , substantially longer than the typical dynamical time scales at the jet formation area . that time scale is surprisingly similar to the time scale defined by the observed knot separation and velocity . of course , it is to early to draw firm conclusions from such a tentative agreement . the reconnection time - scale is governed by the magnetic diffusivity for which we have applied our self - consistent model of turbulent magnetic diffusivity @xcite . it is left to further simulations of the asymptotic collimated jet beam to check whether the detected variation in outflow speed and mass flux ( figs . [ fig_mflux_i],[fig_mflux_mv],[fig_mflux_v ] ) is sufficient in order to generate strong internal shocks comparable to the observed knots . in this section we compare the overall collimation behavior of the different simulation runs . in previous studies @xcite we quantified the degree of outflow collimation @xmath100 by comparing the mass flow rates in axial and lateral direction , @xmath101 in tab . [ tab_all ] we provide mass flow rate within four differently sized volumes , @xmath102 , considering different sub - grids of the whole computational domain , i.e.cylinders of radius and height @xmath103 , and @xmath104 . we calculate the mass fluxes _ ratio _ normalized to the size of the area threaded by the mass flows in r- and z - direction , @xmath105 . the ratio of grid extension @xmath106 as displayed in our figures converts into a ratio of cylinder surface areas of @xmath107 . as a general measure of collimation degree , we give an average value @xmath108 which may consider also the dynamical state of the simulation run . compared to our previous work on pure disk wind collimation , where the relative mass flux of the collimating disk wind is a sufficient quantitative measure of collimation the situation in the present setup is not exactly the same . naturally , the stellar wind mass flux emerges close to the outflow axis and will tend to stay close to it . thus , a high stellar mass flux will naturally cause a more collimated mass flow . in order to compare the collimation degree within the large volume , it is essential to check the evolutionary state of the flow . in all of our simulation runs the initial corona has completely swept out of the grid . in this case , all of the mass flux we measure has been launched by the disk wind . exceptions are simulations a10 , a13 , and a12 , where a relict of the initial bow shock are still visible in the outer layers . the mass loss rates in these examples have to taken with care , at least the values from the larger volumes . run a16 is evolving very slowly in the outer part and it was not possible within reasonable cpu time to evolve the disk wind to larger distances from the disk surface . the degree of collimation derived from the simulations is in general different for the the different volumes , @xmath109 . in most cases , the degree of mass flux collimation increases along the flow which is exactly the signature of mhd self - collimation . the outflow starts as un - collimated disk / stellar wind , and reaches the outer grid boundaries as collimated beam . a good example is simulation a7 , where the collimation of mass flow changes from ration @xmath110 for the inner region to @xmath111 in the outer parts . the same arguments hold for simulations a8 , a15 , a14 . a similar behavior is seen e.g. in simulations a10 , a13 , a14 , however , the collimation in mass flux around the largest volume is low . in these cases the outermost flow structure has not yet evolved into a steady state and either parts of the initial bow shock or the initial corona are still in the computational domain . a counter example is simulation a12 , which stays un - collimated in mass flux , although the magnetic field distribution looks collimated . simulation a12 is exceptional for its high mass flux launched from the disk surface . thus , the disk wind starts super - alfvnic and super - seed the magnetosonic speed quickly . the standard blandford - payne magneto - centrifugal acceleration is not very efficient in this case , which also results in only little induction of toroidal magnetic field and , thus , lorentz force , see fig,[fig_lorentz ] . the radial component of the perpendicular lorentz force component is negative , i.e. directed inwards , however , too weak to balance the strong inner centrifugal force , resulting in a weak collimation of the outer part . table [ tab_all ] shows also the time evolution of the collimation degree . example a15 demonstrates how the collimation degree progresses in time . from earlier times ( t=550 ) to later times ( @xmath112 ) collimation increases as the outflow evolves into a new dynamical state over the whole numerical grid . the same hold for example a14 and others . examples a7 , a8 , a9 are calculated for a high stellar wind mass load , thus the time - evolution of the outer disk wind does not play a big role concerning collimation . this outflows reach a high degree of collimation early in time . at time @xmath113 . [ fig_lorentz ] , title="fig:",width=302 ] at time @xmath113 . [ fig_lorentz ] , title="fig:",width=302 ] the mass loss rates from star and disk are prescribed as boundary condition . a higher mass load in the stellar wind component would naturally result in a higher degree of collimation in the mass flux as this mass is just launched closer to the outflow axis . simulations a2 and a3 are good examples . the stellar wind mass loss is 2 - 3 times the disk wind mass flux . the resulting mass flux ratio is similar , thus this outflow is rather well collimated hydrodynamically . however , in spite the quite good collimation in mass flux , the magnetic field structure is not collimated with almost ( spherical ) radially expanding poloidal field lines ( see fig . [ fig_final1 ] ) . the field structure ( and thus the poloidal velocity field ) has a conical shape and is thus un - collimated . similar as for simulation a12 ( see above ) the disk wind is too weak in order to collimate the stellar wind to a high degree . simulations a12 and a13 have the same star - disk magnetic field profile , however with a different mass load by a factor of ten . run a12 reaches the same dynamical evolutionary state as a13 , but earlier at @xmath114 instead of @xmath115 . the less magnetized outflow a13 is stronger collimated . run a14 has the same mass load as a13 , but the double magnetic flux resulting in similar collimation degree . simulations a10 and a13 differ only by the stellar wind mass load ( factor two ) , however this mass load is low and does not result in a variation of the overall degree of collimation . note that for both simulations , the outer area of the outflow has not yet evolved in a steady state and a relict from the initial bow shock are still visible . from the technical point , the simulation of the case where magnetic dipole and disk magnetic field are anti - aligned ( a3 , a4 , a10 , a13 , a12 , a14 ) are computationally less expensive and were therefore able to evolve to considerably longer time . for the aligned cases ( a9 , a8 , a7 , a16 , a15 , a2 ) only the less magnetized evolve into a quasi stationary state in the outer ( disk ) outflow in reasonable time . we mention in particular run a15 with a reasonably strong disk magnetic flux which enables a collimating disk wind ( compare to a16 and a7 ) . simulations a9 , a8 and a7 demonstrate that a low mass flux disk wind mass flux can not evolve into a collimated disk wind even if the magnetic flux is relatively high if it is dominated by a strong central stellar jet . the most promising model setup in order to explain strong stellar jets from a star - disk magnetosphere are those with relatively strong disk wind and disk magnetic flux . the stellar wind dominated simulations may give a high degree of collimation , however they collimate to too small radii . stellar magnetic flux dominated simulations tend to stay un - collimated . clcccllllllll i d & @xmath116 & @xmath117 & @xmath50 & @xmath118 & @xmath119 & @xmath120 & @xmath121 & @xmath122 & @xmath123 & @xmath124 + a9 & @xmath125 & 0.1 & 10.0 & 740 & 2.37 , 5.82 & 4.72 , 3.87 & 6.42 , 1.64 & 7.44 , 1.09 & 0.41 , 1.2 , 3.9 , 6.8 & + & & & & 700 & 2.90 , 5.99 & 4.73 , 4.10 & 6.18 , 1.87 & 7.23 , 1.11 & 0.48 , 1.2 , 3.3 , 6.5 & 6.7 + & & & & 600 & 2.93 , 5.12 & 5.87 , 3.23 & 6.99 , 1.48 & 7.06 , 1.23 & 0.57 , 1.8 , 4.7 , 5.7 & + a8 & @xmath126 & 0.04 & 5.0 & 1010 & 2.67 , 5.40 & 4.64 , 3.76 & 6.95 , 1.49 & 7.41 , 1.18 & 0.49 , 1.2 , 4.7 , 6.3 & + & & & & 900 & 2.68 , 5.38 & 4.72 , 3.66 & 6.97 , 1.46 & 7.43 , 1.11 & 0.50 , 1.3 , 4.8 , 6.7 & 6.5 + & & & & 800 & 2.66 , 5.39 & 4.58 , 3.77 & 6.90 , 1.45 & 7.48 , 1.08 & 0.49 , 1.2 , 4.8 , 6.9 & + a7 & @xmath127 & 0.02 & 5.0 & 1450 & 1.88 , 6.37 & 2.28 , 5.51 & 5.20 , 1.31 & 6.77 , 1.48 & 0.30 , 0.41 , 4.0 , 4.6 & + & & & & 1350 & 3.02 , 5.26 & 5.99 , 2.46 & 7.41 , 1.07 & 7.51 , 1.05 & 0.57 , 2.4 , 6.9 , 7.2 & 7.0 + & & & & 700 & 2.80 , 5.44 & 4.60 , 3.86 & 6.92 , 1.51 & 7.42 , 1.02 & 0.52 , 1.2 , 4.6 , 7.3 & + a16 & @xmath128 & 0.02 & 5.0 & 1300 & 0.23 , 0.25 & 0.52 , -0.02 & 0.63 , 0.30 & 0.81 , 0.11 & & + & & & & 1200 & 0.19 , 0.27 & 0.55 , 0.04 & 0.79 , -0.14 & 0.38 , 0.13 & & + & & & & 1000 & 0.37 , 0.02 & 0.52 , -0.5 & 0.49 , -0.54 & 0.59 , -0.20 & & + a15 & @xmath129 & 0.04 & 3.0 & 2100 & 0.22 , 0.83 & 0.40 , 0.91 & 0.71 , 0.93 & 1.40 , 0.76 & 0.27 , 0.44 , 0.76 , 1.8 & + & & & & 1900 & 0.23 , 0.82 & 0.41 , 0.90 & 0.72 , 0.90 & 1.41 , 0.87 & 0.28 , 0.46 , 0.8 , 1.6 & 2.0 + & & & & 550 & 0.28 , 0.80 & 0.65 , 0.72 & 1.56 , 0.95 & 1.04 , 2.39 & 0.35 , 0.9 , 1.6 , 0.44 & + a2 & @xmath130 & 0.01 & 5.0 & 700 & 2.34 , 5.93 & 3.21 , 5.34 & 4.04 , 4.79 & 4.95 , 4.30 & 0.40 , 0.60 , 0.84 , 1.2 & 1.2 + & & & & 650 & 2.35 , 5.92 & 3.24 , 5.33 & 4.10 , 4.75 & 5.14 , 4.20 & 0.40 , 0.61 , 0.86 , 1.2 & + a3 & @xmath131 & -0.01 & 5.0 & 500 & 2.60 , 5.73 & 3.59 , 5.25 & 5.05 , 4.33 & 6.45 , 3.25 & 0.45 , 0.68 , 1.2 , 2.0 & + & & & & 480 & 2.59 , 5.73 & 3.61 , 5.24 & 5.10 , 4.32 & 6.50 , 3.23 & 0.45 , 0.69 , 1.2 , 2.0 & 2.0 + & & & & 450 & 2.58 , 5.73 & 3.61 , 5.23 & 5.18 , 4.12 & 6.63 , 3.05 & 0.45 , 0.69 , 1.3 , 2.2 & + a4a & @xmath132 & -0.1 & 3.0 & 3600 & 6.64 , 1.75 & 7.54 , 1.07 & 8.14 , 0.79 & 8.41 , 0.54 & 3.8 , 7.1 , 10.3 , 15.6 & + & & & & 3300 & 6.31 , 2.12 & 7.46 , 1.07 & 7.94 , 0.69 & 8.14 , 0.60 & 3.0 , 7.0 , 11.5 , 13.6 & 12 + & & & & 2000 & 6.85 , 1.46 & 7.33 , 1.18 & 7.72 , 1.10 & 8.39 , 0.35 & 4.7 , 6.2 , 7.0 , 24.0 & + & & & & 80 & 5.71 , 2.86 & 8.04 , 2.83 & 15.0 , 9.92 & .0004 , .02 & 2.0 , 2.8 , 1.5 , - & + a4b@xmath133 & @xmath134 & -0.1 & 3.0 & 80 & 5.96 , 1.34 & 7.17 , 1.77 & 16.5 , 12.7 & - , - & 4.4 , 4.1 , 0.73 , - & + a10 & @xmath135 & -0.1 & 3.0 & 550 & 1.11 , 0.80 & 1.57 , 0.70 & 2.08 , 0.78 & 3.10 , 3.16 & 1.4 , 2.2 , 2.7 , 0.98 & + & & & & 500 & 1.11 , 0.80 & 1.59 , 0.69 & 2.11 , 0.81 & 3.17 , 3.96 & 1.4 , 2.3 , 2.6 , 0.80 & 3.0 + & & & & 400 & 1.13 , 0.79 & 1.64 , 0.66 & 2.26 , 0.92 & 3.63 , 7.33 & 1.5 , 2.5 , 2.5 , 0.50 & + a13 & @xmath136 & -0.1 & 3.0 & 620 & 0.81 , 0.74 & 1.29 , 0.64 & 1.76 , 0.69 & 2.81 , 2.37 & 1.10 , 2.02 , 2.55 , 1.19 & + & & & & 600 & 0.82 , 0.74 & 1.30 , 0.63 & 1.78 , 0.70 & 2.78 , 2.54 & 1.11 , 2.06 , 2.54 , 1.10 & 2.5 + & & & & 350 & 0.84 , 0.71 & 1.41 , 0.58 & 2.09 , 1.11 & 3.92 , 9.19 & 1.18 , 2.43 , 1.88 , 0.43 & + a12 & @xmath137 & -0.1 & 3.0 & 380 & 6.25 , 9.10 & 7.66 , 11.1 & 8.86 , 12.9 & 11.1 , 15.0 & 0.69 , 0.69 , 0.69 , 0.74 & + & & & & 350 & 6.26 , 9.10 & 7.66 , 11.1 & 8.87 , 12.9 & 11.8 , 15.4 & 0.69 , 0.69 , 0.69 , 0.77 & 0.8 + & & & & 300 & 6.26 , 9.10 & 7.66 , 11.1 & 8.91 , 13.1 & 13.5 , 14.9 & 0.69 , 0.69 , 0.68 , 0.91 & + a14 & @xmath136 & -0.2 & 6.0 & 660 & 0.79 , 0.81 & 1.15 , 0.81 & 1.60 , 0.77 & 2.31 , 1.72 & 0.98 , 1.4 , 2.1 , 1.34 & + & & & & 500 & 0.75 , 0.80 & 1.16 , 0.76 & 1.65 , 0.72 & 2.32 , 2.51 & 0.94 , 1.53 , 2.3 , 0.9 & 2.0 + & & & & 350 & 0.78 , 0.78 & 1.28 , 0.71 & 1.74 , 1.00 & 3.62 , 7.14 & 1.0 , 1.8 , 1.7 , 0.5 & + a@xmath138 & @xmath134 & -0.2 & 6.0 & 80 & 5.86 , 1.46 & 6.83 , 1.44 & 13.0 , 18.0 & - , - & 4.0 , 4.7 , 0.72 , - & + we have performed axisymmetric mhd simulations of jet formation from accretion disks surrounding a magnetised star . the simulations start from an equilibrium state of the star - disk corona ( hydrostatic density distribution , force - free field ) . our physical grid size is @xmath139 inner disk radii corresponding to about @xmath140au for @xmath30 . disk and stellar surface are taken as a time - independent boundary condition for the outflow mass loss rates and the magnetic flux profile . the major goal of this paper was to investigate the long - term interrelation between a stellar dipolar field and a disk field and how that affects the outflow collimation . certain combinations of stellar versus disk magnetic flux and field directions were considered namely the cases of an aligned resp . anti - aligned magnetic field direction . the stellar magnetic field has an important impact on the jet formation process by providing additional magnetic flux , additional central ( magnetic ) pressure component , excess angular momentum in the jet launching region . it might disturb the outflow axisymmetry , and/or trigger a time - variation in outflow rate in the case of an inclined central stellar magnetosphere . our results of mhd simulations of a superposed stellar & disk magnetosphere in general demonstrate the de - collimation of the disk jet by the central stellar magnetosphere , respectively the collimation of the stellar wind by the surrounding disk jet . the interplay between disk wind and stellar wind , may result in flares like coronal mass ejections , generating a variation in the mass loading and velocity of the asymptotic jet . we find variations in mass load by a factor of four and in velocity by a factor of two . the time scale for such flaring events is _ long _ , i.e. several hundred inner disk rotations and of the order of 10.000 days . we therefore hypothesized whether such flaring events and the corresponding hydrodynamic variations may be responsible for generating internal shocks in the asymptotic jet visible as knots . in summary we conclude that strong protostellar jets must be generated by collimating disk winds with reasonable mass load . if the two - component system is dominated by the stellar wind , collimation is either too weak ( for low mass load ) or too high ( for high mass load ) . it is only a disk jet which can provide a collimated outflow over a reasonable range in radius . i thank the lca team and m. norman and d. clarke for the possibility to use the zeus code . comments and suggestions by an unknown referee which helped to improve the presentation of this work are acknowledged . anderson , j.m . , li , z .- y . , krasnopolsky , r. , blandford , r.d . 2006 , apj , 653 , l33 blandford , r.d . , & payne , d.g . , 1982 , mnras , 199 , 883 cabrit , s. , the accretion - ejection connexion in t tauri stars : jet models vs. observations , in : j. bouvier & i. appenzeller ( eds . ) , star - disk interaction in young stars , iau symposium 243 , 2007 , p.203 camenzind , m. 1990 , magnetized disk - winds and the origin of bipolar outflows , in : g. klare ( ed . ) rev . 3 , springer , heidelberg , p.234 casse , f. , & keppens , r. 2002 , apj , 581 , 988 fendt , c. 2006 , apj , 651 , 272 fendt , c. , camenzind , m. , & appl , s. 1995 , a&a , 300 , 791 fendt , c. , & camenzind , m. 1996 , a&a , 313 , 591 fendt , c. , zinnecker , h. 1998 , a&a , 334 , 750 fendt , c. , & elstner , d. 1999 , a&a , 349 , l208 fendt , c. , & elstner , d. 2000 , a&a , 363 , 208 fendt , c. , & cemeljic , m. 2002 , a&a , 395 , 1045 fendt , c. , & ouyed , r. 2004 , apj 608 , 378 ferreira , j. 1997 , a&a , 319 , 340 ferreira , j. , dougados , c. , cabrit , s. , 2006 , a&a , 453 , 785 goodson , a.p . , winglee , r.m . , bhm , k-.h . 1997 , apj , 489 , 199 goodson , a.p . , bohm , k .- h . , winglee , r.m . 1999 , apj , 524 , 142 hawley , j.f . , & stone j.m . 1995 , comp . physics comm . , 89 , 127 hayashi , m.r . , shibata , k. , matsumoto , r. 1996 , apj , 468 , l37 hirose , s. , uchida , y. , shibata , k. , matsumoto , r. 1997 , pasj , 49 , 193 kato , s.x . , kudoh , t. , & shibata , k. 2002 , apj , 565 , 1035 keppens , r. , goedbloed , j. p. 1999 , a&a , 343 , 251 kigure , h. , & shibata , k. 2005 , apj , 634 , 879 krasnopolsky , r. , li , z .- y . , & blandford , r. 1999 , apj , 526 , 631 kudoh , t. , matsumoto , r. , & shibata , k. 1998 , apj , 508 , 186 kker , m. , henning , t. , rdiger , g. 2003 , apj , 589 , 397 ; erratum : apj , 614 , 526 kuwabara , t. , shibata , k. , kudoh , t. , & matsumoto , r. 2000 , pasj , 52 , 1109 kuwabara , t. , shibata , k. , kudoh , t. , matsumoto , r. 2005 , apj , 621 , 921 kwan , j. , tademaru , e. 1988 , apj , 332 , l41 lai , d. 1999 , apj , 524 , 1030 lai , d. 2003 , apj , 591 , l119 matsakos , t. , tsinganos , k. , vlahakis , n. , massaglia , s. , mignone , a. , trussoni , e. 2008 , a&a , 477 , 521 matt , s. , goodson , a.p . winglee , r. , & bhm , k .- h . 2002 , apj , 574 , 232 matt , s. , winglee , r. , & bhm , k .- h . 2003 , mnras , 345 , 660 matt , s. , pudritz , r.e . 2005 , apj , 632 , l135 matt , s. , pudritz , r.e . 2008 , apj , 678 , 1109 meliani , z. , casse , f. , sauty , c. 2007 , a&a , 460 , 1 michel , f.c . 1969 , apj , 158 , 727 miller , k.a . , & stone , j.m . 1997 , apj , 489 , 890 ouyed , r. , & pudritz , r.e . , 1997 , apj , 482 , 712 ouyed , r. , clarke , d. , & pudritz , r.e . 2003 , apj , 582 , 292 pfeiffer , h.p . , lai , d. 2004 , apj , 604 , 766 pudritz , r.e . , & norman , c.a . 1983 , apj , 274 , 677 pudritz , r.e . , rogers , c.s . , ouyed , r. , 2006 , mnras , 365 , 1131 pudritz , r.e . , ouyed , r. , fendt , ch . , brandenburg , a. 2007 , in : b. reipurth , d. jewitt , & k. keil ( eds . ) , protostars & planets v , university of arizona press , tucson , 2007 , p.277 raga , a.c . , de colle , f. , kajdic , p. , esquivel , a. , canto , j. 2007 , a&a , 465 , 879 ramsey , j.p . , & clarke , d.a . , jets from keplerian disks and the role of the equation of state , poster contr . , conference `` cores , disks , jets & outflows in low & high mass star forming environments '' , banff 2004 , see www.ism.ucalgary.ca/meetings/banff/posters.html romanova , m. , ustyugova , g. , koldoba , a. , lovelace , r. 2002 , apj , 578 , 420 romanova , m. , ustyugova , g. , koldoba , a. , lovelace , r. 2004 , apj , 610 , 920 sauty , c. , & tsinganos , k. 1994 , a&a , 287 , 893 shu , f. , najita , j. , ostriker , e. , wilkin , f , . ruden , s. , lizano , s. 1994 , apj , 429 , 781 stone , j.m . , & norman , m.l . 1992 , apjs , 80 , 753 stone , j.m . , & norman , m.l . 1992 , apjs , 80 , 791 uchida , y. , low , b.c . 1981 , journal of astroph . and astron . , 2 , 405 uchida , y. , shibata , k. 1984 , pasj , 36 , 105 uchida , y. , & shibata , k. 1985 , pasj , 37 , 515 uzdensky , d.a . , knigl , a. , litwin , c. 2002 , apj , 565 , 1191 ustyugova , g.v . , koldoba , a.v . , romanova , m.m . , chechetkin , v.m . , & lovelace , r.v.e . 1995 , apj , 439 , l39 ustyugova g.v . , koldoba a.v . , & romanova m.m . , chechetkin v.m . , lovelace r.v.e . 1999 , apj , 516 , 221 zanni , c. , ferrari , a. , rosner , r. , bodo , g. , massaglia , s. 2007 , a&a , 469 , 811 zinnecker , h. , mccaughrean , m.j . , rayner , j.t . 1998 , nature , 394 , 862

we consider the propagation of fronts in a periodically driven flowing medium . it is shown that the progress of fronts in these systems may be mediated by a turnstile mechanism akin to that found in chaotic advection . we first define the modified ( `` active '' ) turnstile lobes according to the evolution of point sources across a transport boundary . we then show that the lobe boundaries may be constructed from stable and unstable _ burning invariant manifolds_one - way barriers to front propagation analogous to traditional invariant manifolds for passive advection . because the burning invariant manifolds ( bims ) are one - dimensional curves in a three - dimensional ( ) phase space , their projection into-space exhibits several key differences from their advective counterparts : ( lobe ) areas are not preserved , bims may self - intersect , and an intersection between stable and unstable bims does not map to another such intersection . these differences must be accommodated in the correct construction of the new turnstile . as an application , we consider a lobe - based treatment protocol for protecting an ocean bay from an invading algae bloom . [ sec : lead_paragraph ] _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ fronts in flowing media are the combination of two complementary physical phenomena . front propagation is governed by geometric evolution , built around the structure of rays and perpendicular fronts . fluid flows exhibit structures that are sinewy , tenuous , and generically chaotic even for simple flows . here we study the interplay of these two dynamics . a central concept in both front propagation and fluid flows is that of _ transport_moving something from here to there . a laser pulse traveling through fiber optic cable transports information , while an ocean wave transports energy and surfers . while some instances of fluid transport have a similar linear character , e.g. , the oceanic conveyor belt that moves heat around the globe , we will be more interested in fluid transport as a means to mixing , a process that depends on the chaotic nature of the flow . this study examines how this chaotic transport in fluids is augmented by propagation . our approach draws from the existing geometric theory of transport in fluid flows known as the _ turnstile mechanism_. this mechanism is a first - order description of chaotic mixing which describes the exchange of fluid packets between two adjacent regions . we develop a reformulation of this theory that naturally encompasses the addition of front propagation . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

1305.5005

we build upon the foundation subjects of fluid flow and front propagation , both central dynamical mechanisms in nature . exemplars of fluid flow include oceanic and atmospheric flow , industrial mixing , and microfluidics . we might also include such systems as phase space flow and swarms of moving agents . well - studied cases of front propagation include optical and acoustic wavefronts , phase transition fronts , and chemical reaction fronts . again we can broaden our view to include growth of algae blooms , spread of infectious disease , and dissemination of information and beliefs . it is often the case that the substrate supporting the propagation of a front is not static , and certainly may be some form of flowing fluid . the combination of these two elemental dynamics yields a variety of new and interesting phenomena while maintaining features of each component . recent experiments have combined chemical reaction front propagation with laboratory scale flows @xcite . the flows were generated using magneto - hydrodynamic forcing in a quasi - two - dimensional fluid . the reaction fronts were generated through belousov - zhabotinsky chemistry . a variety of phenomena were investigated including mode - locking and front - pinning . direct numerical simulations @xcite have supported these experiments . here we take a different perspective . invariant manifolds are key to understanding passive advection in time - independent and time - periodic fluid flows @xcite . recent experiment and theory @xcite has shown that these invariant manifolds survive , but are modified by , the addition of front propagation to the dynamics . the original ( advective ) invariant manifolds undergo a bifurcation whereby they split into two nearby _ oriented _ invariant manifolds , called _ burning invariant manifolds _ ( bims ) . such manifolds are oriented in that they are _ one - sided _ barriers to the progress of fronts in fluid flows . the fundamental geometric construct for understanding chaotic transport in passive flows is the turnstile mechanism @xcite . the turnstile is built from segments of advective stable and unstable invariant manifolds . since bims are relevant for front propagation , this suggests that a turnstile - type mechanism may also be at work in active fluid flows . we begin in section [ sec : turnstile_review ] with a review of the turnstile mechanism in passive advection . then section [ sec : dynamical_system ] presents the dynamical system for fronts in flows . section [ sec : burnstile ] defines the new burning lobes in terms of point stimulations . we then recall the necessary features of bims from previous works ( section [ sec : bim_basics ] ) . sections [ sec : small_v_0 ] and [ sec : large_v_0 ] examine the modifications necessary for our new turnstile given small and large front propagation speeds , respectively . we end by using this burning turnstile structure to prescribe an efficient treatment of an oceanic algae bloom ( section [ sec : reaction_suppression ] ) . time - independent , two - dimensional , incompressible flows are well characterized by the structure of their fixed points and separatrices . the separatrices are invariant curves under the flow . they divide the fluid trajectories into classes with qualitatively distinct behaviors ( fig . [ fig : passive_broken_separatrix](a ) ) . unlike the invariant tori that are also present in these flows , separatrices divide trajectories that exponentially diverge or converge . the introduction of a time - periodic perturbation leads to the splitting of each separatrix into distinct stable and unstable invariant manifolds ( fig . [ fig : passive_broken_separatrix](b ) ) . these manifolds are not invariant with respect to the time - varying flow , but rather the map @xmath2 obtained by integrating the flow over one forcing cycle . as a model system for illustrating these structures , we use the alternating vortex chain ( avc ) in a channel geometry @xcite , @xmath3 ) \cos(\pi y ) , \\ u_y(x , y , t ) & = - \cos(\pi[x + b \sin ( \omega t - \phi ) ] ) \sin(\pi y ) , \end{aligned}\ ] ] where @xmath4 , and @xmath5 , @xmath6 , and @xmath7 are dimensionless parameters that describe the periodic forcing amplitude , frequency , and phase , respectively . ( we use the phase @xmath7 to choose a particular poincar section . ) this model has free - slip boundary conditions ( bcs ) . we will use this same model in following sections when discussing active fluids . ( b ) periodic lateral perturbation breaks each separatrix into stable ( red ) and unstable ( blue ) manifolds . @xmath8 for small perturbations and short times , the unforced separatrices are still fairly good classifiers of particle trajectories . however , these separatrices , are traversed by certain trajectories . this is sometimes referred to as a `` leaky separatrix '' . importantly , this is not `` leakiness '' in the sense of diffusion . rather , there is structure to the leakiness . one iteration of the map causes a localized packet of fluid to be transported across from left to right , and a corresponding packet from right to left . to formalize these ideas , we first choose a transport boundary that divides the fluid in two allowing us to define a flux . in this paper , we focus on flow in a channel geometry and choose a transport boundary that defines a left- and right - hand side ( @xmath9 , @xmath10).(starting in sec . [ sec : burnstile ] , we specialize to transport from left to right ) . given this transport boundary , we define lobes as sets of points that cross this boundary from left to right , or right to left , upon one iteration of the map @xmath11 . the two sets , before and after the map , are referred to as lobes . the escape lobes @xmath12 and capture lobes @xmath13 are defined as @xmath14 we use the same notation to denote the map applied to a set of points @xmath15 . using this notation and the invertibility of @xmath11 , it is somewhat more intuitive to write , @xmath16 noting that @xmath17 and @xmath18 . in figure [ fig : arbitrary_transport_boundary ] , we choose the advective separatrix as our transport boundary ( vertical line in the middle ) . examining the action of one map iteration , we find eight regions of interest . two disjoint regions map to two other disjoint regions going left to right ( @xmath19 ) , and similarly going right to left ( @xmath20 ) . notice that these regions are bounded both by segments of the separatrix and also by segments of its forward and reverse iterates , which do not coincide with the separatrix . also , note the overlapping of regions and the amount of total area involved . while this analysis of single - step transport is correct , it can be simplified by a different choice of transport boundary . ) . right to left transport : blue filled regions map to blue outlined regions ( @xmath20 ) . @xmath21 while the above definition yields well - defined lobes given _ any _ transport boundary , it is standard practice to construct this boundary from segments of stable and unstable invariant manifolds @xcite . in doing so , we describe the transport process using `` native '' structures . two hyperbolic fixed points ( on opposite sides of the channel ) are denoted @xmath22 and @xmath23 . the stable and unstable manifolds attached to these points are denoted @xmath24 and @xmath25 , respectively . the finite segments connecting these fixed points to the intersection point @xmath26 are denoted @xmath27 $ ] and @xmath28 $ ] . the intersection point is required to be a _ primary intersection point _ @xcite ; that is , @xmath27 \cap w^u[z^b , p_0 ] \ \{z^a , z^b , p_0\}=\emptyset$ ] . the transport boundary is then defined as @xmath27 \cup w^u[z^b , p_0]$ ] . in fig . [ fig : advective_turnstile ] , we illustrate the lobes derived from this new transport boundary . importantly , these lobe boundaries are comprised solely of segments of these same invariant manifolds . each lobe is also composed of a smaller number ( here just one ) of disjoint regions having smaller total area . the @xmath29 and @xmath30 lobes share a point in their boundary and , loosely speaking , the complementary transport of these packets corresponds to a rotation ( and shift ) about this point . consequently , this exchange is referred to as a _ turnstile mechanism _ @xcite . generalizing the connection between the evolution of sets and bounding invariant manifolds to the case of front propagation is one of the main results of this paper . ] we assume that the two - dimensional fluid is partitioned into regions that can be identified as either `` burned '' or `` unburned''we use `` burned '' generically to refer to the region bounded by an outward propagating front . that is , these regions are separated by well - defined one - dimensional ( codimension one ) curves or _ fronts_. this is known as the `` thin ( sharp ) front '' , or `` geometric optics '' regime @xcite . for instance , if our front is an autocatalytic chemical reaction front , this limit requires that the diffusion rate is small compared with the reaction rate . we assume that each front element ( infinitessimal piece of the one - dimensional front ) propagates perpendicular to the front at a speed @xmath31 measured in the comoving fluid frame , thereby expanding the burned region . to be clear , we require that this speed @xmath31 be isotropic , homogeneous , and time - independent . note that we have implicitly assumed that the front elements are non - interacting ( fronts do not change their behavior when faced with a collision ) , and their motion does not depend on the front curvature . finally , we assume that there is no effect of the front propagation on the underlying fluid flow . in the case of a chemical reaction front , this means that the reaction is not sufficient to change the fluid density , viscosity , etc . this is a good approximation in such experiments as . the evolution of a reacted region can be modeled using the modified fkpp equation @xcite , @xmath32 where @xmath7 is the local fraction of reaction products , @xmath33 is the reaction time - scale , @xmath34 describes the reaction kinetics , and @xmath35 is the diffusion constant . we require the sharp front limit of this equation so that the fronts dividing burned from unburned regions are well defined . we define the map @xmath36 , which takes a point stimulation in @xmath37 and returns the reacted domain obtained by integrating eq . ( [ eqn : fkpp ] ) ( in the sharp front limit ) for one forcing period . here @xmath38 is the collection of subsets of @xmath37 . the reaction dynamics propagates this point stimulation into some finite region that is all the while stretched and folded by the flow dynamics . we extend the map @xmath39 to act on regions by defining @xmath40 such that @xmath41 . the intuition is simply that if stimulation @xmath42 evolves to region @xmath43 and @xmath44 to @xmath45 , then stimulating both @xmath42 and @xmath44 will result in the reacted domain @xmath46 . note that @xmath39 acting on sets is not one - to - one , i.e. , two distinct burned regions can evolve forward to the same burned region . this differs from the passive case in which @xmath11 acting on sets was invertible ( sec . [ sec : turnstile_review ] ) . as shown in , the sharp - front fkpp equation is equivalent to the g - equation , @xmath47 where only the level set @xmath48 , and not the value of this field , has any physical meaning . this level set represents the boundary between reacted and unreacted regions . while one can solve this pde effectively by focusing on the vicinity of the level set , we wish to make an explicit dimension reduction . instead of integrating eq . ( [ eqn : geqn ] ) , we shall use an equivalent formulation @xcite in which each front element evolves independently under the following three - dimensional ode , @xmath49 [ eqn:3dode ] the variables @xmath50 and @xmath51 denote the position and orientation of each front element . the prescribed fluid velocity is given by @xmath52 . again , @xmath31 is the front propagation speed in the comoving fluid frame . for compactness , we use the variables @xmath53 and @xmath54 to indicate the tangent to the front element and the normal direction ( propagation direction ) , respectively . finally , @xmath55 and repeated indices are summed . this ode representation allows us to apply the tools of low - dimensional dynamical systems theory . this ode can also be derived more intuitively . the total translational motion of a front element is the vector sum fluid velocity plus the front propagation velocity in the fluid frame ( eq . ( [ eqn:3dodea ] ) ) . the change in orientation is determined entirely geometrically . equation ( [ eqn:3dodeb ] ) describes the angular velocity of a material line embedded in the fluid . we define @xmath56 , the evolution of front elements , by integrating eq . ( [ eqn:3dode ] ) for one forcing period . note that @xmath57 is the active extension of @xmath11 ( the advection map for a point ) . the `` 2 '' and `` 3 '' subscripts emphasize that these maps operate on two- and three - dimensional phase spaces . finally , we also use @xmath58 to denote the iterate of the curve @xmath59 which is the outward oriented boundary curve ( in @xmath0-space ) of burned region @xmath60 . here we define the turnstile relevant for active fluids in terms of the behavior of sets , much like was done for the advective turnstile . the essential differences arise due to the everywhere - expanding property of the map @xmath39 . because of the asymmetry of the map @xmath39 ( unburned fluid can become burned but not vice versa ) there exists an asymmetry in the lobe definitions depending on whether the burned region is to the left or the right of the transport boundary . for the remainder of this paper , we consider left - to - right propagation . the right - to - left case is defined analogously . we begin with the escape process . since advection maps points to points , any given point begins and ends either right or left of the transport boundary , and so crossing ( e.g. , escaping ) is well defined . however with the addition of front propagation , points map to regions under @xmath39 and so a choice arises ; is a point stimulation , which immediately gives rise to a burned region , said to escape when _ all _ of the region crosses the transport boundary ? or instead when _ any _ of the region crosses ? looking to physical examples such as autocatalytic chemical reactions , fires , algae blooms , infections , etc . , it seems reasonable that the more interesting definition is when _ any _ crosses ; we are generally not concerned whether an _ entire _ wildfire crosses the firebreak . this motivates the following definitions of escape lobes in the active system ( compare eqs . ( [ eqn : passive_lobe_defs_pts ] ) ) , @xmath61 the escape lobe @xmath30 is the set of all points left of the transport boundary such that the evolution of each point stimulation under @xmath39 has some intersection with the area right of the transport boundary . the escape lobe @xmath62 is the set of all points right of the transport boundary that may be reached in one iterate by some point stimulation on the left . see fig . [ fig : escape ] . the capture process is slightly subtler due to the asymmetry in the burning process ( compare eqs . ( [ eqn : passive_lobe_defs_pts ] ) ) , @xmath63 the capture lobe @xmath29 is the set of all points right of the boundary such that the evolution of this point stimulation under @xmath39 lies completely on the left . in other words , @xmath29 contains all those stimulations on the right that do not contribute to the subsequent burning of the @xmath10 . the capture lobe @xmath64 is the set of all points on the left not reachable in one period by any stimulation on the left , i.e. reachable only by a stimulation on the right . see fig . [ fig : capture ] . we reiterate that , just as in the passive case , these set - based definitions of lobes are well - defined with respect to an arbitrary transport boundary . however , we again wish to make use of the dynamic structure of the system in choosing the transport boundary in a natural way . the burning lobe definitions are phrased in terms of mapping properties of point stimulations . experimentally , this definition makes good sense because a point s lobe membership is based on the results of an elementary experiment the stimulation of just this point ( or as close to it as possible ) . however , making use of the extension of @xmath39 to regions , we can rephrase the definition of the burning lobes . the first two , @xmath62 and @xmath65 are relatively intuitive , @xmath66 where @xmath67 indicates the set complement of @xmath60 . in order to similarly describe the other two lobes , we need a map inverse . since @xmath39 is many - to - one , we must _ choose _ an inverse . given a final burned region @xmath68 there are an infinite number of initial regions @xmath60 for which @xmath69 . two useful senses of inverse are derived from considering extreme cases . we define the _ loose _ and _ tight _ inverses respectively as , @xmath70 the loose preimage is the set of all stimulations that intersect the target , while the tight preimage is the set of only those stimulations that map entirely into the target . it follows that the tight preimage is a subset of the loose one , @xmath71 . ( this motivates the overline versus underline notation . ) given any @xmath60 , the loose preimage always has non - zero area while the tight preimage may be empty . note that @xmath72 and @xmath73 form a disjoint partition of the entire space . we may now write , @xmath74 we now review the necessary features of burning invariant manifolds @xcite . while incompressible flows have only hyperbolic ( su ) and elliptic ( ss ) fixed points , all fixed - point stability types ( sss , ssu , suu , uuu ) are possible in front element dynamics ( eq . ( [ eqn:3dode ] ) ) . for the purposes of this paper , we will concern ourselves only with the suu and ssu stability types , and only with invariant manifolds of their singleton dimensions . we will refer to these one - dimensional objects as stable and unstable _ burning invariant manifolds _ ( bims ) , respectively . previous theoretical and experimental studies have demonstrated that unstable bims are one - sided barriers to front propagation in fluid flows , similar to traditional advective invariant manifolds ( fig . [ fig : time_indep_conts_w_bims ] ) . while in many ways bims function like their passive counterparts , there are some important differences . while advective invariant manifolds are codimension 1 , bims are codimension 2 since they exist a three - dimensional phase space . ( note that the figures in this work portray _ projections _ of bims into @xmath1-space . ) it is then not obvious how bims can function as barriers . the answer is that only a subset of curves @xmath75 correspond to fronts . ( here @xmath76 is the @xmath1 euclidean length along the curve . ) the _ front - compatibility criterion _ requires that @xmath77 everywhere equals the angle @xmath78 describing the local tangent to the curve s @xmath1-projection , @xmath79 . any parameterized curve in @xmath1-space , including the boundary curve of a burned region , uniquely lifts to a curve in @xmath0-space that satisfies the front - compatibility - criterion . we shall use the term _ front _ for both the curve in @xmath1-space and its lift to @xmath0-space . it has been shown that bims satisfy the front - compatibility criterion and so are ( locally ) fronts . as such , bims are also subject to the _ no - passing lemma_no front can overtake another front from behind . when the leading front is an invariant one ( a bim ) , the other is bounded by it . because unstable bims are transversely stable , not only do they bound fronts , but fronts also converge to them making unstable bims evident in observations of the dynamics . a similar effect is observed in passive flows @xcite . just as fronts have a burning direction , so do bims . a front near a bim with the same burning direction will not pass through . conversely , a front with burning direction opposite a bim can pass through with no singularity in its motion . this is why bims are referred to as _ one - sided _ barriers . in the full @xmath0-space , bim are smooth , non - intersecting curves . however , when projected into the @xmath1-plane , bims have cusps and self - intersections . these cusps are physically meaningful in that fronts can wrap around them . therefore a bim loses its bounding behavior at a cusp . having established that bims are barriers to front propagation , just as advective invariant manifolds are barriers to fluid transport , it might seem that the analogy is complete and all that remains is to construct an active transport boundary from bim segments . however , in the next few sections we illustrate several new features of the active system that require consideration before the analogy is made complete . we proceed by first analyzing the @xmath80 regime . consider finite segments of a generic advective invariant manifold tangle . then there exists a @xmath31 small enough such that taking the advective manifolds into the left- and right - burning bims by a continuous deformation generates no new intersections in these finite segments ( see fig . [ fig : advective_turnstile_split ] ) . examining this simplest topological case reveals two important features in the composition of the active transport boundary . as @xmath31 increases from @xmath81 , each stable and unstable advective invariant manifold splits into nearby left- and right - burning bims ( fig . [ fig : advective_turnstile_split ] ) . in joining segments of these bims to form a transport boundary , each offers a choice of left- or right - burning . , bims follow near the advective manifolds for a long length . @xmath82 figure [ fig : coorientation ] illustrates these four choices . in case ( a ) , the bim segments are both rightward burning , corresponding to a burned region impinging from the left . case ( c ) corresponds to the boundary of a region on the right . because in each of these two cases , the two bim segments agree on a burning direction , we call these configurations of joined segments _ co - oriented_. in cases ( b ) and ( d ) , the bim segments chosen are not co - oriented , and do not correspond to the oriented boundary of a burned @xmath9 or @xmath10 . more precisely , consider the two front elements involved in the intersection and their respective directions of propagation @xmath83 . for each , denote by @xmath84 the direction tangent to the corresponding bim segment and oriented away from the intersection . if the two cross - products @xmath85 are of opposite sign , the segments are co - oriented . ] if we were to choose a non - co - oriented configuration ( say case ( b ) ) , then a burned region abutting this curve from the left ( right ) will burn through the stable ( unstable ) bim segment immediately . consequently , neither this bim segment , nor its iterate would bound the iterated burned region . therefore , our first new requirement of the active transport boundary is that it be composed of bim segments that are co - oriented . for the remainder of this paper we analyze fronts progressing from left to right as in fig . [ fig : coorientation](a ) . in the advective turnstile , an intersection between stable and unstable invariant manifolds must map to another such intersection . this fact allows us to neatly equate the boundary segments of a lobe iterate with the iterates of the initial boundary segments . in the active system , an intersection in the @xmath1-projection of two bims is not an intersection in @xmath0-space . generically , the bim projections meet with different orientations and so two _ different _ front elements participate in their @xmath1-intersection . more importantly , these two front elements evolve forward splitting apart in @xmath1-space ( fig . [ fig : scissor_and_con](a ) ) . this might seem to foil an active generalization of the turnstile . however , the bounding property of both advective and active invariant manifolds is not a pointwise property , but a property of curve segments . this implies that the functional intersection is not necessarily lost ; we can define an _ intersection continuation _ by tracking the continuous evolution of the intersection of the two bims . figure [ fig : scissor_and_con](a ) shows that while the two front elements initially involved in the intersection always separate , a natural continuation of this intersection can be defined . this continuation outruns each of the two front elements by a factor of @xmath86 in the comoving fluid frame . figure [ fig : scissor_and_con](a ) is compatible with two different burned regions . in fig . [ fig : scissor_and_con](b ) , the burned region has a _ concave corner _ while the region in fig . [ fig : scissor_and_con](c ) has a _ convex corner _ ( compare fig . [ fig : coorientation](a , c ) ) . in order to understand the evolution of these burned regions using the 3d ode for front propagation , we need to properly join the two original front segments at the corner . remembering that fronts are curves in @xmath0-space , it is clear that the two front segments alone form a discontinuous curve . we join these two front segments with a _ connection _ @xmath87 that is simply a straight line connecting the @xmath51 values of the two front elements at the corner . since @xmath88 , there are two distinct intervals between any two @xmath51 values ; the physically correct choice is the shorter of the two intervals . the two segments joined with the connection are now a continuous front in @xmath0-space . in fig . [ fig : scissor_and_con](b ) , the connection evolves into the arc of a circle and is immediately burned past by both front segments . further , the front segments partially overtake each other sending finite portions into the reacted region . the boundary of the final reacted region @xmath89 is a subset of the evolved front , @xmath90 . in fig . [ fig : scissor_and_con](c ) , the connection unfolds in the same way , but it remains on the boundary of the reacted region , @xmath91 . this connection , functioning like a point stimulation made at the convex corner , is a boundary between burned and unburned that is not the iterate of , or continuation of , either original front segment ; it generates `` new boundary '' . also notice that the intersection continuation outruns this point stimulation . this concept will lead to the second requirement of the active transport boundary . having now described front element splitting and intersection continuation and the effect of concavity on the physical relevance of the connection , we apply this to pips between the bims . first we establish some notation for pips , comparing the passive and active case . recall that for advective dynamics , a point @xmath92 is an _ advective pip _ if @xmath93 \cap w^u[z^b , p ] \ \backslash \ \{z^a , z^b , p\}=\emptyset$ ] ( see fig . [ fig : advective_turnstile ] ) . the image and preimage of a advective pip @xmath26 are denoted @xmath94 and @xmath95 , respectively . for any advective pip @xmath96 , @xmath97 . we now consider pips in the active case . the tuple of two intersecting front elements @xmath98 , where @xmath99 , is a _ burning pip _ if @xmath100 ) \cap \pi_{xy}(w^u[z^b , p^u ] ) \ \backslash \ \pi_{xy}(\{z^a , z^b , p\})=\emptyset$ ] where @xmath101 denotes projection into the @xmath1-plane . in fig . [ fig : ps_and_qs ] we have labeled the burning pips @xmath102 and @xmath103 ( solid dots ) . we write @xmath104 ( @xmath105 ) to indicate the stable ( unstable ) component of the @xmath106th burning pip @xmath107 , and similarly for @xmath103 . since a burning pip @xmath26 splits , @xmath108 , @xmath109 , @xmath110 , and @xmath111 . generically , @xmath112 . however , continuation of intersections takes @xmath113 and @xmath114 . in fig . [ fig : ps_and_qs ] unfilled dots denote iterates of the two components of each burning pip ( solid dots ) along either the stable or unstable bim . and @xmath103 ( solid dots ) represent two coincident front elements . these elements are mapped forward and backward ( open dots ) . @xmath82 for instance , examine the evolution of @xmath115 in fig . [ fig : ps_and_qs ] . because of front element splitting , @xmath116 . instead , the two front elements @xmath117 and @xmath118 that compose @xmath115 map to @xmath119 and @xmath120 . notice that the two points @xmath119 and @xmath120 have mapped , respectively , behind the local unstable and stable bims at @xmath26 . unlike the passive case , in which the boundary of a lobe exactly maps to the boundary of the iterated lobe , in the active case , the stable ( unstable ) boundary segment of a lobe maps to a stable ( unstable ) segment that is either shorter or longer than the stable ( unstable ) boundary of the subsequent lobe . we use this knowledge , and the fact that pips alternate between concave and convex , to derive the following relations . @xmath121 ) & = w^s[{\ensuremath m_3}(p^s_{-1 } ) , { \ensuremath m_3}(q^s_{-1})]\\ & \supsetneq w^s[p_{0},q_{0}]\\ { \ensuremath m_3}(w^u[p_{-1},q_{-1 } ] ) & = w^u[{\ensuremath m_3}(p^u_{-1 } ) , { \ensuremath m_3}(q^u_{-1})]\\ & \subsetneq w^u[p_{0},q_{0}]\\ { \ensuremath m_3}(w^s[q_{-1},p_{0 } ] ) & = w^s[{\ensuremath m_3}(q^s_{-1 } ) , { \ensuremath m_3}(p^s_{0})]\\ & \subsetneq w^s[q_{0},p_{1}]\\ { \ensuremath m_3}(w^u[q_{-1},p_{0 } ] ) & = w^u[{\ensuremath m_3}(q^u_{-1 } ) , { \ensuremath m_3}(p^u_{0})]\\ & \supsetneq w^u[q_{0},p_{1 } ] \end{aligned}\ ] ] we now show that constructing a transport boundary from bim segments joined by a concave burning pip ensures that the @xmath62 and @xmath65 burning lobes are bounded by bim segments . in reference to fig . [ fig : ps_and_qs ] , we construct the transport boundary from @xmath27 $ ] and @xmath28 $ ] and the concave connection @xmath87 ( fig . [ fig : scissor_and_con](b ) ) . this transport boundary ( @xmath122 ) is the bounding front of @xmath9 , and so by evolving @xmath122 forward , we can understand the boundary of the evolution of @xmath9 . because we chose a concave pip , @xmath123 . we know by the invariance of bims and fig . [ fig : ps_and_qs ] that this transport boundary will map forward to @xmath124 \cup { \ensuremath m_3}(\mathcal{c } ) \cup w^u[z^b , { \ensuremath m_3}(p^u_0)]$ ] . we also know that the subsection of this front @xmath125 \cup { \ensuremath m_3}(\mathcal{c } ) \cup w^u[p_1 , { \ensuremath m_3}(p^u_0)]$ ] will lie behind the boundary of the burned region @xmath126 \cup w^u[z^b , p_1]$ ] ( see fig . [ fig : lhs_forward ] ) . recalling the set definition of the @xmath62 lobe ( eq . ( [ eqn : active_lobe_defs_extendede0c0 ] ) ) , we see in fig . [ fig : lhs_forward ] that @xmath127 is bounded by @xmath128 $ ] and @xmath129 $ ] . it is similarly straightforward to see that @xmath130 is bounded by @xmath131 $ ] and @xmath132 $ ] . and @xmath122 are evolved forward showing the @xmath62 and @xmath65 lobes . @xmath82 if a convex pip is chosen instead ( @xmath26 in fig . [ fig : wrong_pip_bad_c0 ] ) , the convex connection @xmath87 used to join @xmath27 $ ] and @xmath28 $ ] leads to the following problem . the connection evolves into the arc @xmath133 ( fig . [ fig : wrong_pip_bad_c0 ] inset ) , which lags behind the continuation @xmath42 . the space between @xmath133 and the local stable and unstable segments near @xmath42 is part of @xmath9 yet unburned , and therefore part of the @xmath65 lobe . thus the boundary of @xmath65 includes the arc @xmath133 , which is not a bim segment . therefore , in order to maintain the property that the @xmath62 and @xmath65 lobe boundaries are bim segments , we must impose the second constraint that the transport boundary is defined by a _ concave _ burning pip . burned , and iterate ( filled green ) . small unburned region ( zoomed inset ) has one boundary arc which is not a bim . @xmath134 is a uniform flow to the right . ( b , c , e , f ) green regions are preimages of @xmath60 . black curves represent two ways of mapping @xmath59 backwards . reference curves ( gray ) show original region . ] the correspondence between burned region evolution ( eq . ( [ eqn : fkpp ] ) ) and bounding front propagation ( eq . ( [ eqn:3dode ] ) ) has already been established . we used this correspondence to understand the nature of the @xmath62 and @xmath65 lobes . here we establish an analogous correspondence between the loose and tight preimages of a region @xmath60 and the bounding curves so that we may similarly understand the @xmath30 and @xmath29 lobes . figure [ fig : preimages ] illustrates sets @xmath60 , @xmath72 and @xmath135 as well as the relevant curves . we begin with the loose preimage . consider the set of stimulations which do not reach the target set @xmath60 , @xmath136 . the boundary of @xmath137 maps into the boundary of @xmath60 with orientation reversed , @xmath138 . application of an inverse and second reversal yields @xmath139 . finally , since @xmath137 is the complement of the set of interest , @xmath140 . thus we have the following relation between the boundary of the loose preimage and @xmath60 through the front element dynamics , @xmath141 equality is obtained when no local ( swallowtails ) or global front intersections develop in @xmath142 ( equivalently in @xmath143 ) . as we see comparing figs . [ fig : preimages](b ) and [ fig : preimages](e ) , a concave corner is certain to produce a swallowtail and break the equality . the tight preimage relation is simpler to state , @xmath144 similarly with equality if no front intersections are developed in @xmath145 . with the correspondence between bounding fronts and region preimages ( fig . ( [ fig : preimages ] ) ) established , we can determine the @xmath30 and @xmath29 lobes with the assistance of the invariant manifolds . in fig . [ fig : lhs_backward](a ) , we define the transport boundary @xmath122 using a concave pip . the @xmath30 lobe depends on the loose preimage @xmath146 through eq . ( [ eqn : active_lobe_defs_extendeden1cn1 ] ) . let us examine this preimage . in the previous section , we established the relation eq . ( [ eqn : loose_boundary_relation ] ) between the boundary of a preimage and inverse images of boundaries . here this relation gives , @xmath147 figure [ fig : lhs_backward](b ) illustrates the inverse of the transport boundary @xmath148 and the region it bounds ( green ) . since @xmath122 has a concave pip , @xmath148 is not forced to generate a swallowtail . our earlier assumption about @xmath31 being small enough to not generate any new intersections ensures that the entire curve @xmath148 is free of self intersections . thus we have the equality @xmath149 . this ensures that the loose preimage of @xmath10 is exactly the white region in fig . [ fig : lhs_backward](b ) . finally intersecting with @xmath9 ( eq . ( [ eqn : active_lobe_defs_extendeden1cn1 ] ) ) , we see the location of @xmath30 and that its boundaries are the bim segments @xmath150 $ ] and @xmath151 $ ] . on the other hand , @xmath29 depends on the tight preimage of the other side , @xmath152 ( eq . ( [ eqn : active_lobe_defs_extendeden1cn1 ] ) ) . equation ( [ eqn : tight_boundary_relation ] ) gives , @xmath153 therefore , @xmath152 is exactly the green shaded region in fig . [ fig : lhs_backward](b ) . intersecting with @xmath10 , we see that @xmath29 is not bounded by bim segments @xmath154 $ ] and @xmath155 $ ] ( see fig . [ fig : lhs_backward](b ) inset ) . it is instead bounded by @xmath156 $ ] , @xmath157 $ ] and @xmath158 . in the same way , we previously found that when a convex pip is used , the @xmath65 lobe is bounded in part by @xmath133 , a non - bim segment . and @xmath122 are evolved backward ( loose preimage ) showing the @xmath30 and @xmath29 lobes . @xmath30 lobe is bounded by bims . @xmath29 lobe has non - bim boundary . @xmath82 if one is only interested in the escape process , i.e. transport of the burned region to the right of the @xmath122 , both concave and convex pips are equally suitable , in that @xmath30 and @xmath62 are both bounded entirely by bims . however , for the capture process , i.e. entrainment of unburned fluid to the left of the @xmath122 , either the lobe @xmath29 ( for a concave pip ) or the lobe @xmath65 ( for a convex pip ) will be bounded in part by a non - bim segment . we shall preferentially consider concave pips in the remainder of this paper , both because the lobe @xmath65 is more directly visible in experiments and because the relevant @xmath122 ( concave ) can be realized through the dynamical development of a burned region . as can be seen in figs . [ fig : capture_lobe ] and [ fig : escape_lobe ] , the passive relations @xmath159 and @xmath160 are softened to the following inclusions , [ eqn : inclusion_loose ] @xmath161 the @xmath62 inclusion follows directly from the definition eq . ( [ eqn : active_lobe_defs_ptse ] ) . however , the @xmath65 inclusion depends on the fact that @xmath162)$ ] includes @xmath131 $ ] , the stable boundary of @xmath65 . we may reobtain @xmath62 by intersecting eq . ( [ eqn : inclusion_loosea ] ) with the @xmath10 , @xmath163 lobe and its bounding segments . @xmath82 lobe and its bounding segments . we have shown that for appropriately small @xmath31 , there are new criteria that need to be taken into account when constructing a burning turnstile . these criteria arise from breaking the @xmath164 symmetry . in this section we examine the case of larger @xmath31 , where new intersections arise . in classical front propagation theory , formation of swallowtail catastrophes is generic @xcite ( fig . [ fig : swallowtail ] ) . a front with a concave segment has a point of greatest curvature or , equivalently , minimum radius of curvature . a swallowtail catastrophe occurs when this point propagates forward to its local center of curvature . as the front propagates further still , it develops a self - intersecting structure reminiscent of a swallow s tail . in the case of a front separating burned and unburned regions , the swallowtail must lie entirely within the burned region . that is , the front segments that make up the swallowtail are no longer boundaries between burned and unburned regions . swallowtail formation is also generic for fronts propagating in flowing media . the flow will influence the curvature of the front , thereby promoting or preventing formation of swallowtails . many flows will tend to generate and regenerate areas of concavity in propagating fronts , thus facilitating continued swallowtail formation . given the basic turnstile geometry in fig . [ fig : advective_turnstile ] , we might expect to see swallowtails first appear in the boundary of the @xmath29 lobe . ) . ] this is because the lobe is long and narrow , creating a segment of high curvature at the tip . as the right - burning stable bim deviates from the advective manifold ( with increasing @xmath31 ) , the bim is forced inward , increasing the curvature at the tip until a swallowtail is forced . ) . ] let us consider the effect of such a swallowtail on the @xmath29 lobe . this lobe is defined as the set of all point stimulations that cross the transport boundary completely . suppose a stimulation within the triangular region ( point @xmath165 in fig . [ fig : cn1swallowtail_w_region ] ) were to evolve completely across the @xmath122 . then the last front element to pass through the stable bim would be tangent to it and co - oriented . hence , according to the no - passing lemma , this stimulation could not pass through entirely a contradiction . the swallowtail region is therefore not a part of the @xmath29 lobe . the area remaining is shown in fig . [ fig : cn1swallowtail_w_region ] ( inset ) . the first effect of swallowtails , then , is that the @xmath29 area is diminished by the encroaching swallowtail . the second effect is that the boundary of the @xmath29 lobe is reduced ; here , the middle segment of the stable bim is no longer on the lobe boundary ( fig . [ fig : cn1swallowtail_w_region ] inset ) . in general , a single bounding bim segment is replaced by the joining of two segments . evolving this reduced area forward , we find the situation in the upper - left of fig . [ fig : cn1swallowtail_w_region ] . the region bounded by stable and unstable bims is still the @xmath65 lobe nothing in this region is reached by any stimulation on the @xmath9 . however , notice that a small portion also remains unburned by the evolution of @xmath29 . we can understand this by looking at fig . [ fig : scissor_and_con](c ) ; the intersection continuation runs ahead of the burned region leaving an unburned gap . in this case , the intersection continuation is lost when the swallowtail itself is removed by its catastrophe ; at this point , we are left with the evolution of two distinct fronts ( the boundary of the evolving @xmath29 lobe and the now unfolded segment of stable bim ) . the arc that bounds the unburned gap in fig . [ fig : scissor_and_con](c ) is in fig . [ fig : cn1swallowtail_w_region ] geometrically distorted by the flow , reversing its concavity . our previous inclusion relation for the evolution of @xmath29 ( eqs . ( [ eqn : inclusion_looseb ] ) ) has thus been weakened to : if @xmath166 , then @xmath167 . in the same spirit , the @xmath65 lobe may form a swallowtail as the unstable bim deviates inward from the advective unstable manifold ( fig . [ fig : four_swallowtails ] ) . this swallowtail region represents fluid that has been crossed by two separate front segments of the evolving @xmath9 . as @xmath65 is the region that is not burned by any stimulation on the @xmath9 , this swallowtail region is excluded from the @xmath65 lobe . the unstable bim segment that evolves to form this swallowtail previously bounded part of the @xmath29 lobe . however note that this segment is oriented into @xmath29 . therefore , evolving @xmath29 forward , the whole unstable bim segment ( whether it forms a swallowtail or not ) is immediately burned through by @xmath168 . therefore , this swallowtail on the _ unstable _ manifold does not itself negate the inclusion relation @xmath169 . just as we saw in the @xmath29 lobe , the existence of a @xmath65 swallowtail only removes lobe boundary and so this boundary is still composed of bim segments . adjusting the flow appropriately , the advective @xmath62 lobe becomes longer , thinner and begins to fold . the active @xmath62 lobe can develop a swallowtail in the concave folding region ( fig . [ fig : four_swallowtails ] ) . this area is `` burned twice '' and so is still within the @xmath62 lobe . the inclusion relation remains @xmath170 . the @xmath30 burning lobe can be similarly modified . as in the @xmath29 case , this swallowtail unfolds in forward time . however , since stimulations in @xmath30 need only intersect the @xmath10 , the swallowtail is included in @xmath30 . the inclusion relation is also unaffected . unlike advective pips , which exist for any amount of periodic perturbation , it is possible for burning pips to be pulled apart in certain parameter regimes for instance , when the front propagation speed is large enough . we can understand the basic intuition geometrically . figure [ fig : v0_wipes_out_lobes ] shows bims for the same flow with four increasing values of @xmath31 . in the left - most case ( @xmath171 ) , the bims degenerate to the advective invariant manifolds . all four turnstile lobes have the same area . this is because of area - preservation of the map ( incompressibility of the fluid ) and zero - net - flux across the turnstile . as @xmath31 is increased , the stable and unstable bims of interest are to the left and right of their respective advective fixed points . also , the distance from each fixed point increases , squeezing the capture lobes to zero area , and finally destroying the central pip . note that this loss of area is a mechanism distinct from the swallowtail discussed above . it is also interesting to see the heteroclinic connection in fig . [ fig : v0_wipes_out_lobes](c ) . as previously discussed , intersections of fronts in @xmath1-space are not intersections in @xmath0-space . in this case , since @xmath26 is a tangency between bim projections , this is a true intersection of bims and thus @xmath26 generates a heteroclinic orbit . we can understand the pip existence issue from another perspective . in the time - independent flow the stable and unstable bims are separated by some finite gap ( see fig . [ fig : forcing_overcomes_v0_gap](a ) and appendix [ appx : no_time_indep_lobes ] for details ) . sufficient forcing causes oscillations in the bims on this same length scale enough to generate intersections and lobes . the dependence of pip existence and lobe area on forcing strength appears to be more complex than on burning speed . for instance in fig . [ fig : forcing_overcomes_v0_gap ] we see that the pips gained by some forcing in ( c ) are then lost by greater forcing in ( d ) . in short , there are several distinct ways in which a system may gain or lose a burning pip . the physical implications of this warrant further investigation . in this paper , we have considered fluids within a channel geometry . a stimulation made in this channel with no flow will propagate outwards as a circle with radius growing at speed @xmath31 until it runs into the channel wall . a burned region that encounters the channel wall finds nothing beyond to burn , and so its bounding front stops progress in that direction . while this is straightforward to understand , its implementation is worth noting . numerically , we supply the channel with a thin boundary layer , extending the domain just beyond the channel wall . the particular flow @xmath52 chosen beyond the boundary is not important . upon entering this boundary layer , a front element s velocity perpendicular to the channel is quickly attenuated , reaching zero at the outer limit . in this way , front elements that enter the boundary layer do not leave . in the ideal limit the boundary layer is infinitely thin , and front elements that enter this layer can be identified with points on the channel wall . the result is that when lobes ( such as the @xmath62 lobe in fig . [ fig : channel_wall_bc_off ] ) collide with the channel wall , the channel wall itself forms a segment of the lobe boundary . lobe boundary . there is growing interest in the behavior of biological and chemical systems in fluid flows @xcite . for instance , recent studies have examined the interplay between rapid population growth in plankton and the underlying ocean currents @xcite . here we describe a scenario in which knowledge of the burning turnstile can be used to more efficiently mitigate the impact of an impinging front . imagine an ocean bay that we wish to protect from an invading algal bloom . suppose that the surface ocean flow is well - modeled by a 2d periodically perturbed flow @xcite . a separatrix may be generated by an ocean current parallel to the coastline and across the mouth of the bay . additionally , diurnal tides are a reasonable source of periodic perturbation . this combination produces a turnstile mechanism for ocean water to enter and exit the bay . as we have discussed in this paper , this turnstile may be modified to account for the addition of propagating fronts . an active constituent in the ocean , such as plankton , would be transported into the bay by the advective turnstile while simultaneously growing outward , pushing past the advective invariant manifolds to the bims defining the burning turnstile . suppose the goal is to protect the bay by applying an algaecide to the surface of the water in synchrony with the tide . furthermore , imagine that this substance is 100% effective , but is active for only a short time , possibly due to photo - degradation . importantly , treatment is often costly or toxic to other ocean life , so we wish to apply only as little as possible . we do not presume the details accurately reflect a real - life scenario . the intention is to illustrate a plausible motivation for suppressing the invasion of a front . there are several strategies that may be suggested . first is to apply a strip of algaecide spanning the mouth of the bay in a rectangle . while this protocol and simple to implement , it does not take advantage of our understanding of the dynamical structure , and as a result either fails to prevent the algae invasion , or uses an excess of suppressant . a better strategy attempts to make use of oceanic flow structures . in particular , one could focus the treatment on the advective @xmath30 or @xmath62 lobe . enlarging this treatment area uniformly around the advective lobe yields an effective and more efficient strategy . as a first approximation , one could enlarge outward a distance @xmath173 . while this is a reasonable suggestion , it fails to account for the fact that as the front evolves , its continuous propagation beyond the corresponding passive trajectories places the front into regions of flow different from those experienced by the passive trajectories . in other words , the gradient of the flow feeds back into the evolution of the front and in this way leads to lobes that differ from simple uniform enlargements . during the course of one tidal cycle , the algae s growth pushes it just beyond an advective stable manifold . this grants the algae access to a region of fluid that was entirely inaccessible to passive trajectories . this small breach can then proliferate into a significant bloom . we propose that the best strategy , given the assumptions , is to apply the algaecide to an burning escape lobe @xmath30 or @xmath62 . this protocol eliminates exactly that algae which may transgress or has transgressed the transport boundary . which of the two is the most efficient ( smallest area ) depends on the details of the flow . figure [ fig : reaction_suppression ] illustrates these three protocols ( and control with no suppression ) . the rectangular protocol clearly fails to prevent the rightward propagation of the front , despite the rectangular area being fully twice that of the larger escape lobe @xmath62 . in contrast , both lobe - based protocols are able to prevent the front propagation , using significantly less suppressant . ( note that the discrete grid introduces some noise , which we accommodate by increasing the suppression regions by about 2 cell widths . ) for advective transport , one can readily show that the @xmath122 defined by a pip connecting stable and unstable manifolds minimizes the flux across the @xmath122 . a small perturbation to such a @xmath122 results in an advective flux that is larger than ( or equal to ) the original flux . we conjecture that an analogous statement is true concerning the @xmath122 for front propagation . this question , however , is more subtle than in the passive case and will be addressed in a future publication . we have demonstrated how the turnstile , a fundamental construct in the theory of passive fluid transport , is generalized to fronts propagating in a flowing fluid . this burning turnstile is originally defined in terms of the evolution of point stimulations in the active fluid . the modified turnstile exhibits several features distinct from its passive counterpart . first , the burning turnstile is bounded by bims , _ one - sided _ barriers to front propagation . second , to form a meaningful transport boundary from finite segments of stable and unstable bims , the bims relative orientations must be consistent . third , the burning pips used to join the stable and unstable bims are only intersections of the bims when projected to the @xmath1 plane . fourth , this implies that a segment of the boundary of either @xmath29 or @xmath65 does not consist of bim segments , depending on whether the pip is concave or convex , respectively . finally , the areas of the lobes are not preserved and the relations between @xmath174 and @xmath62 and between @xmath168 are @xmath65 are more complicated than the advective case . increasing the front propagation speed can change the topology of the lobes , most notably by generically introducing swallowtails into the lobe boundaries . these swallowtails excise a portion of a lobe s bim boundary . a large enough propagation speed can even remove a burning pip entirely . if no such pips remain , the turnstile formulation can no longer be applied . to illustrate the utility of this theoretical framework , we applied it to a simple model of an ocean bay invaded by algae . this shows that efficient protection of the bay depends on the geometric constructs discussed herein the burning lobes . the authors gratefully acknowledge discussions with tom solomon , whose experimental work inspired this paper . we also benefited from a talk of solomon s in which he proposed an earlier and alternative formulation of lobes in chaotic advection - reaction - diffusion systems ( unpublished ) . the present work was supported by the us national science foundation under grants phy-0748828 and cmmi-1201236 . 10 m. s. paoletti and t. h. solomon . experimental studies of front propagation and mode - locking in an advection - reaction - diffusion system . , 69:819 , 2005 . m. e. schwartz and t. h. solomon . chemical reaction fronts in ordered and disordered cellular flows with opposing winds . , 100:028302 , jan 2008 . dylan bargteil and tom solomon . barriers to front propagation in ordered and disordered vortex flows . , 22 , 2012 . a. pocheau and f. harambat . effective front propagation in steady cellular flows : a least time criterion . , 73:065304 , jun 2006 . a. pocheau and f. harambat . front propagation in a laminar cellular flow : shapes , velocities , and least time criterion . , 77:036304 , mar 2008 . m. abel , a. celani , d. vergni , and a. vulpiani . front propagation in laminar flows . , 64:046307 , sep 2001 . m. cencini , a. torcini , d. vergni , and a. vulpiani . thin front propagation in steady and unsteady cellular flows . , 15:679688 , 2003 . s. wiggins . . springer - verlag , new york , 1992 . john mahoney , dylan bargteil , mark kingsbury , kevin mitchell , and tom solomon . invariant barriers to reactive front propagation in fluid flows . , 98:44005 , 2012 . kevin a. mitchell and john mahoney . invariant manifolds and the geometry of front propagation in fluid flows . , 22 , 2012 . r. s. mackay , j. d. meiss , and i. c. percival . transport in hamiltonian systems . , 13:55 , 1984 . t. h. solomon and j. p. gollub . chaotic particle transport in time - dependent rayleigh - bnard convection . , 38:62806286 , dec 1988 . v. rom - kedar and s. wiggins . transport in two - dimensional maps . , 109:239 , 1990 . it should be pointed out that the turnstile mechanism is a piece of the larger theory of lobe dynamics . the goal of this paper is to firmly establish the turnstile mechanism for propagating fronts . a treatment of the multi - time - step dynamics is left to a future paper . p. d. ronney . some open questions in premixed turbulent combustion . in j. buckmaster , editor , _ modeling in combustion science_. springer - verlag , berlin , 1994 . r. a. fisher . the wave of advance of advantageous genes . , 7:355 , 1937 . a. n. kolmogorov , i. g. petrovskii , and n. s. piskunov . a study of the diffusion equation with increase in the amount of substance , and its application to a biological problem . , 1:1 , 1937 . f. a. williams . . addison - wesley , redwood city , ca , 1985 . m. oberlack and a. f. cheviakov . higher - order symmetries and conservation laws of the g - equation for premixed combustion and resulting numerical schemes . , 66:121140 , 2010 . we say _ locally _ because unlike the curve in 3d corresponding to the perimeter of a compact burned region , a bim can not form a topological circle . s. wiggins v. rom - kedar , a. leonard . an analytical study of transport , mixing and chaos in an unsteady vortical flow . 214:347394 , 1990 . v. i. arnold . . springer , berlin , 1983 . it is important to note that since ray optics is invertible , reversing the propagation direction ( or time ) can cause the swallowtail to unfold . this phenomenon is found in the stable bims ( see fig . [ fig : cn1swallowtail_w_region ] ) . we are making use of the stability of this perestroika . that is , the structural change is actually occurring in time , but according to ref . @xcite , we can see the same morph with a generic parameter sweep ( such as @xmath31 ) . t. tel , a. de moura , c. grebogi , and g. karolyi . chemical and biological activity in open flows : a dynamical system approach . , 413:91196 , 2005 . z. neufeld and e. hernandez - garcia . . imperial college press , 2009 . mathias sandulescu , cristobal lopez , emilio hernandez - garcia , and ulrike feudel . biological activity in the wake of an island close to a coastal upwelling . , 5(3):228237 , sep 2008 . shawn c. shadden , francois lekien , jeffrey d. paduan , francisco p. chavez , and jerrold e. marsden . the correlation between surface drifters and coherent structures based on high - frequency radar data in monterey bay . , 56(3 - 5):161 172 , 2009 . it is well - known that there are no transverse intersections and hence no advective lobes in time - independent 2d flows . these flows also have no burning lobes , however the reasoning behind this is somewhat different . assume for the remainder of this section that the fluid flow is time - independent . as previously shown @xcite , bims obey the front - compatibility criterion ; @xmath175 where @xmath76 is a parameterization of the bim by the euclidean distance along the bim . this is a condition on spatial structure . additionally , bims are , by definition , time - invariant for time - independent flows . a front element that begins anywhere on the bim travels in time along the bim spatially@xmath176 . these two proportionalities combine to yield @xmath177 . this implies @xmath178 . this is our main constraint on bims in time - independent flows . if @xmath179 , there are no solutions . if @xmath180 , there is one solution @xmath181 . if @xmath182 , there are two solutions which we illustrate in fig . [ fig : rdot_propto_ghat ] . consider the ( hypothetical ) intersection of two bim segments at a burning pip . if at this intersection point there is one constraint solution , both bim segments pass through the same @xmath0 point . by uniqueness of solutions to odes , these trajectories are the same and so there are no burning lobes . if there are two constraint solutions , the two bim segments pass through the two different points @xmath186 and @xmath187 . if they pass through the same point , the previous argument applies again . therefore the intersection must be transverse and the two front elements involved must bear the relation in fig . [ fig : rdot_propto_ghat ] . finally notice that a burning pip , such as @xmath26 or @xmath188 in fig . [ fig : ps_and_qs ] ) , does not have this configuration . hence we reach a contradiction . this demonstrates that time - independent flows do not allow transverse pips or burning lobes .

we prove the asymptotic formulas for the transition densities of isotropic unimodal convolution semigroups of probability measures on under the assumption that its lvy khintchine exponent varies slowly . we also derive some new estimates of the transition densities and green functions .

1606.04178

let @xmath1 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath3 . in the recent paper @xcite , the estimates of the transition densities @xmath4 were studied under the assumption that @xmath3 has lower and upper matuszewska indices strictly between @xmath5 and @xmath6 . in fact , more detailed information can be obtained whenever stronger assumption about the behaviour of @xmath3 is imposed . namely , in @xcite , it is proved that if @xmath3 varies regularly at infinity with index @xmath7 then @xmath8 and @xmath9 where @xmath10 moreover , the asymptotics implies that @xmath3 varies regularly at infinity with index @xmath7 . a similar result was proved also for @xmath3 regularly varying at the origin . the natural question arises : what are the asymptotic behaviour and estimates of the semigroup for the endpoints @xmath11 . regarding estimates , the case @xmath12 was investigated for subordinate brownian motions in the recent paper @xcite . however , the case @xmath13 seems to be hardly studied . in ( * theorem 5.52 ) there are estimates of @xmath14 for small values of time variable for a geometric stable process , that is when @xmath15 for some @xmath16 . to the authors best knowledge the asymptotics were never studied before . this article is an attempt to fill that gap in the theory . we provide a solution for a large class of slowly varying symbols including geometric stable and iterated geometric stable cases . namely , we study processes with symbols belonging to de haan class @xmath17 associated with a function @xmath18 slowly varying at infinity , for definition see . for an isotropic unimodal lvy process @xmath1 on @xmath2 with the lvy khintchine exponent @xmath19 , which is equivalent to regular variation with index @xmath20 at the origin of the density of lvy s measure of the process ( see theorem [ thm:5 ] ) , we derive several asymptotic results describing the behaviour of the transition density @xmath21 and the green function @xmath22 . to be more precise , in section [ sec:3 ] , we prove that @xmath23 we also study @xmath24 the density of the lvy measure of the process @xmath25 . in particular , implies that @xmath26 for the green function @xmath27 of @xmath25 , if the process is transient ( what always holds if @xmath28 ) , we show that @xmath29 in particular , extends the result of @xcite where the geometric stable and iterated geometric stable cases were studied . we also investigate the asymptotics of @xmath14 when @xmath30 gets large provided @xmath18 is a slowly varying bounded function . in corollary [ cor:2 ] we show that @xmath31 we want to emphasize that if @xmath32 is sufficiently small then @xmath33 may be _ this new phenomena requires an additional factor not present in the case when @xmath3 is regularly varying at infinity with index strictly larger than zero . we also obtain the corresponding results when the lvy khintchine exponent @xmath3 belongs to @xmath34 for some function @xmath18 slowly varying at zero . namely , @xmath35 and @xmath36 moreover , we prove several estimates of the transition density and the green function under various assumptions for unimodal and isotropic processes which improve the existing results for subordinate brownian motions . if the lvy khintchine exponent @xmath3 belongs to de haan class associated with a bounded function @xmath18 slowly varying at infinity we show very precise two - sided estimates which complement the results obtained in @xcite . namely , we show that there are positive @xmath37 and @xmath38 such that for all @xmath39 and @xmath40 for @xmath41 means that there @xmath42 such that @xmath43 for all @xmath41 ] @xmath45 to our best knowledge this is the first result of this type when the lvy khintchine exponent is slowly varying at infinity there are many articles exhibiting the local behaviour of the transition density of lvy processes for small time and space variables which are valid under certain assumptions on the density of the lvy measure , see e.g. @xcite . namely , if @xmath46 is comparable at the origin with a decreasing @xmath47 having the matuszewska index at the origin strictly between @xmath5 and @xmath6 then there are positive @xmath37 and @xmath38 such that for all @xmath39 and @xmath40 @xmath48 one of our contributions is to discover that for processes with the lvy khintchine exponent belonging to de haan class associated with a bounded function slowly varying at infinity the rle of @xmath49 is taken by @xmath50 . to illustrate the results we provide examples with slowly varying symbols to which our method applies . these examples suggest that a unified form of estimates may be difficult to discover within the class of processes with slowly varying symbols since the estimates exhibit quite a lot of irregularity . in the case of subordinate brownian motions we prove some extensions of the results of @xcite where estimates of @xmath4 were obtained in terms of the symbol @xmath3 of the process under weak scaling assumptions . namely , @xmath3 satisfies lower and upper scaling conditions at infinity with indices strictly between @xmath5 and @xmath6 if and only if there is @xmath37 such that for all @xmath51 and @xmath39 , if @xmath52 then @xmath53 let us recall that for a subordinate brownian motion we have @xmath54 where @xmath55 is a bernstein function which determines the distribution of the underlying subordinator via laplace transform . it turns out that suitable weak scaling properties imposed on the derivative @xmath56 lead to the following extension of , there is @xmath57 such that for all @xmath51 and @xmath58 , if @xmath59 then @xmath60 the last result is particularly interesting when @xmath3 is slowly varying at infinity since can not hold in this case . moreover , under some additional assumptions on the subordinator we can also show the converse . that is , if is satisfied then @xmath56 has some scaling properties . for subordinate brownian motion , the case when @xmath3 is regularly varying at @xmath61 with index @xmath6 , a function @xmath62 plays a similar rle as @xmath63 in this article . for details we refer to the recent paper @xcite where the estimates were obtained under suitable scaling assumptions of @xmath64 . we can conclude that our present paper together with @xcite shed some light on the situation when the lvy khintchine exponent @xmath3 is slowly varying or @xmath6-regularly varying at infinity . these cases seem to be more complicated and still far from being completely understood . the paper is organized as follows . in section [ sec:2 ] , we recall some basic facts regarding unimodal isotropic lvy processes . in section [ sec:3 ] , we present an extension of uniform tauberian theorems we apply to derive asymptotics of @xmath4 under the assumption that the lvy khintchine exponent is slowly varying and belongs to de haan class . we also study the asymptotics of the green function at the origin and at infinity . in section [ sec:5 ] , we derive general upper and lower bounds for @xmath4 and the green function under various assumptions , usually involving scaling properties of some functions determined by @xmath3 . next , for subordinate brownian motion we show two - sided estimates of the resolvent kernel @xmath65 under assumptions on scaling properties of the derivative of the laplace exponent of the subordinator . then we use the obtained estimates to derive estimates of @xmath4 . finally , in section [ sec:6 ] we give examples illustrating the behaviour of @xmath4 for processes with slowly varying lvy khintchine exponents . let @xmath66 be an isotropic pure jump lvy process in @xmath2 , i.e. @xmath25 is a cdlg stochastic process with a distribution denoted by @xmath67 such that @xmath68 almost surely , the increments of @xmath25 are independent with a radial distribution @xmath69 on @xmath70 . this is equivalent with radiality of the lvy measure @xmath24 and the lvy khintchine exponent @xmath3 . in particular , the characteristic function of @xmath71 has a form @xmath72 where @xmath73 we are going to abuse the notation by setting @xmath74 for @xmath75 to be equal to @xmath76 for any @xmath77 with @xmath78 . the same rule applies to any radial function appearing in this paper . since the function @xmath3 is not necessarily radially monotonic , it is conveniently to work with @xmath79 defined for @xmath80 by @xmath81 } \psi(s).\ ] ] let us recall that for @xmath82 ( see ( * ? ? ? * theorem 2.7 ) ) @xmath83 a borel measure @xmath84 is isotropic _ unimodal _ if it is absolutely continuous on @xmath85 with a radial and radially non - increasing density . a lvy process @xmath86 is isotropic unimodal if @xmath87 is isotropic unimodal for each @xmath51 . we consider a subclass of isotropic processes consisting of isotropic unimodal lvy processes . they were characterized by watanabe in @xcite as those having the isotropic unimodal lvy measure . hence , the lvy measure is absolutely continuous with respect to the lebesgue measure . its density , denoted by @xmath88 , is radially non - increasing . a remarkable property of these processes is ( see ( * ? ? ? * proposition 2 ) ) @xmath89 for all @xmath80 . in this article , we mainly consider isotropic unimodal lvy processes with the lvy khintchine exponent @xmath3 slowly varying at infinity or zero . let us recall that a function @xmath90 , for some @xmath91 , is called _ slowly varying at infinity _ if for each @xmath92 @xmath93 the set of functions slowly varying at infinity is denoted by @xmath94 . a positive function @xmath18 defined in a right neighbourhood of zero is slowly varying at zero if @xmath95 is in @xmath94 . the set of functions slowly varying at zero is denoted by @xmath96 . the following property of a function slowly varying at infinity appears to be very useful ( see @xcite , see also ( * ? ? ? * theorem 1.5.6 ) ) . for every @xmath97 and @xmath98 there is @xmath99 such that for all @xmath100 @xmath101 given a function @xmath18 slowly varying at infinity , by @xmath102 we denote a class of functions @xmath103 such that for all @xmath92 @xmath104 the collection @xmath102 is called _ de haan class at infinity _ determined by @xmath18 . similarly we define de haan class at the origin for @xmath105 . a positive function @xmath106 defined in a right neighbourhood of zero belongs to @xmath107 if @xmath108 in sections [ sec:4 ] and [ sec:5 ] , we consider a class of functions larger than @xmath94 . namely , we say that @xmath109 satisfies weak lower scaling condition at infinity if there are @xmath110 , @xmath111 and @xmath112 $ ] such that for all @xmath113 and @xmath114 @xmath115 then we write @xmath116 . similarly , @xmath106 satisfies the weak upper scaling condition at infinity if there are @xmath117 , @xmath111 and @xmath118 such that for all @xmath119 and @xmath114 @xmath120 then @xmath121 . finally , for a function @xmath122 its laplace transform is defined by @xmath123 whereas the laplace stieltjes transform @xmath124 we have @xmath125 . in this section we consider an isotropic unimodal lvy process @xmath126 on @xmath2 having the lvy khintchine exponent @xmath3 in @xmath17 for some function @xmath18 slowly varying at infinity . we describe an asymptotic behaviour of the transition density @xmath14 close to the origin when @xmath127 is small ( see theorem [ thm:2 ] ) . in the case when @xmath18 is bounded we also give the asymptotics at the origin as @xmath127 gets large ( see theorem [ thm:3 ] ) . we start by proving a slightly more general tauberian theorem . [ thm:1 ] let @xmath128 be a family of non - decreasing and non - negative functions on @xmath129 such that there are two families of positive functions @xmath130 and @xmath131 satisfying @xmath132 we assume that 1 . there are @xmath133 such that for all @xmath134 and @xmath135 @xmath136 2 . there are @xmath133 and @xmath137 such that for all @xmath138 and @xmath135 if @xmath139 and @xmath140 then @xmath141 3 . there is @xmath142 such that for all @xmath92 @xmath143 then @xmath144 for any pair @xmath145 we define a tempered distribution @xmath146 by setting for @xmath147 @xmath148 let us recall that the space @xmath149 consists of schwartz functions on @xmath150 restricted to @xmath129 and @xmath151 consists of tempered distributions which are supported by @xmath152 ( see @xcite for details ) . the upper bounds on @xmath153 imply @xmath154 by ( ii ) we have @xmath155 and @xmath156 hence , we may estimate @xmath157 i.e. the family @xmath158 is equicontinuous . next , for @xmath159 , we consider a function @xmath160 . we have @xmath161 therefore , by and ( iii ) , we obtain @xmath162 let @xmath163 be the linear span of the set @xmath164 . since @xmath163 is dense in @xmath149 and the family @xmath158 is equicontinuous on @xmath149 , from we obtain that for any @xmath147 , @xmath165 to conclude the proof of theorem we consider a specific function @xmath106 . for a given @xmath98 , let @xmath166 be such that @xmath167 and @xmath168 then @xmath169 } \phi_+(s / r ) { \ : \rm d } q_t(s ) \\ & \leq \frac{1}{r q_t\big(r^{-1}\big ) } \int_{[0 , \infty ] } \phi_+(s / r ) { \ : \rm d } q_t(s ) , \end{aligned}\ ] ] thus @xmath170 hence , @xmath171 similarly , taking @xmath172 such that @xmath173 and @xmath174 we can show that @xmath175 which ends the proof . essentially the same proof gives the following theorem . [ thm:8 ] let @xmath128 be a family of non - decreasing and non - negative functions on @xmath129 such that there are two families of positive functions @xmath130 and @xmath131 satisfying @xmath176 we assume that 1 . there are @xmath177 such that for all @xmath134 and @xmath178 @xmath136 2 . there are @xmath177 and @xmath179 such that for all @xmath138 if @xmath139 , @xmath140 and @xmath178 then @xmath180 3 . there is @xmath142 such that for all @xmath92 @xmath181 then @xmath182 to get the asymptotics of @xmath14 as @xmath127 approaches zero we are going to apply theorem [ thm:1 ] to @xmath183 with @xmath184 and @xmath185 . [ prop:1 ] if @xmath186 then there are @xmath187 such that @xmath188 moreover , @xmath189 let @xmath190 . first , we prove the following claim . [ clm:3 ] there are @xmath191 such that for all @xmath192 and @xmath135 , if @xmath193 then @xmath194 indeed , by and , there are @xmath195 and @xmath111 such that for all @xmath192 , if @xmath196 then @xmath197 and @xmath198 let @xmath193 . if @xmath199 then @xmath200 otherwise , @xmath201 and we can estimate @xmath202 since , by , we have @xmath203 the estimate follows . next , let us consider @xmath204 we observe that by the fubini tonelli s theorem @xmath205 since @xmath206 by the second application of the fubini tonelli s theorem we get @xmath207 where in the last step we used polar coordinates . we now claim that [ clm:1 ] for @xmath208 @xmath209 for the proof of the claim , by , we write @xmath210 by , the integrand in is uniformly bounded by an integrable function . since for a fixed @xmath211 , @xmath212 we obtain @xmath213 hence , by the dominated convergence theorem we obtain . notice , that by we also get @xmath214 provided that @xmath193 and @xmath135 . we now turn to the proof of . observe that @xmath215 . hence , @xmath216 by monotonicity of the laplace stieltjes transform of @xmath217 we infer that for any @xmath218 @xmath219 therefore , @xmath220 thus , by taking @xmath221 tending to @xmath222 we get . finally , since @xmath153 is non - decreasing , by , @xmath223 therefore , if @xmath224 , by , @xmath225 for @xmath226 , by we estimate @xmath227 hence , by and we get @xmath228 which finishes the proof . [ thm:2 ] let @xmath25 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath19 for some @xmath229 . then @xmath230 in the proof we use an argument from ( * ? ? ? * theorem 1.7.2 ) . for any @xmath231 , we have @xmath232 since the function @xmath233 is non - increasing , we get @xmath234 thanks to proposition [ prop:1 ] , we may use theorem [ thm:1 ] to get @xmath235 hence , by the first inequality in @xmath236 by taking @xmath237 , @xmath238 and letting @xmath239 to zero we obtain @xmath240 similarly , using the second inequality in , we show that @xmath241 [ thm:5 ] let @xmath25 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath3 and the lvy density @xmath24 . let @xmath242 . the following statements are equivalent : 1 . @xmath186 ; 2 . there is @xmath243 , @xmath244 3 . there is @xmath243 , @xmath245 the implication ( i ) @xmath246 ( ii ) follows by theorem [ thm:2 ] . next , ( ii ) @xmath246 ( iii ) is a consequence of the fact that @xmath247 vaguely on @xmath85 . to prove that ( iii ) implies ( i ) , first we consider the case @xmath248 . by , we can write @xmath249 thus , for any @xmath250 @xmath251 we observe that , by and the dominated convergence , @xmath252 hence , @xmath253 applying the second mean value theorem we obtain @xmath254 to deal with the second integral in , we use monotonicity of @xmath24 . namely , we have @xmath255 for some non - negative measure @xmath84 . hence , by the fubini theorem , @xmath256 finally , we get @xmath257 since @xmath258 was arbitrary , we conclude that @xmath259 which finishes the proof for @xmath248 . for @xmath260 , we consider the lvy measure @xmath261 corresponding to the one - dimensional projection of @xmath25 . we claim that @xmath262 indeed , using spherical coordinates we can write @xmath263 where @xmath264 is the surface measure of the unite sphere in @xmath265 . since @xmath266 it is enough to show that @xmath267 by the change of variables , we have @xmath268 since @xmath269 for a fixed @xmath211 we obtain @xmath270 therefore , by we can use the dominated convergence theorem to conclude that @xmath271 by the same line of reasoning as in the proofs of proposition [ prop:1 ] and theorems [ thm:1 ] , [ thm:2 ] and [ thm:5 ] one can show the corresponding results if the lvy - khintchine exponent belongs to de haan class at the origin . only one modification is needed to prove that ( iii ) implies ( i ) . namely , one should consider an asymptotically equal ( at the origin ) lvy - khintchine exponent corresponding to the lvy density @xmath272 ( compare with the proof of ( * ? ? ? * theorem 7 ) ) . [ thm:5inf ] let @xmath25 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath3 and the lvy density @xmath24 . let @xmath273 . the following statements are equivalent : 1 . @xmath274 ; 2 . there is @xmath243 , @xmath275 3 . there is @xmath243 , @xmath276 in the case of large @xmath277 , we again use @xmath153 given by , but this time with @xmath278 and @xmath279 . in this section we assume that @xmath18 is a _ bounded _ function slowly varying at infinity . we start by the following observation . [ lem:1 ] suppose @xmath280 , for some @xmath111 , is such that @xmath281 then for each @xmath190 there are @xmath282 such that for all @xmath283 , if @xmath284 then @xmath285 let @xmath286 and @xmath287 . first , we show , that for all @xmath288 @xmath289 fix @xmath114 . if @xmath290 $ ] then @xmath291 therefore , if @xmath292 for @xmath293 then @xmath294 next , we observe that for all @xmath284 @xmath295 therefore , by , there is @xmath195 such that for all @xmath51 @xmath296 which finishes the proof . [ prop:2 ] suppose @xmath19 for some bounded @xmath242 . then there are @xmath297 such that for all @xmath134 and @xmath178 @xmath298 in particular , @xmath299 the proof follows the same line as of proposition [ prop:1 ] . let @xmath300 . first , we show that for every @xmath137 there are @xmath301 such that for all @xmath192 and @xmath178 , if @xmath302 then @xmath303 observe that there are @xmath304 such that for all @xmath192 if @xmath302 then @xmath305 indeed , if @xmath306 then is a consequence of and . if @xmath307 then by we can estimate @xmath308 and since by @xmath309 we get . next , for @xmath186 we apply ( * ? ? ? * theorem 3.8.6(b ) ) to show that @xmath310 therefore , by lemma [ lem:1 ] , there are @xmath282 such that for all @xmath311 , @xmath312 thus by , whenever @xmath302 we obtain @xmath313 by taking @xmath178 for @xmath239 sufficiently small to satisfy @xmath314 , we conclude . now , let @xmath315 . by applying to the formula we get @xmath316 for all @xmath178 and @xmath193 . again , with the help of we can see that the integrand in is uniformly bounded by an integrable function . since for a fixed @xmath211 , @xmath317 by the dominated convergence theorem we obtain the following claim . for @xmath190 @xmath318 now , the formulas together with imply . finally , let us recall the estimate @xmath319 if @xmath224 , we may use to get @xmath320 for @xmath226 , by we estimate @xmath321 hence , by we obtain @xmath322 which finishes the proof . [ thm:3 ] let @xmath25 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath19 for some bounded @xmath242 . then @xmath323 the argument is analogous to the proof of theorem [ thm:2 ] . one needs to use proposition [ prop:2 ] and theorem [ thm:8 ] instead of proposition [ prop:1 ] and theorem [ thm:1 ] . in this section we study the potential measure @xmath27 associated to a _ transient _ isotropic unimodal lvy process @xmath25 , that is @xmath324 where @xmath325 is the standard measure @xmath326 and @xmath327 is a borel set . we set @xmath328 . we also use the same notation @xmath27 for the density of the part of the potential measure which is absolutely continuous with respect to the lebesgue measure . we have @xmath329 and we write @xmath330 . in the following theorem we provide the asymptotic behaviour of the potential measure @xmath27 . [ greenasymp ] let @xmath25 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath19 for some @xmath242 . then for all @xmath113 @xmath331 let @xmath332 hence , by the fubini tonelli s theorem @xmath333 by , the fubini tonelli s theorem and integration in polar coordinates give @xmath334 hence , @xmath335 next , by and , for any @xmath336 there are @xmath337 such that if @xmath338 then @xmath339 and @xmath340 for @xmath341 , we have @xmath342 since the process is transient the right - hand side is finite . because @xmath343 , by we get @xmath344 if @xmath345 , by and , we can estimate @xmath346 in particular , the integrand in restricted to @xmath347 is bounded by an integrable function . since @xmath348 , by @xmath349 thus by the dominated convergence theorem and we obtain @xmath350 now , by applying de haan tauberian theorem ( * ? ? ? * theorem 3.9.1 ) we conclude that @xmath351 [ cor:2 ] assume that @xmath186 for some @xmath229 . then @xmath352 we use theorem [ greenasymp ] and the line of reasons from theorem [ thm:2 ] . similarly one can prove the following proposition . [ cor:2i ] assume that @xmath274 for some @xmath353 . then @xmath354 in the rest of this section we study the asymptotic behaviour of the green function . the proofs below rely on theorem [ thm : equig ] , but for future reference , it is convenient to include them in this section . let us recall that for any integrable radial function @xmath355 @xmath356 where @xmath357 is the bessel function of the first kind defined for @xmath358 and @xmath134 by the complex integral @xmath359 for each @xmath92 , let us denote by @xmath65 a @xmath360-resolvent kernel , that is @xmath361 for @xmath362 . for any borel set @xmath327 we set @xmath363 [ thm:6 ] suppose @xmath364 . let @xmath25 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath3 . then @xmath365 if and only if @xmath366 , for some @xmath367 . in particular , implies that @xmath368 if @xmath366 , for some @xmath369 , then by ( * ? ? ? * corollary 4 ) we obtain . hence , it is enough to prove the converse . assume that holds . we are going to use a version of drasin shea jordan theorem proved in ( * ? ? ? * theorem 1 ) . first , let us observe that , by , @xmath27 is unbounded . next , for each @xmath51 , in view of theorem [ thm : equig ] the characteristic function of @xmath71 is integrable . hence , following the argument given in ( * ? ? ? * lemma iii.3 ) we obtain @xmath370 since @xmath371 we also have @xmath372 moreover , @xmath373 is integrable and its fourier transform equals to @xmath374 . thus , by , we can write @xmath375 if we set @xmath376 then for @xmath211 @xmath377 where for two functions @xmath378 , by @xmath379 we denote their mellin convolution , that is @xmath380 now , in view of , we have @xmath381 since @xmath373 is integrable and non - increasing the function @xmath382 is bounded , thus @xmath383 let @xmath384 be the mellin transform of @xmath385 , i.e. for @xmath386 @xmath387 whenever the integral converges . by the well - known asymptotics for the bessel functions of the first kind , there is @xmath195 such that for all @xmath211 @xmath388 hence , the integral is absolutely convergent on the strip @xmath389 . moreover , by ( * ? ? ? * , ( 1 ) ) , we can calculate @xmath390 finally , in view of theorem [ thm : equig ] and , the function @xmath106 has bounded decrease . for detailed study of functions with bounded decrease we refer to ( * ? ? ? * section 2.1 ) . next , we calculate @xmath391 therefore , by ( * ? ? ? * proposition 1 ) and ( * ? ? ? * section 5 ) , the hypothesis of ( * ? ? ? * theorem 1 and theorem 2 ) are satisfied and the theorem follows . [ thm:7 ] suppose @xmath364 . let @xmath25 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath3 . then @xmath392 if and only if @xmath393 for some @xmath394 . in particular , implies that @xmath395 if @xmath396 for some @xmath369 , then in view of ( * ? ? ? * corollary 3 ) we get . therefore , it is enough to prove the converse . again , we are going to use a version of drasin shea jordan theorem proved in ( * ? ? ? * theorem 2 ) . since for each @xmath92 , the @xmath360-resolvent kernel is integrable and its fourier transform equals @xmath397 , by , for @xmath211 we have @xmath398 by and , there are @xmath399 and @xmath195 such that for all @xmath400 , we can estimate @xmath401 which in view of is integrable on @xmath402 . next , for @xmath403 we have @xmath404 which is integrable on @xmath405 $ ] since @xmath27 is integrable on @xmath2 . hence , the integrand in has integrable majorant independent of @xmath360 . therefore , by the dominated convergence theorem , as @xmath360 approaches zero we obtain @xmath406 from this point on , the proof follows the same line as in theorem [ thm:6 ] and is omitted . in this section we study concentration functions @xmath407 defined @xmath211 by the formula @xmath408 where @xmath409 for @xmath410 and @xmath211 we set @xmath411 we write @xmath412 . there is a connection between @xmath413 , the concentration function @xmath414 and the lvy khintchine exponent @xmath3 , namely in @xcite , it was shown that for all @xmath211 @xmath415 and @xmath416 moreover , for all @xmath211 and @xmath119 we have @xmath417 indeed , @xmath418 next , we observe that for all @xmath211 and @xmath119 , @xmath419 from the other side , for all @xmath211 and @xmath119 , we have @xmath420 because @xmath24 is radially monotonic we can estimate @xmath421 the function @xmath422 controls the increments of the lvy khintchine exponent @xmath3 . indeed , we have the following lemma . [ k1 ] let @xmath290 $ ] . then for all @xmath362 , @xmath423 we start by observing that for all @xmath135 @xmath424 let @xmath84 be a measure supported on @xmath425 defined by @xmath426 hence , by and the integration by parts we obtain @xmath427 therefore , by , @xmath428 from the other side , by the change of variables and the fubini tonelli s theorem @xmath429 the function @xmath430 naturally appears in estimates of the lvy measure . indeed , there is @xmath195 , depending only on @xmath431 , such that for all @xmath362 @xmath432 however , for all @xmath211 @xmath433 in general @xmath430 is not comparable to @xmath422 . instead , we have the following lemma . [ lem:3 ] if @xmath260 then there is @xmath195 such that for all @xmath211 @xmath434 we have @xmath435 next , by taking @xmath436 we can write @xmath437 where in the last step we have used spherical coordinates in @xmath265 . by the change of variables , @xmath438 since there is @xmath195 such that for @xmath439 @xmath440 by the fubini tonelli s theorem @xmath441 [ cor:1 ] @xmath430 is bounded if and only if @xmath422 is bounded . if @xmath422 is bounded the conclusion about @xmath430 follows from the estimate . conversely , if @xmath430 is bounded then by , @xmath442 now , applying lemma [ lem:3 ] we can easily deduce that @xmath422 must be bounded . [ sp1 ] suppose that there are @xmath195 and @xmath443 such that for all @xmath119 and @xmath211 @xmath444 then @xmath445 for the proof , let us observe that , by , for @xmath446 @xmath447 thus @xmath448 which , by , is bounded by a constant multiply of @xmath449 . corollary [ sp1 ] can be applied for self - decomposable processes . indeed , by ( * ? ? ? * theorem 15:10 ) there is a non - increasing function @xmath450 such that @xmath451 . therefore , for @xmath119 and @xmath211 @xmath452 [ slow10 ] if @xmath453 then @xmath454 . let us notice that by and lemma [ k1 ] , for @xmath290 $ ] we have @xmath455 thus , in view of lemma [ lem:3 ] , we need to show that @xmath456 given @xmath98 , let @xmath457 be such that for @xmath458 @xmath459 by and the monotonicity of @xmath460 we have @xmath461 again , by monotonicity of @xmath460 we have @xmath462 thus @xmath463 [ nslow10 ] there are @xmath464 and @xmath465 such that @xmath466 if and only if @xmath467 in particular , implies that @xmath468 . if is satisfied , then there are @xmath469 such that for @xmath470 @xmath471 therefore , for @xmath470 @xmath472 which implies that the function @xmath473 is non - decreasing . hence , @xmath474 belongs to @xmath475 . finally , by , @xmath3 belongs to @xmath476 for some @xmath112 $ ] . conversely , if @xmath477 then @xmath474 belongs to @xmath478 for some @xmath479 $ ] . in particular , there is @xmath480 such that for all @xmath481 @xmath482 since @xmath483 is non - decreasing , by we have @xmath484 which finishes the proof . in this section we prove estimates of the transition density @xmath4 and the green potentials @xmath485 under the assumption that the process @xmath25 is isotropic and unimodal . the main result is theorem [ thm:4 ] . the special case , subordinate brownian motion , is considered in section [ sec:5.1 ] . for @xmath486 , by @xmath487 we denote the ball in @xmath2 of radius @xmath488 centred at the origin . the key tool in proving lower bounds is the following lemma proved in @xcite . [ hklb ] let @xmath25 be an isotopic unimodal lvy process on @xmath2 . then for all @xmath51 and @xmath362 @xmath489 where for a borel set @xmath258 we have set @xmath490 the second result essential for our argument is an improvement of ( * ? ? ? * theoreme 2.1 ) where one - dimensional symmetric lvy processes were considered . for a related estimates in the case of one - dimensional feller processes see ( * ? ? ? * lemma 7 ) . [ tau ] there exist @xmath491 such that for any isotropic lvy process @xmath492 for all @xmath493 . for @xmath494 we set @xmath495 and @xmath496 . by ( * ( 3.2 ) ) , there exists a constant @xmath195 depending only on the dimension @xmath431 such that for any @xmath283 @xmath497 in particular , by taking @xmath498 , in view of we obtain @xmath499 therefore , by the markov property , for @xmath500 , @xmath501 hence , for all @xmath500 @xmath502 given @xmath51 , let @xmath503 . then @xmath504 to prove the lower bound we consider , for @xmath505 ( to be specified later ) , @xmath506 observe that for any @xmath507 $ ] @xmath508 hence for @xmath509 if @xmath510 we have @xmath511 as well . this implies , for any @xmath512 , @xmath513 and @xmath514 . next , by the markov property we have , for @xmath515 , @xmath516 since @xmath71 is isotropic @xmath517 for any @xmath518 : @xmath519 . therefore @xmath520 again , by isotropicity of @xmath521 we have @xmath522 for any @xmath505 and by @xmath523 for @xmath524 . this ends the proof of the lower bound by . an immediate corollary to lemma [ hklb ] and proposition [ tau ] is the following proposition . [ hklb1 ] there exist constants @xmath525 such that for any isotropic unimodal lvy process @xmath25 and for all @xmath526 , @xmath527 and @xmath528 we have @xmath529 and @xmath530 the constants @xmath531 depends only on the dimension @xmath431 . since the constant in depends only on @xmath431 , the first estimate follows from lemma [ hklb ] and proposition [ tau ] . now , using we can estimate @xmath532 our method for obtaining upper bounds is based on the following elementary observation : there is a positive constant @xmath533 depending only on the dimension @xmath431 , such that for all @xmath493 @xmath534 as the first consequence , we obtain the following theorem . [ gub3 ] let @xmath25 be an isotropic unimodal lvy process on @xmath2 . then there is @xmath195 such that for all @xmath51 and @xmath535 @xmath536 the constant @xmath533 depends only on the dimension @xmath431 . let @xmath537 be an infinitesimal generator of @xmath25 , that is @xmath538now we consider a function @xmath539 such that @xmath540 and @xmath541 , for @xmath542 and @xmath543 for @xmath544 or @xmath545 . then for @xmath546 one can show using taylor expansion for @xmath547 , the representation of generator ( see ( * ? ? ? * theorem 31.5 ) ) and monotonicity of @xmath24 ( compare with ( * ? ? ? * the proof of lemma 3 ) ) that @xmath548 since @xmath549 we have @xmath550 for a positive constant @xmath533 depending only on the dimension @xmath431 . therefore , by the dynkin s formula we get @xmath551 hence , in view of we conclude the proof . [ lem : pt_gen1 ] suppose @xmath552 satisfies @xmath553 for all @xmath211 . then there is @xmath195 such that for all @xmath51 and @xmath535 @xmath554 the constant @xmath533 depends only on the dimension @xmath431 . monotonicity of @xmath555 together with imply @xmath556 taking @xmath557 in , we obtain @xmath558 hence , there is @xmath195 depending on @xmath431 such that for any @xmath559 @xmath560 since for any @xmath561 , @xmath562 , we obtain @xmath563 which completes the proof . to obtain a more precise upper bound we impose on the function @xmath564 a condition that there are @xmath243 , @xmath565 and @xmath566 such that for all @xmath567 @xmath568 then there is @xmath569 such that for all @xmath51 and @xmath570 , @xmath571 indeed , by lemma [ lem : pt_gen1 ] , @xmath572 the constant in depends on @xmath573 , @xmath574 , @xmath575 and the dimension @xmath431 . [ gub ] suppose that @xmath576 assume that there is @xmath577 satisfying @xmath578 for all @xmath211 such that there are @xmath195 , @xmath565 and @xmath566 with @xmath579 for all @xmath567 . then there are @xmath195 and @xmath580 such that for all @xmath40 and @xmath535 @xmath581 the constant @xmath533 depends on @xmath573 , @xmath574 , @xmath575 and the dimension @xmath431 . thanks to , we can apply lemma [ lem:1 ] , thus , there are @xmath582 and @xmath457 such that for all @xmath583 , @xmath584 which together with gives @xmath585 now , by , @xmath586 taking @xmath587 , the last integral is finite for @xmath40 . in view of lemma [ k1 ] the natural candidate for @xmath564 is the function @xmath422 . [ rem2 ] suppose that @xmath588 . if @xmath430 is a bounded function then there are @xmath195 and @xmath580 such that for all @xmath40 and @xmath535 @xmath589 the constant @xmath533 depends on the dimension @xmath431 . indeed , by and , the function @xmath422 belongs to @xmath590 . hence , by lemma [ k1 ] and corollary [ cor:1 ] the claim follows from proposition [ gub ] . for @xmath248 , we need to impose better scaling properties of @xmath422 to be able to apply proposition [ gub ] . the main theorem of this section is theorem [ thm:4 ] . [ thm:4 ] let @xmath25 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath19 for some bounded @xmath242 . then there are @xmath591 such that , for all @xmath40 and @xmath592 , @xmath593 the implicit constants , @xmath37 and @xmath38 depend on the dimension @xmath431 and the process @xmath25 . since @xmath19 , by theorem [ thm:5 ] there is @xmath57 such that @xmath594 for all @xmath39 . hence , by corollary [ hklb1 ] , the lower estimate for @xmath14 holds whenever @xmath595 . if @xmath596 and @xmath32 is sufficiently small , we apply theorem [ thm:3 ] . the upper estimate follows from proposition [ gub ] because by we have @xmath597 [ gbound ] suppose @xmath598 . let @xmath25 be an isotropic unimodal lvy process on @xmath2 . then there is @xmath195 such that for all @xmath535 @xmath599 the constant @xmath533 depends only on the dimension @xmath431 . integrating both sides in with respect to @xmath600 we get @xmath601 hence , by lemma [ k1 ] and , @xmath602 now , let us observe that by and monotonicity of @xmath460 , the function @xmath603 is non - decreasing , whereas by and monotonicity of @xmath604 , the function @xmath605 is non - increasing . therefore , @xmath606 and @xmath607 which concludes the proof of the theorem . [ cor:3 ] suppose @xmath608 . assume there are @xmath609 and @xmath610 such that for all @xmath119 and @xmath211 @xmath611 then @xmath612 by the first inequality in and lemma [ lem:3 ] , there is a constant @xmath195 such that @xmath613 , thus the upper estimate follows from theorem [ gbound ] . the second inequality implies that @xmath614 , thus the lower estimate is the consequence of corollary [ hklb1 ] . we conjecture that the upper estimate in corollary [ cor:3 ] is true for any unimodal isotropic process provided the dimension @xmath615 . a consequence of theorem [ gbound ] is theorem [ thm : equig ] , which generalizes ( * ? ? ? * theorem 5 ) to isotropic unimodal lvy processes , provided the dimension @xmath598 . [ thm : equig ] suppose @xmath598 . let @xmath25 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath3 . then there exist @xmath616 such that for all @xmath617 @xmath618 if and only if there are @xmath619 such that @xmath3 satisfies @xmath620 . if @xmath3 satisfies @xmath621 , for some @xmath622 , then the estimate for the green function @xmath27 follows by ( * ? ? ? * theorem 3 ) . conversely , if we assume then by theorem [ gbound ] there is @xmath243 such that for all @xmath617 @xmath623 then , by , there is @xmath243 such that for all @xmath617 @xmath624 hence , by proposition [ nslow10 ] applied to the one - dimensional projection of @xmath25 we obtain the claim . similarly one can prove the following result . [ thm : equig1 ] suppose @xmath598 . let @xmath25 be an isotropic unimodal lvy process on @xmath2 with the lvy khintchine exponent @xmath3 . then there exist @xmath616 such that for all @xmath625 @xmath626 if and only if there are @xmath619 such that @xmath627 satisfies @xmath620 . in this section we consider a pure - jump subordinate brownian motion @xmath1 with the lvy khintchine exponent @xmath628 where @xmath629 is a bernstein function such that @xmath630 . let @xmath631 be a brownian motion on @xmath2 . the process @xmath25 can be constructed as the time changed @xmath632 by an independent subordinator @xmath633 with the laplace exponent @xmath55 , that is for @xmath135 @xmath634 let us recall that there is a measure @xmath84 supported on @xmath129 such that for @xmath80 @xmath635 and @xmath636 the measure @xmath84 is the lvy measure of the subordinator . if @xmath84 is absolutely continuous then we will denote its density by @xmath84 as well . [ k_sbm ] for all @xmath211 @xmath637 the comparability constant depends only on the dimension @xmath431 . we observe that by taking derivative of and fubini tonelli s theorem @xmath638 since @xmath639 we get @xmath640 from the other side , for all @xmath211 @xmath641 thus by fubini tonelli s theorem @xmath642 finally , by @xmath643 which finishes the proof . there is a constant @xmath195 such that for all @xmath51 and @xmath535 @xmath644 the constant @xmath533 depends only on the dimension @xmath431 . this is a consequence of theorem [ gub3 ] and lemma [ k_sbm ] . [ ubound_p_t ] suppose that there exists @xmath645 such that @xmath646 , then there is @xmath195 , dependent on @xmath647 , @xmath575 and @xmath431 , such that for all @xmath51 and @xmath648 @xmath649 by the scaling property , for all @xmath650 , @xmath651 hence , @xmath652 if the lvy measure of the subordinator has a non - increasing density and @xmath653 then there is @xmath195 such that for all @xmath51 and @xmath362 @xmath654 indeed , let us observe that @xmath655 thus @xmath656 is non - decreasing in particular , @xmath63 belongs to @xmath657 . hence , is the consequence of proposition [ ubound_p_t ] . [ r^2k(r ) ] suppose that there exists @xmath645 such that @xmath646 , then there are @xmath658 , depending only on @xmath431 , such that for all @xmath659 @xmath660 observe that @xmath661 the lower bound follows from the monotonicity of @xmath63 , while the upper bound was already proved in lemma [ ubound_p_t ] . now , the conclusion follows from lemma [ k_sbm ] . [ prop : glambda_up ] let @xmath92 . then there is @xmath195 such that for all @xmath535 @xmath662 if the process @xmath25 is transient , the estimate is also valid for @xmath663 . since @xmath664 is non - decreasing , by the mean value theorem , for all @xmath665 @xmath666 multiplying both sides of by @xmath667 , integrating with respect to @xmath668 and finally applying the estimate , we obtain @xmath669 since the function @xmath670 is non - increasing , we can estimate @xmath671 hence , @xmath672 since @xmath673 , and the process @xmath25 is transient , the same argument proves for @xmath663 . let @xmath674 . suppose that @xmath55 is a special bernstein function , that is a bernstein function such that @xmath675 is again a bernstein function . then there is @xmath195 such that for all @xmath570 @xmath676 for the proof , we notice that @xmath677 is increasing because @xmath55 is a special bernstein function ( see ( * ? ? ? * lemma 4.1 ) ) . therefore the claim follows by proposition [ prop : glambda_up ] . let us comment , that the estimate has already been proved in @xcite , however the method used was different . [ g_lambda ] let @xmath25 be a subordinate brownian motion on @xmath2 with the lvy khintchine exponent @xmath678 . let @xmath679 . suppose that there exists @xmath645 such that @xmath680 , then there is @xmath681 , dependent on @xmath682 and @xmath647 such that for all @xmath683 @xmath684 if the process is transient and there exists @xmath645 such that @xmath685 then holds for @xmath686 . if additionally @xmath687 , for some @xmath369 , then there is @xmath688 such that for all @xmath683 @xmath689 for @xmath690 we set @xmath691 let @xmath692 . since @xmath693 , @xmath694 if @xmath695 , the proof of the upper estimate follows by proposition [ prop : glambda_up ] . suppose @xmath696 . since @xmath697 , we have @xmath698 because @xmath699 belongs to @xmath700 , we can estimate @xmath701 now , applying , we obtain @xmath702 from the other side , by and , we arrive at @xmath703 combining both estimates of @xmath704 we obtain @xmath705 hence , by proposition [ prop : glambda_up ] , for @xmath683 @xmath706 where @xmath707 in the transient case @xmath708 is finite , thus the estimate is also valid in the case @xmath709 . next , we additionally assume that @xmath699 belongs to @xmath710 , for some @xmath394 . let @xmath711 . then @xmath712 let us observe that if @xmath713 then analogous calculation to shows that @xmath714 for any @xmath715 . since @xmath65 is non - increasing we have @xmath716 hence @xmath717 by the mean value theorem and monotonicity of @xmath55 and @xmath63 , @xmath718 since @xmath699 belongs to @xmath710 , by , for @xmath719 we get @xmath720 hence , we obtain @xmath721 now , setting @xmath722 we obtain @xmath723 finally , for @xmath724 , by monotonicity of @xmath55 and @xmath63 we have @xmath725 to get the lower estimate for @xmath726 we may use proposition [ hklb1 ] and the fact that the density of the lvy measure is everywhere positive . a version of theorem [ g_lambda ] for @xmath709 was proved in ( * ? ? ? * proposition 4.5 ) where in addition to scaling properties of @xmath63 ( slightly different than ours ) it was also assumed that the potential density a of the subordinator was decreasing on @xmath425 . [ p_t_bound ] let @xmath25 be a subordinate brownian motion on @xmath2 with the symbol @xmath727 . suppose that there exists @xmath728 and @xmath394 such that @xmath729 , then for all @xmath51 and and @xmath362 , if @xmath730 and @xmath648 @xmath731 in view of proposition [ ubound_p_t ] , it is enough to show the lower estimate . first , we find a lower bound of the lvy measure . let @xmath732 and @xmath733 , then by monotonicity of the density of the lvy measure @xmath734 next , we find @xmath221 , small enough , such that @xmath735 . indeed , by lemma [ r^2k(r ) ] , @xmath736 , hence the application of @xmath737 provides the desired @xmath221 . therefore @xmath738 this implies that @xmath739 and the conclusion is a consequence of proposition [ hklb1 ] . [ nuapprox ] let @xmath740 . suppose the density of the lvy measure @xmath84 of the subordinator is non - increasing . then there is @xmath243 such that for all @xmath535 @xmath741 first , we observe that for @xmath51 @xmath742 since @xmath743 and @xmath84 is non - increasing , the estimate follows by ( * ? ? ? * lemma 5 ) applied to @xmath744 and @xmath745 . by , we have @xmath746 by and the monotonicity of @xmath747 , @xmath748 similarly , by and monotonicity of @xmath84 @xmath749 which ends the proof . [ thm:9 ] let @xmath25 be a subordinate brownian motion on @xmath2 with the lvy khintchine exponent @xmath678 . suppose that the density of the lvy measure @xmath84 of the subordinator is non - increasing and @xmath615 . then the following are equivalent : 1 . [ thm:9:1 ] there exist @xmath750 such that @xmath751 . [ thm:9:2 ] there are @xmath752 and @xmath753 such that for all @xmath51 and @xmath362 , if @xmath59 and @xmath754 then @xmath755 3 . [ thm:9:3 ] there are @xmath195 and @xmath753 such that if @xmath754 then @xmath756 since @xmath656 is non - decreasing , @xmath757 . therefore , ( i ) @xmath246 ( ii ) is the consequence of theorem [ p_t_bound ] . the implication ( ii ) @xmath246 ( iii ) follows because @xmath758 vaguely on @xmath85 , see e.g. the proof of ( * ? ? ? * theorem 6 ) . it remains to prove that ( iii ) implies ( i ) . first , let us observe that by proposition [ nuapprox ] there is a constant @xmath243 such that for @xmath211 @xmath759 therefore , for @xmath760 @xmath761 hence , @xmath762 is non - increasing for @xmath763 , thus @xmath764 and the proof is completed . let @xmath765 then , for @xmath766 , @xmath767 thus @xmath768 . moreover , for all @xmath769 and @xmath770 @xmath771 hence , by theorem [ thm:4 ] there are @xmath580 and @xmath772 such that for all @xmath40 and @xmath773 @xmath774 by theorem [ gub3 ] and proposition [ hklb1 ] , we have @xmath775 . [ exmp:2 ] let @xmath776 be lvy process with the lvy khintchine exponent @xmath777 for @xmath778 $ ] and @xmath633 a subordinator with the laplace exponent @xmath779 for @xmath780 $ ] . then @xmath781 where @xmath782 has the lvy khintchine exponent @xmath783 thus @xmath19 for @xmath784 by theorem [ thm:4 ] , there are @xmath580 and @xmath57 such that for all @xmath40 and @xmath39 @xmath785 let @xmath25 be a lvy process with the lvy khintchine exponent @xmath789 let @xmath790 be the heat kernel for the standard isotropic @xmath574-stable lvy process . then we have @xmath791 recall that @xmath792 first , let us note that @xmath793 and @xmath794 . hence , by ( * theorem 1 and remark 2 ) , theorem [ gub3 ] and proposition [ hklb1 ] , there is @xmath795 such that for all @xmath796 and @xmath362 @xmath797 let @xmath798 be two incomplete gamma functions . then @xmath799 for @xmath800 , we have @xmath801 moreover , for @xmath802 @xmath803 whereas , for @xmath804 we have @xmath805 hence , for @xmath806 , @xmath807 . \end{cases}\ ] ] let us notice , that if @xmath808 then @xmath809 and @xmath810 . \end{cases}\ ] ] let @xmath25 be the lvy process with the lvy khintchine exponent @xmath813 by @xmath814 we denote the transition density of the geometric @xmath574-stable process and by @xmath790 we denote the transition density of the standard isotropic @xmath574-stable process . if @xmath633 is an independent of @xmath815-stable subordinator , then @xmath816 for @xmath817 and @xmath526 , by example 3 , we have @xmath818 , hence @xmath14 is comparable to @xmath819-stable transition density , that is for all @xmath51 and @xmath820 @xmath821 let @xmath822 be chosen in example 3 . in fact we may take @xmath823 . for @xmath824 we write @xmath825 by example 3 , we have @xmath826 again by example 3 , we get @xmath827 and @xmath828 observe that for @xmath829 , we have @xmath830 and @xmath831 in view of lemma [ k_estimate ] , for @xmath832 and @xmath824 , we have @xmath833 if @xmath834 then @xmath835 , for some @xmath836 hence @xmath837 combining all the estimates , we conclude that for @xmath838 and @xmath839 we obtain @xmath840 if @xmath841 and @xmath839 then @xmath842 and finally , if @xmath843 and @xmath51 we have @xmath844 let us observe that for @xmath845 we have @xmath846 for all @xmath362 , and @xmath847 for all @xmath51 .

detailed chemical abundances are presented for seven m31 outer halo globular clusters ( with projected distances from m31 greater than 30 kpc ) , as derived from high resolution integrated light spectra taken with the hobby eberly telescope . five of these clusters were recently discovered in the pan - andromeda archaeological survey ( pandas)this paper presents the first determinations of integrated fe , na , mg , ca , ti , ni , ba , and eu abundances for these clusters . four of the target clusters ( pa06 , pa53 , pa54 , and pa56 ) are metal - poor (~<~-1.5 $ ] ) ,-enhanced ( though they are possibly less-enhanced than milky way stars at the level ) , and show signs of star - to - star na and mg variations . the other three globular clusters ( h10 , h23 , and pa17 ) are more metal rich , with metallicities ranging from = -1.4 $ ] to . while h23 is chemically similar to milky way field stars , milky way globular clusters , and other m31 clusters , h10 and pa17 have moderately low [ ca / fe ] , compared to milky way field stars and clusters . additionally , pa17 s high [ mg / ca ] and [ ba / eu ] ratios are distinct from milky way stars , and are in better agreement with the stars and clusters in the large magellanic cloud ( lmc ) . none of the clusters studied here can be conclusively linked to any of the identified streams from pandas ; however , based on their locations , kinematics , metallicities , and detailed abundances , the most metal - rich pandas clusters h23 and pa17 may be associated with the progenitor of the giant stellar stream , h10 may be associated with the sw cloud , and pa53 and pa56 may be associated with the eastern cloud . [ firstpage ] galaxies : individual(m31 ) galaxies : abundances galaxies : star clusters : general globular clusters : general galaxies : evolution

1501.04626

the andromeda galaxy ( m31 ) is the nearest large neighbour to the milky way ( mw ) . its relative proximity offers a unique opportunity to study the stellar populations of a nearby , large galaxy without being hindered by interstellar extinction . though the stars in m31 are considerably fainter than mw field stars , they can still be resolved for photometric and low resolution spectroscopic observations while entire globular clusters ( gcs ) can be targeted for integrated light ( il ) spectroscopy . such observations contribute to studies of the formation and evolution of m31 and its satellite system . deep surveys of m31 s outer regions , such as the isaac newton telescope ( int ) survey @xcite and the pan - andromeda archaeological survey ( pandas ; @xcite ) have demonstrated that there is a significant amount of coherent substructure in the outer halo ; these coherent streams indicate that m31 s halo is currently being assembled by accretion of dwarf satellites . @xcite present maps of the outer halo , demonstrating that there are a multitude of streams , plumes , and clouds of metal - poor stars , along with a significant metal - rich population that is mostly located in the giant stellar stream ( gss ) south of m31 . as much as 42% of the most metal - poor stars ( @xmath3 < -1.7 $ ] ) lie in coherent substructure for metal - rich stars ( @xmath3 > -0.6 $ ] ) this percentage rises to 86% . the rest of the outer halo stars are distributed in a `` smooth '' component . the presence of metal - rich and metal - poor coherent streams implies that m31 is currently accreting multiple dwarf satellites ( see , e.g. , @xcite for accretion simulations that lead to coherent streams ) . the locations and metallicities of the outer halo field stars can be used to infer the nature of these dwarf galaxies . 1 . the metal - rich gss and the surrounding metal - rich features indicate that m31 has accreted at least one fairly massive ( @xmath5 lmc - mass ) galaxy @xcite . star formation histories indicate that most of the metal rich stars formed more than 5 gyr ago , hinting that the progenitor may have been an early - type dwarf @xcite . 2 . intermediate [ fe / h ] stars are also found in coherent substructures , notably the sw cloud @xcite , which has its own gcs ( pa7 , pa8 , and pa14 ; @xcite ) and possibly its own h@xmath6roman1 gas @xcite . estimates based on metallicity and total brightness suggest that the sw cloud progenitor was also a fairly massive dwarf galaxy ( with a mass comparable to the fornax dwarf spheroidal ; @xcite ) . 3 . the most metal - poor streams are gas - free ( based on 21 cm observations ; @xcite ) and likely originated in lower mass dwarf spheroidal systems . the outer halo gcs also suggest that m31 has had a fairly active accretion history . of the @xmath7 gcs discovered in pandas , many appear to lie along stellar streams . @xcite demonstrated that the positions of these gcs are correlated with the positions of the streams , with a @xmath8% chance that these gcs are located along the streams by chance . pa7 and pa8 are kinematically associated with each other @xcite and with the sw cloud itself @xcite . @xcite also show that many of the gcs which appear to lie along streams could be associated , based on their kinematics . @xcite have also demonstrated that at least one outer halo gc ( g002 ) has likely been accreted from a dwarf satellite . taken as a whole , the entire outer halo gc system is also different from the mw s gc population . for instance , @xcite show that m31 has more luminous and faint gcs than the mw they suggest that m31 could have acquired these extra gcs through accretion . @xcite also demonstrate that the entire gc system has clear signs of rotation in both the gcs spatially associated with streams and those that are unassociated with streams . this rotation suggests either 1 ) that m31 experienced a merger with a galaxy large enough to bring in a substantial fraction of the outer halo gcs , or 2 ) that the parent dwarf galaxies were accreted from a preferred direction such that their angular momenta were correlated , as is presently seen for a separate plane of dwarf satellites @xcite . ultimately , the observations of outer halo stars and gcs indicate that m31 has recently experienced a merger with at least one massive dwarf and multiple lower mass dwarfs . detailed chemical abundances ( from high resolution spectroscopy ) are an effective way to isolate and identify coeval groups , a process known as `` chemical tagging '' ( e.g. , @xcite ) . with abundances of @xmath1 , iron - peak , and neutron capture elements , metal rich ( @xmath3 \ga -1.5 $ ] ) stars and gcs from dwarf galaxies can be distinguished from those in massive galaxies ( e.g. @xcite ) . unlike gcs associated with the mw and its dwarf satellites , however , distant extragalactic targets must be observed through their integrated light ( il ) . while metallicity and @xmath1-abundances can be determined from lower resolution il spectra ( e.g. @xcite ) , the detailed chemical abundances necessary for chemical tagging analyses require high resolution il spectroscopy ( see , e.g. , @xcite ; @xcite ) . as in the mw ( @xcite ) , if the m31 outer halo gcs originated in lower mass dwarf galaxies then they could have distinct abundance patterns from mw field stars , depending on metallicity . high resolution il spectroscopy has been tested extensively on galactic gcs ( @xcite ; @xcite ; @xcite ) , demonstrating that 1 . il abundances trace the individual stellar abundances and represent cluster averages when abundances do not vary between stars 2 . abundance determinations from high resolution il spectroscopy are more precise than from lower resolution studies 3 . certain abundance ratios are very stable to uncertainties in the underlying stellar populations ( e.g. interloping field stars , uncertainties in cluster age , microturbulence relations , etc . ) , such as [ ca / fe ] 4 . partially resolved photometry down to the horizontal branch ( hb ) reduces systematic uncertainties by constraining [ fe / h ] and hb morphology . thus , a high resolution il spectroscopic analysis of m31 gcs can provide precise and accurate abundances that will be suitable for chemical tagging . high resolution il spectroscopic techniques have been applied to nearby gcs in the local group ( in m31 , @xcite ; the lmc , @xcite and @xcite ; and in dwarf galaxies , @xcite ) and in the large elliptical galaxy ngc 5128 @xcite . at the distance of m31 , the size and brightness of its gcs make them ideal targets for high resolution il spectroscopy . this paper presents a high resolution il chemical tagging analysis of seven outer halo m31 gcs , five of which were first observed in pandas . this is the first chemical analysis for these five pandas clusters , at any resolution . section [ sec : data ] presents the data and the analysis methods , while section [ sec : modelatms ] describes the process for generating synthetic stellar populations for each cluster . the il abundances are given in section [ sec : abunds ] , and the implications for the formation of m31 s outer halo are then discussed in section [ sec : discussion ] . finally , the results are summarized in section [ sec : conclusions ] . the primary goal of this high resolution spectroscopic investigation of m31 gcs is to study the nature of the stars and clusters in m31 s outer halo . the targets were therefore restricted to gcs with projected distances from the centre of m31 that are @xmath9 kpc ; priority was given to the gcs with the largest projected radii . observational constraints required that the targets were sufficiently bright to observe in a reasonable amount of time and to avoid stochastic sampling issues as a result , the target gcs are on the bright end of the m31 outer halo gc distribution ( see @xcite ) . seven outer halo gcs were targeted ; their properties are summarized in table [ table : pandastargets ] . two of these target clusters ( h10 and h23 ) were discovered by @xcite ; the other five were discovered in the pandas programme @xcite . the locations of these gcs on a density map of metal - poor outer halo stars ( from m. irwin , _ private communication _ ) is shown in figure [ fig : pandasclusterlocations ] . note that none of the targets fall precisely onto any of the major stellar streams or overdensities in the outer halo . nonetheless , pa56 sits on a faint north - western extension of the eastern cloud , and is likely associated with this feature ( mackey et al . 2015 , _ in prep . _ ) . while h23 has been tentatively linked to stream d based on its relative proximity in projection @xcite , its radial velocity renders a genuine association unlikely @xcite . [ 1]d .. #1 @lccd1d1ccc@ cluster & ra ( j2000 ) & dec ( j2000 ) & & & photometric@xmath10 & spatial association & references + & & & & & [ fe / h ] & with stream ? & + h10 & @xmath11 & @xmath12 & 15.7 & 78.5 & -1.84 & n & 1 , 2 + h23 & @xmath13 & @xmath14 & 16.8 & 37.1 & -1.54 & stream d ? & 1 , 2 + pa06 & @xmath15 & @xmath16 & 16.5 & 93.7 & mp & n & 2 , 3 , 4 + pa17 & @xmath17 & @xmath18 & 16.3 & 53.9 & ? @xmath19 & n & 2 , 3 + pa53 & @xmath20 & @xmath21 & 15.4 & 95.9 & mp & n & 2 , 3 , 4 + pa54 & @xmath22 & @xmath23 & 15.9 & 95.8 & mp & n & 2 , 3 , 4 + pa56 & @xmath24 & @xmath25 & 16.8 & 103.3 & mp & n & 2 , 3 , 4 + & & & & & & + & & & & & & & + + 1 = @xcite , 2 = @xcite , 3 = @xcite , 4 = mackey et al . , _ in prep._. + @xmath26 projected distances are from the centre of m31 . + @xmath10 photometric metallicity estimates are used to constrain the parameters of the input isochrones ; see section [ sec : modelatms ] . [ fe / h ] estimates are based on galactic gc fiducial fits . mp indicates a metal - poor cluster ( @xmath3 \la -1.5 $ ] ) ; fiducial fits will be given in mackey et al . , _ in prep._. + @xmath19 pa17 does not have an _ hst _ cmd , and there is therefore no _ a priori _ information about its metallicity or hb morphology . + @xcite present detailed tests of systematic abundance errors that occur because of uncertainties in the underlying stellar populations . these tests include : * uncertainties in isochrone parameters such as age and [ fe / h ] * properties of evolved horizontal branch ( hb ) and asymptotic giant branch ( agb ) stars , particularly the hb morphology * assumptions about the stellar populations ( e.g. microturbulence relations , initial mass functions , mass segregation ) * unusual stars like interloping field stars , long period variables , etc . these errors can be mitigated if gc properties are constrained with photometry or lower resolution analyses over a broader wavelength range . observing priority was therefore given to those targets with partially resolved photometry . six of the seven target gcs have _ hubble space telescope _ ( _ hst _ ) photometry down to the hb ( @xcite , mackey et al . 2015 , _ in prep . this photometry is used to place constraints on the best fitting isochrones ( see section [ sec : modelatms ] ) . the targets were observed with the hobby - eberly telescope ( het ; @xcite ) at mcdonald observatory in fort davis , tx in 2011 and early 2012 . the high resolution spectrograph ( hrs ; @xcite ) was utilized with the 3@xmath27 fibre and a slit width of 1@xmath27 , yielding an instrumental spectral resolution of @xmath28 . with the 600 gr / mm cross disperser set to a central wavelength of 6302.9 , wavelength coverages of @xmath29 and @xmath30 were achieved in the blue and the red , respectively . the 3@xmath27 fibre provided coverage of the clusters past their half - light radii ; the additional sky fibres ( located 10@xmath27 from the central object fibre ) provided simultaneous observations for sky subtraction . exposure times were calculated to obtain a total s / n = 80 ( per resolution element ) , although not all targets received sufficient time to meet this goal . the details of the observations are shown in table [ table : pandasobservations ] . [ 1]d,#1 @llccccd3d3d3@ cluster & observation & exposure & s / n@xmath26 & s / n@xmath26 & @xmath31 & & & + & dates & time ( s ) & ( 5500 ) & ( 7000 ) & ( @xmath32 ) & & & + h10 & 2011 jan 2 , 10 , 11 , 22 , 23 , 28 , 30 & 19180 & 82 & 140 & 2.3 & -351.9,1.5 & -352,9 & 6.6,0.4 + & & & & & & & & + h23 & 2011 jul 5 , 7 , 10 , 11 , aug 2 , 4 , sep 23 & 16050 & 65 & 82 & 4.1 & -373.3,0.1 & -377,11 & 6.2,0.4 + & & & & & & & & + pa06 & 2011 sep 29 , oct 1 , 2 , 4 , 6 , 17 , 18 , 19 , 21 , 23 & 29278 & 65 & 105 & 2.6 & -341.4,0.7 & -327,15 & 5.6,0.4 + & & & & & & & & + pa17 & 2012 jan 16 , 18 , 21 & 8100 & 41 & 50 & 2.6 & -260.0,1.0 & -279,15 & 6.1,0.5 + & & & & & & & & + pa53 & 2011 aug 1 , sep 20 , 26 & 8100 & 115 & 148 & 2.7 & -270.8,0.9 & -253,10 & 12.0,0.4 + & & & & & & & & + pa54 & 2011 aug 25 , sep 24 , 25 , 27 & 10800 & 82 & 130 & 2.2 & -344.9,0.8 & -336,8 & 7.5,0.4 + & & & & & & & & + pa56 & 2011 oct 7 , 17 , 18 , 20 , 21 , 23 , 24 , 30 , & 30623 & 65 & 82 & 2.4 & -241.4,1.7 & -239,8 & 6.4,0.4 + & nov 19 , dec 30 , & & & & & & & + & 2012 feb 10 , 11 & & & & & & & + & & & & & & & & + & & & & & & & & + + literature radial velocities are from @xcite + @xmath26s / n ratios are per resolution element , and assume that there are 2.7 pixels per resolution element for hrs . + @xmath10the coverage radii are based on the half - light radii in @xcite and @xcite all gcs are covered past their half - light radii . + data reduction was performed in in the image reduction and analysis facility program ( iraf ) . as in @xcite , bias removal was not performed to avoid adding noise to the individual spectra , and variance weighting was used during aperture extraction to remove cosmic rays . sky spectra ( from the separate sky fibres ) were replaced with continuum fits with the emission lines added back in ( according to the sky line identifications from the uves quality control sky spectrum website ) and were then subtracted from the object spectra . telluric standards were also observed for removal of atmospheric absorption features . in order to avoid removing broad , blended features , the target spectral orders were normalized with continuum fits to an extremely metal - poor star ( cs29502 - 092 ; see @xcite ) . low - order polynomial fits were then necessary to fully normalize the target spectra . the individual observations were cross - correlated with a high resolution , high s / n arcturus spectrum ( from the arcturus atlas ; @xcite ) to determine radial velocities . heliocentric velocities were determined for each individual observation ; the final , averaged heliocentric velocities are shown in table [ table : pandasobservations ] . the quoted errors represent the dispersion between observations . the individual , rest - frame spectra were combined with average sigma - clipping rejection routines to remove any remaining cosmic rays ; each spectrum was weighted by flux during the combination . the combined spectra were once again cross - correlated with the arcturus template spectrum to determine cluster dispersions ( see @xcite for a description of how this is done ) . these velocity dispersions are also shown in table [ table : pandasobservations ] . examples of the final , combined spectra are shown in figure [ fig : pandasgcspectra ] . for these chemical tagging analyses , spectral lines of fe , na , mg , ca , ti , ni , ba , and eu are utilized . @xcite demonstrated that for galactic gcs reliable fe , ca , ti , ni , and ba abundances can be determined from equivalent widths ( ews ) , while na , mg , and eu abundances can be determined with spectrum syntheses . note that in this analysis the ba lines are synthesized instead of using ews , due to the lower s / n ratios of the target spectra . ews for fe , ca , ti , and ni lines were measured with the program daospec @xcite . @xcite verified that daospec measurements of il spectral lines compare well with measurements from other methods for gcs with a range of velocity dispersions . given the low s / n of some of the pandas targets , many of the daospec ews were verified or refined by hand . il spectral lines have contributions from stars with a variety of line strengths in particular , il lines with ews @xmath33 m can have contributions from bright red giant stars with ews @xmath34 m . these strong lines are difficult to model ( see the discussions in paper i and @xcite ) and are typically removed from individual stellar analyses . for this reason , as a conservative estimate , all lines stronger than @xmath35 m were removed from this analysis unless absolutely necessary . table [ table : linelist ] shows the spectral lines that were used to derive abundances in the analysis , their atomic data , and the measured il ews for each cluster . the line list is based on the standard il lines lists of @xcite and @xcite , with supplements from the red giant branch ( rgb ) stellar line lists of @xcite and @xcite . lines that were measured with spectrum syntheses are indicated . the full line lists to synthesize all lines in a 10 window ( for spectrum syntheses ) are the ones used in @xcite : they consist of the ew line list , with supplements from the vienna atomic line database ( vald ; @xcite ) , the kurucz database , and the national institute of standards and technology ( nist ) database . molecular features from the kurucz database were included when indicated in the arcturus atlas ( for ch , cn , and mgh ) . isotopic and hyperfine structure ( hfs ) components for the ba@xmath6roman2 lines are from @xcite while the eu@xmath6roman2 components are from @xcite and @xcite . the hfs components are included in the syntheses ; while they do not significantly affect the strengths of the lines , they can affect the line profiles . [ cols="<,^,^,^,^,^,^,^,^,^,^ " , ] + total errors are conservatively estimated by adding the individual errors from @xcite in quadrature , based on the appropriate [ fe / h ] and hb morphology . + the [ fe / h ] abundances were determined with ews , as described in section [ subsec : daospec ] . because 1 ) there are fewer fe@xmath6roman2 lines and 2 ) fe@xmath6roman2 is more susceptible to systematic effects @xcite , [ fe@xmath6roman1/h ] is adopted as the representative [ fe / h ] , despite potential non - lte ( nlte ) effects . note that for all clusters fe@xmath6roman1 and fe@xmath6roman2 are in reasonably good agreement . three clusters , h10 , h23 , and pa17 , are more metal - rich than @xmath3~=~-1.5 $ ] , while the other four are metal - poor . the most metal - poor clusters all have fewer detectable lines ( @xmath36 fe@xmath6roman1 lines and 1 - 2 fe@xmath6roman2 lines ) than the more metal - rich gcs , leading to larger random errors in [ fe / h ] . most gc systems in dwarf galaxies tend to be metal - poor , with more massive dwarf galaxies possessing more metal - rich gcs ( e.g. @xcite ) . the comparatively high metallicities of h10 , h23 , and pa17 therefore suggest that they formed in a galaxy more massive than the fornax ( for ) dwarf spheroidal . it is clear from tables [ table : pandastargets ] and [ table : abunds ] that the metal - rich pandas gcs have smaller projected distances from the centre of m31 than the more metal - poor gcs . based on a low resolution survey of inner halo , bulge , and disk m31 gcs , @xcite found evidence for an abundance gradient in the inner regions of m31this gradient seems to flatten to @xmath3 \sim -1.8 $ ] at distances @xmath37 kpc . h23 and pa17 are therefore more metal - rich than expected given their large distance from the centre of m31 ; however , metal - rich field stars _ have _ been identified in streams in m31 s outer halo @xcite . the @xmath1-elements ( mg , ca , and ti ) form primarily through captures of @xmath38he nuclei during hydrostatic burning in massive stars , while fe forms both during core collapse supernovae of massive stars _ and _ during the detonation of white dwarfs whose progenitors were lower mass stars . comparisons between the abundances of @xmath1-elements and fe vs. [ fe / h ] provide an indication of how the chemical contributions from different types of stars have changed in a specific environment . because the trends in [ @xmath1/fe ] with [ fe / h ] differ between massive and dwarf galaxies , the abundance ratios can be used to link stars and/or gcs to their birth environments . magnesium abundances for the pandas gcs are presented in table [ table : abunds ] . for most clusters [ mg / fe ] is determined from the 5528 and 5711 lines however , pa17 s [ mg / fe ] is only determined from the 5711 line because its 5528 line is too strong ; similarly , the 5711 line is too weak in the most metal - poor cluster , pa06 , and in the gc with the highest velocity dispersion , pa53 . comparisons with mw , m31 , and dwarf galaxy field stars and clusters are shown in figure [ fig : pandasmg ] . field stars in the most massive , nearby , well - studied dwarfs that host gcs are also shown , including the lmc and the sgr and for dwarf spheroidals ; lower mass galaxies such as sculptor and carina are not included since they do not have gcs . the average and/or integrated abundances of mw , m31 , lmc , sgr , and for gcs are shown separately from the field stars . the unusual mw gcs palomar 1 ( pal 1 ; @xcite ) and ruprecht 106 ( rup 106 ; @xcite ) and the m31 gc g002 @xcite are given different symbols , since their chemical abundances indicate that they may have originated in dwarf galaxies ( though their host galaxies have not yet been identified ) . note that star - to - star mg variations _ have _ been observed in massive , metal - poor galactic clusters ( @xmath3 \la -1.2 $ ] ; @xcite ) ; in those clusters , mg variations can be as large as @xmath39 dex ( e.g. @xcite , @xcite ) . the most metal - poor pandas gc , pa06 , has a low [ mg / fe ] ratio for its metallicity , similar to m15 ( see @xcite ) ; this suggests that strong star - to - star mg variations may exist in pa06 . the integrated [ mg / fe ] ratios in the other metal - poor gcs roughly agree with the galactic gcs . the integrated mg abundances of the @xcite m31 gcs are systematically lower than the pandas gcs at the metal poor end however , this may be due to the masses of the observed clusters . many of the colucci et al . targets have larger velocity dispersions and higher total luminosities , indicating that they are more massive than these pandas gcs . if this is the case , the gcs in the colucci et al . sample may harbour stronger mg variations ( e.g. @xcite ) . @xcite _ do _ find a correlation between integrated [ mg / fe ] and total @xmath40 for the metal poor gcs , such that the more massive gcs have lower integrated [ mg / fe]all five of the most metal poor ( @xmath3 < -1.2 $ ] ) pandas gcs are consistent with the colucci et al . relation for all gcs . because of their high metallicities the two most metal - rich pandas gcs , h23 and pa17 , are not likely to host significant star - to - star variations in mg , though they may host multiple populations ( see section [ subsubsec : namg ] ) . the [ mg / fe ] ratios of h10 and h23 agree with galactic field stars . note that h10 s [ mg / fe ] is higher than the value presented by @xcite , which was derived from slightly different lines ; considering random and systematic uncertainties , however , the two are consistent . pa17 has a [ mg / fe ] ratio of 0.8 dex , which is higher than most of the field stars and all of the gcs . the syntheses of pa17 s 5711 line are shown in figure [ fig : synthmg]this line has a similar strength in all 3 exposures , suggesting that the line strength is not due to an improperly removed cosmic ray . note that there are no other mg@xmath6roman1 features in the observed spectrum ( other than the strong 5528 line ) to verify this high abundance . with its high [ mg / fe ] , pa17 falls at the upper end of the galactic and lmc field stars . this high [ mg / fe ] ratio suggests a greater contribution from massive star ejecta , either from a top - heavy imf or from inhomogeneous mixing and stochastic sampling of supernova ejecta . alternatively , high mg could indicate contributions from a rapidly rotating massive star that went supernova ( e.g. @xcite ) . calcium and titanium mainly form explosively in core collapse supernovae and generally behave like the other @xmath1-elements , though both elements also form in type ia supernovae @xcite . the integrated [ ca@xmath6roman1/fe@xmath6roman1 ] abundances are determined based on ews of @xmath41 spectral lines . all ca@xmath6roman1 line ews were verified by hand , and any uncertain lines were removed from the analysis . spectrum syntheses were also performed on several ca@xmath6roman1 lines to verify the ew - based abundances . the ti abundances are based on fewer lines ( 1 - 3 ) , but both [ ti@xmath6roman1/fe@xmath6roman1 ] and [ ti@xmath6roman2/fe@xmath6roman2 ] generally agree with [ ca / fe ] . @xcite determined that the [ ti / fe ] ratios are more sensitive to systematic uncertainties in the underlying stellar populations ; thus , only [ ca / fe ] is compared with other targets . the pandas cluster [ ca / fe ] ratios are compared to mw and dwarf galaxy field stars in figure [ fig : pandascafss ] and to mw , m31 , and dwarf galaxy gcs in figure [ fig : pandascagcs ] . most of the pandas gcs are ca - enhanced , though pa17 s [ ca / fe ] is roughly solar . all the pandas clusters look very similar to the lmc and sgr gcs ; most of the for clusters are more metal - poor than the pandas gcs . s [ ca / fe ] agrees well with @xcite and is slightly lower than the average mw and m31 clusters . pa17 s low [ ca / fe ] could be understood several ways . as a @xmath1-element , low [ ca / fe ] could be indicative of the slow star formation ratios or top - light imfs seen in dwarf galaxies ( e.g. @xcite)however , pa17 s high [ mg / fe ] ratio suggests that this interpretation is not so simple . regardless of the cause , pa17 s ca abundance looks most similar to pal 1 , ter 7 , pal 12 , and the lmc field stars and clusters . if mg and ca have the same nucleosynthetic site then the [ mg / ca ] ratio should not change with [ fe / h ] . in galactic field stars this is more or less what is observed : [ mg / ca ] is roughly solar except at the lowest metallicities where the dispersion is very high . the dispersion at low metallicities is typically ascribed to inhomogeneous mixing ( see , e.g. , @xcite ) . however , metal - rich dwarf galaxy stars appear to have different [ mg / ca ] ratios from mw stars . in fact , the nucleosynthetic sites of mg and ca are _ not _ the same : mg is mainly produced during the hydrostatic burning of massive stars ( e.g. @xcite ) while ca is formed explosively in both type ii and ia supernovae ( e.g. @xcite , @xcite , @xcite ) . figure [ fig : pandasmgca ] shows that the metal - rich field stars in the lmc and for have higher [ mg / ca ] ratios than mw stars at the same metallicity this discrepancy is typically ascribed to the different nucleosynthetic sites of mg and ca ( e.g. @xcite ) . however , stars in sgr have low [ mg / ca ] ratios ( except possibly for stars in the sgr streams ; @xcite ) ; @xcite attribute these low ratios to a top - light imf , where the lack of massive stars means that less mg is produced than ca . most of the pandas gcs overlap with galactic and dwarf galaxy stars ; however , pa17 s [ mg / ca ] ratio is higher than the galactic stars , in agreement with the lmc stars . again , star - to - star mg variations could affect the integrated [ mg / ca ] ratios of the metal - poor gcs , complicating comparisons with field stars . however , if the integrated [ mg / fe ] ratios are affected by multiple populations , one might expect the integrated [ mg / fe ] and [ mg / ca ] ratios to be _ lower _ than the `` primordial '' values ( because mg is expected to decrease in the second population ) . only pa06 has a lower [ mg / ca ] than the majority of the galactic stars , in agreement with the integrated [ mg / ca ] of m15 , which has known , _ strong _ star - to - star mg variations that affect the integrated [ mg / fe ] @xcite . thus , pa06 seems to be similar to m15 , and may host a significant second , mg - deficient population . the other metal - poor gcs agree well with the galactic stars and gcs , with the exception of pa56 , which has a slightly high [ mg / ca ] . note that most of the pandas gcs have higher [ mg / ca ] ratios than the m31 gcs again , this is likely because the gcs in the @xcite sample are more massive than the pandas gcs , and therefore may have stronger star - to - star mg variations . the gcs more metal rich than @xmath3 \sim -1.2 $ ] are not likely to host significant mg variations . h10 and h23 have roughly solar [ mg / ca ] ratios , in agreement with the galactic field stars and clusters ( though @xcite find a lower [ mg / ca ] for h10 because of their lower [ mg / fe ] ) . pa17 is higher than the galactic field stars and clusters by @xmath42 , in agreement with the lmc and fornax field stars . this high [ mg / ca ] is driven by the high [ mg / fe ] ( determined from spectrum syntheses of the 5711 line ) and the low [ ca / fe ] ( from ews of four clean lines ) . again , the high mg suggests that pa17 is enriched with the products of the most massive stars , which may be due to imf effects , stochastic sampling of supernova ejecta , or rotating supernova progenitors ( @xcite also note that rotating massive stars will not produce as much ca , compared to non - rotating stars ) . despite its agreement with the lmc field stars , pa17 is distinct from the lmc gcs , because the [ mg / ca ] ratios of the metal rich lmc gcs are not high like the lmc field stars . for the gc with the lowest [ mg / ca ] ratio ( ngc 1718 ) , @xcite suggest an inhomogeneous mixing scenario , where ngc 1718 formed out of material with no contributions from the highest mass stars . however , the three metal rich , [ mg / ca]-poor lmc gcs have intermediate ages ( 1 - 3 gyr ) , and it may not be entirely appropriate to compare these intermediate - age gcs with the older pandas gcs . regardless of the disagreement with the lmc gcs , pa17 s high [ mg / ca ] ratio is most similar to the lmc and fornax field stars . sodium abundances are derived from spectrum syntheses of the 6154 and 6160 lines , which are all weaker than 100 m in the pandas il spectra . sodium lines can be particularly sensitive to nlte effects ; however , the 6154/6160 lines are not expected to have significant nlte corrections ( @xmath43 dex ) at subsolar metallicities and for weak line strengths ( e.g. @xcite ) . ) can have ews @xmath44 m in the most metal - rich clusters . ] no corrections were applied to the [ na / fe ] abundances , which are shown in table [ table : abunds ] . figure [ fig : pandasna ] shows comparisons between the pandas gcs and the field stars and gcs associated with the various galaxies . several of the pandas gcs have [ na / fe ] ratios that are higher than mw and dwarf galaxy field stars . pa06 , pa17 , and pa54 all have high [ na / fe ] ratios ( @xmath45 dex ) which places them above the galactic and dwarf galaxy stars , while the h10 , pa53 and pa56 ratios are mildly high ( @xmath46 dex ) . these high values agree well with the il values of the galactic and old lmc gcs and with many of the averaged galactic gcs . in galactic gcs the [ na / fe ] ratio is affected by star - to - star abundance variations within the clusters , which are often observed as na / o anticorrelations ( see @xcite ) . these abundance variations are typically interpreted as arising from two chemically distinct stellar populations within the gcs , which may indicate two separate _ generations _ of stars ( see the review by @xcite ) . several galactic gcs have more centrally concentrated na - enhanced , o - deficient populations ( e.g. m13 and 47 tuc ; @xcite ) . il spectra of the central regions ( such as the galactic gcs in @xcite ) will likely be dominated by the second generation ( na - enhanced , o - deficient ) populations . it is therefore probable that ( like the galactic gcs ) the pandas gcs are na - enhanced because of the presence of multiple populations . it is possible that h23 s solar [ na / fe ] is due to the larger coverage of the il spectrum ( see table [ table : pandasobservations ] ) , which may include light from the less centrally concentrated first generation stars ; alternatively , h23 may have more `` primordial '' stars . again , the @xcite [ na / fe ] ratio for h10 is lower than the one derived here however , the ratios are derived from only two lines in both cases and are consistent within random and systematic errors . thus , the [ na / fe ] ratios indicate that these pandas gcs are likely to be `` classical '' gcs under the definition of ( * ? ? ? * i.e. the gc stars exhibit the na / o anticorrelation ) . note that in this paper only na abundances are derived for the pandas gcs , though o abundances would also be affected by multiple populations . however , the o lines are fairly weak in this wavelength range , and are not easily detectable in such low s / n spectra . lines in other regions ( e.g. the near infrared ) would provide more robust integrated o abundances . na and mg are expected to form primarily in massive stars ( e.g. @xcite ) ; in the absence of multiple population effects , the abundances of na and mg are therefore expected to correlate . the mw field stars with @xmath3 \ga -2 $ ] show a clear trend of increasing [ na / mg ] with [ fe / h]@xcite interpret this as a signature of increasing na yields with metallicity . however , the metal poor ( @xmath47 \la -1.2 $ ] ) galactic and pandas gcs show the opposite trend : in pa06 and m15 [ na / mg ] is very high , though [ na / mg ] decreases to subsolar values by @xmath3 \sim -1.5 $ ] . the gcs therefore behave in a very different way from the field stars . again , this is likely to be a result of multiple populations , combined with the metallicity dependency of the mg / al anticorrelation ( e.g. @xcite ) . for these lower mass gcs , at low metallicities the star - to - star mg variations are likely to be strong in these ( fairly bright ) gcs , driving the integrated [ mg / fe ] down and [ na / mg ] up . with increasing [ fe / h ] , the integrated [ mg / fe ] returns to its `` primordial '' value , and the [ na / mg ] ratio stabilizes . the [ na / mg ] abundances of the pandas gcs therefore also suggest that most of the gcs host multiple populations . note that the behaviour of the massive m31 clusters ( from @xcite ) is similar at the metal poor end , though their [ na / mg ] ratios are higher at the metal rich end again , this may be because the massive , metal rich gcs have stronger mg variations . pa17 remains an unusual object , because its similarly high [ na / fe ] and [ mg / fe ] ratios place its [ na / mg ] in agreement with the mw field stars , despite its disagreement in [ mg / fe ] . if pa17 s high [ mg / fe ] is caused by increased supernovae ejecta from massive and/or rapidly rotating stars , then [ na / fe ] would also be enhanced ; thus , pa17 may have an additional integrated na enhancement that is caused by contributions from massive stars . nickel is expected to form in core collapse and type ia supernovae , along with fe @xcite . the [ ni / fe ] ratios were determined with ews , even though there are few ( @xmath48 ) ni@xmath6roman1 lines however , these lines were carefully checked to ensure that they were clean ( i.e. uncontaminated by cosmic rays , sky lines , or noise ) and were properly measured . the ni abundances are given in table [ table : abunds ] and are compared to field stars and other gcs in figure [ fig : ni ] . the pandas clusters all have [ ni / fe ] ratios that are consistent with mw and dwarf galaxy stars however , h23 and pa17 both have slightly low ni abundances , in better agreement with dwarf galaxy stars and gcs . in particular , pa17 s [ ni@xmath6roman1/fe@xmath6roman1 ] ratios agree well with the pal 12 and ter 7 ratios . the ba and eu abundances are determined through spectrum syntheses . only two ba@xmath6roman2 lines were considered : the 5853 and 6141 lines the 6496 line was removed because of possible nlte effects which lead to abundance overestimates @xcite . no molecular lines are included in these syntheses , since none are identified in those regions in the arcturus atlas . the eu@xmath6roman2 abundances are determined through syntheses of the 6645 line and include ch and cn molecular lines ( see @xcite ) ; these syntheses are shown in figure [ fig : syntheu ] . hfs and isotopic components were included for the ba@xmath6roman2 and eu@xmath6roman2 lines , as described in section [ subsec : daospec ] . @xcite demonstrated that the [ ba / eu ] ratio is often more stable than the individual [ ba / fe ] and [ eu / fe ] ratios to uncertainties in the underlying population . comparisons of [ ba / eu ] ratios are shown in figure [ fig : pandasbaeu ] , and illustrate that metal - rich dwarf galaxy stars have higher [ ba / eu ] ratios than mw field stars . pa06 and pa54 only have upper limits on the eu abundance , which translates into a lower limit in [ ba / eu ] . the estimated r - process - only ratio ( from @xcite ) is also indicated in figure [ fig : pandasbaeu ] . the pandas gcs agree with the mw field stars and gcs , with the exception of pa17 and pa56 . pa56 has a low [ ba / eu ] , much like m15 . both ba and eu have been known to vary within some of the most massive metal poor gcs ( such as m15 ; see , e.g. , @xcite , @xcite ) ; neutron star mergers have been suggested as the origin of these star - to - star heavy element variations ( e.g. @xcite ) . pa56 s il abundances may also be affected by the presence of heavy element dispersions with the gcs . table [ table : abunds ] shows that pa56 has a moderate ba abundance and a high eu abundance , which leads to a [ ba / eu ] that is below the r - process only estimate ; this indicates that pa56 may host stars with a significant heavy element dispersion . pa17 s high [ ba / eu ] is in agreement with the galactic gcs within its @xmath2 errors , though it agrees best with the dwarf galaxy stars and gcs , pal 1 , and pal 12 . the high [ ba / eu ] ratios in dwarf galaxies are typically interpreted as an excess of s - process over r - process elements . its moderately high [ ba / eu ] ratio indicates that pa17 has received chemical contributions from agb stars . [ eu/@xmath1 ] is another popular chemical tagging indicator since dwarf galaxy stars and gcs have higher [ eu/@xmath1 ] ratios than mw stars and gcs at a given [ fe / h]this has been interpreted as a sign of an additional r - process site ( e.g. @xcite ) or a top - light imf ( e.g. @xcite ) . figures [ fig : pandaseuca ] and [ fig : pandaseucagcs ] show the [ eu / ca ] ratios ( which serve as [ eu/@xmath1 ] indicators ) in different environments . again , only upper limits are available for pa06 and pa54 . in general the clusters are in agreement with the mw and dwarf galaxy field stars and gcs , with the exception of pa56 , which has a high [ eu / ca ] , similar to m15again , this is likely a signature of star - to - star eu variations . the detailed abundances of stars are dictated by the chemical composition of the interstellar medium in their host galaxy . chemical comparisons between the pandas gcs and the field stars and gcs associated with other galaxies therefore provide clues about the nature of a gc s birth environment , such as whether a cluster formed in a massive galaxy or a dwarf galaxy . chemical comparisons can also determine if the gcs are chemically similar to _ particular _ galaxies , streams , or other gcs . if these gcs originated in dwarf galaxies , the abundances in this paper can therefore be used to infer the nature of the dwarfs that are currently being accreted into m31 s outer halo . the more metal - rich clusters all have higher metallicities than expected given their large projected distances from the centre of m31 . however , each gc has a unique chemical signature . pa17 : : : without _ hst _ photometry , pa17 s abundances are more uncertain than the other gcs ; however , the high [ na / fe ] ratio suggests that pa17 is a `` classical '' gc ( under the @xcite definition ) with signs of a na spread ( and therefore probably an o spread as well ) . pa17 is the most metal - rich of the target clusters , at @xmath3 \sim -0.9 $ ] . despite the lack of _ hst _ photometry , systematic errors in [ fe / h ] are likely to be @xmath50 dex . pa17 s metallicity alone indicates that it formed in a fairly massive galaxy its unusual location far away in the outer halo ( @xmath51 kpc ; @xcite ) suggests that it formed in a dwarf galaxy like the lmc or sgr . + pa17 s low [ ca / fe ] and high [ mg / ca ] and [ ba / eu ] ratios indicate that pa17 is more _ chemically _ similar to the lmc stars and clusters than to those associated with the mw ( and presumably to those in m31 ) , even when systematic offsets are considered . as discussed earlier , the [ @xmath1/fe ] and [ ba / eu ] ratios indicate whether a cluster formed in an environment enriched by type ia supernova and agb star products . combined with its metallicity , pa17 s detailed chemical abundance ratios therefore indicate that it formed in a galaxy which had sufficient mass to have a fairly high star formation rate or to populate the high mass end of the imf . however , its high [ mg / fe ] ( and possibly its high [ na / fe ] ) indicate that pa17 may have been enriched in ejecta from the most massive stars . as discussed in sections [ subsec : pandasalpha ] and [ subsec : pandaslight ] this suggests that pa17 s progenitor galaxy had more massive stars than a typical dwarf ( i.e. it had a top - heavy imf ) , was enriched by supernova ejecta from a rapidly rotating massive star , and/or was uniquely enriched from the products of a supernova due to inhomogeneous mixing in the host galaxy . + ultimately , pa17 s integrated abundances are chemically distinct from those of the metal - rich galactic gc 47 tuc , and are similar to pal 1 , the intermediate age lmc clusters , and the accreted sgr clusters pal 12 and ter 7 . + h10 and h23 : : : h10 and h23 are the third and second most metal - rich gcs in the sample of pandas gcs , respectively . based on the same arguments as for pa17 , this indicates that the clusters likely originated in lmc- or sgr - like dwarf galaxies . unlike pa17 , h23 s chemical abundance ratios are indistinguishable from mw field stars and clusters and m31 gcs at the same [ fe / h ] . h10 s [ ca / fe ] is mildly low while its [ mg / ca ] and [ eu / ca ] ratios are mildly high this [ ca / fe ] ratio could indicate that h10 formed in a dwarf galaxy with a knee in the [ ca / fe ] vs. [ fe / h ] relationship near @xmath3 \sim -1.5 $ ] , similar to lmc intermediate - aged clusters . it is more difficult to tag metal - poor gcs ( @xmath52 \la -1.5 $ ] ) with chemical abundances , since the chemistries of metal - poor stars in dwarf galaxies and massive galaxies are not always significantly different . the analysis is further complicated by potential star - to - star na , mg , ba , and eu variations within the gcs . pa53 and pa56 : : : these clusters are fairly close to each other in projection , and have very similar radial velocities and metallicities ; it is therefore possible that these clusters are physically associated with one another and were accreted from the same dwarf . their discrepant mg , ba , and eu ratios do not preclude the possibility that they are related , as the integrated abundances may be affected by star - to - star variations within the clusters . both gcs are slightly enhanced in na ( with [ na / fe]@xmath46 ) , which is likely a signature of the na / o anticorrelation . pa56 also has a very low [ ba / eu ] ( below the r - process only yields ) , similar to m15 , which supports the idea that it may have star - to - star heavy element variations . both clusters are metal - poor ( @xmath3 \sim -1.7 $ ] ) and have [ ca / fe ] ratios slightly lower than the mw field stars , similar to the fornax field stars . + pa54 : : : although pa54 is extremely close to pa53 in projection , their discrepant radial velocities imply that these clusters are not likely to be associated , nor is pa54 likely to be associated with pa56 . however , pa54 has very similar abundance ratios as pa53 and pa56 , even with only upper limits on [ eu / fe ] . this indicates that pa54 may have formed in similar conditions . + pa06 : : : pa06 is the most metal - poor pandas gc in this analysis , though it is still more metal - rich than the galactic gc m15 . again , pa06 s high [ na / fe ] indicates the presence of a na / o anticorrelation additionally , its low [ mg / fe ] hints at a mg / al anticorrelation . pa06 is similar to the metal - poor gcs in the mw ( particularly m15 ) and in the dwarf galaxies : it is metal - poor and @xmath1-enhanced . the weakness of its spectral lines means that only an upper limit can be obtained for eu@xmath6roman2 . the presence of streams in the metal - poor density map of m31 s outer halo ( see figure [ fig : pandasclusterlocations ] ) implies that some metal - poor ( i.e. low mass ) dwarf galaxies are currently being accreted . the chemical ratios of pa06 , pa53 , pa54 , and pa56 are all consistent with an accretion origin in at least one metal - poor dwarf galaxy . none of these gcs have been associated with streams based on their positions . given their lack of association with any streams , if pa06 and pa54 were accreted from dwarf satellites they may be remnants of ancient accretion events ; @xcite estimate that @xmath53% of the metal - poor halo ( @xmath3 \la -1.7 $ ] ) resides in a `` smooth '' component , which may include these gcs . recall that pa53 and pa56 are located near the eastern cloud . their similar kinematics and chemical compositions hint that pa53 and pa56 are associated with each other however , the numbers of stars around the clusters ( mackey et al . 2015 , _ in prep . _ ) , their metallicities , and their detailed chemical abundances indicate that pa53 and pa56 were also likely associated with the dwarf galaxy that created the eastern cloud . in this case , the eastern cloud must turn back toward m31 , encompassing pa53 . this further suggests that the eastern cloud s progenitor must have been massive enough to form gcs , though not sufficiently massive to produce a significant metal rich stellar population . the presence of the metal - rich giant stellar stream ( gss ) , the associated h@xmath6roman1 gas , the outer halo stellar mass , and the number of metal - rich gcs in the outer halo indicate that m31 likely experienced a minor merger with a sgr or lmc - mass galaxy ( e.g. @xcite , @xcite , @xcite ) . @xcite find that 86% of the most metal - rich stars ( @xmath3 \ga -0.5 $ ] ) in the outer halo are associated with substructure from an accreted companion . though this number decreases with metallicity , the fraction of accreted stars still remains high : 78% of the stars in pa17 s metallicity bin ( @xmath54 \la -0.5 $ ] ) are expected to be associated with coherent , accreted structures , while 58% of the stars in h10 and h23 s metallicity bin ( @xmath55 \la -1.1 $ ] ) are expected to be in streams . it is therefore likely that these comparatively metal rich gcs were associated with dwarf satellites . h23 has been tentatively linked to a stream near the gss ( stream d ; see @xcite ) based on its position , though its kinematics indicate that such an association is unlikely ; neither h10 nor pa17 have been linked to any visible stellar streams , though _ hst _ imaging of fields around h10 and h23 reveal extremely metal - rich populations of field stars , suggesting that the gcs may be located on low surface brightness streams @xcite . h10 is located near the sw cloud , an overdensity of stars to the southwest of m31 . km s@xmath56 ) from pa14 ( @xmath57 km s@xmath56 ; @xcite ) and high [ fe / h ] ( @xmath58 compared to @xmath59 ; @xcite ) make it unlikely that pa17 was associated with the sw cloud progenitor . ] with photometry , @xcite estimate that the cloud hosts a metal - rich population ( @xmath3 \sim -1.3 $ ] ) , and that the progenitor galaxy was a fairly bright dwarf ( @xmath60 , which is slightly fainter than for and sgr ) . three gcs ( pa7 , pa8 , and pa14 ) appear to be kinematically associated with each other and with the sw cloud @xcite . the sw cloud does extend to the southeast , and h10 lies at the end of this extension ( see figure 2 in @xcite ) . h10 s radial velocity ( @xmath61 km s@xmath56 ) agrees with the stream s radial velocity and that of the gcs , and even agrees with the velocity gradient noted by bate et al . ( where the northernmost gc , pa7 , is moving toward the mw faster than pa8 and pa17 ) . furthermore , h10 s @xmath3 \sim -1.4 $ ] agrees very well with the cloud and its gcs , and its detailed abundances support the idea that it could have originated in a dwarf galaxy with the mass of for or sgr . thus , it is likely that h10 is associated with pa7 , pa8 , pa14 , and the sw cloud . as mentioned above , h23 has been spatially associated with stream d based on its position , though its velocity disagrees with the other gcs in the stream . its metallicity , @xmath3 = -1.1 $ ] , now further suggests that h23 is not likely to be associated with that stream , since stream d is primarily composed of metal - poor stars with @xmath3 < -1.1 $ ] @xcite . it is more likely that h23 came from the progenitor of the gss , along with pa17 . @xcite perform n - body simulations to reproduce the gss and other stellar debris . these simulations indicate that a massive progenitor was accreted during multiple orbits , culminating in a recent final accretion ( @xmath62 myr ago ) . many of the metal - rich pandas gcs might have been stripped early on , and may no longer be associated with any bright , coherent substructure . given their metallicities and other chemical abundances , it is possible that h23 and pa17 both formed in the gss progenitor . if h23 and pa17 were born in the same dwarf galaxy , their [ ca / fe ] ratios clearly imply a `` knee '' at @xmath3 \sim -1.1 $ ] to -1.3 , similar to the lmc stars and clusters . this paper has presented integrated fe , na , mg , ca , ti , ni , ba , and eu abundances of seven outer halo m31 gcs , five of which were discovered in the pan - andromeda archaeological survey . these detailed il chemical analyses of pandas clusters have identified gcs in an extragalactic system that may have been accreted from dwarf galaxies . detailed investigations of the chemical abundance ratios of individual targets such as this are only possible with _ high resolution _ spectroscopy . the pandas cluster abundances suggest that these outer halo m31 gcs may have been accreted from multiple dwarf galaxies : * the metal - rich gc pa17 is chemically more similar to the lmc stars and clusters than to mw field stars and clusters , suggesting that it originated in an lmc - like dwarf galaxy . s abundances are indistinguishable from mw field stars and clusters , though its metallicity and location suggest it may have originated in a massive dwarf galaxy . pa17 and h23 may have been accreted along with the progenitor of the metal - rich giant stellar stream . + * the intermediate metallicity gc h10 has a location , metallicity , and radial velocity that agree well with the sw cloud and its gcs , which may have originated in a sagittarius or fornax sized progenitor @xcite . s chemical abundance ratios support the suggestion that it may have formed in a dwarf galaxy . + * pa53 , pa54 , and pa56 have abundances and metallicities that are more typical of an intermediate mass dwarf galaxy like for . this suggests that they are currently being accreted from at least one metal - poor dwarf galaxy and could be associated with one or more of the coherent , metal - poor streams observed in pandas . based on their chemistries and radial velocities pa53 and pa56 could be associated and are likely to have been accreted from the dwarf galaxy that created the eastern cloud . despite its proximity in projection , it is unlikely that pa54 is associated with either pa53 or pa56 . + * pa06 s metallicity makes it a difficult target for chemical tagging analyses , since the chemistries between dwarf and massive galaxies have likely not had sufficient time to diverge at @xmath3 \sim -2 $ ] . several of its integrated abundances do not agree with mw or dwarf galaxy stars , suggesting that strong star - to - star chemical variations are present in the cluster . + thus , this chemical tagging analysis is consistent with the observation that m31 s outer halo gc system is being built up by accretion of dwarf satellites . in addition , this detailed abundance analysis provides additional information on the nature of these progenitor systems that is , at present , very difficult to obtain from the field populations . the authors thank don vandenberg for his suggestion to investigate the rgb slope calibration , for providing the isochrones in _ hst _ magnitudes , and for all his comments . the authors also thank judy cohen for providing a helpful referee report that contributed to this analysis and george wallerstein for reading the manuscript . cms acknowledges funding from the natural sciences & engineering research council ( nserc ) , canada , via the vanier cgs program . kav acknowledges funding through the nserc discovery grants program . adm is grateful for support by an australian research fellowship ( grant dp1093431 ) from the australian research council . the hobby - eberly telescope ( het ) is a joint project of the university of texas at austin , the pennsylvania state university , stanford university , ludwig - maximilians - universitt mnchen , and georg - august - universitt gttingen . the het is named in honour of its principal benefactors , william p. hobby and robert e. eberly . the authors wish to thank the night operations staff of the het for their assistance and expertise with these unusual observations . this work has made use of basti web tools . it is well established that the slope of a cluster s rgb is correlated with cluster metallicity ( e.g. @xcite ) . this appendix discusses the calibration of rgb slope vs. [ fe / h ] , utilizing high quality _ hst _ cmds of galactic gcs from the acs survey of galactic globular clusters @xcite . for this calibration 15 galactic gcs with high quality cmds and well - populated rgbs that span a wide range in [ fe / h ] were selected ; priority was given to clusters with low foreground reddening . using the original f606w and f814w _ hst _ magnitudes , the distance moduli from @xcite were used to overplot clusters , and distance moduli and reddening values were adjusted so that cluster hbs overlapped . the average rgb colours were then determined at two magnitudes ( @xmath63 and @xmath64 ) , and rgb slopes were calculated . these points are shown in figure [ fig : allrgbs ] , while the slopes are listed in table [ table : rgbslopes ] . uncertainties in rgb slope were estimated based on the uncertainties in colour at the two magnitudes . the metal - poor clusters have small uncertainties in rgb colour , though this translates into large uncertainties in slope because the rgbs are steeper . the metal - rich clusters have very uncertain rgb colours , but because the rgbs are flatter the uncertainty in slope is much lower . note that at higher [ fe / h ] ( @xmath65 ) a linear fit to the rgb is no longer a good approximation to the actual shape of the rgb thus , this calibration is likely to break down at the metal rich end . for each cluster the [ fe / h ] ratios from @xcite were adopted ; [ fe / h ] uncertainties of 0.05 dex were assumed . figure [ fig : mwrgbslopes ] shows the relationship between rgb slope and cluster [ fe / h ] . the linear least squares fit to the points is shown with the solid line , while the dashed lines show the uncertainty in the fit . the trend is clear : metal - poor gcs have steeper rgbs than more metal - rich gcs . non - standard chemical abundance mixtures ( in , e.g. , c , n , o or @xmath1-elements like mg or si ) can affect the shape of the rgb ( e.g. @xcite ) , which would affect where a cluster falls in the plot . the adopted [ @xmath1/fe ] ratio is particularly important : the qualitative effects of [ @xmath1/fe ] on the f606w , f814w rgb slope are illustrated in figure [ fig : alphaisos ] using the victoria - regina isochrones from @xcite and the colour transformations of @xcite . figure [ fig : alphaisosa ] shows that at a given [ fe / h ] , clusters with @xmath66 = 0 $ ] have steeper rgbs than clusters with @xmath66 = + 0.4 $ ] ; this offset is greater at higher metallicity . furthermore , figure [ fig : alphaisosb ] illustrates that the slope difference can be mimicked by lowering [ fe / h ] by 0.3 dex while maintaining @xmath66 = + 0.4 $ ] ( note that in figure [ fig : alphaisosb ] the dotted isochrones have been shifted by 0.02 magnitudes so that both the @xmath1-enhanced and @xmath1-normal isochrones can be seen ) . the blue dot - dashed line in figure [ fig : mwrgbslopes ] shows the effects of lowering [ @xmath1/fe ] by 0.4 dex , which is approximated by lowering the [ fe / h ] by 0.3 dex ( also see @xcite ) . the rgbs of the pandas clusters are shown in figure [ fig : pandasrgbs ] , while the rgb slopes and spectroscopic metallicities are shown in figure [ fig : rgbslopes ] . the photometric uncertainties , sparseness of the cmds , and difficulties in determining distance moduli ( particularly for pa56 ) make the rgb slopes more uncertain for the pandas clusters , compared to the galactic gcs . however , all pandas gcs agree with the mw rgb slope relation in figure [ fig : mwrgbslopes ] within their errors . this relationship illustrates that the spectroscopic metallicities are consistent with the observed rgb slopes in the _ hst _ cmds . all of the clusters in figure [ fig : rgbslopes ] lie above the mw relation closer to the low [ @xmath1/fe ] value , hinting at the possibility of low [ @xmath1/fe ] ratios in the pandas targets . @lcccd5@ cluster & & [ fe / h ] & & + m92 & & @xmath67 & & -11.429,1.172 + m15 & & @xmath68 & & -11.765,1.238 + m53 & & @xmath69 & & -10.929,1.077 + ngc 5286 & & @xmath70 & & -10.870,1.066 + m2 & & @xmath71 & & -9.346 , 0.799 + m13 & & @xmath72 & & -9.217 , 0.778 + m3 & & @xmath73 & & -9.091 , 0.758 + m5 & & @xmath74 & & -8.299 , 1.181 + ngc 1261 & & @xmath75 & & -7.937 , 1.087 + ngc 1851 & & @xmath76 & & -7.491 , 0.976 + ngc 6362 & & @xmath77 & & -6.557 , 0.760 + 47 tuc & & @xmath78 & & -3.766 , 1.031 + ngc 6652 & & @xmath78 & & -4.090 , 1.187 + m69 & & @xmath79 & & -3.883 , 1.086 + ngc 5927 & & @xmath80 & & -7.491 , 0.595 + @lcccd5@ cluster & & [ fe / h ] & & + pa06 & & @xmath81 & & -12.0482,2.658 + pa54 & & @xmath82 & & -11.173,1.405 + pa56 & & @xmath83 & & -10.811,2.982 + pa53 & & @xmath84 & & -10.526,1.238 + h10 & & @xmath85 & & -8.621,0.813 + h23 & & @xmath86 & & -7.353 , 1.268 +

we investigate the effect of a hybrid electron population , consisting of both thermal and non - thermal particles , on the synchrotron spectrum , image size , and image shape of a hot accretion flow onto a supermassive black hole . we find two universal features in the emitted synchrotron spectrum : ( i ) a prominent shoulder at low frequencies that is weakly dependent on the shape of the electron energy distribution , and ( ii ) an extended tail of emission at high frequencies whose spectral slope depends on the slope of the power - law energy distribution of the electrons . in the low - frequency shoulder , the luminosity can be up to two orders of magnitude greater than with a purely thermal plasma even if only a small fraction of the steady - state electron energy is in the non - thermal electrons . we apply the hybrid model to the galactic center source , sgr a . the observed radio and ir spectra imply that at most of the steady - state electron energy is present in a power - law tail in this source . this corresponds to no more than of the electron energy injected into the non - thermal electrons and hence into the thermal electrons . we show that such a hybrid distribution can be sustained in the flow because thermalization via coulomb collisions and synchrotron self - absorption are both inefficient . the presence of non - thermal electrons enlarges the size of the radio image at low frequencies and alters the frequency dependence of the brightness temperature . a purely thermal electron distributions produces a sharp - edged image while a hybrid distribution causes strong limb brightening . these effects can be seen up to frequencies hz and are accessible to radio interferometers . to appear in _ the astrophysical journal _

astro-ph0004195

the mechanisms of particle heating and acceleration , and the emission spectra from the resulting particle energy distributions , are of great importance in the theory of collisionless hot accretion flows onto compact objects . discussions in the literature have focused on the physics of electron heating and acceleration ( begelman & chiueh 1988 ; bisnovatyi - kogan & lovelace 1997 ; quataert & gruzinov 1999 ; gruzinov & quataert 1999 , medvedev 2000 ) , the efficiency of particle thermalization ( ghisellini , guilbert , & svensson 1988 ; ghisellini , haardt , & fabian 1993 ; mahadevan & quataert 1997 ; ghisellini , haardt , & svensson 1998 ; nayakshin & melia 1998 ) , and the generation of hybrid thermal - nonthermal electron energy distributions in these plasmas ( see , e.g. , coppi 1999 and references therein ) . despite a substantial amount of work , many issues remain unresolved , primarily because of our incomplete understanding of physical processes such as magnetic reconnection , mhd turbulence , and collective plasma modes . these questions are especially relevant for an optically - thin advection - dominated accretion flow ( adaf ) . an adaf is an example of a hot , rarefied , magnetic plasmas with low radiative efficiency ( ichimaru 1977 ; narayan & yi 1994 , 1995b ; abramowicz et al . 1995 ; see narayan , mahadevan , & quataert 1998b and kato , fukue , & mineshige 1998 for reviews ) . a basic property of the nearly collisionless plasma in an adaf is that the coulomb coupling between the electrons and ions is weak so that energy transfer from the ions to the electrons is inefficient . in addition , it is commonly assumed that the viscously generated energy primarily heats the heavier species , the ions , and that the electrons retain a thermal distribution throughout the flow ( e.g. , narayan & yi 1995b ; mahadevan 1997 ) . there are , however , processes such as mhd turbulence , pair production ( e.g. , through pion decay ) , and electron - proton coupling which can both heat the electrons and generate non - thermal distributions . quataert and gruzinov ( 1999 ; see also gruzinov & quataert 1999 ) considered two processes specific to mhd turbulence that accelerate particles in magnetic collisionless plasmas : landau damping by electric fields parallel to the local magnetic field and transit - time damping by time - varying magnetic fields . they found that the assumption of negligible electron heating / acceleration is valid only for weak magnetic fields , i.e. , when the ratio of the gas pressure to total pressure @xmath8 is larger than a critical value @xmath9 . the value of @xmath9 is very uncertain and is around 0.9 . for @xmath10 , turbulence primarily accelerates the protons , while for stronger magnetic fields , the results are inconclusive . shocks and pion decay can also lead to non - thermal electrons in the accretion flow . several processes , such as coulomb collisions and synchrotron self - absorption , can potentially lead to thermalization of particles in accretion flows ( mahadevan & quataert 1997 ; ghisellini et al . mahadevan & quataert ( 1997 ) showed that coulomb collisions are ineffective in thermalizing the electrons in an adaf . however , they argued that , for sufficiently high mass - accretion rates , the electrons in the plasma can be thermalized by synchrotron self - absorption . nayakshin and melia ( 1998 ) showed that considerable deviations from a maxwell - boltzmann distribution can be sustained in a plasma with low source compactness when coulomb collisions , comptonization , and pair processes are taken into account . in view of the difficulty of calculating the heating , cooling and thermalization of particles from first principles , many authors have restricted their models to either purely maxwellian or purely non - thermal ( extended power laws or monoenergetic ) electron distributions . only recently have there been attempts towards explaining spectra of accreting black holes with models including hybrid thermal / non - thermal distributions of electrons . some models have physically motivated distributions , such as non - thermal electrons produced by decaying pions ( mahadevan 1998 ) , whereas others invoke more _ ad hoc _ distributions to fit the data ( e.g. , beckert & duschl 1997 ; falcke & biermann 1999 ) . the reverse process of trying to constrain the energy distributions of particles in accreting plasmas by comparing models to data , however , faces issues of uniqueness which can be addressed only by a more comprehensive study of hybrid models . in this paper , we consider generalized electron distributions consisting of a dominant maxwellian plus a small non - thermal power - law component of varying slope and energy content , and study the synchrotron emission from the resulting hybrid plasmas . we identify the characteristic signatures of the non - thermal electrons on the emitted radio synchrotron spectrum of an accretion flow and on its image as observed with a radio telescope . we also discuss to what extent the observed effects could be used to determine the details of the underlying non - thermal electron distribution . this work is relevant for interpreting observations of low luminosity agns such as the galactic center source , sagittarius a@xmath3 ( sgr a@xmath3 ) . in @xmath11 we review the basic properties of adafs , hybrid plasmas , synchrotron radiation , and radiative transfer . in @xmath12 we present a series of models of supermassive black holes with low accretion rates . we apply these models to sgr a@xmath3 and derive constraints on the fraction of the electron energy that can be present in a non - thermal form . in @xmath13 , we study the correspondence between the energy distribution of the electrons and the resulting spectrum . in @xmath14 we study the energetics of a hybrid flow and calculate the heating , cooling and thermalization rates of the non - thermal electrons . we summarize our conclusions in @xmath15 . we present in an appendix approximate analytic expressions for the contribution of a non - thermal particle population to the synchrotron spectrum of an accretion flow . we begin by reviewing some of the basic properties of the optically - thin branch of adafs . adafs are quasi - spherical , hot , magnetic accretion flows in which the accreting plasma is too rarefied to cool efficiently by radiative processes . the viscously dissipated energy is therefore advected into the black hole or other compact object at the center ( see narayan et al . 1998b and kato et al . 1998 for reviews ) . in the limit where the fraction of the viscous energy advected inward is independent of radius , a self - similar analytic solution for the thermodynamic quantities of the accreting gas can be obtained ( narayan & yi 1994 , 1995b ) . we make use of this solution in appendix a. for the numerical calculations presented in the rest of the paper , we use more accurate global solutions to obtain the run of electron temperature and density with radius . these solutions are calculated by the methods described in narayan , kato , & honma ( 1997b ) , chen , abramowicz , & lasota ( 1997 ) , and popham & gammie ( 1998 ) . the magnetic field strengths are obtained by assuming that the ratio of gas pressure to total pressure ( sum of gas and magnetic pressure ) is equal to a specified value @xmath8 . since the focus of this paper is on massive black holes in galactic nuclei with low mass - accretion rates , with specific applications to the black hole in our own galactic nucleus , sgr a@xmath3 , we scale masses in units of @xmath16 solar mass , i.e. , @xmath17 and radii in units of the schwarzschild radius , i.e. , @xmath18 , where @xmath19 we scale the mass accretion rate in units of @xmath20 , i.e. , @xmath21 where the eddington mass accretion rate is @xmath22 in defining the eddington rate , we assume a standard radiative efficiency of @xmath23 . ( this is purely for the purposes of the definition ; the actual radiative efficiency can be very different from @xmath24 ) . for the calculations presented here we use adaf models in which the viscosity parameter is set to @xmath25 , the equipartition parameter to @xmath26 , and the ratio of viscous electron heating to proton heating to @xmath27 . note that @xmath8 differs from the usual plasma parameter @xmath28 , which is the ratio of the gas pressure to the magnetic pressure ; the value of @xmath29 assumed in our adaf models corresponds to a plasma @xmath30 ( quataert & narayan 1999 ) . we assume that a large fraction of the electrons in the plasma are in a thermal distribution with temperature @xmath31 and that the rest of the electrons are in a non - thermal distribution , usually with a power - law form . we denote the number density of electrons in the thermal population by @xmath32 and in the non - thermal population by @xmath33 . we denote the emissivities of the thermal and power - law electron populations by @xmath34 and @xmath35 respectively , and the corresponding absorption coefficients by @xmath36 and @xmath37 . for the thermal electron population , we use the relativistic maxwell - boltzmann distribution given by @xmath38,\ ] ] where @xmath39 is the electron lorentz factor , @xmath28 is the relativistic electron velocity , and @xmath40 is the dimensionless electron temperature . the modified bessel function of second order @xmath41 arises from the normalization of the maxwellian . similarly , for the non - thermal electron population we use a power - law distribution extending from @xmath42 to infinity , @xmath43 the number density of thermal electrons @xmath32 is a function of the flow radius and is determined by the global adaf solutions . we determine @xmath33 at each radius by assuming that the steady - state energy in the non - thermal distribution is equal to a fraction @xmath44 of the energy in the thermal distribution , with @xmath44 constant in radius . although the calculations presented in @xmath12 are all carried out with this assumption , generalizations to radially - dependent @xmath45 as well as a discussion of the energetics of such a flow will be presented in @xmath46 . note that we implicitly assume that the non - thermal electron population does not affect the dynamics or the thermal properties of the flow ; this will be justified in @xmath46 . the energy density of a maxwell - boltzmann distribution of electrons at temperature @xmath47 was derived by chandrasekhar @xmath48)$ ] to be @xmath49 where @xmath50\ ] ] is a coefficient that varies from 3/2 for a non - relativistic electron gas to 3 for fully relativistic electrons , and @xmath51 are the modified bessel functions of the n@xmath52 order . for the present purposes , we use a simplified expression for @xmath53 which has an error of less than @xmath54 at all temperatures ( gammie & popham 1998 ) : @xmath55 the number density of the non - thermal distribution is then @xmath56 this normalization of the power law population typically corresponds to @xmath57 depending on the electron temperature and the power law index . the distributions considered above correspond to the steady state that results from the competition between heating / acceleration and cooling by radiation . we discuss in some detail in @xmath46 the energy equations for the thermal and non - thermal electron populations . here we simply note that the synchrotron cooling timescale of electrons moving with a lorentz factor @xmath39 scales as @xmath58 , and hence electrons in the high energy tail of a power law distribution cool most rapidly . as a result , if electrons are injected with a power - law distribution with index @xmath59 @xmath60 $ ] and cool only by synchrotron emission , the synchrotron cooling causes the power law index of the steady state distribution to be @xmath61 above a certain @xmath62 , called the cooling break , thus causing the steady state distribution to fall off more steeply at higher electron energies . mahadevan & quataert ( 1997 ) calculated @xmath62 in an adaf by comparing the cooling timescale to the inflow timescale and found that at sufficiently high accretion rates , the break occurs at a very low lorentz factor , @xmath63 . the lorentz factors of interest to us are invariably larger than @xmath62 . therefore , the values of @xmath64 we consider below in @xmath12 and @xmath13 are always equal to @xmath65 , so that the injected energy distribution @xmath66 is harder by one power of @xmath39 than the steady state energy distribution , @xmath67 . corresponding to @xmath44 , we can define another quantity @xmath68 that measures the fraction of electron energy @xmath69 into a power law distribution . if @xmath70 , then @xmath68 can be significantly greater than @xmath44 . if we assume a distribution from @xmath71 to some @xmath72 , @xmath68 is greater than @xmath44 by a factor @xmath73 . however , it is thought that the acceleration mechanisms typically encountered in astrophysics , such as shock acceleration or acceleration via mhd turbulence , inject energy into particles with @xmath74 such that the steady state distribution has @xmath75 . in this case there is little dependence on @xmath72 , and @xmath76 which is not very different from unity . although we expect @xmath74 and @xmath77 for most systems , for completeness we consider models in the range @xmath78 . the synchrotron emissivity of a relativistic electron moving with a lorentz factor @xmath39 in a magnetic field of strength @xmath79 is given by ( rybicki & lightman 1979 ) @xmath80 here , @xmath81 is the non - relativistic cyclotron frequency , @xmath82 is the angle between the direction of the magnetic field and the velocity of the electron , and @xmath83 with @xmath84 the modified bessel function of order 5/3 , @xmath85 , and @xmath86 . for a thermal distribution of electrons , the total emissivity for a given angle @xmath82 is obtained by integrating equation ( [ eq : sngl ] ) over the maxwellian distribution ( pacholczyk 1970 ) , @xmath87 where @xmath88 and @xmath89 the limiting behaviour of @xmath90 for small and large @xmath91 was derived by pacholczyk @xmath92 and petrosian @xmath93 , respectively . mahadevan , narayan , & yi ( 1996 , hereafter mny96 ) integrated equation ( [ eq : ixm ] ) over the angle @xmath82 for an isotropic distribution of electron velocities and provided a fitting function for the emissivity in the ultrarelativistic to mildly relativistic regimes : @xmath94 with @xmath95 given by @xmath96 the best fit values of the coefficients @xmath97 , and @xmath98 for different temperatures are given in mny96 . the coefficients tend to unity in the ultrarelativistic limit . finally , the synchrotron absorption coefficient @xmath36 is related to the emissivity via kirchoff s law , @xmath99 where @xmath100 is the black body source function . for the total emissivity of electrons in a power law distribution , we use the expression given in rybicki & lightman ( 1979 ) and average over angles , @xmath101 where @xmath33 is defined in terms of @xmath32 as above and @xmath102 the corresponding absorption coefficient is @xmath103 with @xmath104 the equation of radiative transfer for a time - independent , spherically symmetric flow is ( e.g. , mihalas 1978 ) @xmath105 i(r , \mu , \nu ) = j(r,\nu ) - \alpha(r , \nu ) i(r , \mu , \nu),\ ] ] where @xmath106 is the cosine of the angle between the ray and the radial direction , @xmath107 is the radial coordinate , @xmath108 is the frequency , and @xmath109 and @xmath110 are the emission and absorption coefficients defined above . one can simplify this equation by taking plane parallel rays of varying impact parameters ( perpendicular distances of rays to the central line of sight ) through the flow and solving the equation along these rays ( mihalas 1978 ) . the equation then becomes @xmath111 where now @xmath59 is the line element along the ray and the coefficients @xmath112 and @xmath113 correspond to radiation coming towards and going away from an external observer , respectively . in our problem , @xmath114 and @xmath115 . rewriting the equation in terms of the source function @xmath116 and optical depth @xmath117 , where @xmath118 is the point of intersection of the ray with the outer boundary of the flow , equation ( [ eq : radtr2 ] ) becomes @xmath119 where the combined source function is @xmath120 in the numerical calculations reported below , we integrate equation ( [ eq : radtr4 ] ) using the formal solution and the appropriate boundary conditions for a non - illuminated atmosphere ( see mihalas 1978 ) . we carry out the integral along rays with impact parameters up to @xmath121 schwarzschild radii , beyond which the temperature of the electrons becomes too low for significant synchrotron emission . because of the very steep dependence of the synchrotron emissivity on photon frequency , magnetic field strength , and electron temperature and density , we use an adaptive step size for the radiative transfer integral . the total flux is obtained by integrating over all impact parameters . we validated the implementation of our numerical algorithm by comparing its output to analytic solutions of the radiative transfer equation in uniform media . the discrepancy between the numerical and analytic solutions was @xmath122 for the cases considered . in all model adaf spectra published so far , the transport of radiation has been calculated using an approximate method based on concentric shells ( narayan , barret , & mcclintock 1997a ) . in figure 1 , we compare , for a typical flow , the exact spectrum obtained by formally solving the above transfer equation to that obtained with the previous approximate method . we see that there is a fairly good agreement between the two methods , but with some differences . the most prominent difference is a shifting of the peak toward higher frequencies in the exact calculation , as well as some broadening . this is probably due to the poorer resolution of the concentric shell method which becomes a limiting factor close to the black hole . there is also a slight offset at lower frequencies , probably again due to poor resolution . note that there is a second peak at high frequencies ( @xmath123 hz ) in the spectrum calculated with the approximate method . this is caused by the inverse compton scattering of soft photons , a process which is not included in the radiative transfer code described here . the remaining features of the two spectra are quantitatively consistent . in this section , we study the effects of an extended power - law electron distribution on the synchrotron emission spectra of adafs . figure 2 shows the various components of the spectrum of a typical hybrid model , with @xmath124 , @xmath125 , @xmath126 , and @xmath127 . compared to the spectrum of a pure thermal model ( dashed line ) , we see two primary effects due to the power - law electrons : ( i ) there is a prominent shoulder of optically thick emission at low frequencies and ( ii ) there is an extended power - law tail of optically thin emission at high frequencies . in between these two features there is a region of the spectrum where the thermal peak dominates . these features were first identified by mahadevan ( 1998 ) for a specific model . we find that they are universal for hybrid models . to understand the results , we note that the spectra of hybrid populations are determined by a competition between @xmath35 and @xmath34 and between @xmath37 and @xmath36 , each having different radial and frequency dependences . since the emission at different frequencies arises from different radii in the flow , the relative importance of @xmath37 to @xmath36 and @xmath35 to @xmath34 at that particular frequency and radius determines the local behaviour of the spectrum . in the less steep segment of the low - frequency shoulder , the emission from the power - law population ( which is more efficient at low frequencies ) exceeds that of the thermal electrons , while absorption is still mostly dominated by the more numerous thermal electrons . as a result , this segment of the spectrum assumes a shape roughly described by the rather unusual source function @xmath128 . at still lower frequencies , where the emission comes from larger radii in the flow , the contribution of the non - thermal electrons to the absorption becomes non - negligible and the spectrum falls more steeply with decreasing @xmath108 . in the region of the thermal peak , we have both @xmath129 and @xmath130 , with @xmath131 for frequencies smaller than the peak frequency and @xmath132 for frequencies higher than the peak frequency . the spectrum is essentially the same as for a purely thermal model ( dashed line ) . ( but , note that for sufficiently high @xmath44 , the synchrotron spectrum assumes an entirely non - thermal character and this region too can be dominated by the power - law electrons , with the peak luminosity becoming higher and the peak flattened ) . beyond the thermal peak , there is no self - absorption either by thermal or power - law electrons @xmath133 and the emission is optically thin . here we find an extended power - law tail dominated entirely by the non - thermal population . this segment of the spectrum has the familiar form @xmath134 which gives a rising spectrum with increasing frequency for @xmath135 and a falling spectrum for @xmath77 . figures 3a - d show how the spectrum of a hybrid model depends on the various parameters . in figure 3a , the energy content of the power - law population @xmath44 is varied from @xmath136 ( purely thermal ) to @xmath137 ; we see that the range and normalization of the low - frequency shoulder increase linearly with increasing @xmath44 . the normalization of the high - frequency tail also depends linearly on the number of electrons in the power - law distribution and thus varies in an obvious way with @xmath44 . in figure 3b , we fix @xmath44 at @xmath138 and vary the power - law index @xmath64 . note that the overall luminosity of both the low - frequency shoulder and the high - frequency tail show considerable dependence on @xmath64 . this is not unexpected . when @xmath64 is small , there are more particles both at intermediate lorentz factors @xmath139 , which give rise to the low - frequency shoulder , as well as at large @xmath39 , which emit the high - frequency tail ( see @xmath13 ) . therefore , small values of @xmath64 give more emission in both these segments of the spectrum . the high - frequency tail has a spectral slope equal to @xmath140 and therefore has a strong dependence on @xmath64 . two striking qualitative results in figures 3a and 3b need to be highlighted . first , even very small values of the non - thermal energy content @xmath44 give rise to significant excess super - thermal emission at low frequencies . for instance , even if only @xmath138 of the total electron energy is in non - thermal electrons , there can be several orders of magnitude higher luminosity at frequencies below the thermal peak . even more striking is the fact that the spectral shape of the low frequency shoulder is nearly independent of @xmath44 and @xmath64 , when both these parameters are independent of radius , while the normalization depends on both ( see appendix a for an analytic expression for the spectral slope of this segment ) . this leads to a degeneracy between @xmath44 and @xmath64 in models of the low - frequency emission , so that it is not possible to distinguish between different combinations of @xmath44 and @xmath64 , by studying spectral data below the thermal peak only . the degeneracy can be lifted with data on the high - frequency tail whose slope has a strong dependence on @xmath64 . we finally consider the effect of changing the black hole mass and the mass - accretion rate . we find that the relative excess emission due to non - thermal electrons increases rather weakly with increasing black hole mass ( figure 3c ) . this is because the thermal peak moves to lower frequencies for higher @xmath141 ( mahadevan 1997 ) and at these frequencies the effect of non - thermal electrons is more prominent . the non - thermal contribution to both the low and high frequency emission increases only weakly with increasing @xmath142 ( figure 3d ) . over the last decade , many models have been developed to explain the radio spectrum of sgr a@xmath3 . this source is believed to be an accreting supermassive black hole at the center of our galaxy . nearly all the published models invoke synchrotron radiation from relativistic or quasi - relativistic electrons . melia ( 1992 ) considered emission by thermal electrons in a spherical accretion flow and showed that the resulting cyclo - synchrotron emission is consistent with the broad features of the observed spectrum . narayan , yi , & mahadevan ( 1995 ) and narayan et al . ( 1998a ) developed adaf models of sgr a@xmath3 which included rotation , viscosity and a two - temperature plasma , and obtained similar results , again with a purely thermal distribution of electrons . beckert & duschl ( 1997 ) and falcke & biermann ( 1999 ) considered non - thermal models , while mahadevan ( 1998 ) analyzed a specific hybrid model in which the non - thermal electrons are produced by pion decay . this section is a generalization of mahadevan s work . in figure 4 we apply our hybrid emission model to sgr a@xmath3 . we take @xmath143 ( eckart & genzel 1997 ; ghez et al . 1998 ) and adjust @xmath142 in order to fit approximately the thermal peak . the data shown are the same as in narayan et al . ( 1998a ) . figure 4 shows four models with @xmath144 , @xmath145 ; @xmath146 , @xmath147 ; @xmath124 , @xmath148 ; and @xmath149 , @xmath150 . we first note that the agreement with data at low frequencies is significantly better with these hybrid models ( dashed lines ) than with a purely thermal model ( solid line ) , as was first shown by mahadevan ( 1998 ) . the two major results pointed out in @xmath151 are evident in this figure . first , there is no unique solution for the parameters @xmath64 and @xmath44 . the four hybrid models shown in figure 4 give indistinguishable spectra below the thermal peak , rendering it impossible to determine @xmath64 and @xmath44 from low - frequency spectral data alone . second , @xmath44 is extremely small in all models . a very small fraction of the energy in non - thermal electrons , with @xmath44 at most @xmath138 , is sufficient to produce all the observed emission at low frequencies . this means that the non - thermal electrons are a minor perturbation on the electron population , which is itself a minor perturbation on the more dominant ion population . we are therefore consistent when we compute the gas dynamics with a purely thermal model and ignore the non - thermal electrons for the dynamics . further justification of this assumption as well as implications of this constraint are discussed in @xmath46 . we can obtain additional constraints on the parameters @xmath44 and @xmath64 in sgr a@xmath3 by studying the infrared data . sgr a@xmath3 is quiet at infrared wavelengths , with a current upper limit of @xmath152 erg @xmath153 on the luminosity at 2.2 @xmath154 m ( eckart & genzel 1997 ; ghez et al . since @xmath155 decreases with increasing frequency for @xmath77 , we can not use ir data to constrain @xmath44 very strongly in these cases . on the other hand , if @xmath135 , electrons in an extended power - law distribution produce significant emission in the infrared . for example , if @xmath144 , we find that the maximum allowed fraction of energy in non - thermal electrons is @xmath156 . tighter bounds on the infrared flux will constrain the parameters @xmath44 and @xmath64 even more strongly . we note that imposing a maximum lorentz factor @xmath72 on the power - law electron distribution also has the effect of suppressing the high - frequency emission . therefore , for @xmath157 , we could alternatively use the ir data to constrain @xmath72 rather than @xmath44 . the @xmath158 m emission is produced predominantly by electrons with @xmath159 ( see @xmath13 ) placing a maximum lorentz factor at @xmath160 , if @xmath157 and @xmath161 . finally , we have considered adaf models with strong outflows following the ideas described in blandford & begelman ( 1999 ) , di matteo et al . ( 1999 ) , and quataert & narayan ( 1999 ) . we find that the constraints on @xmath44 and @xmath72 obtained above do not depend strongly on the presence of winds . we now investigate the effect of a power - law electron population on the size and shape of the radio image of an adaf . as a typical example , we consider an adaf model of sgr a@xmath3 with @xmath126 and @xmath162 and take a hybrid electron distribution with @xmath163 , @xmath148 which agrees well with the observed spectrum as shown in figure 4 . preliminary size measurements of sgr a@xmath3 are available at two frequencies ( lo et al . 1998 , krichbaum et al . 1998 ) and shape measurements may be possible in the near future . we note that although we use the parameters for a specific source , the qualitative results apply equally well to other massive black holes with different accretion rates . we define the image size as twice the radial distance from the center of the image to the point at which the specific intensity falls to half the value of the central intensity , i.e. , the fwhm of the radio map . figure 5 shows the predicted fwhm of the image as a function of frequency . the dashed curve corresponds to the case when all the electrons are thermal and the solid line to the case of a hybrid energy distribution . the two data points for sgr a@xmath3 are from lo et al . ( 1998 ) at 7 mm and krichbaum et al . ( 1998 ) at 1.4 mm . note that our radiative transfer code is one dimensional and can only handle spherical models , though adafs in general are oblate and are likely to appear elliptical in projection ( narayan & yi 1995a ) . in the comparison with observations , we use only the measured long axes of the images . the effect of power - law electrons is to enlarge the image size at long wavelengths @xmath164 , where the excess non - thermal emission , which extends outward of the surface of unit optical depth , is most prominent . this is a general result which holds true for all model parameters we have studied . the other related result is the steepening of the dependence of the image size on the frequency at long wavelengths . while for thermal electrons we find @xmath165 when we include power law electrons , we find @xmath166 this is in agreement with the preliminary result obtained by lo et al . ( 1998 ) when they combined their measurement of the intrinsic size at 7 mm with the results of krichbaum et al . ( 1998 ) at 1.4 mm , although the thermal and hybrid slopes are not distinguishable with the current data . we also calculate the effective brightness temperature @xmath167 at each frequency as @xmath168 . figure 6 shows the result . for the thermal model , @xmath167 measures the electron temperature at the photosphere . since the photosphere moves out with decreasing frequency , the brightness temperature falls . for the hybrid model , at lower frequencies , the contribution of the non - thermal electrons is large , the spectrum is non - thermal , and the resulting brightness temperature increases . this is another clear signature of a synchrotron - emitting hybrid electron population . we finally study the effect of the non - thermal electrons on the shape of the radio image . figure 7 shows the variation of the brightness temperature @xmath167 across the source , for the thermal and hybrid models of sgr a@xmath3 described above . we consider two wavelengths , 3.6 cm and 7 mm . the most striking feature of the hybrid case is the limb brightening seen at long wavelengths , a feature which is absent in the thermal models . thus , if the accretion flow contains non - thermal electrons , its image would look like a shell rather than a disk . this is again due to the different radial dependences of absorption ( which is predominantly thermal ) and emission ( which is mostly non - thermal ) at the frequencies where the non - thermal shoulder appears in the energy spectrum . because the total absorption falls off more steeply away from the center than the total emissivity , the image appears brighter for a range of impact parameters away from the center than at the center . in addition , due to the overall increase in the intensity of the emerging radiation in this same frequency range , the image looks brighter overall , with the brightness temperature @xmath167 of the hybrid case increasing to twice the thermal value in the center of the image at 3.6 cm . the limb brightening , along with the enhanced overall brightness at long wavelengths are signatures of a non - thermal population which may be accessible to observations . unfortunately , in the case of sgr a@xmath3 , precisely at the wavelengths where the effects are strongest , interstellar scattering blurs the observed image ( e.g. , lo et al . 1998 ) . it may be worthwhile to look for these effects in other sources where the scattering is less severe . the numerical studies presented in @xmath12 show that the presence of a small population of power - law electrons causes universal modifications to the spectrum : it introduces a shoulder at low frequencies and a power - law tail at high frequencies . we now attempt to associate each of these features of the hybrid spectrum with a specific range of lorentz factors of the non - thermal electrons . in doing so , we address three important issues . the first is the question of degeneracy : why is the low frequency shoulder in the spectrum degenerate to a combination of the power - law index @xmath64 and the energy content @xmath44 of the non - thermal electrons ? understanding the source of non - uniqueness is especially important in trying to extract the underlying electron distributions from spectral data and determining what we can conclude @xmath169 about these distributions . second , since introducing even a very small fraction of power - law electrons results in significant enhancement of the luminosity at low frequencies , existing radio data can be used to constrain the non - thermal energy fraction in these accretion flows . if only a part of the electron energy distribution is responsible for the excess emission , data can further constrain these specific parts of the particle distribution . the third question we address here is related to the absence of significant ir emission from the accretion flow around sgr a@xmath3 ( eckart & genzel 1997 ) , in the frequency range where the extended optically - thin non - thermal emission is expected . absence of emission at these wavelengths provides information about the underlying electron distributions by constraining either the power - law index @xmath64 or the maximum lorentz factor @xmath170 electrons can attain in these accretion flows . determining @xmath170 in a particle distribution may provide a better understanding of particle acceleration and cooling processes and timescales in hot accretion flows . we first investigate the relationship between the lorentz factor @xmath39 of an electron and the dominant frequency of the synchrotron emission it produces in the presence of a dominant thermal electron population . at very low frequencies , synchrotron emission comes from large radii in the flow where the temperatures are low , the non - thermal emission and absorption are efficient , and @xmath171 . however , at higher frequencies near and above the low - frequency shoulder in the spectrum , @xmath130 , as discussed earlier . for these frequencies , we can thus neglect @xmath37 . we make this key simplification only in this section . it simplifies the radiative transfer equation and allows us to compute separately the contribution to the intensity from each value of @xmath39 in the non - thermal distribution . we proceed by writing all quantities as integrals over the power - law electron distribution , @xmath172 @xmath173 and @xmath174 where @xmath175 is assumed to be a separable function of @xmath107 and @xmath39 given by @xmath176 ( note that , for this analysis , the non - thermal distribution does not have to be a power - law , though we have assumed this in order to compare this analysis directly to our numerical results ) . substituting equations ( 29)-(32 ) into the radiative transfer equation and rearranging terms , we obtain @xmath177 ~n^{\rm{pl}}_\gamma(\gamma ) d\gamma = 0.\ ] ] for this integral to vanish for any non - thermal electron distribution , the integrand must be identically zero , thus giving @xmath178 we solve this equation for a wide range of lorentz factors , @xmath179 , by choosing a specific radial profile of the non - thermal electron density @xmath180 corresponding to @xmath181 . we study the contribution of each lorentz factor to the different parts of the spectra . the total intensity for the power - law distribution takes the form @xmath182 and therefore , in order to study the true contribution of each @xmath39 to the total spectrum , we plot the integrand @xmath183 as a function of frequency for each lorentz factor . the result for @xmath184 , and 1000 ( for @xmath124 ) are shown in figure 8a . the curves give a good estimate of the individual contribution of each lorentz factor to the total spectrum . most of the contribution to the low - frequency shoulder comes from electrons with @xmath185 . emission from electrons with @xmath186 is already lower by an order of magnitude at those frequencies , and the emission completely dies off beyond @xmath187 . contribution to the high - frequency tail , on the other hand , starts around @xmath188 and increases with increasing electron energy . to get a more quantitative idea of which range of lorentz factors contributes to the three distinct regions of the spectrum , we show in figure 8b the above integrand as a function of @xmath39 for 3 frequencies , namely @xmath189 hz ( in the low - frequency shoulder ) , @xmath190 hz ( in the thermal peak ) , and @xmath191 hz ( in the high - frequency tail . we see that the emission at @xmath189 hz is primarily from electrons with @xmath192 ; the emission at @xmath190 hz is mostly from @xmath193 ( thermal electrons , while the emission at @xmath191 hz comes mostly from @xmath194 . to quantify this effect further , we plot in figure 8c the minimum and maximum lorentz factors for which the value of the integrand drops to half its maximum value for each frequency this gives us exactly the contributing range of lorentz factors to the emission at each frequency in the low - frequency shoulder and the high - frequency tail . we may summarize the results as follows . the low - frequency shoulder is caused by a narrow range of electron lorentz factors , @xmath195 the narrowness of the range explains why the spectra are degenerate to different combinations of @xmath64 and @xmath44 : a power - law distribution with small values of @xmath64 and small @xmath44 has nearly the same number of electrons in this narrow range of lorentz factors as one with a larger @xmath64 and a larger @xmath44 , thus producing the same emission in the shoulder . the high - frequency tail , however , is produced by electrons with a wide range of lorentz factors @xmath196 with higher frequencies coming from higher @xmath39 . intermediate lorentz factors @xmath197 emit predominantly at frequencies around the thermal peak and are overpowered by the thermal emission , and therefore do not have an observable feature in the hybrid spectrum . in this section , we study the energy flow through the thermal and non - thermal electron populations as well as the energy exchange between these two populations through synchrotron self - absorption and coulomb collisions . starting with the energy equations for the two populations and a parametrization of the energy input into each , we calculate the steady - state energy content of both as a function of radius assuming that the shapes of the distributions are known a priori . note that if the acceleration mechanism is known , the particle distributions can be calculated exactly ( see , e.g. , nayakshin & melia 1998 ) . here we simply assess the feasibility of a steady - state thermal / power - law hybrid distribution . we then extend our discussion of hybrid synchrotron spectra to cases where the parameter @xmath44 is not constant but is allowed to vary with radius . we start by writing the energy equations for the two populations . neglecting advection and diffusion which we estimate to be minor corrections , the energy balance for the power - law population reads @xmath198 while the thermal population obeys @xmath199 where @xmath200 and @xmath201 are the steady state energy contents of the thermal and power - law populations respectively , @xmath202 is the fraction of viscous energy that heats the thermal electrons , and @xmath203 is the corresponding fraction injected into the power - law population ( @xmath204 , where @xmath68 is defined in @xmath205 ) ; the time derivatives are set to zero in steady - state . the energy exchange ( thermalization ) timescale due to coulomb collisions of high energy electrons with the thermal bath is denoted by @xmath206 . the radiation term in equation ( [ eq : ennt ] ) corresponds to the energy loss by the non - thermal electrons via optically thin synchrotron emission ; because the non - thermal absorption coefficient is negligible throughout the flow , the non - thermal electrons do not gain energy by absorption of synchrotron radiation . the corresponding term in equation ( [ eq : enth ] ) , on the other hand , describes the heating of the thermal electrons by the local radiation field that is in excess of the blackbody limit . when the radiation energy density is a blackbody , there is locally no heating or cooling , but as the radiation energy density is above this limit due to emission by the non - thermal electrons , the thermal electrons are heated by absorbing this total emission . note that this is the same energy exchange mechanism invoked in the synchrotron boiler process ( ghisellini , guilbert , & svensson 1988 ; ghisellini , haardt , & fabian 1993 ; ghisellini , haardt , & svensson 1998 ) . the transport of energy due to non - local synchrotron self - absorption is likely to be small and have little effect on the spectra as the photon mean free path becomes very short very rapidly along a radial path and thus the radiation emitted in the optically thick regions of the flow is reabsorbed locally . this is due to the very steep radial dependence of the synchrotron absorption coefficient . there are several issues we would like to address regarding the relative importance of thermal and non - thermal electrons in the flow and the energy exchange between the two populations . first , the quantity @xmath207(r)$ ] determines as a function of radius the relative rate of heating of non - thermal electrons compared to thermal electrons . second , the energy exchange between the populations ( heating of the thermal electrons by non - thermal electrons ) proceeds both via the coulomb term and the absorption of the synchrotron photons emitted by the power - law populations ( the second term of equation ( [ eq : enth ] ) . therefore the magnitude of these terms relative to the viscous heating of the thermal electrons needs to be assessed . finally , the importance of the thermalization of the power - law distribution due to the coulomb term also needs to be understood . we first estimate the relative magnitude of the terms in equation ( [ eq : ennt ] ) . the ratio of the energy loss of the power - law electrons due to coulomb collisions ( the last term ) to the energy emitted in synchrotron photons ( the second to last term ) determines whether coulomb collisions or synchrotron self - absorption is the dominant mechanism by which the non - thermal electrons cool . the hybrid spectra presented in this paper have been computed under the assumption that the power - law electrons lose energy primarily through synchrotron emission , and this needs to be checked for consistency . for a given electron velocity @xmath28 in the power - law distribution , the energy exchange rate with the thermal electrons is given by ( nayakshin & melia 1998 ) @xmath208 where @xmath209 is the coulomb logarithm , @xmath210 is the average thermal lorentz factor , @xmath211 is the thomson mean - free time and @xmath212 the electron density . the rate of energy loss of an electron of lorentz factor @xmath39 emitting synchrotron radiation in a region of magnetic field b is ( rybicki & lightman 1979 ) @xmath213 where @xmath214 is the time for the electron to lose half its energy . using the analytic expressions for the electron density and magnetic field strength given in @xmath215 and substituting values for the coefficients and parameters appropriate for sgr a@xmath3 , we find for an electron of a given @xmath39 : @xmath216 where r is the radius in scwarzschild units as usual . most of the emission in the low - frequency shoulder in the spectrum originates from around @xmath217 and the radiation is produced by electrons with @xmath218 ( cf . 8c and the appendix ) . we thus see that the ratio in equation ( [ eq : ratio ] ) is at most a few percent . this shows that coulomb collisions play a negligible role in the cooling of non - thermal electrons . it also shows that that the heating of thermal electrons by coulomb collisions is unimportant compared to energy exchange via synchrotron emission and absorption . we therefore neglect the coulomb term in the following calculations . the power - law electrons then obey a simple energy balance between the injected energy through viscous dissipation and the energy loss through synchrotron emission . the latter simply is the total frequency- and angle - integrated synchrotron emissivity of the non - thermal electrons because this population is optically thin to synchrotron emission . therefore , once @xmath219 is specified , it is possible to calculate the parameter @xmath44 , which describes the steady - state energy content of the power - law electrons , as a function of radius . conversely , if we specify @xmath45 as we did in the previous sections , it is possible to compute the energy @xmath203 that would need to be injected into the non - thermal electrons as a function of radius . we first study the latter case for @xmath220 constant , as this is the assumption we have made in most of this paper . figure 9 shows @xmath221 as a function of radius for one of the hybrid models of sgr a * with @xmath222 , and @xmath124 . the figure demonstrates that the energy input into the non - thermal electrons never exceeds @xmath223 of the heating of the thermal electrons and in fact does not exceed @xmath5 at those radii that contribute to the low - frequency shoulder in the spectrum @xmath224 . thus , at no radius in the flow do the power - law electrons become energetically more important than , or even comparable to , the thermal electrons . this may seem like a surprising result considering that the non - thermal radiation clearly dominates over the thermal emission in the low - frequency shoulder of the spectrum and in the high - frequency tail . however , the synchrotron emissivities of the thermal and power - law populations are different by many orders of magnitude at these frequencies . ( optically thin synchrotron emission from a power - law population peaks at a much higher intensity and at a lower frequency than the thermal emission for the same energy density and magnetic field strength . ) therefore , even when the non - thermal population has less energy content than the thermal population , and even when a fraction of the non - thermal radiation is absorbed by the thermal electrons at frequencies where the flow is optically thick , the escaping radiation can still be dominated by the non - thermal emission and the intensity can be much above the blackbody limit . this is the situation in the low - frequency shoulder of the spectrum . in the high - frequency tail , the only particles that are energetic enough to produce the radiation are the non - thermal electrons , and since the flow is optically thin all the radiation escapes freely . incidentally , the fact that the energy input into the non - thermal electrons does not exceed @xmath223 of the heating of the thermal electrons ( and even this level is reached only over a small range of radius close to the black hole ) shows that the dynamics of the flow is not affected by the presence of non - thermal electrons . we also consider the other mechanism of energy exchange between the two populations , namely synchrotron self - absorption by thermal electrons . the second term in equation ( [ eq : ennt ] ) measures the maximum energy that can be radiated by the non - thermal electrons as a function of radius . this term also represents an upper limit on the energy that can be transferred to the thermal population via self - absorption . this term is bounded from above by @xmath225 according to equation ( [ eq : ennt ] ) . since @xmath221 never exceeds @xmath223 as shown in figure 9 , heating of the thermal electrons by absorbing synchrotron emission from non - thermal electrons can @xmath226 introduce a @xmath223 correction to the thermal energy equation ; the correction is only a few percent at large radius . this again justifies neglecting the contribution of the non - thermal electrons to the thermodynamic properties of the flow . finally we note that the calculations in the previous sections considered a simple parametrization of the non - thermal population via the quantity @xmath44 , which measures the fraction of the electron energy in steady state that is present in non - thermal electrons . now , @xmath44 is a secondary quantity whose value depends on the balance between non - thermal heating and cooling . it would perhaps be more useful to parametrize the model by specifying the non - thermal heating parameter @xmath203 . to this end , we consider models in which the energy injection rate varies as a power - law in radius , @xmath227 , and show results for @xmath228 between 0 and 1 . we pick the normalization of @xmath203 at the inner edge of the flow such that the resulting @xmath45 is comparable to the values used in @xmath12 and is in accord with the results of figure 9 . in figure 10a , we show @xmath45 for @xmath124 and three different choices of q : 0 , @xmath229 , and @xmath230 . for none of the three cases do we see very large variations in the resulting @xmath45 versus radius ; we find that @xmath45 rises with radius for @xmath231 and falls for larger values of @xmath228 . figure 10b shows the spectra corresponding to the three models of @xmath203 . we see that the two universal features of a hybrid spectrum , the low - frequency shoulder and the high - energy tail , are present for all values of @xmath228 . thus , super - thermal emission is not an artifact of specific assumptions or parameter values of our models but is indeed a robust signature of a hybrid population . the slope of the low - frequency shoulder depends very mildly on the particular model of @xmath232 . in this paper we considered hot accretion flows around supermassive black holes , using the adaf model as a typical example . such hot flows are expected to occur at low mass - accretion rates . we assume that a fraction of the viscous dissipation energy in the accretion flow goes into accelerating electrons to a non - thermal power - law distribution . we find that a power - law tail of high energy electrons can be sustained in such flows ; neither coulomb collisions nor synchrotron self - absorption is able to thermalize the power - law electrons . assuming that there are no other thermalizing mechanisms ( e.g. , collective plasma modes ) , we calculate the resulting hybrid thermal / non - thermal spectrum from such a plasma . the presence of even a population of non - thermal electrons gives rise to two universal and prominent features in the synchrotron spectrum : a low - frequency shoulder and a high - frequency tail ( fig . these features were identified by mahadevan ( 1998 ) who considered a specific mechanism ( via pion decay ) for the production of power - law electrons . even if only a small fraction of the total steady - state electron energy is in the non - thermal power - law component , we find that there is significant super - thermal emission in the low - frequency shoulder and the high frequency tail of the spectrum . furthermore , each of these universal features can be associated with a specific range of lorentz factors of the emitting electrons . the low - frequency shoulder is emitted by electrons with lorentz factors @xmath233 , while the high - energy tail is emitted by electrons with @xmath234 ( fig.8 ) . since the power - law electrons cause significant emission at low frequencies , comparing this low - frequency shoulder to data can provide stringent constraints on the fraction of the electron energy that is present in a non - thermal population . in the case of sgr a@xmath3 ( fig . 4 ) , we conclude that , in steady - state , electrons even with intermediate lorentz factors ( @xmath39 as low as 10 ) make up only a very small fraction of the electron population . since coulomb collisions and synchrotron self - absorption are ineffective in thermalizing these flows , this means that either the energy injected into a non - thermal population , @xmath235 , is small as in the calculations described in @xmath46 or there is some other efficient thermalization mechanism which plays a role in these flows . the study on sgr a@xmath3 needs to be extended to other sources before we can judge how universal the result is . here it is important to emphasize two points regarding the constraints on non - thermal electrons obtained from the spectrum . first , for the part of the spectrum below the thermal peak , there is a mapping between each frequency in the spectrum and the thin shell of radii which predominantly emits at that frequency . this is due to the steep radial dependence of the synchrotron absorption that causes a sharp transition between optically thick and optically thin regions in the flow . therefore , emission at each frequency comes primarily from a specific radius in the flow and can be used to probe the non - thermal content in a shell that extends from that radius out to a somewhat larger radius . ( mathematically , radiation emerging at each frequency comes from the entire range from the radius where the optical depth is unity out to infinity , but because the total emissivity also drops rapidly with radius , the biggest contribution to the intensity integral comes from a relatively narrow shell ) . as a result , the constraints on the non - thermal distribution obtained from each frequency of the low - frequency shoulder apply only to electrons within this shell . inside the radius corresponding to optical depth unity , the emission is heavily self - absorbed and the electron distribution there is inaccessible to observations . second , the constraints on the non - thermal energy content which we have obtained for sgr a@xmath3 would not be invalidated if the observed low - frequency shoulder originated in a separate , extended region such as a jet or an outflow exterior to the accretion flow . in that case , the observed fluxes may be treated as upper limits to the emission from the inner accretion flow . this would imply that the non - thermal emission from the accretion flow is even smaller than in our models and the resulting constraints on the fraction of non - thermal electrons would become only tighter . when the non - thermal electron distribution is a power law , three parameters of the model and three corresponding pieces of physics can be extracted by comparing hybrid emission models to spectral data . the first parameter is @xmath44 , or equivalently @xmath203 ( cf.5 ) , which measures the fraction of the dissipated energy that goes into the non - thermal electrons . the second parameter is @xmath64 , the slope of the power - law energy distribution ( eq . 4 ) . this can be uniquely determined by measuring the spectral slope of the high - frequency tail in the spectrum ( provided there is no competing emission via compton scattering at these wavelengths , cf . 1 ) . the value of @xmath64 may provide an indication of the mechanism by which the electrons in a hot accretion flow are accelerated into a power - law tail . the third parameter , which is more significant for distributions with @xmath135 , is @xmath72 , the maximum lorentz factor to which electrons are accelerated . if this parameter can be determined by an ir cutoff in the spectrum , it would provide information on the ratio of the acceleration timescale to the synchrotron cooling timescale . due to the degeneracy of the low - frequency shoulder to the parameters @xmath44 and @xmath64 ( fig . 4 ) , this segment of the spectrum by itself is not sufficient to constrain the two parameters . however , @xmath44 and @xmath64 may be determined uniquely if we have information on both the low - frequency shoulder and the high - frequency tail . for this to work , there should be no contamination from a jet or outflow to the observed radiation in the low - frequency shoulder or from compton scattering in the high frequency tail . we showed in this paper that the presence of a non - thermal population of electrons also has measurable effects on the shape of the image of the source , and the size of the accretion flow , as a function of frequency ( figs . because the frequency dependence of the image size is strongly affected by the fraction of non - thermal electrons in the flow , measuring the size of accretion flows at multiple wavelengths can provide a quantitative constraint on the non - thermal electron energy content . moreover , the brightness temperature of the source as a whole , as well as the variation of the brightness temperature across the source , at long wavelengths behave differently for a hybrid plasma compared to a purely thermal population of electrons . finally , at the frequencies corresponding to the low - frequency shoulder , we show that there would be limb brightening of the image . all these image - related signatures may help to identify the presence of a non - thermal population of electrons , and the parameters of the corresponding energy distribution . we thank eliot quataert , anthony aguirre , tiziana di matteo , and james cho for many useful discussions and comments on the manuscript . we also thank s. nayakshin , the referee , for many useful comments and suggestions that improved the presentation of the paper . d. p. was supported by a post - doctoral fellowship of the smithsonian institution . this work was supported in part by nsf grant ast 9820686 . in this appendix , we provide analytic formulae for the slope of the spectrum below and above the thermal peak and the relative importance of non - thermal electrons as a function of mass , accretion rate , power law index @xmath64 , and non - thermal energy content @xmath44 . for convenience , we use the self - similar solution of adafs to obtain these analytic estimates . we begin by presenting the self - similar adaf solution developed by narayan @xmath236 yi ( 1994 , 1995b ) . the self - similar solution describes the local properties of the accretion flow as a function of the black hole mass @xmath237 , the mass accretion rate @xmath238 , the radius @xmath239 , the viscosity parameter @xmath110 , the ratio of gas pressure to the total pressure @xmath28 , and the fraction of viscously dissipated energy that is advected inwards @xmath240 . in terms of the scalings introduced in @xmath215 , the height - averaged electron number density @xmath212 , the magnetic field strength b , and the dimensionless proton temperature @xmath241 of the accretion flow are : @xmath242 @xmath243 and @xmath244 where @xmath245 and @xmath246 the coefficients @xmath247 and @xmath248 are defined in narayan @xmath236 yi ( 1995b ) and are related to the adiabatic index of the gas and the fraction of advected energy @xmath240 . for the cases of interest here , @xmath249 and @xmath250 the remaining parameters were specified in @xmath215 . we now use the self - similar solution to obtain expressions for the cyclotron frequency and the harmonic number for synchrotron emission . since we focus our applications to galactic nuclei and very large mass black holes where the synchrotron frequencies of interest are in the radio regime , we scale the frequency as @xmath251 the magnetic field strength is then @xmath252 and thus fundamental cyclotron frequency defined in @xmath253 becomes : @xmath254 the ion temperature retains its virial value throughout the flow and is given , in dimensionless units @xmath255 , by : @xmath256 when we set the parameter values to those used in the simulations . the behaviour of the electron temperature on the other hand is more complicated and depends on the detailed balance of heating and cooling at each radius . qualitatively , electron temperature starts out virial and equal to the ion temperature at large radii @xmath257 , but increases less steeply than the ion temperature at smaller radii due to cooling by mainly synchrotron emission . because it is not possible to derive the exact electron temperature as a function of radius analytically , we provide instead numerical fits to the average radial dependence of electron temperature @xmath258 since synchrotron absorption coefficient is a steep function of radius , we can assume that most of the optically thick emission comes from a narrow range of radii around the radius with optical depth equal to unity . due to the complicated nature of the expression for the thermal synchrotron absorption which does not allow analytic integration , we numerically determine the dependence of this @xmath259 radius , @xmath260 on mass , mass accretion rate and frequency . the best fit gives @xmath261 substituting @xmath260 into @xmath262 and @xmath263 we get @xmath264 and @xmath265 finally , we compute the harmonic number defined in @xmath253 : @xmath266 @xmath267 which provides all the necessary synchrotron expressions as a function of black hole mass , mass accretion rate and frequency . we can now derive analytic estimates for the contribution of the non - thermal emission to the low - frequency shoulder and the high - frequency tail . in the optically thick shoulder , self - absorption of the synchrotron emission is done predominantly by the thermal electrons , and we find that absorption due to non - thermal electrons are negligible down to a frequency of @xmath268 as discussed above . thus the source function will take the form @xmath269 where @xmath34 is synchrotron emissivity per unit volume given in @xmath253 . for the analytic approximations , we will use a simplified form of the thermal emissivity . we first take the fully relativistic limit derived first by pacholczyk , which is equivalent to setting @xmath97 and @xmath98 equal to 1 in @xmath270 . we further neglect the second and third terms of the sum altogether by setting @xmath271 and @xmath98 equal to zero as these terms provide only a very small correction unimportant for the present purposes . thus , in simplified form , the ratio of the two emissivities is given by @xmath272 where @xmath273 and all other quantities are as defined as before . the non - thermal emission dominates over the thermal emission in the spectrum when @xmath274 . this ratio can be easily computed using the expressions for @xmath275 , and @xmath276 evaluated at the @xmath277 radius given above . the power law emission beyond the thermal peak is optically thin and hence the excess in this region depends sensitively on the power - law index as expected and is simply given by the optically thin synchrotron emission from the power - law electrons . the amount of emission is determined simply by @xmath35 . we finally note that , for the small values of @xmath44 considered here , thermal emission dominates at the thermal peak . the strong dependence of the position and the normalization of the thermal peak on @xmath278 and @xmath279 is discussed by mahadevan ( 1997 ) . the shape of the low - frequency shoulder is dominated by the emission of the power - law population and absorption by the thermal electrons . in order to determine the spectral slope of this segment , we therefore first calculate the hybrid source function @xmath280 at the surface of unit optical depth . to convert this into a luminosity , we multiply the source function by the area of the @xmath281 surface . if we write the shoulder of the spectrum as @xmath282 , the spectral slope @xmath59 is then given by @xmath283 which gives the canonical value of @xmath284 for the mass and mass - accretion rate of sgr a@xmath3 and scales ( weakly ) according to the expression above for other masses and accretion rates . one can immediately see that the spectral slope depends very weakly on p , only as @xmath285 , thus demonstrating again the independence of the spectral slope of the low - frequency shoulder on the shape of the electron energy distribution . the slope of the optically thin power law emission is a well - known result , @xmath286 which in @xmath155 for @xmath135 , flat for p=3 and falls off for @xmath77 . it is independent of the black hole mass and the accretion rate .

we consider a class of fully - nonlinear fermi - pasta - ulam ( fpu ) lattices , consisting of a chain of particles coupled by fractional power nonlinearities of order . this class of systems incorporates a classical hertzian model describing acoustic wave propagation in chains of touching beads in the absence of precompression . we analyze the propagation of localized waves when is close to unity . solutions varying slowly in space and time are searched with an appropriate scaling , and two asymptotic models of the chain of particles are derived consistently . the first one is a logarithmic kdv equation , and possesses linearly orbitally stable gaussian solitary wave solutions . the second model consists of a generalized kdv equation with hlder - continuous fractional power nonlinearity and admits compacton solutions , i.e. solitary waves with compact support . when , we numerically establish the asymptotically gaussian shape of exact fpu solitary waves with near - sonic speed , and analytically check the pointwise convergence of compactons towards the limiting gaussian profile .

1307.3837

the problem of analyzing the response of a nonlinear lattice to a localized disturbance arises in many applications , such as the study of stress waves in granular media after an impact @xcite , the excitation of nonlinear oscillations in crystals by atom bombardment @xcite , or the response of nonlinear transmission lines to a voltage pulse @xcite . several important dynamical phenomena can be captured by the fermi - pasta - ulam ( fpu ) model @xcite consisting of a chain of particles coupled by a pairwise interaction potential @xmath3 . the dynamical equations for a spatially homogeneous fpu chain read @xmath4 where @xmath5 is the displacement of the @xmath6th particle from a reference position . system ( [ nc ] ) can be rewritten in terms of the relative displacements @xmath7 and particle velocities @xmath8 as follows @xmath9 the dynamical evolution of localized solutions of ( [ ncrelham ] ) is strongly influenced by the properties of the interaction potential @xmath3 . in its most general form , the interaction potential satisfies @xmath10 where @xmath0 and @xmath11 . in the work @xcite , the dispersive stability of the zero equilibrium state is proved for @xmath12 and @xmath13 , i.e. for sufficiently weak nonlinearities near the origin . more precisely , the amplitude ( i.e. supremum norm ) of the solution of the fpu lattice ( [ ncrelham ] ) goes to @xmath14 when @xmath15 for all initial conditions sufficiently small in @xmath16 , where @xmath16 denotes the classical banach space of bi - infinite summable sequences . in contrast , in many situations nonlinear effects are strong enough to compensate dispersion , yielding the existence of coherent localized solutions of the fpu lattice ( [ ncrelham ] ) such as solitary waves propagating at constant speed , or time - periodic breathers ( see e.g. @xcite for a review ) . the first existence result for solitary waves in a general class of fpu lattices was obtained by friesecke and wattis @xcite , when @xmath3 has a local minimum ( not necessarily strict ) at the origin and is superquadratic at one side ( see also @xcite and references therein ) . in addition , the existence of solitary waves near the so - called long wave limit was established in @xcite for smooth ( @xmath17 ) potentials @xmath3 . more precisely , for @xmath18 and @xmath19 ( i.e. @xmath20 in ( [ potential - v ] ) ) , there exists a family of small amplitude solitary waves parameterized by their velocity @xmath21 , where @xmath22 defines the sound velocity " of linear waves . these solutions take the form @xmath23 where @xmath24 and @xmath25 . in particular , these solitary waves decay exponentially in space and broaden in the limit of vanishing amplitude . equivalently , one has @xmath26 where @xmath27 , @xmath28 , and @xmath29 is a solitary wave solution of the korteweg de vries ( kdv ) equation @xmath30 more generally , the solutions @xmath31 \rightarrow \mathbb{r}$ ] of the kdv equation ( [ kdv ] ) yield solutions of the fpu system of the form ( [ approxkdv ] ) , valid on a time scale of order @xmath32 @xcite . in addition , the nonlinear stability of small amplitude fpu solitary waves was proved in @xcite , as well as the existence and stability of asymptotic @xmath33-soliton solutions @xcite . these results allow to describe in particular the propagation of compression solitary waves in homogeneous granular chains under precompression @xcite . another interesting case corresponds to fully - nonlinear interaction potentials , where @xmath34 ( which corresponds to a vanishing sound velocity , that is , @xmath35 ) and @xmath3 has a local minimum at the origin . a classical example is given by the hertzian potential @xmath36 with @xmath0 , where we denote by @xmath37 the heaviside step function . this potential describes the contact force between two initially tangent elastic bodies ( in the absence of precompression ) after a small relative displacement @xmath38 @xcite . the most classical case is obtained for @xmath39 and corresponds to contact between spheres , or more generally two smooth non - conforming surfaces . more recently , granular chains involving different orders of nonlinearity have attracted much attention , see @xcite and references therein . in particular , experimental and numerical studies on solitary wave propagation have been performed with chains of hollow spherical particles of different width @xcite and chains of cylinders @xcite , leading to different values @xmath1 in the range @xmath40 ( see also @xcite for other systems with @xmath1 close to unity ) . the propagation of stationary compression pulses in the fpu lattice ( [ nc ] ) with potential ( [ vhertz ] ) for @xmath39 was first analyzed by nesterenko @xcite . these results rely on a formal continuum limit and provide approximate solitary wave solutions with compact support . an alternate continuum limit problem has been introduced in @xcite for arbitrary values of @xmath0 , leading to different ( compactly supported ) approximations of solitary waves . the existence of exact solitary wave solutions of the fpu lattice ( [ nc ] ) with potential ( [ vhertz ] ) follows from the general result of friesecke and wattis @xcite mentioned previously ( see also @xcite ) . the width of these solitary waves is independent of their amplitude due to the homogeneous nonlinearity of the hertzian potential . in addition , the fully - nonlinear character of the hertzian potential induces a doubly - exponential spatial decay of solitary waves @xcite . while the above analytical results provide useful informations on strongly localized solitary waves , they are not entirely satisfactory for several reasons . first of all , the existence result of @xcite does not provide an approximation of the solitary wave profile , and the approximations available in the literature @xcite rely on a long wave " assumption that is not justified ( for example , the solitary waves considered in @xcite are approximately localized on five particles ) . in addition , the dynamical properties of solitary waves in fully - nonlinear fpu lattices are not yet understood . indeed , no mathematical results are available concerning their stability , the way they are affected by lattice inhomogeneities , or the existence of @xmath33-soliton solutions . another interesting problem is to characterize the excitation of one or several solitary waves from a localized initial perturbation @xcite . for @xmath41 and small amplitude long waves , this problem can be partially analyzed in the framework of kdv approximation by using the inverse scattering transform methods @xcite , but such reduction is presently unavailable for fully - nonlinear fpu lattices . these questions are important for the analysis of impact propagation in granular media , and more generally for the design of multiple impact laws in multibody mechanical systems @xcite . in this paper , we attack the problem by considering a suitable long wave limit of fully - nonlinear fpu lattices . we consider the fpu lattice ( [ nc ] ) with the homogeneous fully - nonlinear interaction potential @xmath42 with @xmath0 . obviously , all solutions @xmath43 of the fpu lattice ( [ nc ] ) with the potential ( [ vhom ] ) are also solutions of the hertzian fpu lattice ( [ nc ] ) and ( [ vhertz ] ) . the problem can be rewritten in terms of the relative displacements in the following way @xmath44 where we denote @xmath45 and @xmath46 is the discrete laplacian . for approximating the temporal dynamics of ( [ ncrel ] ) in a continuum limit , fully - nonlinear versions of the boussinesq equation considered in @xcite possess serious drawbacks , since they may lead to blow - up phenomena in analogy with the classical bad " boussinesq equation @xcite . in section [ bouss ] , we numerically show that these models introduce artificial dynamical instabilities with arbitrarily large growth rates , which suggests ill - posedness of these equations @xcite . instead of using a boussinesq - type model , we then formally derive a logarithmic kdv ( log - kdv ) equation as a modulation equation for long waves in fully - nonlinear fpu lattices , obtained in the limit @xmath2 ( section [ deriv ] ) . the log - kdv equation takes the form @xmath47 and provides approximate solutions @xmath48 of the original fpu lattice ( [ ncrel ] ) for @xmath49 , @xmath50 , and @xmath51 . the log - kdv equation ( [ kdv-1 ] ) admits gaussian solitary wave solutions ( section [ trav ] ) , which have been previously identified as solutions of the stationary logarithmic nonlinear schrdinger equation ( log - nls ) in the context of nonlinear wave mechanics @xcite . closer to our case , gaussian homoclinic solutions have been also found to approximate the envelope of stationary breather solutions in newton s cradle ( i.e. system ( [ nc ] ) and ( [ vhertz ] ) with an additional on - site potential ) in the limit @xmath2 @xcite . in section [ trav ] , we numerically check that solitary wave solutions of the hertzian fpu lattice with velocity @xmath52 converge towards gaussian approximations when @xmath53 is fixed and @xmath2 . these solitary waves have velocities close to unity , which corresponds to the value of sound velocity in the linear chain with @xmath54 . in addition , we check that the fpu solitary waves are well approximated by the compacton solutions derived in @xcite when @xmath55 $ ] . to go beyond the stationary regime , we check numerically that the gaussian approximation captures the asymptotic shape of a stable pulse forming after a localized velocity perturbation in the hertzian fpu lattice ( [ nc])-([vhertz ] ) with @xmath56 ( section [ gauss ] ) . consistently with the above dynamical simulations , we prove in section [ lstab ] the linear orbital stability of gaussian solitary waves for the log - kdv equation . our analysis makes use of a suitable convex conserved lyapunov function , but negative index techniques developed in recent works @xcite for kdv - type equations would also apply . the link between gaussian solitary waves and compactons is made explicit is section [ comprev ] , where we check the pointwise convergence of the compacton solutions of @xcite towards gaussian profiles when @xmath2 . in addition , following the methodology developed in section [ deriv ] , we derive from the fully - nonlinear fpu lattice a generalized kdv equation with hlder - continuous nonlinearity ( h - kdv ) : @xmath57 when @xmath2 , the h - kdv equation ( [ kdv-2 ] ) is consistent with the fpu lattice in the sense that each solution to this equation almost " satisfies ( [ ncrel ] ) up to a small residual error . equation ( [ kdv-2 ] ) admits explicit compacton solutions whose form is close to the compactons obtained in @xcite with the use of a boussinesq type model . when @xmath2 , these solutions converge towards the gaussian solitary waves studied in section [ trav ] , and thus they provide an ( asymptotically exact ) approximation of fpu solitary waves with near - sonic speed . this result sheds a new light on the compacton approximations for fpu solitary waves heuristically derived in the literature @xcite . another interest of the h - kdv equation lies in the ( non - differentiable ) hlder - continuous nonlinearity @xmath58 which allows for the existence of compactons . this type of degeneracy is quite different from the classical feature of compacton equations which incorporate degenerate nonlinear dispersion @xcite . we finish this paper with a summary of our results and a discussion of several open questions concerning the qualitative dynamics of the log - kdv and h - kdv equations and their connections with fully - nonlinear fpu chains ( section [ alt ] ) . fully nonlinear boussinesq equations have been introduced in @xcite as formal continuum limits of fpu chains with hertzian - type potentials . in @xcite , the continuum limit is performed on system ( [ nc ] ) with potential ( [ vhom ] ) describing particle displacements , whereas @xcite considers system ( [ ncrel ] ) for relative displacements . in what follows , we discuss the continuum limit introduced in @xcite , which takes a slightly simpler form than the system derived in @xcite . the fully nonlinear boussinesq equation introduced in @xcite takes the form @xmath59 where @xmath60 denotes an approximation of a solution @xmath61 of ( [ ncrel ] ) . the right side of ( [ b ] ) is obtained by keeping the first two terms of the formal taylor expansion of the discrete laplacian in ( [ ncrel ] ) @xmath62 this truncation is purely formal , but a numerical justification is presented in @xcite in the particular case of solitary wave solutions . more precisely , the solitary waves @xmath63 , @xmath64 of ( [ ncrel ] ) are numerically compared with solitary wave solutions @xmath65 , @xmath66 of ( [ b ] ) . for this class of solutions , equation ( [ b ] ) reduces to a fourth order ordinary differential equation , which can be integrated twice and leads to @xmath67 whereas equation ( [ ncrel ] ) reduces to the differential advance - delay equation @xmath68 with @xmath69 . the wave velocity can be normalized to unity due to a scaling invariance of the fpu system ( [ nc ] ) with homogeneous potential ( [ vhom ] ) ( or hertzian potential ( [ vhertz ] ) ) , namely each solution @xmath61 generates a one - parameter family of solutions @xmath70 with @xmath71 ( the same scaling invariance exists in system ( [ b ] ) ) . according to the numerical computations presented in @xcite , the solitary wave of the differential advance - delay equation ( [ diffadvdel - eq ] ) is well approximated by the compactly supported solitary wave of the differential equation ( [ compacton - eq ] ) for @xmath72 , and the discrepancy increases with @xmath1 . the compacton solution of ( [ compacton - eq ] ) found in @xcite takes the form @xmath73 where @xmath74 in what follows , we reexamine the consistency of ( [ ncrel ] ) and ( [ b ] ) from a dynamical point of view , by analyzing the spectral stability of compactons . linearizing ( [ b ] ) at the compacton @xmath75 in the reference frame travelling with unit velocity , we use the ansatz @xmath76 , where @xmath77 is the spectral parameter and @xmath78 is the perturbation term . we arrive at the spectral problem @xmath79 where @xmath80}(z)$ ] and @xmath81 denotes the characteristic function . one can notice that @xmath82 and @xmath83 vanish at the end points @xmath84 of the compact support of @xmath75 . we look for eigenvectors in the hilbert space @xmath85 since @xmath86 and @xmath87 are discontinuous at @xmath84 , this yields the condition @xmath88 this allows us to reduce the eigenvalue problem ( [ b - lin ] ) to the compact interval @xmath89 $ ] with boundary conditions ( [ b - bc ] ) , and approximate the spectrum with the standard finite difference method ( we have used second - order difference approximations for derivatives and @xmath90 grid points ) . if there exist eigenvalues with @xmath91 , then the solitary wave is spectrally unstable . if all the eigenvalues are located at the imaginary axis , then the solitary wave is called spectrally stable . figure [ fig - spectrum ] shows the complex eigenvalues @xmath77 of the spectral problem ( [ b - lin])-([b - bc ] ) for @xmath92 . the spectrum is invariant under @xmath93 and @xmath94 ( note the presence of a small number non - symmetric eigenvalues , which originate from numerical errors ) . we find the existence of unstable eigenvalues for all values of @xmath0 considered , and the eigenvalues approach the real line far from the origin ( this part of the spectrum is not visible in the first two panels of figure [ fig - spectrum ] ) . consequently , these results imply the spectral instability of the compacton ( [ compacton ] ) in system ( [ b ] ) . note that the usual notion of instability may not be well defined , since the evolution problem ( [ b ] ) may not be well posed . indeed , our numerical results indicate that the spectrum of ( [ b - lin])-([b - bc ] ) is unbounded in the positive half - plane ( in fact at both sides of the imaginary axis ) , and thus the linearized evolution problem may be ill - posed . we conjecture that ill - posedness occurs also in system ( [ b ] ) , in analogy with ill - posedness results recently obtained in @xcite for certain nonlinear degenerate dispersive equations . along these lines , it is interesting to consider the limit case of the spectral problem ( [ b - lin])-([b - bc ] ) when @xmath2 . since @xmath95 for all @xmath96 and @xmath97 , the limiting spectral problem possesses constant coefficients and is defined on the entire real line . using the fourier transform , one can compute the ( purely continuous ) spectrum explicitly , which yields @xmath98 this limit case is represented in figure [ fig - spectrum ] ( red curves ) . the spectrum being unbounded in the positive half - plane , the corresponding linear evolution problem is then ill - posed . the above instability phenomena are not physically meaningful since the solitary waves are known to be stable from simulations of impacts in hertzian chains @xcite . in equation ( [ cont - spectrum ] ) obtained in the limit @xmath99 , these instabilities occur for short wavelengths ( with @xmath100 ) , whose dynamical evolution can not be correctly captured by the continuum limit ( [ b ] ) . in the next section , we derive a different asymptotic model free of such artificial instabilities . eigenvalues of the spectral problem ( [ b - lin])-([b - bc ] ) ( blue dots ) for @xmath101 ( top left ) , @xmath102 ( top right ) , and @xmath103 ( bottom ) . the red dotted curves show the continuous spectrum ( [ cont - spectrum ] ) obtained in the limit case @xmath2 . , title="fig : " ] eigenvalues of the spectral problem ( [ b - lin])-([b - bc ] ) ( blue dots ) for @xmath101 ( top left ) , @xmath102 ( top right ) , and @xmath103 ( bottom ) . the red dotted curves show the continuous spectrum ( [ cont - spectrum ] ) obtained in the limit case @xmath2 . , title="fig : " ] eigenvalues of the spectral problem ( [ b - lin])-([b - bc ] ) ( blue dots ) for @xmath101 ( top left ) , @xmath102 ( top right ) , and @xmath103 ( bottom ) . the red dotted curves show the continuous spectrum ( [ cont - spectrum ] ) obtained in the limit case @xmath2 . , title="fig : " ] a classical model introduced to correct artificial short wavelength instabilities of ( [ b ] ) corresponds to the regularized boussinesq equation @xmath104 ( see e.g. @xcite ) . this model has the inconvenience of altering the spatial decay of solitary waves in the case of fully - nonlinear interaction potentials . indeed , looking for traveling wave solutions @xmath65 , @xmath66 and integrating equation ( [ rb ] ) twice , one obtains @xmath105 after setting two integration constants to @xmath14 . equation ( [ rbs ] ) admits nontrivial symmetric homoclinic solutions @xmath106 satisfying @xmath107 , corresponding to solitary wave solutions of ( [ rb ] ) . these solutions decay exponentially in space , which is too slow compared with the superexponential decay of the solitary wave solutions of ( [ diffadvdel - eq ] ) . in order to pass to the limit @xmath2 for long waves , it is convenient to rewrite ( [ ncrel ] ) in the form @xmath108 where @xmath109 ( uniformly in @xmath110 on bounded intervals ) when @xmath2 . for @xmath111 , we have @xmath112 for all @xmath113 and system ( [ ncrel2 ] ) reduces to a semi - discrete linear wave equation . in that case , the scaling in ( [ approxkdv ] ) ( with @xmath114 ) yields a linearized kdv equation for the envelope function @xmath115 . to analyze the limit @xmath2 , we assume the same type of scaling for the solution @xmath110 , i.e. we search for solutions depending on slow variables @xmath49 and @xmath50 , where @xmath116 is a small parameter . we look for solutions of the form @xmath117 in contrast with ( [ approxkdv ] ) , the leading term @xmath118 is assumed of order unity and the remainder term in ( [ scal ] ) is @xmath119 . of course , due to the scaling invariance of the fpu lattice ( [ ncrel ] ) for @xmath0 , solutions with arbitrarily small or large amplitudes can be deduced from any solution of the form ( [ scal ] ) . from the scaling ( [ scal ] ) and using a taylor expansion and the chain rule , we obtain @xmath120 and @xmath121 to evaluate the right side of ( [ ncrel2 ] ) , we use the expansion @xmath122 where the logarithmic remainder term accounts for the possible vanishing of @xmath123 . setting now @xmath124 and using ( [ oper1 ] ) , we obtain @xmath125 with this choice of @xmath126 , the left- and right - hand sides of equation ( [ ncrel2 ] ) have the same order @xmath127 according to expansions ( [ oper2 ] ) and ( [ nl ] ) . substituting these expansions in ( [ ncrel2 ] ) yields @xmath128 then neglecting the higher order terms and integrating with respect to @xmath129 leads to @xmath130 where the integration constant has been fixed to @xmath14 in order to cover the case when @xmath131 . we shall call equation ( [ logkdv ] ) the logarithmic kdv ( log - kdv ) equation . it can be rewritten @xmath132 where the potential @xmath133 reads @xmath134 equation ( [ consistency ] ) shows that the log - kdv equation is consistent with the nonlinear lattice ( [ ncrel ] ) , i.e. each solution of ( [ logkdv ] ) is almost a solution of ( [ ncrel ] ) up to a small residual error . note that if @xmath123 is a solution of ( [ logkdv ] ) , so is @xmath135 . in addition , equation ( [ logkdv ] ) admits a nonstandard galilean symmetry involving a rescaling of amplitude , i.e. each solution @xmath123 generates a one - parameter family of solutions @xmath136 ( \xi , \tau ) : = e^c\ , v(\xi - c\ , \tau , \tau ) , \quad c\in \mathbb{r}.\ ] ] in particular , all travelling wave solutions of the log - kdv equation ( [ logkdv ] ) can be deduced from its stationary solutions . this property is inherited from the scaling invariance of the fpu system ( [ ncrel ] ) ( this point will be detailed in section [ trav ] ) . equation ( [ logkdv ] ) falls within the class of generalized kdv equations . systems in this class possess three ( formally ) conserved quantities @xcite , namely the mass @xmath137 the momentum @xmath138 and the energy @xmath139 well - posedness results for the cauchy problem associated with ( [ logkdvpot ] ) are known when @xmath140 is a @xmath141 function @xcite , but the existing theory does not apply to our case where @xmath142 diverges logarithmically at the origin . looking for solutions of the log - kdv equation ( [ logkdv ] ) depending only of @xmath129 , one obtains the stationary log - kdv equation @xmath143 integrating once under the assumption @xmath144 , one obtains @xmath145 this equation can be seen as a ( one - dimensional ) stationary log - nls equation @xcite . the potential @xmath133 in ( [ slognls ] ) has a double - well structure with a local maximum at @xmath146 ( see figure [ graphew ] ) , hence there exists a pair of ( symmetric ) homoclinic orbits to @xmath14 and a continuum of periodic orbits . the homoclinic solutions have the explicit form @xmath147 with @xmath148 note that these ( gaussian ) homoclinic solutions decay super - exponentially , but do not decay doubly - exponentially unlike the solitary wave solutions of the differential advance - delay equation ( [ diffadvdel - eq ] ) @xcite . [ 1][bl ] @xmath123 [ 1][bl ] @xmath149 graph of the double - well logarithmic potential @xmath133 . ] the homoclinic solution ( [ homlog ] ) yields an approximate gaussian solitary wave solution of the fpu lattice ( [ ncrel ] ) with velocity equal to unity @xmath150 with @xmath151 figure [ soliton ] compares the solitary wave solution of the differential advance - delay equation ( [ diffadvdel - eq ] ) computed numerically with the analytical approximations corresponding to the compactly supported solitary wave ( [ compacton ] ) and the gaussian solitary wave ( [ rgauss ] ) . the numerical approximations of solitary wave solutions of ( [ diffadvdel - eq ] ) were obtained using the algorithm described in @xcite , based on a reformulation of ( [ diffadvdel - eq ] ) as a nonlinear integral equation and the method of successive approximations ( see also @xcite for variants of this method ) . figure [ convergence ] shows the relative error ( in @xmath152 norms ) between solitary wave solutions of the differential advance - delay equation ( [ diffadvdel - eq ] ) and the two approximations ( [ compacton ] ) and ( [ rgauss ] ) as a function of @xmath1 . the gaussian solitary wave provides a worse approximation compared to the compactly supported solitary wave , but both approximation errors converge to zero as @xmath153 . we have in addition @xmath154 , hence the absolute errors between the exact solitary wave @xmath155 and the two approximations converge to zero similarly to the relative errors plotted in figure [ convergence ] . [ 1][bl ] @xmath156 [ 1][bl ] @xmath157 solitary wave solution ( blue dotted line ) of the differential advance - delay equation ( [ diffadvdel - eq ] ) in comparison with the compactly supported solitary wave ( red solid line ) given by ( [ compacton ] ) and the gaussian solitary wave ( green dashed line ) given by ( [ rgauss ] ) for @xmath103 ( left ) and @xmath158 ( right ) . , title="fig : " ] solitary wave solution ( blue dotted line ) of the differential advance - delay equation ( [ diffadvdel - eq ] ) in comparison with the compactly supported solitary wave ( red solid line ) given by ( [ compacton ] ) and the gaussian solitary wave ( green dashed line ) given by ( [ rgauss ] ) for @xmath103 ( left ) and @xmath158 ( right ) . , title="fig : " ] [ 1][bl ] @xmath1 [ 1][bl ] -90 @xmath159 relative error ( in supremum norm ) between the solitary wave solution @xmath155 of the differential advance - delay equation ( [ diffadvdel - eq ] ) and either the compactly supported solitary wave @xmath160 given by ( [ compacton ] ) ( blue dots ) or the gaussian solitary wave @xmath161 defined by ( [ rgauss ] ) ( green dots ) . both approximation errors converge to zero as @xmath153 ( absolute errors converge to zero similarly ) . ] so far , we have computed a solitary wave solution of the fpu lattice ( [ ncrel ] ) with unit velocity and have checked its convergence towards the gaussian approximation ( [ approxsolvone ] ) when @xmath2 . we shall now examine the convergence of solitary waves with velocities different from unity . using the galilean invariance of ( [ logkdv ] ) , the homoclinic solution ( [ homlog ] ) yields two ( symmetric ) families of solitary wave solutions of the log - kdv equation ( [ logkdv ] ) @xmath162 parameterized by the wave velocity @xmath163 . these profiles yield the approximate solitary wave solutions of the original fpu lattice ( [ ncrel ] ) @xmath164 where we have set @xmath165 , @xmath166 introduced an additional phase shift @xmath167 and used the fact that @xmath168 . one can observe that the width of the approximate solitary wave ( [ approxsol ] ) diverges as @xmath169 when @xmath2 . moreover , similarly to solitary wave solutions of the fpu lattice ( [ ncrel ] ) , the wave width remains constant if @xmath1 is fixed and the wave amplitude @xmath170 ( or equivalently the wave velocity @xmath171 ) is varied . in addition , approximation ( [ approxsol ] ) can be rewritten @xmath172 where the renormalized gaussian profile @xmath173 is defined in ( [ rgauss ] ) . one can notice that @xmath174 if @xmath175 is fixed @xmath2 . consequently , approximation ( [ approxsolbis ] ) is close to a rescaling of ( [ approxsolvone ] ) through the invariance @xmath176 of ( [ ncrel ] ) . this observation illustrates why the ( nonstandard ) galilean invariance ( [ scaling ] ) of the log - kdv equation is inherited from the scaling invariance of the fpu lattice ( [ ncrel ] ) . similarly , the solitary wave solution @xmath155 of the differential advance - delay equation ( [ diffadvdel - eq ] ) yields the family of solitary wave solutions of ( [ ncrel ] ) @xmath177 one can distinguish two contributions to the error between the gaussian approximation ( [ approxsolbis ] ) and the exact solution ( [ exactvqqe ] ) , one originating from the profile function @xmath173 and the other from the wave amplitude . from the numerical results of figure [ convergence ] , we know that @xmath178 when @xmath2 . in addition , fixing @xmath175 and considering wave velocities ( [ velform ] ) close to unity when @xmath2 , we have seen that @xmath179 . consequently , the exact solitary wave ( [ exactvqqe ] ) with velocity ( [ velform ] ) converges uniformly towards the gaussian approximation ( [ approxsolbis ] ) when @xmath2 . note that the convergence result above concerns solitary waves with velocities converging towards unity , i.e. the value of sound velocity in the linear chain with @xmath54 . this restriction is due to the specific scaling ( [ scal ] ) assumed for solutions described by the log - kdv equation . on the contrary , the exact fpu solitary wave ( [ exactvqqe ] ) with fixed velocity @xmath180 becomes degenerate when @xmath2 , since the wave amplitude goes to @xmath14 if @xmath181 and diverges if @xmath182 . the compactly supported solitary wave ( [ compacton ] ) yields approximate solutions @xmath183 of the fpu system ( [ ncrel ] ) taking the form @xmath184 . this approximation can be compared to the exact fpu solitary wave @xmath110 defined by ( [ exactvqqe ] ) . the results of figure [ convergence ] show that @xmath185 as @xmath186 . consequently , one can infer that the relative error @xmath187 converges to zero when @xmath171 is fixed and @xmath2 . in section [ trav ] , we have computed solitary wave solutions of the fpu lattice ( [ ncrel ] ) and have checked their convergence towards the gaussian approximation ( [ approxsol ] ) when @xmath2 . these results are valid in a stationary regime and for prescribed wave velocities converging towards unity . to complete this analysis , we shall study the formation of solitary waves from a localized perturbation of given magnitude , and compare their profiles to gaussian approximations when @xmath1 is close to one . in what follows , we numerically integrate the fpu system ( [ nc ] ) with hertzian potential ( [ vhertz ] ) for different values of @xmath0 . we consider the hertzian potential ( [ vhertz ] ) rather that the symmetrized potential ( [ vhom ] ) because of its relevance to impact mechanics . in addition , the differential equations ( [ nc ] ) are easier to integrate numerically due to the absence of dispersive wavetrains for the above initial condition . we consider a lattice of @xmath188 particle with free - end boundary conditions . computations are performed with the standard ode solver of the software scilab . we consider a velocity perturbation of the first particle ( at @xmath189 ) , corresponding to the initial condition @xmath190 due to the scale invariance of system ( [ nc ] ) and ( [ vhertz ] ) , all positive initial velocities for @xmath191 yield a rescaled solution of the form @xmath192 . the front edge of the solution evolves into a solitary wave whose profile becomes stationary for large enough times , at least on the timescales of the simulations . when @xmath1 is sufficiently close to unity , one observes that the asymptotic velocity of the solitary wave is close to unity ( this property is also true for any initial velocity @xmath193 ) . the solitary wave is of compression type ( i.e. with @xmath194 ) , hence it also defines a solution of the fpu lattice ( [ ncrel ] ) with the symmetrized potential ( [ vhom ] ) and can be compared with approximation ( [ approxsol ] ) . [ 1][bl ] @xmath6 [ 1][bl ] @xmath195 formation of a localized wave resulting from the initial condition ( [ ic ] ) in the hertzian fpu lattice ( [ nc ] ) and ( [ vhertz ] ) with @xmath196 . top left : solution profile @xmath61 at @xmath197 ( note the absence of dispersive wavetrains behind the localized pulse ) . top right : the pulse profile at @xmath198 ( black curve ) is compared with the gaussian approximation ( [ approxsol ] ) ( blue curve ) . a small mismatch between the two curves is visible at this stage ( note the slightly asymmetric shape of the numerical solution ) . the value of @xmath199 has been fixed so that the exact and approximate solitary waves have the same amplitude for large values of @xmath200 . bottom : same comparison at @xmath201 , where the approximate and numerical solutions almost perfectly coincide . , title="fig : " ] formation of a localized wave resulting from the initial condition ( [ ic ] ) in the hertzian fpu lattice ( [ nc ] ) and ( [ vhertz ] ) with @xmath196 . top left : solution profile @xmath61 at @xmath197 ( note the absence of dispersive wavetrains behind the localized pulse ) . top right : the pulse profile at @xmath198 ( black curve ) is compared with the gaussian approximation ( [ approxsol ] ) ( blue curve ) . a small mismatch between the two curves is visible at this stage ( note the slightly asymmetric shape of the numerical solution ) . the value of @xmath199 has been fixed so that the exact and approximate solitary waves have the same amplitude for large values of @xmath200 . bottom : same comparison at @xmath201 , where the approximate and numerical solutions almost perfectly coincide . , title="fig : " ] formation of a localized wave resulting from the initial condition ( [ ic ] ) in the hertzian fpu lattice ( [ nc ] ) and ( [ vhertz ] ) with @xmath196 . top left : solution profile @xmath61 at @xmath197 ( note the absence of dispersive wavetrains behind the localized pulse ) . top right : the pulse profile at @xmath198 ( black curve ) is compared with the gaussian approximation ( [ approxsol ] ) ( blue curve ) . a small mismatch between the two curves is visible at this stage ( note the slightly asymmetric shape of the numerical solution ) . the value of @xmath199 has been fixed so that the exact and approximate solitary waves have the same amplitude for large values of @xmath200 . bottom : same comparison at @xmath201 , where the approximate and numerical solutions almost perfectly coincide . , title="fig : " ] the results are shown in figure [ pulse1_01 ] for @xmath196 . the parameter @xmath202 in the gaussian approximation ( [ approxsol ] ) is determined from the relation @xmath170 , where @xmath203 is the exact solitary wave amplitude obtained by integrating ( [ nc ] ) with initial data ( [ ic ] ) . the approximation of the solitary wave profile is very accurate in the stationary regime , as shown by the bottom panel . in addition , the measured velocity of the numerical solitary wave solution @xmath204 can be compared with the velocity @xmath205 of the approximate solitary wave ( [ approxsol ] ) , which yields a relative error @xmath206 around @xmath207 . discrepancies appear between the profiles of the numerical solution and the gaussian approximation for larger values of @xmath1 , as already noticed in figures [ soliton ] and [ convergence ] . in addition , we find a relative error @xmath208 between numerical and approximate wave velocities around @xmath209 for @xmath210 and @xmath211 for @xmath103 . as a conclusion , while some quantitative agreement is still obtained for @xmath212 between the numerical solution and the gaussian approximation , the latter becomes unsatisfactory for @xmath103 . the numerical results of section [ gauss ] indicate the long time stability of the solitary wave solutions of system ( [ nc ] ) with hertzian potential ( [ vhertz ] ) that form after a localized perturbation . therefore , the stability of the gaussian solitary waves ( [ sol ] ) appears as a necessary ( of course not sufficient ) condition to establish the validity of the log - kdv equation ( [ logkdv ] ) as a modulation equation for the fpu system with hertzian potential . in this section , we prove the linear orbital stability of solitary waves of the log - kdv equation . we perform the analysis for the stationary gaussian solution @xmath213 defined by ( [ homlog ] ) . by the scaling transformation ( [ scaling ] ) , the stability result extends to the entire family of solitary waves @xmath214 with @xmath53 . the log - kdv equation ( [ logkdvpot ] ) can be written in the hamiltonian form @xmath215 associated with the energy ( [ energy ] ) . the stationary gaussian solution @xmath213 is a critical point of the energy @xmath216 , i.e. @xmath217 . the hessian operator evaluated at this solution reads @xmath218 and corresponds to a schrdinger operator with a harmonic potential . equation ( [ kdvham ] ) linearized at @xmath213 reads @xmath219 in what follows , we formulate ( [ lkdvham ] ) as a differential equation in suitable function spaces and derive a linear stability result based on the energy method for kdv - type evolution equations @xcite . the spectral properties of @xmath220 are well known @xcite . the operator @xmath220 is self - adjoint in @xmath221 with dense domain @xmath222 its spectrum consists of simple eigenvalues at integers @xmath223 , where @xmath224 ( the set of natural numbers including zero ) . in particular , @xmath220 has a simple eigenvalue @xmath225 with eigenspace spanned by @xmath226 , a simple zero eigenvalue with eigenspace spanned by @xmath227 , and the rest of its spectrum is bounded away from zero by a positive number . the discreteness of the spectrum comes from the fact that the harmonic potential of the schrdinger operator @xmath220 is unbounded at infinity , which implies that the embedding @xmath228 is compact ( see e.g. @xcite , p. 43 - 44 ) and @xmath220 has a compact resolvent in @xmath221 . hereafter we denote by @xmath229 the usual @xmath230-scalar product . the operator @xmath231 inherits a double non semi - simple zero eigenvalue , with generalized kernel @xmath232 spanned by the eigenvector @xmath233 and the generalized eigenvector @xmath234 . the algebraic multiplicity of this eigenvalue is @xmath235 because equation @xmath236 has no solution in @xmath237 ( since @xmath238 ) . the double zero eigenvalue is linked with the existence of a two - parameter family of solitary waves of the log - kdv equation ( [ logkdv ] ) parameterized by the location and velocity of the waves . it induces in general a secular growth of the solutions of ( [ lkdvham ] ) along the eigenvector @xmath233 , linked with a velocity change of perturbed solitary waves . in order to prove a linear stability result , we thus have to project ( [ lkdvham ] ) onto the invariant subspace under @xmath239 associated with the nonzero part of the spectrum . following a classical computation scheme @xcite , the spectral projection onto @xmath232 takes the form @xmath240 with @xmath241 ( one can readily check that @xmath242 commutes with @xmath239 ) . note in passing that @xmath203 is well defined because @xmath243 . now , splitting the solutions @xmath244 of ( [ lkdvham ] ) into @xmath245 with @xmath246 , one obtains the following equivalent system @xmath247 @xmath248 in order to prove the linear ( orbital ) stability of the gaussian solitary wave , we have to show the lyapunov stability of the equilibrium @xmath249 of the linear evolution equation ( [ eq1 ] ) for a suitable topology . let us recall that @xmath250 , hence @xmath251 belongs to the codimension-@xmath235 subspace of @xmath237 defined as @xmath252 since @xmath220 is positive on @xmath253 and @xmath254 , we can define @xmath255 . due to the fact that @xmath220 is positive - definite on @xmath256 and @xmath257 , @xmath258 defines a norm on @xmath259 ( roughly a weighted @xmath260-norm ) . denote by @xmath261 the completion of @xmath259 with respect to the norm @xmath262 . this norm defines a convex conserved lyapunov function for system ( [ eq1 ] ) since @xmath263 for simplicity , let us choose an initial data in the schwartz space @xmath264 of rapidly decreasing @xmath265 functions : @xmath266 thanks to property ( [ conserv ] ) , we get from the energy method @xcite and standard bootstrapping arguments ( see e.g. chapter 11.1.4 of @xcite ) a unique global solution of ( [ eq1])-([cilin ] ) which is infinitely smooth in time and space . we have @xmath267 and @xmath268 for all @xmath269 , which shows the lyapunov stability of the equilibrium @xmath249 of ( [ eq1 ] ) in @xmath261 . therefore , we have proved the linear orbital stability of the gaussian solitary wave . the lyapunov stability of the equilibrium @xmath249 implies the absence of eigenvalue of @xmath270 with positive real part , since such an eigenvalue would lead to exponential growth of the solution along a corresponding eigenvector . in addition , the spectrum of @xmath239 is invariant by @xmath93 since @xmath239 possesses a reversibility symmetry , i.e. anticommutes with the symmetry @xmath271 . this implies that all the eigenvalues of @xmath239 lie on the imaginary axis . this result contrasts with the instability of the compacton solutions of the fully nonlinear boussinesq equation numerically analyzed in section [ bouss ] . moreover , it is consistent with the absence of solitary wave instabilities observed in section [ gauss ] . the absence of eigenvalues of @xmath270 with positive real part could be also obtained from the recent works @xcite . this result follows from the main theorems in @xcite if the number of negative eigenvalues of @xmath220 is equal to one and @xmath272 where @xmath273 . assumption ( [ last - assumption ] ) is satisfied since @xmath274 ( recall @xmath275 ) and @xmath276 the results of figures [ soliton ] and [ convergence ] indicate that compacton approximations converge towards solitary wave solutions of the differential advance - delay equation ( [ ncrel ] ) when @xmath2 . it seems delicate to establish this result directly from the methodology described in section [ bouss ] , where equation ( [ compacton - eq ] ) is heuristically obtained by truncation of the differential advance - delay equation ( [ diffadvdel - eq ] ) . however , it is instructive to compare analytically the compacton approximation @xmath277 defined in ( [ compacton ] ) and the gaussian approximation ( [ rgauss ] ) when @xmath1 is close to unity , since our numerical results indicate that both profiles become very close ( see figure [ soliton ] , right panel ) . to check the consistency of ( [ compacton ] ) and ( [ rgauss ] ) , we note that the compact support @xmath278 $ ] of approximation ( [ compacton ] ) extends to the entire real line as @xmath2 . furthermore , let @xmath279 , @xmath280 , and perform the expansions @xmath281 and @xmath282 for all fixed @xmath283 . from expansions ( [ expa ] ) and ( [ expb ] ) , it follows that the renormalized compacton @xmath284 converges towards the gaussian solution ( [ homlog ] ) for any fixed @xmath283 when @xmath2 . one possible approach for the justification of compactons consists in deriving an asymptotic model consistent with fpu lattice ( [ ncrel ] ) and supporting compacton solutions , in analogy with the derivation of the log - kdv equation . such a model may be free of the artificial instabilities introduced by the boussinesq equation ( [ b ] ) , and may lead to a well - posed evolution problem in a suitable function space and for appropriate initial data . in what follows we show that a generalized kdv equation with nonsmooth fractional power nonlinearity can be derived from ( [ ncrel ] ) using the method of section [ kdvloga ] . for this purpose , we note that the nonlinearity @xmath285 defined by ( [ deffa ] ) satisfies @xmath286 ( uniformly in @xmath110 on bounded intervals ) when @xmath2 . consequently , the generalized kdv equation @xmath287 is consistent with system ( [ ncrel ] ) at the same order as the log - kdv equation , and converges towards the log - kdv equation when @xmath2 . we call equation ( [ rac ] ) the hlderian kdv ( h - kdv ) equation due to the presence of the hlder - continuous nonlinearity @xmath58 . this approximation of the logarithmic nonlinearity is reminiscent of results of @xcite obtained for a stationary hlderian nls equation close to a logarithmic limit . note that the h - kdv equation ( [ rac ] ) admit the same three conserved quantities ( [ mass ] ) , ( [ momentum ] ) , and ( [ energy ] ) with potential @xmath133 replaced by @xmath288 moreover , equation ( [ rac ] ) admits a nonstandard galilean invariance similarly to the log - kdv equation . more precisely , any solution @xmath123 of ( [ rac ] ) generates a one - parameter family of solutions @xmath289 defined by @xmath290(\xi,\tau ) : = ( 1-\mu ) ^{\frac{\alpha}{1-\alpha } } v\left((1- \mu ) ^{\frac{1}{2}}(\xi - c\ , \tau),(1-\mu ) ^{\frac{3}{2 } } \tau\right ) , \ \ \ c\ , \left ( 1-\frac{1}{\alpha } \right),\ ] ] and parameterized by @xmath291 $ ] . one can check that this symmetry reduces to the galilean invariance ( [ scaling ] ) when @xmath2 . let us check that the h - kdv equation admits compacton solutions . their existence is due to the non - differentiable hlder - continuous nonlinearity , in contrast to classical compacton equations where degenerate nonlinear dispersion plays a central role @xcite . the stationary h - kdv equation integrated once reads @xmath292 ( the integration constant has been set to @xmath14 ) , or equivalently @xmath293 . this equation is integrable and the potential @xmath294 has a double - well structure with a local maximum at @xmath146 . this property implies the existence of a pair of symmetric homoclinic orbits to @xmath14 corresponding to compactons . using the change of variable @xmath295 equation ( [ racs ] ) is mapped to the form ( [ compacton - eq ] ) which possesses compacton solutions given by ( [ compacton ] ) . consequently , equation ( [ racs ] ) admits stationary compacton solutions @xmath296 where @xmath297 in addition , we have as above @xmath298 for any fixed @xmath283 , i.e. the compacton approximation ( [ compactonkdv ] ) converges towards the gaussian solitary wave approximation ( [ homlog ] ) when @xmath2 . the compacton approximation @xmath284 defined in ( [ compacton ] ) and obtained using the boussinesq equation ( [ b ] ) differs from the compacton ( [ compactonkdv ] ) deduced from the h - kdv equation . however , both approximations are equivalent when @xmath2 since they converge towards the gaussian profile ( [ homlog ] ) . in fact , an infinity of compacton approximations could be constructed , depending on the approximation of the logarithmic nonlinearity introduced in ( [ deffam1 ] ) . using the galilean invariance of ( [ rac ] ) , the compacton ( [ compactonkdv ] ) yields two ( symmetric ) families of compactly supported solitary waves of the h - kdv equation ( [ rac ] ) @xmath299 , \ \ \ \mu=\tilde{c}\ , ( 1-\frac{1}{\alpha}),\ ] ] parameterized by the wave velocity @xmath300 $ ] . from the expression ( [ scal ] ) , these profiles yield the approximate compacton solutions of the fpu lattice ( [ ncrel ] ) @xmath301,\ ] ] where we have set @xmath302 , @xmath52 , @xmath303 , introduced an additional phase shift @xmath167 and used the fact that @xmath168 . for any fixed value of @xmath199 and @xmath283 , we have @xmath304 . in this limit , the compacton ( [ approxsoltrav ] ) converges towards the gaussian approximation ( [ approxsol ] ) , therefore it is also close to the exact fpu solitary wave ( [ exactvqqe ] ) . we have obtained two generalized kdv equations ( the log - kdv equation ( [ logkdv ] ) and the h - kdv equation ( [ rac ] ) ) as formal asymptotic limits of fermi - pasta - ulam lattices with hertzian - type potentials , when the nonlinearity exponent @xmath0 goes to unity and slowly varying profiles are considered . using numerical computations , we have checked that fpu solitary waves converge towards gaussian solitary waves and compacton solutions to these kdv equations when @xmath2 and for near - sonic wave speeds . in addition , we have illustrated numerically the formation of stable solitary waves after a localized velocity perturbation in the hertzian fpu system ( [ nc ] ) and ( [ vhertz ] ) when @xmath56 , a limit in which the propagating pulse becomes nearly gaussian . the linearized log - kdv equation preserves the spectral stability of solitary waves , which is lost when using other formal ( boussinesq - type ) continuum models . while our study does not yield a complete proof of the asymptotic behaviour of exact fpu solitary waves when @xmath2 , it provides nevertheless an asymptotic framework to explain classical formal compacton approximations @xcite , whose justification remained unclear up to now . it would be interesting to examine the dynamical properties of the log - kdv and h - kdv equations for different classes of initial conditions . relevant questions include local well - posedness ( or ill - posedness ) , derivation of a priori bounds , global well - posedness ( or blow - up ) , scattering of some initial data , and nonlinear stability of solitary waves . in our context , the study of the nonlinear orbital stability @xcite or asymptotic stability @xcite of solitary waves rises new difficulties , linked with the lack of smoothness of the energy functional ( [ energy ] ) . in addition , if the well - posedness of the cauchy problem for the log - kdv or h - kdv equation could be established for appropriate initial data , one could then study analytically or numerically their connection with the fpu system ( [ nc ] ) with homogeneous potentials ( [ vhom ] ) . an open question is to check that well - prepared initial data evolve ( up to higher order terms and on long finite times ) according to the log - kdv or h - kdv equation ( in the same spirit as the justification of the classical kdv equation for fpu chains @xcite ) . this problem may be extended to the hertzian fpu system ( [ nc ] ) and ( [ vhertz ] ) , at least close to a solitary wave solution and for a suitable topology ( i.e. using a weighted norm flattening perturbations behind the propagating wave @xcite ) . the construction of appropriate numerical methods for the time - integration of the log - kdv or h - kdv equations is of course also fundamental in this context . another open problem concerns the dynamical stability of the solitary wave solutions of the fpu lattice ( [ ncrel ] ) . our proof of the linear orbital stability of solitary waves for the log - kdv equation could be useful in this context , following the lines of @xcite ( using the linearized log - kdv equation instead of the linearized kdv equation ) . these open questions will be explored in forthcoming works . _ acknowledgements : _ g.j . is grateful to v. acary , b. brogliato , w. craig and y. starosvetsky for stimulating discussions on this topic . part of this work was carried out during a visit of g.j . to the department of mathematics at mcmaster university , to which g.j . is grateful for hospitality . g.j . acknowledges financial support from the french embassy in canada and the rhne - alpes complex systems institute ( ixxi ) . d.p . acknowledges financial support from the nserc . _ author contributions : _ g.j . introduced the time - dependent logarithmic and hlderian kdv equations and their derivations from the hertzian chain model . numerical computations were performed by d.p . ( figures 1 , 3 , 4 ) and g.j . ( figure 5 ) . all authors contributed to the linear stability analysis of gaussian solitary waves . . showed the convergence of the ahnert - pikovsky compacton towards a gaussian for hertz force exponents close to unity . all authors contributed to the writing and editing of the manuscript . 00 e. afshari and a. hajimiri . nonlinear transmission lines for pulse shaping in silicon , _ ieee journal of solid - state circuits 40 _ ( 2005 ) , 744 - 752 . k. ahnert and a. pikovsky . compactons and chaos in strongly nonlinear lattices , _ phys . e 79 _ ( 2009 ) , 026209 . ambrose , g. simpson , j.d . wright , and d.g . yang , ill - posedness of degenerate dispersive equations , _ nonlinearity 25 _ ( 2012 ) , 26552680 . d. bambusi and a. ponno . on metastability in fpu , _ comm . phys . 264 _ ( 2006 ) , 539 - 561 . i. bialynicki - birula and j. mycielski . nonlinear wave mechanics , _ annals of physics 100 _ ( 1976 ) , 62 - 93 . j.l . bona , p.e . souganidis and w.a . stability and instability of solitary waves of korteweg - de vries type , _ proc . lond . a 411 _ ( 1987 ) , 395 - 412 . campbell et al , editors . the fermi - pasta - ulam problem : the first @xmath305 years , _ chaos 15 _ ( 2005 ) . q. dou , j. cuevas , j.c . eilbeck and f.m . breathers and kinks in a simulated breather experiment , _ discrete contin . s 4 _ ( 2011 ) , 1107 - 1118 . dubovsky and a.v . emission of supersonic soliton wave beams - generators of restructuring of nanocrystals under atom bombardment , and the self - organization of a dynamic superlattice of complexes of soliton atomic vibrations , _ phys . of solid state 52 _ ( 2010 ) , 899 - 903 . english and r.l . pego . on the solitary wave pulse in a chain of beads , _ proc . amer . math . soc . 133 _ , n. 6 ( 2005 ) , 1763 - 1768 . . solitary waves on fpu lattices : i. qualitative properties , renormalization and continuum limit , _ nonlinearity 12 _ ( 1999 ) , 1601 - 1627 . linear implies nonlinear stability , _ nonlinearity 15 _ ( 2002 ) , 1343 - 1359 . howland - type floquet theory , _ nonlinearity 17 _ ( 2004 ) , 207 - 227 . proof of stability at low energy , _ nonlinearity 17 _ ( 2004 ) , 229 - 251 . g. friesecke and j.a wattis . existence theorem for solitary waves on lattices , _ commun . phys . 161 _ ( 1994 ) , 391 - 418 . m. herrmann . unimodal wavetrains and solitons in convex fermi - pasta - ulam chains , _ proc . sect . a - math . 140 _ ( 2010 ) , 753 - 785 . e. j. hinch and s. saint - jean . the fragmentation of a line of ball by an impact , _ proc . london , ser . a 455 _ ( 1999 ) , 3201 - 3220 . a. hoffman and c.e . asymptotic two - soliton solutions in the fermi - pasta - ulam model , _ equat . 21 _ ( 2009 ) , 343 - 351 . a. hoffman and c.e . wayne . a simple proof of the stability of solitary waves in the fermi - pasta - ulam model near the kdv limit , _ fields institute communications 64 _ ( 2013 ) , 185 - 192 . travelling waves in the fermi - pasta - ulam lattice , _ nonlinearity 13 _ ( 2000 ) , 849 - 866 . g. james and y. starosvetsky . breather solutions of the discrete @xmath306-schrdinger equation , in _ localized excitations in nonlinear complex systems _ , p. kevrekidis , r. carretero - gonzlez , j. cuevas - maraver , d. frantzeskakis , n. karachalios and f. palmero - acebedo eds . , _ nonlinear systems and complexity 7 _ , springer , 2014 . s. job , f. melo , a. sokolow and s. sen . solitary wave trains in granular chains : experiments , theory and simulations , _ granular matter 10 _ ( 2007 ) , 13 - 20 . k.l . _ contact mechanics _ , cambridge univ . press , 1985 . long wave asymptotics . integrable equations as asymptotic limits of non - linear systems , _ russian math . surveys 44 _ ( 1989 ) , 3 - 42 . t. kapitula and a. stefanov . a hamiltonian krein ( instability ) index theory for kdv - like eigenvalue problems , _ studies in applied mathematics _ ( 2013 ) , doi : 10.1111/sapm.12031 . d. khatri , d. ngo and c. daraio . highly nonlinear solitary waves in chains of cylindrical particles , _ granular matter 14 _ ( 2012 ) , 63 - 69 . r.s . solitary waves in a chain of beads under hertz contact , _ phys . lett . a 251 _ ( 1999 ) , 191 - 192 . a. mielke and c. patz . dispersive stability of infinite - dimensional hamiltonian systems on lattices , _ applicable analysis 89 _ ( 2010 ) , 1493 - 1512 . t. mizumachi . asymptotic stability of lattice solitons in the energy space , _ commun . phys . 288 _ ( 2009 ) , 125 - 144 . t. mizumachi . asymptotic stability of @xmath33-solitary waves of the fpu lattices , _ archive for rational mechanics and analysis 207 _ ( 2013 ) , 393 - 457 . _ dynamics of heterogeneous materials _ , springer verlag , 2001 . d. ngo , s. griffiths , d. khatri and c. daraio . highly nonlinear solitary waves in chains of hollow spherical particles , _ granular matter 15 _ ( 2013 ) , 149 - 155 . nguyen and b. brogliato . _ multiple impacts in dissipative granular chains _ , lecture notes in applied and computational mechanics 72 , springer , 2014 . pego and m.i . asymptotic stability of solitary waves , _ commun . phys . 164 _ ( 1994 ) , 305 - 349 . spectral stability of nonlinear waves in kdv - type evolution equations , _ spectral analysis , stability , and bifurcation in modern nonlinear physical systems _ , wiley - iste , 2013 . m. renardy and r.c . _ an introduction to partial differential equations _ , springer verlag , new york , 2004 . p. rosenau . dynamics of nonlinear mass - spring chains near the continuum limit , _ phys . lett . a 118 _ ( 1986 ) , 222 - 227 . p. rosenau and j. hyman . compactons : solitons with finite wavelength , _ phys . lett . 70 _ ( 1993 ) , 564 - 567 . g. schneider and c.e . counter - propagating waves on fluid surfaces and the continuum limit of the fermi - pasta - ulam model . in b. fiedler , k. grger and j. sprekels , editors , _ international conference on differential equations appl . 5 _ ( 1 ) , 69 - 82 ( 1998 ) . _ asymptotic analysis of soliton problems _ , lect . notes in math . 1232 , springer verlag , 1986 . k. sekimoto . newton s cradle versus nonbinary collisions , _ phys . 104 _ ( 2010 ) , 124302 . shubin , _ partial differential equations _ vol . 7 , encyclopaedia of mathematical sciences , springer , 1994 . a. stefanov and p.g . kevrekidis . on the existence of solitary traveling waves for generalized hertzian chains , _ j. nonlinear sci ( 2012 ) , 327 - 349 . d. sun , c. daraio and s. sen . nonlinear repulsive force between two solids with axial symmetry , _ phys . e 83 _ ( 2011 ) , 066605 . d. sun and s. sen . nonlinear grain - grain forces and the width of the solitary wave in granular chains : a numerical study , _ granular matter 15 _ ( 2013 ) , 157 - 161 . z. yang and x. wang . blowup of solutions for the bad " boussinesq - type equation , _ j. math . appl . 285 _ ( 2003 ) , 282 - 298 . p. zhidkov . _ korteweg - de vries and nonlinear schrdinger equations _ , lect . notes in math . 1756 , springer verlag , 2001 .

we argue that conservation laws based on the _ local _ matter - only stress - energy - momentum tensor ( characterized by energy and momentum per unit _ volume _ ) can not adequately explain a wide variety of even very simple physical phenomena because they fail to properly account for gravitational effects . we construct a general _ quasi_local conservation law based on the brown and york _ total _ ( matter _ plus _ gravity ) stress - energy - momentum tensor ( characterized by energy and momentum per unit _ area _ ) , and argue that it _ does _ properly account for gravitational effects . as a simple example of the explanatory power of this quasilocal approach , consider that , when we accelerate toward a freely - floating massive object , the kinetic energy of that object increases ( relative to our frame ) . but how , exactly , does the object acquire this increasing kinetic energy ? using the energy form of our quasilocal conservation law , we can see precisely the actual mechanism by which the kinetic energy increases : it is due to a bona fide gravitational energy flux that is exactly analogous to the electromagnetic poynting flux , and involves the general relativistic effect of frame dragging caused by the object s motion relative to us . * quasilocal conservation laws : why we need them * paul l. mcgrath , richard j. epp and robert b. mann _ department of physics and astronomy , university of waterloo , waterloo , ontario n2l 3g1 , canada _ _ perimeter institute for theoretical physics , waterloo , ontario n2l 2y5 , canada _ _ pmcgrath@uwaterloo.ca , repp@perimeterinstitute.ca , rbmann@sciborg.uwaterloo.ca_

1206.6512

a conservation law ought to explain the change in some physical quantity contained inside a volume of space ( e.g. , total energy ) in terms of related fluxes passing through the bounding surface of that volume . the standard approach to constructing conservation laws is based on the identity : @xmath5 , where @xmath6 is the matter stress - energy - momentum tensor , and @xmath7 is a vector that determines the type of conservation law , viz . , energy , momentum , or angular momentum . in the context of general relativity , matter energy - momentum is locally covariantly conserved , i.e. , @xmath8 , and the identity reduces to @xmath9 . as is well known , the problem with this local conservation law is that the right hand side is , in general , not zero ( or even a covariant divergence ) , resulting in a bulk term in the integrated conservation law that spoils what a conservation law ought to be . this problem is essentially gravitational in nature . there are several ways to see this : ( 1 ) the offending bulk term disappears when the spacetime has a suitable symmetry , i.e. , admits a killing vector , @xmath7 , but dynamically interesting spacetimes ( e.g. , ones containing gravitational effects due to objects in motion ) generically do not . the idea of relying on a spacetime symmetry to construct a conservation law is a throwback to pre - general relativity days , e.g. , special relativity , where spacetime is maximally symmetric . the same goes for relying on asymptotic spacetime symmetries , where we are still in essentially a special relativistic mindset . given that gravity _ is _ nontrivial spacetime geometry , an approach relying on spacetime symmetries can not hope to properly incorporate gravitational effects in general . ( 2 ) the local conservation law above is _ homogeneous _ in @xmath6 . in any matter - free region it is vacuous , even if that region contains interesting gravitational physics , e.g. , gravitational waves . the local conservation law is essentially blind to gravitational physics . we need a conservation law that is nontrivial even when @xmath10 . ( 3 ) because of the equivalence principle , gravitational effects , e.g. , gravitational energy , are not localizable , so we have no hope of capturing gravitational physics with a conservation law based on a local stress - energy - momentum tensor . for example , there is no such thing as a local gravitational energy density ( energy per unit volume ) , that when integrated over a volume gives the total gravitational energy in that volume ( see , e.g. , 20.4 of reference @xcite ) . ( 4 ) a bulk term in a local conservation law can be a symptom of the presence of fields that are not being accounted for in the stress - energy - momentum tensor . for example , in the standard poynting theorem , the @xmath11 bulk term is present because @xmath6 excludes the charged matter field that is the source of the electromagnetic field , and represents an energy transfer mechanism between the electromagnetic field and the charged matter field . we contend that the bulk term in the local conservation law is , similarly , a result of @xmath6 not properly accounting for the physics of the gravitational field . a solution to this problem is to move from local to _ quasi_local conservation laws , which _ can _ properly account for the gravitational physics . in this paper we construct a general quasilocal conservation law based not on the local _ matter _ stress - energy - momentum tensor , but on the brown and york quasilocal _ total _ stress - energy - momentum tensor ( matter _ plus _ gravity ) @xcite . here , quasilocal " means that the differential conservation law is integrated not over the history of a volume of space , but over the history of the _ boundary _ of that volume . we focus on the case of energy conservation , and show that in the quasilocal approach , the quasilocal analogue of the offending @xmath12 bulk term becomes a _ surface flux _ term , which immediately solves the main problem mentioned in the opening paragraph above . moreover , this surface flux term has two components : ( i ) the first component is a stress times strain " term that can always be made to vanish by a suitable choice of frame , called a _ rigid quasilocal frame _ ( or rqf ) @xcite . the reader need not be familiar with rqfs to read this paper ; it is sufficient to mention that , regarding point ( 1 ) in the previous paragraph , an rqf satisfies a certain quasilocally projected " form of the timelike killing vector condition for stationary spacetimes that allows us to move just far enough away from the spacetime symmetry mindset to include generic ( i.e. , non - stationary ) spacetimes in conservation laws for energy , momentum , and angular momentum . ( ii ) the second component is an acceleration times momentum " term . this term is familiar from classical mechanics , and represents the rate at which the kinetic energy of an object increases due to one s acceleration toward it . motivated by a simple equivalence principle argument , we show that this second term is actually a _ gravitational _ energy flux involving the general relativistic effect of frame dragging . we thus show precisely how quasilocal conservation laws resolve the bulk term problem in local conservation laws by properly accounting for the physics of the gravitational field . our paper is organized as follows . in [ paradox ] we introduce a very simple example of energy conservation in the context of special relativity , for the purpose of having a concrete example with which to illustrate the development of the general ideas . we consider a variant of bell s spaceship paradox in which a box accelerates rigidly in a transverse , uniform electric field . obviously , the electromagnetic energy inside the box increases , but how would co - moving observers explain this increase ? paradoxically , only _ half _ of the increasing energy comes from a net poynting flux , and , according to the local energy conservation law , the other half comes from a bulk acceleration times momentum " term integrated over the volume of the box . in [ local ] we examine local conservation laws in general , with a particular focus on the role played by the @xmath12 bulk term in our paradox example . this provides a point of comparison for [ quasilocal ] , in which we construct a general quasilocal conservation law and argue how it properly accounts for gravitational physics . we also apply the energy form of this quasilocal conservation law to the general relativistic version of our paradox example to concretely illustrate how , what we would normally think of as a bulk acceleration times momentum " term , is actually a gravitational energy flux entering through the boundary of the box . in [ conclusions ] we present a complementary summary and conclusions , and argue that quasilocal conservation laws are necessary to understand more deeply a wide variety of phenomena , including the simple example of dropping an apple . [ trajectory ] consider a right cylinder of length @xmath13 and cross - sectional area @xmath14 , whose axis is parallel to the @xmath15-axis of an inertial reference frame in flat spacetime with minkowski coordinates @xmath16 . the cylinder is immersed in a constant , uniform electric field of magnitude @xmath17 in the positive @xmath18-direction , and thus contains an electromagnetic energy density equal to @xmath19 . now let us give the cylinder constant proper acceleration in the positive @xmath15-direction so that the length ( and volume ) of the cylinder remain fixed for co - moving observers . it is well - known from bell s spaceship paradox that , in order for the length of the cylinder to remain constant for co - moving observers , the top of the cylinder ( represented by the hyperbola on the right in the diagram ) must experience _ less _ proper acceleration , namely @xmath20 , compared to @xmath21 , the proper acceleration experienced at the bottom of the cylinder ( the hyperbola on the left ) @xcite . the following two facts , also illustrated in the diagram , will be important to us : ( 1 ) the straight lines passing through the origin represent a natural choice for the surfaces of simultaneity for the co - moving observers : on any such simultaneity , all of the accelerating co - moving observers see the _ same _ instantaneous velocity , @xmath22 , relative to observers at rest in the inertial frame . in other words , this is a constant @xmath22 " time foliation , with @xmath22 monotonically increasing with time . ( 2 ) co - moving observers at different positions along the length of the cylinder will see proper time flowing at different rates relative to one another . that is , between two co - moving simultaneities , if a proper time @xmath23 elapses for an observer at the bottom of the cylinder , a _ greater _ proper time , @xmath24 elapses for an observer at the top . with these facts in mind , let us consider the proper electric and magnetic fields seen by the co - moving observers . let @xmath25 ( respectively , @xmath26 ) denote the proper time of observers at the bottom ( respectively , top ) of the cylinder , and @xmath27 denote the natural choice of co - moving spatial cartesian unit vectors . since the observers are moving in a direction perpendicular to the electric field , they will see , in addition to a stronger electric field , a magnetic field ; specifically , @xmath28 and @xmath29 , where @xmath30 and @xmath31 . it is important to note that since these fields depend only on @xmath22 , all observers on any given co - moving simultaneity will see the _ same _ instantaneous electric and magnetic fields . they will thus see the _ same _ proper poynting vector , @xmath32 , and the _ same _ proper electromagnetic energy density , @xmath33 . notice that , according to this expression for @xmath34 , the total electromagnetic energy inside the cylinder is clearly increasing with time . we ask : how would the co - moving observers explain this increasing electromagnetic energy ? one might expect that the change in energy between two co - moving simultaneities is just equal to the net poynting flux into the cylinder over that time interval . as we will see , this is only half correct . consider a pair of infinitesimally separated co - moving simultaneities labeled by proper time @xmath25 and @xmath35 as experienced by observers at the bottom of the cylinder . ( a bit of thought shows that it does not matter whose proper time we use to parameterize the simultaneities . ) between these two simultaneities the total electromagnetic energy inside the cylinder changes by an amount @xmath36 where @xmath37 is the volume of the cylinder ( which is constant for the co - moving observers ) . in this calculation we have made use of the relation @xmath38 for an observer at the bottom of the cylinder , who is experiencing constant proper acceleration @xmath21 . naively , we ought to be able to arrive at the same result by considering the net poynting flux crossing the boundary of the cylinder . before we calculate this , it is interesting to understand the mechanism by which the poynting vector carries energy into the cylinder . as noted earlier , on any given co - moving simultaneity the relative velocity , @xmath22 , is constant along the cylinder and , thus , the proper poynting vector is the same at the top and bottom of the cylinder . since no flux leaves or enters the sides of the cylinder , this suggests that the net poynting flux is zero . recall , however , that because of the acceleration , proper time advances more quickly at the top of the cylinder relative to the bottom . this results in a greater proper time - integrated flux entering the top of the cylinder compared to that leaving the bottom of the cylinder in other words , a net accumulation of electromagnetic energy . net electromagnetic energy enters the cylinder because of the time dilation effect associated with the acceleration . with the magnitude of the poynting vector given above , we find @xmath39 observe that the net poynting flux , equation ( [ e(2 ) ] ) , accounts for only _ half _ of the change in electromagnetic energy inside the cylinder , equation ( [ e(1 ) ] ) . apparently , we have a paradox . to understand the missing piece of this puzzle , let us first consider a simple problem in classical mechanics . imagine an object with rest mass @xmath40 from the point of view of a reference frame moving towards it with instantaneous speed @xmath22 . in this reference frame , we would consider the object to have total energy @xmath41 , where @xmath42 is the object s instantaneous momentum relative to the frame . now , if our frame is accelerating at a proper rate @xmath21 , it is easy to show that , in a proper time @xmath43 , the object s total energy changes by an amount @xmath44 . thus , the total energy of the object increases solely as a result of the acceleration of our frame relative to the object . the resolution to the paradox follows the same line of reasoning : as the cylinder accelerates relative to the existing momentum inside , there is an increase in energy proportional to this momentum times the acceleration . the momentum in this case belongs to the electromagnetic field contained in the cylinder , and is proportional to the poynting vector : @xmath45 ( as a consistency check , observe that as the electromagnetic energy , @xmath46 , inside the cylinder increases , so does the electromagnetic momentum , @xmath47 , such that @xmath48 times the invariant mass , @xmath49 remains constant , as expected , and equal to the electromagnetic energy inside the cylinder when @xmath50 . ) now , for the problem at hand , the contribution to the energy change due to this effect is just @xmath51 notice that this energy change combined with the accumulation of poynting flux due to the acceleration - induced time dilation along the length of the cylinder , @xmath52 , now precisely matches that calculated from the energy density , i.e. , @xmath53 . thus we have found the missing piece of the puzzle in the apparent paradox . it is not sufficient to consider the change in energy inside the cylinder based solely on the accumulation of poynting flux across the surface ; one must also take into account a momentum times acceleration " term due to the frame accelerating relative to an existing momentum . the inclusion of this _ bulk _ term then resolves the apparent paradox but comes with a cost it does away with the usual _ change in energy equals net energy flux through the boundary " _ picture of a conservation law . let us now look more generally at the notion of conservation laws in the context of both special and general relativity . consider a smooth four - dimensional manifold , @xmath54 , endowed with a lorentzian spacetime metric , @xmath55 , and associated covariant derivative operator , @xmath56 . in the presence of a non - zero matter stress - energy - momentum tensor , @xmath6 , we can construct the identity @xmath57 this identity gives a differential conservation law for the current @xmath58 , whose physical interpretation depends on the choice of the weighting vector , @xmath7 . integrating both sides of this identity over a finite four - volume , @xmath59 , we have @xmath60.\end{aligned}\ ] ] on the left hand side , @xmath61 and @xmath62 are the initial and final three - dimensional spatial volume end caps " of @xmath59 , with timelike future - directed unit normal vector @xmath63 . on the right hand side , @xmath64 is a three - dimensional timelike worldtube spanning the boundaries of the end caps , @xmath65 and @xmath66 , with spacelike outward - directed unit normal vector , @xmath67 , and induced lorentzian three - metric @xmath68 . in general , the left hand side of equation ( [ bulkintegrated ] ) gives the change in a quantity contained inside a spatial volume ( e.g. , electromagnetic energy ) over some time interval , while the first term on the right hand side corresponds to matter fluxes across the boundary of that spatial volume during that time interval ( e.g. , electromagnetic poynting flux ) . this is the desired form of a conservation law . however , there are two additional terms on the right hand side that are four - dimensional bulk integrals . the first of these involves @xmath69 , which is zero in general relativity ; however in special relativity it will in general not vanish when @xmath6 does not include all of the matter fields ( e.g. , when it does not include the electric four - current source of an electromagnetic field ) . the second term is present when @xmath7 is not a killing vector and , as it turns out , is crucial in resolving the apparent paradox above . to gain a better understanding of the terms in equation ( [ bulkintegrated ] ) let us look at an example . consider a general electromagnetic field , which can be decomposed as @xmath70 } + \epsilon^{ab}_{\phantom{ab}c } b^c,\end{aligned}\ ] ] where @xmath71 is the four - velocity of a volume - filling , three - parameter family of observers who see proper electric and magnetic fields @xmath72 and @xmath73 , respectively , and @xmath74 is the spatial three - volume element orthogonal to the observers worldlines . denoting the spatial three - metric orthogonal to the observers worldlines as @xmath75 , the electromagnetic stress - energy - momentum tensor can be decomposed as @xmath76 \right\}\end{aligned}\ ] ] also , it follows from maxwell s equations that @xmath77 where @xmath78 is the electric four - current source of the electromagnetic field . for simplicity we will work with a congruence that has zero twist , i.e. , one for which the four - velocity @xmath71 is hypersurface orthogonal , and assume that @xmath79 on @xmath61 and @xmath62 . this reduction in generality does not affect our conclusions it just makes the formulas simpler and more transparent . choosing as our weighting vector @xmath80 , equation ( [ bulkintegrated ] ) becomes an energy conservation equation , the individual terms of which are @xmath81 here @xmath82 and @xmath83 are the proper electromagnetic energy density and poynting vector , respectively , and @xmath78 is the electric four - current vector , as before . @xmath84 is the proper electromagnetic momentum density , and @xmath85 is the spatial , three - dimensional maxwell stress tensor . @xmath86 is the observers four - acceleration , and @xmath87 is the observers spatial , three - dimensional strain rate tensor , measuring the expansion and shear of their congruence . inserting equations ( [ energy]-[bulktransfer2 ] ) into equation ( [ bulkintegrated ] ) gives a general relativistic version of poynting s theorem for a hypersurface orthogonal congruence of observers : @xmath88.\end{aligned}\ ] ] notice that the standard form of poynting s theorem is recovered when @xmath71 is a timelike killing vector , in which case the last two bulk terms on the right hand side , coming from equation ( [ bulktransfer2 ] ) , vanish . ( the bulk term involving @xmath89 remains , however , and represents an energy transfer between the electromagnetic field and its electric four - current source . ) when @xmath71 is _ not _ a killing vector , however , these two bulk terms provide additional mechanisms for energy transfer . the @xmath90 term represents change in energy due to the frame accelerating relative to an existing electromagnetic momentum in the system ( as was seen in resolving the apparent paradox of [ paradox ] ) . the @xmath91 term represents the maxwell stress , @xmath92 , doing work against the strain , @xmath93 , of the three - parameter congruence . since we would like to generalize beyond the case where @xmath71 is a killing vector , both of these energy transfer mechanisms will be important . we will now illustrate the use of the local conservation law , equation ( [ bulkintegrated ] ) , to resolve the apparent paradox discussed in [ paradox ] , generalized slightly to allow for a time - dependent acceleration along the @xmath15-axis .- axis is chosen purely for notational simplicity . ] moreover , to facilitate comparison with the general relativistic calculation in the next section ( on _ quasi_local conservation laws ) , we will switch from a right circular cylinder of length @xmath13 and cross - sectional area @xmath14 to a round sphere of areal radius @xmath94 . the reason for this switch is that the general relativistic calculation involves the extrinsic curvature of the boundary of our spatial volume , and the sharp corners of a cylinder introduce an unnecessary technical complication . let us denote the minkowski coordinates of an inertial reference frame in flat spacetime by @xmath95 , @xmath96 . let the embedding @xmath97 define an accelerated , timelike worldline , @xmath98 , where @xmath25 is proper time , and let @xmath99 denote the unit vector tangent to @xmath98 . we will construct a three - parameter family of accelerated observers in the neighborhood of this worldline using the coordinate transformation @xmath100 here @xmath101 are coordinates adapted to the congruence of observers : the spherical coordinates @xmath102 are parameters that label the observers worldlines , and @xmath25 is a time parameter along the worldlines which in general is _ not _ proper time , except for the observer at @xmath103 ( i.e. , there is a nontrivial lapse function , @xmath104 , which will be given shortly ) . also , @xmath105 are the usual direction cosines in a spherical coordinate system , and @xmath106 is a fermi - walker transported spatial triad defined on @xmath98 that is orthogonal to @xmath107 . specializing to the case relevant to our apparent paradox , we let @xmath98 ( i.e. , the observer at @xmath103 ) undergo proper acceleration @xmath108 along the @xmath109 ( i.e. , @xmath15 ) axis , in which case our tetrad has the form : @xmath110 where @xmath111 . a short calculation reveals that the metric in the observer - adapted coordinate system is @xmath112 where @xmath113 is the metric on the unit round sphere , and @xmath114 is the lapse function . the observers four - velocity is defined as @xmath115 , with resulting four - acceleration @xmath116 . comparing with the discussion of the paradox given in [ paradox ] , in the context of an accelerating cylinder , the nontrivial lapse function , @xmath117 , is the analogue of our earlier relation @xmath24 , i.e. , when @xmath118 , proper time passes more quickly for observers above the plane @xmath119 relative to those below it . also , the magnitude of the observers proper acceleration , @xmath120 , is the analogue of our earlier relation @xmath20 , i.e. , when @xmath118 , observers above the plane @xmath119 experience less proper acceleration compared to those below it . finally , by inspection of the metric in equation ( [ adaptedcoordmetric ] ) , it is clear that the observers are accelerating _ rigidly _ , in the sense that the orthogonal distance between all nearest neighbor pairs of observers remains constant in time , as in the bell spaceship paradox ( despite the time - dependent acceleration ) . thus , the three - dimensional strain rate tensor , @xmath93 in equation ( [ bulktransfer2 ] ) , is zero . ( this is a very special case . in general , it is not possible to find a volume - filling congruence of observers for which @xmath121 , a point we will return to in the next section . ) as in [ paradox ] , we now introduce a constant , uniform electric field of magnitude @xmath17 in the positive @xmath18-direction , perpendicular to the observers motion : @xmath122 ( and @xmath73 and @xmath78 vanish , at least in the region of the observers congruence ) . it follows from equation ( [ emstresstensor ] ) that the electromagnetic stress - energy - momentum tensor is @xmath123 . choosing @xmath80 in equation ( [ bulkintegrated ] ) makes this an energy conservation equation . with @xmath124 and @xmath125 we find that the individual terms in equation ( [ bulkintegrated ] ) are @xmath126 where we have introduced the usual lorentz transformation parameters : @xmath127 and @xmath128 such that @xmath129 . observe that the right hand side of equation ( [ bulkem2 ] ) is due entirely to the @xmath90 term in equation ( [ bulktransfer2 ] ) since , as noted above , the collective rigidity of the observers motion means that the strain rate tensor , @xmath93 , vanishes in this case . in other words , this last contribution is entirely the result of a change in energy due to the frame accelerating with respect to the existing electromagnetic momentum in the system . putting these results together , the conservation law reads @xmath130 where the lapse function in the bulk term , @xmath131see equation ( [ bulkem2 ] ) , has been canceled by that in the volume element , @xmath132 . note also that we have used @xmath133 and @xmath134 , where @xmath135 is the area element on the unit round sphere . in order to compare directly with the results in [ paradox ] , let the time interval be infinitesimal : @xmath136 . using the relation @xmath137 , the left hand side of equation ( [ paradoxconservationlaw ] ) then integrates to precisely the same change in electromagnetic energy , @xmath138 , given in equation ( [ e(1 ) ] ) , except with the cylinder volume , @xmath37 , replaced with the sphere volume , @xmath139 . also , the @xmath23 in the equation now refers to the proper time elapsed for the observer at the center of the sphere , instead of an observer at the bottom of the cylinder . the poynting flux integral over @xmath140 on the right hand side of equation ( [ paradoxconservationlaw ] ) has two terms . the first term ( proportional to @xmath141 ) integrates to zero over the angles , which is analogous to the proper poynting vector in [ paradox ] being the same at the top and bottom of the cylinder such that , at lowest order , flux in equals flux out . at the next order in @xmath94 , however , the nontrivial lapse function is responsible for a @xmath142 term that does _ not _ integrate to zero . as in [ paradox ] , it is this non - isotropic time dilation that allows for a nonzero accumulation of poynting flux . evaluating this integral we find precisely the same change in electromagnetic energy , @xmath52 , given in equation ( [ e(2 ) ] ) , except with @xmath143 and @xmath23 reinterpreted as discussed above . finally , the bulk integral over @xmath59 on the right hand side of equation ( [ paradoxconservationlaw ] ) , which measures the change in energy due to the acceleration of the frame relative to the existing electromagnetic momentum in the system , evaluates to precisely the same change in electromagnetic energy , @xmath144 , given in equation ( [ e(3 ) ] ) , except with @xmath143 and @xmath23 again reinterpreted as discussed above . this example illustrates that any local conservation law constructed from the matter stress - energy - momentum tensor , @xmath6 , as in equations ( [ bulkdifferential ] ) and ( [ bulkintegrated ] ) , in general contains _ two _ terms responsible for the change in a physical quantity inside a volume of space : the first is a three - dimensional surface flux integral over @xmath145 , as one might expect , and the second is a four - dimensional bulk integral over @xmath146 . it is the addition of this bulk integral , which naively we did not expect , that resolves the apparent paradox introduced in [ paradox ] . but is such a bulk integral really necessary ? in the next section we will see how to naturally convert this bulk integral into a surface flux integral , and at the same time generalize the conservation law in equations ( [ bulkdifferential ] ) and ( [ bulkintegrated ] ) to include gravitational effects . we will now argue that the local conservation law given in equations ( [ bulkdifferential ] ) and ( [ bulkintegrated ] ) is defective in two respects . first , even when matter is locally covariantly conserved , i.e. , @xmath8 ( which is always true in general relativity ) , the local conservation law contains a nontrivial bulk term when @xmath6 is present and @xmath7 is not a killing vector . insofar as a generic spacetime does not admit any killing vectors , when @xmath6 is present this bulk term is generically present . but this bulk term violates what we usually think of as a conservation law , i.e. , that the change in some physical quantity over a period of time is equal to some related surface flux integral over that period of time . is there a natural and general way to convert this bulk term into a surface flux term ? to help motivate this question , let us return for a moment to the momentum times acceleration " bulk term that resolves the apparent paradox introduced in [ paradox ] . we imagine being inside an accelerating box in flat spacetime that contains a freely - floating , massive object that appears to be accelerating toward us ; the object s kinetic energy ( relative to us ) increases due to the acceleration of our frame . we ask : where does the increasing kinetic energy come from ? this might sound like a silly question after all , energy is frame - dependent , and we are just changing the frame ! however , the question is perhaps not so silly when we ask it in the context of the equivalence principle . instead of being inside an accelerating box , we could imagine that the box is at rest in a uniform gravitational field , and that the object is experiencing an acceleration toward us due to the force " of gravity . this force " acting through a distance represents an energy transfer mechanism from the gravitational field energy to the kinetic energy of the object . so it might be possible to convert the momentum times acceleration " bulk term into some kind of surface flux term representing _ gravitational _ energy entering the box from the outside . in the context of general relativity , i.e. , when we properly account for the frame dragging produced by the object in motion , we will see that this is exactly what happens . the second defect , which is related to the first , is that the local conservation law given in equations ( [ bulkdifferential ] ) and ( [ bulkintegrated ] ) can not properly account for gravitational effects since it is _ homogeneous _ in @xmath6 . in any vacuum spacetime region where @xmath10 , the local conservation law has nothing to say . for example , we can imagine a vacuum region of space containing gravitational energy that is flowing in or out , to which the local conservation law is completely blind . this is , of course , to be expected , given that gravitational energy is not localizable . it seems that we need a term _ like _ @xmath147 , which can be thought of as a measure of the presence of interesting gravitational physics , e.g. , a non - stationary spacetime , and that this term should be coupled not to @xmath6 , but rather some kind of _ quasi_local stress - energy - momentum tensor that represents both matter _ and _ gravity , so that it can be nonzero even when @xmath10 . we will presently construct a general quasilocal conservation law with precisely these properties , that will remove this second defect and , by its very construction , will also automatically remove the first defect , in an interesting and subtle way , exactly as anticipated in the previous paragraph . let us consider an identity exactly analogous to equation ( [ bulkdifferential ] ) , except defined in the _ three_-dimensional spacetime of the timelike worldtube , @xmath64 , i.e. , the history of the boundary of a three - dimensional system . then , for an arbitrary vector field , @xmath148 , tangent to @xmath64 , we have the identity @xmath149 where @xmath150 is the covariant derivative with respect to the three - metric , @xmath151 , induced in @xmath64 . in place of the four - dimensional _ matter _ stress - energy - momentum tensor , @xmath6 , used in equation ( [ bulkdifferential ] ) , we have inserted the three - dimensional _ total _ ( matter plus gravitational ) stress - energy - momentum tensor , @xmath152 , defined by brown and york @xcite . this _ quasi_local stress - energy - momentum tensor is defined as @xmath153 , where @xmath154 is the gravitational momentum canonically conjugate to the three - metric @xmath155 on @xmath64 , and @xmath156 . @xmath154 , in turn , is equal to @xmath157 , where @xmath158 is the extrinsic curvature of @xmath64 . it will be useful to decompose the brown - york quasilocal stress - energy - momentum tensor into energy , momentum , and stress surface densities , respectively , as : @xmath159},\\ { \mathcal p}_a & = - \frac{1}{c^2 } \sigma_{a}^{\phantom{a}b}u^{c}t^{\mathcal b}_{bc } & { \rm [ momentum / area]},\\ { \mathcal s}_{ab } & = \sigma_{a}^{\phantom{a}c}\sigma_{b}^{\phantom{b}d}t^{\mathcal b}_{cd } & { \rm [ force / length]},\end{aligned}\ ] ] where @xmath71 is the four - velocity of a two - parameter family of observers residing at the boundary of a spatial volume ; i.e. , the worldtube boundary , @xmath64 , is the congruence of the integral curves of @xmath71 . the spatial two - metric , @xmath160 , projects tensors into the space orthogonal to both @xmath71 and @xmath161 , i.e. , tangent to the spatial two - surface the observers reside on . integrating equation ( [ quasilocaldifferential ] ) over a section of the worldtube , @xmath140 , bounded by initial and final spacelike slices , @xmath162 and @xmath163 , we have @xmath164.\ ] ] on the left hand side , @xmath165 is the surface area element on @xmath162 and @xmath163 , and @xmath166 is the timelike future - directed unit vector normal to @xmath162 and @xmath163 ( and tangent to @xmath64 ) . it is interesting to compare this quasilocal conservation law to the local conservation law in equation ( [ bulkintegrated ] ) . the left hand side of both equations gives the change in a physical quantity ( e.g. , energy ) contained in a spatial three - volume . the first key difference is that the left hand side of equation ( [ quasilocalintegrated ] ) is _ quasilocal _ : it is an integral of a _ surface _ density ( e.g. , energy per unit area ) over the two - surface boundary of the three - volume . unlike the local volume density , the quasilocal surface density has no meaning by itself ; only the quasilocal surface density integrated over a closed two - surface is physically meaningful . the second key difference is that the integrated surface density includes contributions from both matter _ and _ gravity ( e.g. , gravitational energy ) @xcite . the first term on the right hand side of equation ( [ quasilocalintegrated])the matter flux term is identical to that in equation ( [ bulkintegrated ] ) ( when @xmath167 is tangent to @xmath64 ) , except its origin is very different . unlike equation ( [ bulkdifferential ] ) , equation ( [ quasilocaldifferential ] ) is purely geometrical , involving only the intrinsic and extrinsic geometry of @xmath64 , with no reference to matter . the matter stress - energy - momentum tensor , @xmath6 , enters equation ( [ quasilocalintegrated ] ) by first applying the gauss - codazzi identity ( @xmath168 , where @xmath169 is the einstein tensor ) , and then applying the einstein equation ( @xmath170 ) . thus , equation ( [ quasilocalintegrated ] ) forbids us from ignoring the fact that matter distorts the spacetime geometry ; for example , it is meaningless to analyze an electromagnetic field in _ flat _ spacetime , as we did in [ local ] . notice also that , because the boundary of a boundary is zero , there is actually _ no _ term in the quasilocal conservation law that is analogous to the matter flux term in equation ( [ bulkintegrated ] ) . the matter flux term in equation ( [ quasilocalintegrated ] ) arises through an entirely different mechanism : it is actually a surface flux analogue of the @xmath89 bulk term in poynting s theorem . the latter term arises when @xmath171 , which is analogous to @xmath172 in the quasilocal conservation law . specifically , @xmath173 represents local energy - momentum transfer from the electromagnetic field to its sources , whereas @xmath174 represents quasilocal energy - momentum transfer from the matter fields to the system " contained in @xmath64 . the last term on the right hand side of the quasilocal conservation law in equation ( [ quasilocalintegrated ] ) is analogous to the last term in equation ( [ bulkintegrated ] ) . to make the comparison more detailed , let us look at energy conservation by taking @xmath175 . using the decomposition in equation ( [ surfacesemcomponents ] ) we find @xmath176 where @xmath177 is the projection of the observers four - acceleration tangent to @xmath64 , and @xmath178 is the strain rate tensor describing the time development ( i.e. , expansion and shear ) of their congruence . the structure of equation ( [ quasilocaltransfer ] ) is identical to that of equation ( [ bulktransfer2 ] ) , except it is one dimension lower : surface , rather than volume , i.e. , quasilocal , rather than local . by the very construction of the quasilocal conservation law , the volume densities @xmath90 and @xmath179 in equation ( [ bulktransfer2 ] ) automatically become surface densities , @xmath180 and @xmath181 , respectively , i.e. , surface fluxes . this removes the first defect of local conservation laws discussed at the beginning of this section . let us first focus on the quasilocal momentum times acceleration " term , @xmath180 . unlike the momentum volume density , @xmath182 , the momentum surface density , @xmath183 , is purely geometrical : it is a component of the extrinsic curvature of @xmath64 closely related to a vector , @xmath184 , that measures the precession rate of the observers local radial " vector , @xmath161 , relative to inertial gyroscopes . as shown in more detail in reference @xcite , the precise relation is : @xmath185 , where @xmath186 is the two - dimensional spatial volume form orthogonal to both @xmath71 and @xmath161 . this allows us to write : @xmath187 which is reminiscent of the normal component of the electromagnetic poynting vector , @xmath188 . in fact , as we argue in more detail in reference @xcite , we believe that @xmath180 is the exact , purely geometrical , operational expression for gravitational energy flux . so the momentum times acceleration " bulk term , @xmath90 , has really become a gravitational energy surface flux term , exactly as anticipated in the equivalence principle argument given near the beginning of this section . in the context of general relativity , the presence of momentum ( matter or gravitational ) flowing through the system causes the observers local radial " vector to precess relative to inertial gyroscopes ( i.e. , a frame dragging effect ) , and the vector cross product between this precession rate ( the gravitational analogue of a magnetic field ) and the observers acceleration ( the gravitational analogue of an electric field ) corresponds to a flow of gravitational energy into the system . of course this frame dragging mechanism of energy transfer would be difficult to confirm experimentally because the frame dragging precession rate is smaller than the momentum density by the factor @xmath189 . nevertheless , the existence of this mechanism is not surprising given that it bears a close resemblance to a similar mechanism in electrodynamics . in the context of special relativity , consider a charged particle accelerating in a uniform electric field ( for simplicity we will ignore radiation reaction , which does nt affect the point of our discussion ) . we ask : where does the particle s increasing kinetic energy come from ? the answer is obviously the field . " but what is the mechanism , exactly ? consider that the velocity field part of the magnetic field due to the particle s motion is localized near the particle , and circulates around its axis of motion . a moment s thought shows that the vector cross product between the external electric field ( causing the particle s acceleration ) and this magnetic velocity field ( moving with the particle ) points toward the axis of motion , and is maximum in the plane containing the particle . in other words , there is a poynting vector representing a flow of energy from the field to the particle . intuitively , one might guess that this answers our question . surprisingly , it seems that a detailed analysis of this basic energy transfer mechanism in classical electrodynamics has been done only very recently @xcite . the authors of this reference verify that our intuition is correct , at least in the ultra - relativistic limit . of the particle s kinetic energy is supplied by the poynting flux ; @xmath190 is supplied by a peculiar interference " effect between the external electric field and the velocity field part of the particle s electric field @xcite . this subtlety is interesting , but it does not affect the spirit of our discussion . ] to the extent that the linearized approximation to general relativity is very similar to electrodynamics , one might expect essentially the same energy transfer mechanism to occur in gravitational physics , when , e.g. , we replace the massive , charged particle accelerating in a uniform electric field with a massive ( charged or uncharged ) particle accelerating in a uniform gravitational field ( or , via the equivalence principle , a frame accelerating toward a freely - floating particle ) . in the context of the linearized , slow motion approximation to general relativity , sometimes called _ gravitoelectromagnetism _ ( for a review , see @xcite ) , precisely such a mechanism has been confirmed ; see , e.g. , @xcite . what s new in this paper is that this frame dragging energy transfer mechanism is confirmed in the full , nonlinear general theory of relativity with arbitrary matter and with no approximations ; the gravitational energy flux , @xmath180 , is an exact , purely geometrical contribution to the energy flux that must be considered to properly explain a wide variety of phenomena , including phenomena as basic as a falling apple . we will compute an explicit example of this energy transfer mechanism in the next section . let us now focus on the quasilocal stress times strain " term , @xmath181 , in equation ( [ quasilocaltransfer ] ) . the boundary observers strain rate tensor , @xmath191 , is the two - dimensional analogue of the three - dimensional strain rate tensor , @xmath93 , in equation ( [ bulktransfer2 ] ) , for a volume - filling set of observers . in the latter case , it is obvious that such a _ three_-parameter family of observers can not , in general , move _ rigidly _ , i.e. , move in such a way as to maintain constant radar - ranging distances between all nearest neighbor pairs of observers . this is because the condition @xmath192 represents six differential constraints on three functions ( the three independent components of the observers four - velocity , @xmath71 ) . in other words , the local conservation law generically contains a nontrivial bulk stress times strain " term , @xmath179 . the same is _ not _ true for a _ two_-parameter family of observers , because there the condition @xmath193 represents only _ three _ differential constraints on the same three functions . in reference @xcite we argue that , in a generic spacetime , it is always possible to find a two - parameter congruence of integral curves of @xmath71 , comprising @xmath64 , that is expansion- and shear - free . moreover , the degrees of freedom of motion left to such a congruence of boundary observers are precisely the same as those for rigid body motion in newtonian space - time , viz . , three translations and three rotations , each with arbitrary time dependence . in other words , in the quasilocal conservation law ( but _ not _ in the local conservation law ) , it is always possible to choose a family of observers such that the stress times strain " term vanishes . we have called such a family of observers a _ rigid quasilocal frame _ , or rqf for short . we refer the reader to references @xcite for a more detailed discussion of the properties of rqfs . for our present purposes we need only appreciate that , since a congruence of rqf observers is expansion- and shear - free , there are no extraneous fluxes entering or leaving the system due merely to a change in the size or shape of the system boundary . so all of the interesting physics lies in the @xmath180 surface flux term discussed above . this is a key advantage to working with a quasilocal conservation law in general , and rqfs in particular . we will compute an explicit example of an rqf in the next section . in summary , the quasilocal conservation law introduced in this section removes two defects of the familiar local conservation law : ( 1 ) the undesirable bulk terms in the local conservation law automatically appear as surface flux terms in the quasilocal conservation law , and ( 2 ) unlike the local conservation law , the quasilocal conservation law accounts for gravitational effects . for example , the quasilocal energy conservation law contains an energy transfer mechanism related to frame dragging that is needed to understand in detail what is actually happening with regards to energy flow when , say , an apple falls . in the next section we will apply the quasilocal energy conservation law to a general relativistic analogue of the apparent paradox introduced in [ paradox ] . to illustrate the explanatory power of the quasilocal approach let us now look at a general relativistic analogue of the apparent paradox introduced in [ paradox ] . we will work with a spacetime whose metric is @xmath55 in coordinates @xmath95 , @xmath96 , and which contains a covariantly constant ( i.e. , uniform ) electric field of magnitude @xmath17 in the positive @xmath18-direction @xcite : @xmath194 where @xmath195 is the length scale of the geometry . it is important to note that the presence of the electric field is intimately connected to the spacetime curvature we are no longer working in flat spacetime . this is readily verified by checking that there exist non - zero curvature invariants such as @xmath196 . alternatively , it is easy to see that this spacetime is a product of two surfaces of constant curvature @xcite . observe that @xmath55 reduces to the minkowski metric in the @xmath197 plane . in this plane let us construct a timelike worldline , @xmath198 , parameterized by proper time , @xmath25 , and defined by the embedding @xmath199 , that represents a fiducial observer undergoing proper acceleration @xmath108 along the @xmath15-axis . as in [ local ] , along @xmath198 we define a tetrad , @xmath107 and @xmath106 , where @xmath200 denotes the unit vector tangent to @xmath198 , and the spatial triad , @xmath106 , is fermi - walker transported along @xmath198 . because @xmath55 equals the minkowski metric on @xmath201 , we can use the same coordinate components for this tetrad as given in equations ( [ tetrad_0])-([tetrad_3 ] ) . around @xmath201 we will construct a two - parameter family ( two - sphere s worth ) of accelerated observers , whose congruence of worldlines comprises a timelike worldtube , @xmath140 . our notation will follow that used in [ local ] : the spacetime metric , @xmath55 , induces on @xmath140 the spacelike outward - directed unit normal vector field , @xmath161 , the lorentzian three - metric , @xmath202 , and the covariant derivative operator , @xmath150 . also , we let @xmath71 denote the observers four - velocity , which is tangent to @xmath140 , and @xmath160 denote the observers spatial two - metric . in [ local ] we constructed a three - parameter ( volume - filling ) family of accelerated observers in flat spacetime in the neighborhood of @xmath198 using the coordinate transformation given in equation ( [ worldline ] ) . for arbitrary fixed @xmath94 , this coordinate transformation actually defines an exact rqf , i.e. , a two - parameter family ( two - sphere s worth ) of observers who , despite their time - dependent acceleration , maintain constant radar - ranging distance from their nearest neighbors they are moving _ rigidly_. indeed , inspection of equation ( [ adaptedcoordmetric ] ) shows that @xmath203 and @xmath204 both vanish , so the observers reckon they are on a surface with constant spatial two - metric @xmath205 , which is a round sphere of areal radius @xmath94 . now because spacetime is locally flat , at lowest order in @xmath94 we can start with the same coordinate transformation given in equation ( [ worldline ] ) , but because of the spacetime curvature , we must add corrections at higher order in @xmath94 to achieve the same rqf , i.e. , to achieve @xmath206 . thus we begin with the ansatz : @xmath207 + \mathcal{o}\left(\frac{r^4}{l^3}\right),\ ] ] where , as before , @xmath208 are coordinates adapted to the observers : the angular coordinates @xmath209 are parameters that label the observers worldlines ; the radial coordinate , @xmath94 , parameterizes the size of the rqf ; and the time coordinate , @xmath25 , represents our choice of simultaneities on @xmath64 . the three arbitrary functions @xmath210 allow us to perturb the observers worldlines to satisfy the three conditions @xmath206 needed to achieve a round sphere rqf of areal radius @xmath94 . ( an rqf of generic shape and size need only satisfy the three weaker conditions , @xmath211 . ) the arbitrary function @xmath212 allows us to perturb the choice of simultaneities on @xmath64 to achieve other natural geometrical conditions , which will be discussed later . for the time being , we will keep the four functions @xmath213 and @xmath214 _ choosing @xmath215 in equation ( [ quasilocalintegrated ] ) makes this a quasilocal energy conservation equation . using grtensorii running under maple , we compute the individual terms in equation ( [ quasilocalintegrated ] ) and find : @xmath216 where @xmath217 is the area element on the unit round sphere , as before , and @xmath218 and @xmath219 are the same lorentz transformation parameters we defined after equation ( [ bulkem2 ] ) . we have also inserted the relation @xmath195 where appropriate . let us analyze each of these terms separately . equation ( [ rqfenergydensity ] ) represents the quasilocal energy density . the first term on the right hand side , at order @xmath220 inside the braces , is a vacuum energy . insofar as it is constant in time , this term does not contribute to changes in the energy inside the sphere , and so can be ignored . the quantity @xmath221 is a function of the angles @xmath222 , but not @xmath25 . so for the same reason just cited , it can also be ignored . the quantity @xmath223 is a homogeneous function of the angular derivatives of the @xmath213 , which is a total derivative . it integrates to zero for _ any _ choice of the arbitrary functions @xmath213 , and so can also be ignored . the quantity @xmath224 is an inhomogeneous function of the angular derivatives of the @xmath213 , which is a linear combination of the three round sphere rqf conditions , @xmath225 . it is zero when we choose a set of functions @xmath213 that satisfy these conditions ( which we will do later ) , and so can also be ignored . the only nontrivial term is the second one . insofar as only the integrated quasilocal energy density is physically meaningful , we can subtract off the spherical harmonics that integrate to zero , leaving the _ effective _ quasilocal energy density : @xmath226 equation ( [ rqfpoyntingdensity ] ) represents the quasilocal poynting flux . the quantity in braces is ( at least up to the order indicated ) equal to the lapse function , @xmath227 , which is contained in @xmath228 . up to the lowest nontrivial order in @xmath94 , this lapse function agrees with that found in [ local ] ; as before , it is this non - isotropic time dilation that allows for a nonzero accumulation of poynting flux in the sphere . note that , up to the order indicated , the poynting flux is independent of our choice of arbitrary functions @xmath213 and @xmath214 . equation ( [ rqfgravitydensity ] ) represents the sum of the quasilocal gravitational flux and the quasilocal stress times strain " term recall equation ( [ quasilocaltransfer ] ) . when we impose the rqf rigidity conditions on the functions @xmath213 , the latter term vanishes ( i.e. , the strain rate tensor , @xmath191 , vanishes ) . this leaves only the quasilocal gravitational flux term , @xmath180 , to consider . note that , up to the order indicated , the gravitational flux is also independent of our choice of arbitrary functions @xmath213 and @xmath214 . putting these results together , with the assumption that the rqf rigidity conditions can be satisfied ( we ll demonstrate this below ) , the quasilocal energy conservation law in equation ( [ quasilocalintegrated ] ) reads , at order @xmath229 , @xmath230 this quasilocal conservation law is to be compared with the local conservation law in equation ( [ paradoxconservationlaw ] ) . as we did for the latter equation , we will take the time interval to be infinitesimal : @xmath231 . using the relation @xmath232 [ compare this with @xmath137 , which differs by a factor of 2 ] , the left hand side of equation ( [ quasilocalintegratedparadox ] ) then integrates to precisely the same change in electromagnetic energy , @xmath138 , given by the local conservation law , and also equation ( [ e(1 ) ] ) ( with the cylinder volume , @xmath37 , replaced with the sphere volume , @xmath139 ) . the quasilocal poynting flux integral over @xmath140 on the right hand side of equation ( [ quasilocalintegratedparadox ] ) is , as mentioned earlier , identical to the local poynting flux integral in equation ( [ paradoxconservationlaw ] ) , and so yields the same change in electromagnetic energy , @xmath52 , as found in that case , and also given in equation ( [ e(2 ) ] ) ( with , again , the cylinder volume replaced with the sphere volume ) . we emphasize once more the role of the non - isotropic time dilation ( the lapse function in parentheses ) that allows for a nonzero accumulation of poynting flux . the quasilocal gravitational flux integral over @xmath140 on the far right hand side of equation ( [ quasilocalintegratedparadox ] ) is , as discussed earlier , the surface flux version of the bulk acceleration times momentum " energy transfer term in the local conservation law . the difference is that the local momentum volume density , @xmath182 , has been replaced with the quasilocal momentum surface density , @xmath233 , and the integration is not over a volume , but the boundary of that volume . we stress again that the actual mechanism of the energy transfer is a general relativistic effect , viz . , the vector cross product between the boundary observers acceleration and the precession rate of their gyroscopes due to the frame dragging caused by the electromagnetic momentum inside the system . all together , then , half of the energy flux is due to the electromagnetic field ( with the flux entering mainly near the poles of the sphere , i.e. , @xmath142 ) , and the other half is due to the gravitational field ( with the flux entering mainly near the equator of the sphere , i.e. , @xmath234 ) . both contribute to a changing electromagnetic energy on the left hand side of the quasilocal conservation law . one might wonder how a flux of _ gravitational _ energy through the boundary of the system becomes _ electromagnetic _ energy inside the system . it seems that in general relativity there is no distinction between the two forms of energy . all forms of energy are equivalent . this is not surprising when we look at the metric in equation ( [ electricspacetime ] ) . the geometry is nontrivial and so presumably contains gravitational energy ( in some nonlocal , or quasilocal sense ) , and the nontrivial geometry in turn is intimately connected to the electric field . they can not be disentangled . all that remains in this section is to verify that we actually _ can _ solve the rqf rigidity conditions , viz . , the three differential constraints @xmath206 on the three functions @xmath210 which ensure that the observers frame is a rigid round sphere of areal radius @xmath94 . as argued in reference @xcite , such rqf solutions always exist in a generic spacetime ; moreover , they are unique up to motions of the rqf equivalent to those of rigid motion in newtonian space - time , viz . , six arbitrary time - dependent degrees of freedom : three translations and three rotations . aiming for the simplest solution , we first set @xmath235 and @xmath236 to zero , which eliminates translations of the rqf away from @xmath198 in the @xmath18 and @xmath237 directions . we next demand that the rqf is not rotating , i.e. , that the twist of the observers congruence is zero , i.e. , that the observers four - velocity , @xmath71 , is hypersurface orthogonal ( as a vector field in @xmath64 ) . this means that , by adjusting @xmath212 appropriately , we can always find a time foliation of @xmath64 ( i.e. , choose the observers surfaces of simultaneity ) such that @xmath71 is orthogonal to them , which is equivalent to the two conditions @xmath238 . using grtensorii we can readily find a solution that satisfies all of these conditions , given by : @xmath239 note that the @xmath240 dependence in @xmath241 is not surprising since the electric field , being parallel to the @xmath242 plane , breaks the azimuthal symmetry the electric field affects the geometry , and hence the coordinate embedding of the rqf . one can show that these results are in agreement with the more generally derived formulas found in reference @xcite . using the standard , local way of constructing conservation laws see equations ( [ bulkdifferential ] ) and ( [ bulkintegrated])we analyzed conservation of energy in the context of a simple example in special relativity , viz . , a box , rigidly accelerating along the @xmath15-axis , that is immersed in a transverse , uniform electric field . according to the local energy conservation law , the electromagnetic energy inside the box increases due to two separate mechanisms : ( 1 ) half of the increasing energy is due to energy flowing in from outside the box via a poynting flux . interestingly , even though the instantaneous proper poynting vector is uniform throughout the box ( suggesting no net flux ) , there is a net proper time - integrated flux due to the acceleration - induced relative time dilation between observers at the top and bottom of the box . ( 2 ) the other half of the increasing energy is _ not _ due to energy flowing in from the outside ; rather , it is a bulk effect due to the co - moving observers accelerating relative to the existing electromagnetic momentum inside the box , i.e. , an acceleration times momentum " energy transfer mechanism familiar from classical mechanics , integrated over the volume of the box . one might wonder if these two energy transfer mechanisms adequately explain what s happening . the answer is : no . first , in special relativity we assume that spacetime is flat , even though there is an electromagnetic field present . this precludes any possible general relativistic effects such as a flux of gravitational energy associated with the curvature of the spacetime caused by the electromagnetic field . secondly , even in the context of general relativity , a _ local _ conservation law can not properly capture all of the gravitational physics ; for example , gravitational energy is not localizable there is no such thing as a gravitational energy per unit volume , so a local conservation law can not tell the whole story . another way to see this problem is to notice that the standard local conservation law is _ homogeneous _ in the matter stress - energy - momentum tensor , and so has nothing to say in a matter - free region of space , even though that region may contain dynamical curvature , e.g. , gravitational waves . to address these shortcomings in general , and in particular see what s really happening with regards to the increasing energy inside the box , we constructed a _ quasi_local conservation law based on the brown & york quasilocal ( matter plus gravitational ) stress - energy - momentum tensor defined on the history of the boundary of a spatial volume see equations ( [ quasilocaldifferential ] ) and ( [ quasilocalintegrated ] ) . using the energy form of this quasilocal conservation law , we analyzed the general relativistic analogue of the simple example described above . we found , again , that the electromagnetic energy inside the box increases due to two separate mechanisms . the first mechanism is identical in form to the poynting flux mechanism described above [ number ( 1 ) ] , but conceptually it has a completely different origin : it is actually analogous to a surface flux version of the @xmath11 bulk term in the standard poynting theorem , except instead of energy being transferred locally from the electromagnetic field to the four - current source of the field , it represents energy being transferred quasilocally from the matter fields to the system " contained inside the box . the second mechanism is , at first sight , conceptually identical to the acceleration times momentum " energy transfer mechanism described above [ number ( 2 ) ] , except the volume integral has been converted to a _ surface flux _ integral over the boundary of the box . going further , we argued that this surface flux is actually a _ gravitational _ energy flux exactly analogous to the electromagnetic poynting flux , @xmath243 , with @xmath244 replaced by the acceleration of the co - moving observers on the boundary of the box , and @xmath245 replaced by the precession rate of their gyroscopes due to the frame dragging caused by the electromagnetic momentum ( in general , matter and gravitational momentum ) flowing through the box . in _ both _ cases , now , energy is entering from outside the box via surface fluxes : half is an electromagnetic energy flux ( entering the box through its top and bottom ) and the other half is a gravitational energy flux ( entering the box through its _ sides _ ) . because an electromagnetic field is intimately intertwined with the spacetime curvature it produces , general relativity does not distinguish between electromagnetic and gravitational energy entering the box both contribute on the same footing to the increasing electromagnetic energy inside the box . we can understand the second , gravitational energy flux mechanism intuitively as follows . imagine being inside an accelerating box in empty space , which contains a freely - floating massive body that appears to be accelerating toward you . from your perspective , its kinetic energy is increasing . where does the increasing kinetic energy come from ? explain " it as simply the acceleration times momentum " energy transfer mechanism familiar from classical mechanics . alternatively , we could invoke the equivalence principle and say that our box is at rest in a uniform gravitational field , and that the kinetic energy is coming from the gravitational field outside the box via some kind of gravitational energy flux passing through the boundary of the box . an analogous mechanism exists in the context of electrodynamics that explains , in detail , how an accelerating charge acquires kinetic energy from the external electric field causing the particle s acceleration @xcite ; moreover , the gravitational version of this mechanism has been shown to exist in the context of the linearized , weak field approximation to general relativity @xcite . in this paper we have established the existence of this very basic , but subtle mechanism in the full , nonlinear general theory of relativity , with arbitrary matter and no approximations . our analysis made use of the concept of _ rigid quasilocal frames _ ( rqfs ) , discussed more fully in references @xcite , in order to properly isolate the relevant energy fluxes passing through the boundary of the box . in conclusion , we have explained the bulk acceleration times momentum " energy transfer mechanism familiar in classical mechanics _ exactly _ in terms of a simple , operationally defined , purely geometrical , general relativistic gravitational energy flux passing through the _ boundary _ of the volume in question . naively , one might argue that since there is no @xmath246 or @xmath247 in acceleration times momentum " , this can not be a general relativistic effect . but it _ is_. it is based on frame dragging ( the gravitational analogue of @xmath245 " in @xmath243 " ) , which is a general relativistic effect . we do nt notice this mechanism in our day - to - day experiences because the typical gyroscopic precession rate vector due to a nearby object in motion is very tiny ; but it is precisely this vector , multiplied by the huge number @xmath189 , that we identify as the momentum " of the object [ more precisely , the quasilocal momentum surface density , rotated by @xmath248 degrees see equation ( [ gravitationalpoynting ] ) ] . this general relativistic gravitational energy flux mechanism is what s _ really _ happening in the bulk acceleration times momentum " energy transfer mechanism in classical mechanics . this deeper understanding would not be possible in the context of local conservation laws , and is a nice example of why we need _ quasi_local conservation laws . this work was supported in part by the natural sciences and engineering research council of canada . 99 c. w. misner , k. s. thorne and j. a. wheeler , _ gravitation _ , isbn 0 - 7167 - 0344 - 0 ( w. h. freeman and company , 1973 ) . j. d. brown and j. w. york , phys . d * 47 * , 4 , 1407 - 1419 ( 1993 ) . r. j. epp , r. b. mann and p. l. mcgrath , class . * 26 * , 035015 ( 2009 ) . r. j. epp , r. b. mann and p. l. mcgrath , _ classical and quantum gravity : theory , analysis , and applications _ , isbn 978 - 1 - 61122 - 957 - 8 ( nova science publishers , 2012 ) , ch . 14 . j. s. bell , _ speakable and unspeakable in quantum mechanics _ , isbn 0 - 521 - 33495 - 0 ( cambridge university press , 1987 ) , pp . . g. n. gaidukov and a. a. abramov , phys . usp . * 51 * , 163 ( 2008 ) . b. mashhoon , _ the measurement of gravitomagnetism : a challenging enterprise _ , isbn 978 - 1 - 60021 - 002 - 0 ( nova science publishers , 2007 ) , pp . 29 . p. krumm and d. bedford , am . . phys . * 55 * , 362 ( 1987 ) . c. j. de matos and m. tajmar , e - print : gr - qc/0107014 ( 2001 ) . b. bertotti , phys . rev . * 116 * , 1331 ( 1959 ) . h. stephani , d. kramer , m. maccallum , c. hoenselaers , and e. herlt , _ exact solutions of einstein s field equations _ , isbn 0 - 521 - 46136 - 7 ( cambridge university press , 2003 ) , pp .